Drought forecast model based on Artificial Neural Networks for Brazilian municipalities

Drought forecast model based on Artificial Neural Networks for Brazilian municipalities | Research Square window.SnipcartSettings = { analytics: { enabled: false } }; (function() { var accessVector = localStorage.getItem('access_vector') || ''; window.dataLayer = window.dataLayer || []; if (accessVector) { window.dataLayer.push({ user: { profile: { profileInfo: { snid: accessVector } } } }); } })(); (function(w,d,s,l,i){w[l]=w[l]||[];w[l].push({'gtm.start':new Date().getTime(),event:'gtm.js'});var f=d.getElementsByTagName(s)[0],j=d.createElement(s),dl=l!='dataLayer'?'&l='+l:'';j.async=true;j.src='https://www.googletagmanager.com/gtm.js?id='+i+dl;f.parentNode.insertBefore(j,f);})(window,document,'script','dataLayer','GTM-K279D39R'); Browse Preprints In Review Journals COVID-19 Preprints AJE Video Bytes Research Tools Research Promotion AJE Professional Editing AJE Rubriq About Preprint Platform In Review Editorial Policies Our Team Advisory Board Help Center Sign In Submit a Preprint Cite Share Download PDF Research Article Drought forecast model based on Artificial Neural Networks for Brazilian municipalities Guilherme Garcia de Oliveira, Nicholas Becker Pires Pi, Laurindo Antonio Guasselli, and 1 more This is a preprint; it has not been peer reviewed by a journal. https://doi.org/ 10.21203/rs.3.rs-4784321/v1 This work is licensed under a CC BY 4.0 License Status: Under Revision Version 1 posted 5 You are reading this latest preprint version Abstract The increase in the frequency of droughts, driven by climate change, implies the need to understand and mitigate these extreme events. In Brazil, there are technical-scientific gaps in relation to climate disaster warnings. The integration of an inventory of droughts that caused losses with remote sensing data, hydrometeorological and climate indexes, using artificial neural networks (ANN) can contribute to a drought forecast. In this study, we developed a monthly forecast model for droughts in Brazilian municipalities using ANN. Precipitation and temperature indexes, in addition to municipal descriptors, for example, the region of the country, the biome, and distance from the oceans and the Amazon, were used as predictor variables in the model. We used an inventory of droughts that caused losses by municipalities (2013–2022) from the Brazilian Integrated Disaster Information System. After model training, we tested the ANN for drought forecasts for lead times of 1–4 months, using seasonal forecast data from the European Center for Medium-Range Weather Forecasts (ECMWF). The overall accuracy of the ANN model for drought simulation was 0.931. The forecast accuracy ranged from 0.922 for a 1-month lead time to 0.757 for 4 months. Remarkably, the model reproduced the spatial pattern of droughts, especially when the output is interpreted as a continuous index of drought risk. We conclude that the trained model is efficient and the results indicate strong potential for drought forecasting and warning, using ANN, remote sensing data, hydrometeorological and climate indexes. Satellite-Derived Precipitation Seasonal Weather Forecast Natural Disaster Drought Warning Figures Figure 1 Figure 2 Figure 3 Figure 4 Figure 5 Figure 6 Figure 7 Figure 8 Figure 9 Figure 10 Introduction The spatial and temporal variability of precipitation has the potential to trigger extreme climatological processes, such as droughts. These events can cause serious economic losses and impacts on society, especially when they occur in regions where the population is more vulnerable to disasters (Blaikie et al. 2005; Hagenlocher et al. 2019 ; Liu and Chen 2021 ). In Brazil, droughts affect more than 40 million people in some years. In the 3-year period between 2020 and 2022, the accumulated loss was more than US $ 35 billion (SEDEC 2023). The frequency of droughts has intensified due to climate change and global warming (Kim and Jehanzaib 2020 ). Recent data reveals that global surface temperatures have warmed by more than 1°C compared to the long-term average from 1951 to 1980 (NASA/GISS 2023). Although meteorological and climatic conditions are dynamic and extreme events are the results of adverse natural processes, the influence of human activities on global warming observed in recent decades is already considered unequivocal (IPCC 2021 ). Changes in climate affect different natural systems and dynamics with severe environmental, social and economic impacts (Howells et al. 2013 ; Shukla et al. 2019). Knowledge about the spatial-temporal distribution of droughts and their trends is essential for effective risk management and the development of mitigation strategies (Mulualem and Liou 2020 ). Developing a drought monitoring, early warning system or seasonal forecast can strengthen a country's resilience to losses caused by droughts (Tanguy et al. 2023 ). Research and products related to drought diagnosis, analysis, forecasting and warning can mitigate the impacts caused by these extreme weather events (Panu and Sharma 2002 ; Canedo Rosso et al. 2018 ), as it allows the planning of different sectors of society, such as agriculture (Trnka et al. 2020 ), energy (He et al. 2019 ), public water supply (Stewart et al. 2020 ) and banks and insurance companies (Huynh et al. 2020 ). The decrease in accumulated precipitation is the main explanation for triggering droughts. Satellite-derived precipitation, including products that merge remote sensing data with measurements from rainfall stations, has the main advantage of extensive spatial coverage (Boluwade 2020 ; Kofidou et al. 2023 ). These datasets make it possible to analyze precipitation over large areas efficiently and without the need for interpolation. Climate disaster monitoring is one of the possible and currently expanding applications (Cheng et al. 2021 ; Palharini et al. 2022 ). Based on the temporal and spatial precipitation series, researchers can understand how previous accumulated precipitation relates to disasters, aiming to warn about extreme events. These studies are of interest to the scientific community and society, with some researchers dedicating themselves to this (Boult et al. 2020 ). According to Zhang et al. ( 2019 ) the inclusion of variables that refer to hydrometeorological dynamics such as rainfall, temperature, potential evapotranspiration, air pressure, wind speed and relative humidity, often results in improvements in drought prediction. Regarding the dependent variable of the forecast model, defining drought is an important step. Usually, researchers use one or more indexes to define the occurrence of drought (Kallis 2008 ; Dikshit et al. 2022 ), such as the Standard Precipitation Index - SPI (McKee et al. 1993 ), Standard Precipitation Evaporation Index - SPEI (Vicente-Serrano et al. 2010 ), and vegetation indices based on remote sensing such as the Normalized Difference Vegetation Index - NDVI or Enhanced Vegetation Index - EVI (Huete et al. 2002 ). However, these indexes do not consider local economic and social dynamics, that is, a drought defined by these parameters does not always result in effective losses for the population. Predicting droughts with the potential to cause losses, for example, in agricultural production, is a complex task as it depends on the location, the dry period, the seasonality of crops, land use management, irrigation techniques, among others (Musolino et al. 2018 ). The losses are strongly linked to the spatial and temporal context. Therefore, an inventory of drought occurrences, with losses certified by authorities or government entities, is a powerful tool that could be used as a dependent variable in the forecast model. In Brazil, the Integrated Disaster Information System - S2iD (SEDEC 2023), is used for municipal authorities to report the occurrence of droughts that result in losses to various economic and social sectors. The processing of this data set and its use as a dependent variable is a distinguishing feature of this research, making it possible to train a model that predicts droughts with the potential to cause losses in each municipality. Drought forecasting models represent a notable challenge due to the intricate interaction of diverse hydrometeorological factors, compounded by the effects of climate change. Among the various approaches, such as statistical, physical and data-based methods, machine learning methods are among the most common for predicting drought index (Belayneh and Adamowski 2013 ; Sundararajan et al. 2021 ). In this approach, artificial neural networks (ANN) often provide excellent results for modeling and predictions in the environmental area, using data obtained through remote sensing, meteorological and climate data as model input variables (Marj and Meijerink 2011 ; Mokhtari and Akhoondzadeh 2019 ). The spatial complexity, evidenced by the non-linear and multivariate property of droughts, highlights the ability of ANN-based models to quickly and effectively capture dynamic relationships, considerably boosting their application (Dikshit et al. 2022 ). Several studies use ANN for drought index forecasts (for examples: Mishra et al. 2006; Santos et al. 2014; Zhang et al. 2017 ; Dikshit et al. 2020 ; Mulualem and Liou 2020 ). Mishra et al. (2006) forecasted SPI at multiple time scales for West Bengal, India, using three approaches (RBF, ANFIS and ANN). Their study revealed ANN produced the best result, with the key finding of using a direct multi-step approach to forecast at higher lead times, instead of a recursive multi-step approach. Santos et al. (2014) tested the ability of neural network approaches to hindcast the spring SPI on a 6-month time scale in Portugal, based on winter large-scale climatic indices. The authors indicate that the ANN showed good prediction capacity, although they identify limitations in relation to modeling extreme SPI values. Zhang et al. ( 2017 ) tested ARIMA, ANN and W-ANN’s applicability, using SPI as the drought index for Haihe River Basin, China. The results obtained by the authors indicated an advantage for the wavelet-based models at lead times of 3 and 6 months. Dikshit et al. ( 2020 ) present SPEI prediction results for a region of Australia, comparing ANN and Support Vector Regression (SVR) models. The drought index was calculated at various time scales (1, 3, 6 and 12 months) using a Climate Research Unit (CRU) dataset. The results indicate that ANN was better than SVR in predicting temporal drought trends. Mulualem and Liou ( 2020 ) developed ANN predictive models incorporating hydro-meteorological, climate, sea surface temperatures, and topographic attributes to SPEI forecast in the Upper Blue Nile basin. The authors showed that accurate results in predicting SPEI values could be achieved by including large-scale climate indexes. In recent years, seasonal forecast meteorological models have been developed. For example, Johnson et al. ( 2019 ) describe SEAS5, the fifth-generation seasonal forecast system of European Center for Medium-Range Weather Forecasts (ECMWF). In summary, the researchers indicate that SEAS5 is a state-of-the-art seasonal forecast system which continues to display a particular strength in the El Niño Southern Oscillation (ENSO) prediction. Ferreira and Reboita ( 2022 ) and Ferreira et al. ( 2022 ) evaluated the performance of ECMWF-SEAS5 in simulating South American precipitation. The authors conclude that ECMWF-SEAS5 has good performance in the climate characterization of South America. However, we did not identify any study that tested coupling an ANN model for drought prediction using a seasonal meteorological forecast dataset, such as ECMWF-SEAS5. This is another scientific gap that we intend to explore in this paper. In this context, and considering the need for advances in relation to forecasting and warning of climate disasters in Brazil, our objective was to develop a monthly ANN model with satellite-derived precipitation and seasonal weather forecast to predict droughts with the potential to cause losses in Brazilian municipalities. Materials and Methods In this study, we used a dataset with coverage across Brazil, considering all 5,570 municipalities in the country as units of analysis and prediction. The time period used for training, cross-validation, and testing of the calibrated models was between 2013 and 2022, considering a monthly data interval. These spatial and temporal scales are associated with the modeled dependent variable, namely, the record in the database recognizing situations of emergency related to droughts in the Integrated Disaster Information System (S2iD) of the National Secretariat for Civil Protection and Defense (SEDEC 2023). In Fig. 1 we illustrate the number of months in which each municipality in Brazil was affected by droughts that resulted in losses, including the water supply, agriculture, livestock and energy sectors. Although all regions present occurrences of droughts, the Northeast Region stands out with a significant number of municipalities being affected in more than 20% of the months of analysis. The methodology of this study is based on the use of artificial neural networks for drought warning based on seasonal precipitation and temperature forecasts. This research presents advances in relation to the methodology presented in Oliveira et al. ( 2023 ), including new predictive variables, increased forecast lead time and expansion of the forecast to the entire territory of Brazil. We organized the study into six main steps (Fig. 2 ), containing: i) processing an inventory of droughts that caused losses; ii-iii) calculation of monthly precipitation and temperature indexes; iv) extraction of spatial descriptors for each municipality; v-vi) training, validation and testing of the ANN model for drought forecasting. In this research, we use the datasets from the list below: S2iD Database (SEDEC 2023); Geospatial base of the Brazilian Institute of Geography and Statistics, including vector files of the limits of Brazilian biomes (IBGE 2019), and the territorial grids of municipalities, microregions and regions of Brazil (IBGE 2023); Climate Hazards Group InfraRed Precipitation with Station Data (CHIRPS), University of California at Santa Barbara (UCSB 2023); National Centers for Environmental Prediction (NCEP) database, NCEP/DOE Reanalysis II product (Kanamitsu et al. 2002 ); ECMWF-SEAS5 Seasonal Forecast (ECMWF 2023). In the first step, we compiled and processed data on drought occurrences from the S2iD Database, producing an inventory of drought occurrences. The database consists of the municipal code, date of the emergency protocol, updating the registration date as the drought prolongs, in addition to detailed information on losses in various economic sectors. The time series was discretized into 120 monthly time intervals, considering the period between January 2013 and December 2022. An algorithm was developed in MATLAB to organize the data in the form of a matrix, where each row corresponds to a municipality, and each column represents a time interval (month/year). In the matrix, each cell (municipality vs. time interval) was assigned the code 1 for the occurrence of drought. In the remaining cells, the code 0 was established (no occurrence). After an exploratory analysis of the data in spatial and temporal contexts, and due to uncertainties in establishing the beginning and end of each drought event, we performed an analysis of consistency and transformation on the occurrence matrix, considering spatial-temporal dependence. The procedure was executed for each cell x m,t , where m is a municipality and t is a time interval, and the following criteria and rules were subsequently applied (in sequential loops): When an occurrence was identified in municipality m at time t ( x m,t = 1), the code 2 was assigned to the cells of the same municipality in the immediately preceding time interval ( x m,t−1 = 2) and the subsequent time ( x m,t+1 = 2). This code represents a possible temporal extension of the drought, and it can be interpreted as a doubtful record; When a doubtful occurrence was identified in municipality m at time t ( x m,t = 2), with more than 50% of the municipalities in its geographic microregion in a situation of confirmed occurrence in the same time interval, the record was changed to drought ( x m,t = 1); When a non-occurrence was identified in municipality m at time t ( x m,t = 0), but with a record of occurrence in the immediately preceding time interval ( x m,t−1 = 1) and subsequent time ( x m,t+1 = 1), the record was changed to drought ( x m,t = 1); When a record of non-occurrence was identified in municipality m at time t ( x m,t = 0), with more than 60% of the municipalities in its geographic microregion in a situation of confirmed occurrence between the time intervals t-1 and t + 1 , the record was changed to drought ( x m,t = 1). The sum of extreme drought events was calculated to establish the number of non-occurrence samples for training the ANN considering a 1:1 ratio, that is, with a balance between occurrence and non-occurrence samples chosen randomly. The antecedent accumulated precipitation is the most important hydrological process in defining drought, establishing a strong correlation with these extreme events. In this study, we selected the CHIRPS product, version 2.0, monthly precipitation with a spatial resolution of 0.05°. In this product, precipitation is estimated by integrating satellite rainfall data with measurements from meteorological stations (UCSB 2023). We developed a MATLAB script that: i) performs spatial clipping of monthly data for the Brazilian territory; ii) calculates the mean value of monthly accumulated precipitation per municipality; iii) exports the data in table format, where each row represents a municipality and each column represents a time interval. The climate normal of average monthly precipitation was calculated using a 30-year baseline period of data, from 1993 to 2022. This data was used to extract the municipal monthly precipitation anomaly time series, according to Eq. 1: $$\:{AP}_{m/a}={P}_{m/a}-{\mu\:}_{m}\:\left(1\right)$$ Where: AP m/a is the precipitation anomaly in mm.month − 1 , for month m , year a ; P m/a is the accumulated precipitation in mm.month − 1 , for month m , year a ; µ m is the average precipitation in mm.month − 1 , for month m , baseline 1993–2022. In addition to the precipitation anomaly, we calculated the SPI index (McKee et al. 1993 ), widely used to characterize meteorological droughts on various time scales. The SPI is closely related to soil moisture on short time scales. On longer time scales, the SPI may be related to groundwater reserves and storage. In the calculation of the SPI, observed precipitation is quantified as a standardized deviation from a selected probability distribution to model the monthly accumulated precipitation time series. After exploratory data analysis, we fit the Gamma distribution to the monthly precipitation series of each municipality in MATLAB. Based on the calculated parameters of the distribution, for each precipitation value in the municipal time series, the cumulative distribution function (CDF) is fitted describing the probability of that value not being exceeded according to the Gamma distribution. The SPI is obtained for each month and each municipality by transforming the probability value obtained in the previous step using the inverse function of the Normal distribution, with a mean of 0 and a standard deviation of 1. SPI values can be interpreted as the number of standard deviations by which the observed anomaly deviates from the long-term average. It is expected that 95% of the time, the SPI will remain between − 2 and + 2. With the aim of incorporating larger time scales to represent not only soil moisture but also groundwater storage and reserves, the same precipitation indexes (accumulated precipitation, precipitation anomaly, and SPI) were calculated for accumulations of 3, 6, and 12 months, resulting in 12 explanatory and dynamic variables (Table 1 ). For example, in Fig. 3 we illustrate the combination of precipitation variables for the month of January 2022. Table 1 Precipitation variables used in the modeling process Typology Variable Acronym Time Intervals Accumulated Precipitation Indexes P-01 P ( t ) P-03 P ( t ) + [...] + P ( t -2) P-06 P ( t ) + [...] + P ( t -5) P-12 P ( t ) + [...] + P ( t -11) Accumulated Precipitation Anomaly Indexes PA-01 PA ( t ) PA-03 PA ( t ) + [...] + PA ( t -2) PA-06 PA ( t ) + [...] + PA ( t -5) PA-12 PA ( t ) + [...] + PA ( t -11) Standardized Precipitation Index (SPI) SPI-01 SPI ( t ) SPI-03 SPI ( t to t -2) SPI-06 SPI ( t to t -5) SPI-12 SPI ( t to t -11) Temperature was used as a predictive variable in the neural network model as it directly influences the evapotranspiration rate. Seasonal forecasts are usually more accurate for the temperature variable, given that the estimation of evapotranspiration results from a complex interaction of meteorological factors such as wind, relative humidity, atmospheric pressure, among others. Since the goal is to adjust a model to have the ability to drought seasonal forecasts, using temperature should be advantageous. We used the NCEP/DOE Reanalysis II product (Kanamitsu et al. 2002 ) to generate the monthly temperature time series in municipalities. The monthly temperature data in raster format were processed in MATLAB using a script developed to: i) perform spatial clipping of monthly data for the Brazilian territory; ii) calculate the mean monthly temperature value per municipality; iii) export the data in table format. We calculated four temperature variables using lag times of 1, 3, 6 and 12 months (Table 2 ). We defined descriptive and spatial attributes that contribute to the modeling process for each Brazilian municipality, allowing the ANN to seek a customized threshold for precipitation and temperature needed to trigger drought in each territorial unit. In total, we established six static variables to describe each municipality, in addition to two dynamic variables related to the temporal/seasonal context (Table 3 ). Table 2 Temperature variables used in the modeling process Variable Time Intervals T-01 T( t ) T-03 [T( t ) + [...] + T( t -2)] / 3 T-06 [T( t ) + [...] + T( t -5)] / 6 T-12 [T( t ) + [...] + T( t -11)] / 12 Table 3 Descriptive, spatial and dynamic temporal variables by municipality Descriptive and spatial variables Dynamic temporal variables Biome identifier (BIO) Region identifier (REG) Frequency of occurrences (FRO) Month of the year (MON) Latitude (LAT) Status - previous situation (STA) Distance from oceans (DOC) Distance from the Amazon (DAM) The Brazilian biomes (BIO) (IBGE 2019) encompass types of vegetation that share similar geographical and climatic conditions, and they can aid in the process of regionalizing precipitation and temperature thresholds in the neural network to identify occurrences of droughts. The regions (REG) (IBGE 2023) incorporate, in addition to the aspect of location in the Brazilian territory, the political-administrative dimension inherent to the emergency decrees of municipalities and federative units. According to the biome and region of the country, we expect the ANN model to identify the appropriate thresholds for precipitation indexes to trigger droughts. The frequency of occurrences (FRO) in each municipality rescues the historical dimension of this type of climatic disaster and its impacts on society. Latitude (LAT) can assist in the modeling process, as this component directly influences global climate patterns and seasonality. The distance from oceans (DOC) and the Amazon (DAM) incorporates the importance of the proximity of areas that contribute to the distribution of moisture to continental areas, directly influencing meteorological and climatic conditions. Monthly identification (MON) allows the neural network to seek customized thresholds according to the time of the year, incorporating the seasonal dimension, while the previous situation (STA) reveals the initial condition of the municipality regarding droughts. We trained an ANN model for drought forecast and warning, specifically a Multi-Layered Perceptron (MLP) with a three-layer architecture, implemented through code developed in the MATLAB script editor. The assumption of a single intermediate layer is based on the existence theorem (Hecht-Nielsen 1990), which demonstrates that an ANN with a single hidden layer approximates any continuous numerical relationship. Training was conducted using the backpropagation algorithm (Rumelhart et al. 1986 ) with the Delta Rule for updating synaptic weights (Widrow and Hoff 1960). We chose the unipolar sigmoid activation function for all neurons. Several ANN models were developed to predict drought from the reported input variables. We tested 27 different combinations of precipitation and temperature predictor variables (Fig. 4 ). The objective of the tests was to verify how the indexes and lag-time windows affect the model's performance. We prepare a set of 129,609 samples, temporally and spatially distributed. The samples were divided: i) training (60%); ii) cross-validation (20%); iii) testing (20%). The training set is used to calibrate synaptic weights. The cross-validation set is used in parallel with training to prevent overfitting of the ANN (Hecht-Nielsen 1990). The testing set is used to assess the model's performance statistics. From the 10 years of data (2013–2022), we define the period from January 2021 to December 2022 as the test set (24 months). The training of the ANN models was performed using a k-fold cross-validation approach (Shao et al. 2013 ; Jiang and Chen 2016). In this study, we set k = 4, that is, the remaining dataset (from January 2013 to December 2020) was divided into four subsets (or folds), each with 24 months. In this manner, four ANN models were developed using different subsets of data samples. In each model round, three subsets were used for training and one for cross-validation. In the end, the ensemble result was considered, with the calculation of the average of the four adjusted ANN’s. Each round of modeling was performed with twenty repetitions using random initiations of the synaptic weights. This procedure was performed to reduce the effect of random initialization on the results. Additionally, we also tested the ideal number of neurons in the intermediate layer and the number of iterations required in each repetition of the model for the neural network to converge. The performance of the ANN was measured using four metrics: i) AUC index, area under the Receiver Operator Characteristic curve (Hanley 1989 ; Fawcett 2006 ); ii) overall accuracy (OA), where the number of correctly classified samples is divided by the total number of samples, obtained from the confusion matrix, which discriminates the output of the ANN into two classes (occurrence or non-occurrence of droughts); precision or positive predictive value, which indicates the fraction of positive predictions that were correct, obtained by dividing the number of True Positive (TP) samples by Predicted Positive (PP); recall or sensitivity, which indicates the proportion of observed droughts that were predicted, obtained by dividing TP by the total number of observed droughts. In the last stage of the study, we tested the best ANN model for droughts forecast. We progressively replaced the precipitation (CHIRPS dataset) and temperature (NCEP dataset) with the ECMWF-SEAS5 Seasonal Forecast (ECMWF 2023), using for lead time prediction of 1, 2, 3, and 4 months. The STA variable (occurrence or non-occurrence of drought in the previous month), when necessary, was replaced by the ANN's forecast for the immediately previous lead time (Fig. 5 ). We tested the performance of the forecast for the period between January 2021 and December 2022, considering all prediction lead times. Results The exploratory analysis of the sample data showed us relevant results on the explanatory variables of precipitation and temperature. In Fig. 6 , we show the average values of the accumulated precipitation (P), precipitation anomaly (PA), Standardized Precipitation Index (SPI) and temperature (T), in drought and non-drought conditions. We extracted the values for lag-times of 1, 3, 6 and 12 months per biome in Brazilian territory. The temperature tends to be higher in drought conditions, except in the Caatinga biome, which has a semi-arid tropical climate. In this region, the temperature has low seasonal variation and droughts occur almost every year, depending in particular on precipitation dynamics. Opposite to this, in the Pampa biome (extreme south of Brazil), the temperature presents much higher values during droughts, in particular, the variable T-01, 3.8 K of average difference in drought and non-drought situations. We found that the variables T-01 and T-03 are the most explanatory to define droughts. In Brazil, the temperature is 2 K higher during droughts considering the average of the variables T-01 and T-03. In Brazil, when droughts occur, precipitation decreases by 49.5% on average, ranging from − 53.9% (P-01) to -45.7% (P-12). Proportionally, variables P-01, P-03 and P-06 are the most explanatory. Once again, in the Caatinga biome we observed the smallest difference in drought and non-drought conditions, with an average decrease of 22.5%. The greatest relative differences in precipitation (P) were observed in the Cerrado and Pantanal biomes. In absolute values, we also highlight the Amazon, with an average decrease of 76 mm (P-01), 265 mm (P-03) and 389 mm (P-06). The biggest difference was observed in the Cerrado, with a decrease of 491 mm in the P-12 variable in droughts (Fig. 6 ). We observed that precipitation anomaly (PA) indexes are strongly related to drought and non-drought conditions in the Pampa biome. For example, the PA-03 variable has an average value of -181 mm in droughts, while in other periods it is + 30 mm. The largest precipitation anomaly was observed in the Amazon, with an average of -332 mm (PA-12). In the Caatinga, the precipitation anomaly presents lower values. For example, the average PA-03 in droughts is only − 24 mm. Regarding the SPI index, we observed that it is more explanatory for the droughts that occurred in the Pampa, Pantanal and Amazon biomes. For example, the SPI-03 variable in Pampa presented an average value of -1.3 in droughts. In the Caatinga, the SPI shows little change in drought and non-drought conditions, except for the SPI-12 variable, with an average of -0.42 in droughts. In general, we observe that precipitation indexes calculated from the CHIRPS product show a significant correlation with the occurrence of drought in Brazil. The P-12 variable was the one with the strongest linear correlation, with a correlation coefficient (r) of -0.592, indicating a significant dependence between the occurrence of drought and the accumulated precipitation in 12-months interval. The weakest linear correlation was observed for the SPI, with values between − 0.153 (SPI-01) and − 0.271 (SPI-12). Temperature variables obtained from the NCEP database showed linear correlations between 0.306 (T-06) and 0.33 (T-01), values that are also statistically significant. In Table 4 we present the performance metrics of ANN models for drought simulation in Brazilian municipalities. ANN-1 was the best model, with OA = 0.931, AUC = 0.952 and precision of 0.805. This model has 24 input variables, 47 neurons in the hidden layer and one output neuron (ANN 24-47-1). Regarding recall, we highlight the ANN-13 model, which obtained a value of 0.79 in this performance metric. The worst performing model was the ANN-20, in which we excluded all P, PA and SPI variables. This model obtained OA = 0.764, AUC = 0.84, precision of 0.383 and recall of 0.639. In general, the ANN models that exclude the precipitation anomaly (PA) variables present the lowest values of OA (average = 0.858), AUC (0.903), precision (0.576) and recall (0.72). Opposite to this, the exclusion of temperature variables (T) from ANN models slightly weakens the performance for drought simulation, with average values of OA = 0.899, AUC = 0.931, precision of 0.689 and recall of 0.756. Regarding lag times, we observed that the ANN models without the variables at lag times of 3 and 6 months show slightly lower performance. These results indicate that the trained ANN's (especially ANN-1) was able to simulate the droughts that cause damage in Brazilian municipalities, using various explanatory variables such as precipitation indexes (accumulated precipitation, precipitation anomaly and SPI) extracted from the CHIRPS product and temperature from the NCEP product, considering lag times from 1 to 12 months. However, only by subjecting the ANN to a situation where rainfall/temperature was not observed in the prediction time interval can we verify if the model has the ability to forecast drought. For this purpose, we tested the combined use of CHIRPS and NCEP products, along with the precipitation and temperature forecast from the ECMWF-SEAS5 Seasonal Forecast, for prediction lead times from 1 to 4 months. We carried out forecast tests for the time interval between July 2021 and June 2022, due to the numerous occurrences of drought, mainly in the Southern Region of Brazil. Table 4 Performance metrics of ANN models for drought simulation in Brazilian municipalities ANN Model Excluded Variable Overall Accuracy (OA) Area Under the ROC Curve (AUC) Precision Recall 1 - 0.931 0.952 0.805 0.784 2 SPI 0.925 0.947 0.786 0.766 3 PA 0.899 0.929 0.681 0.762 4 P 0.894 0.928 0.663 0.762 5 T 0.922 0.946 0.771 0.769 6 Lag-12 0.901 0.929 0.690 0.755 7 Lag-06 0.916 0.941 0.747 0.763 8 Lag-03 0.904 0.933 0.698 0.765 9 Lag-01 0.891 0.925 0.655 0.755 10 PA and SPI 0.849 0.900 0.542 0.708 11 P and SPI 0.882 0.919 0.630 0.737 12 P and PA 0.868 0.908 0.587 0.745 13 T and SPI 0.920 0.945 0.751 0.790 14 T and PA 0.896 0.931 0.672 0.756 15 T and P 0.918 0.943 0.749 0.776 16 Lag-06 and 12 0.831 0.882 0.501 0.698 17 Lag-03 and 12 0.885 0.920 0.639 0.740 18 Lag-01 and 12 0.863 0.907 0.575 0.735 19 Lag-01 and 06 0.877 0.915 0.613 0.742 20 P, PA and SPI 0.764 0.840 0.383 0.639 21 T, PA and SPI 0.856 0.900 0.560 0.700 22 T, P and SPI 0.909 0.938 0.715 0.770 23 T, P and PA 0.873 0.913 0.604 0.728 24 Lag-03, 06 and 12 0.781 0.852 0.410 0.663 25 Lag-01, 06 and 12 0.800 0.863 0.441 0.677 26 Lag-01, 03 and 12 0.808 0.870 0.455 0.672 27 Lag-01, 03 and 06 0.793 0.858 0.428 0.659 For example, Fig. 7 illustrates the precipitation variable P-12 (accumulated precipitation in 12 months) for January 2022, calculated from the integration of CHIRPS and ECMWF data, according to the forecast lead time. We emphasize the visual similarity between the map of 12-month accumulated precipitation (CHIRPS dataset only), Fig. 7 A, and the other four maps, including predicted precipitation from the ECMWF-SEAS5 model (Figs. 7 B- 7 E). Overall, combining the two datasets did not significantly change long-term precipitation accumulation. For January 2022, for example, the linear correlation was 0.956 between the 12-month observed accumulated precipitation (CHIRPS: 02/2021-01/2022), Fig. 7 A, and the version of the same variable with 4-month forecasted precipitation (CHIRPS: 02/2021-09/2021; ECMWF: 10/2021-01/2022), Fig. 7 E. The linear correlation coefficient was 0.993 for the 1-month forecasted precipitation, Fig. 7 B. The mean absolute difference in Brazilian municipalities varied between 49.3 mm (1-month forecast) and 151.7 mm (4-months forecast), considering the 12-month accumulated precipitation (P-12). While the 90th percentile error of the P12 variable calculation is 114.9 mm for a 1-month forecast, the index increases to 297.3 mm when we include the 4-month seasonal forecast from the ECMWF-SEAS5. Overall, as expected, as forecasted precipitation data is added to the 12-month accumulated total, the uncertainty increases, which should have an impact on the drought forecast made by the ANN model. We present in Fig. 8 some examples of the application of the ANN model for January 2022, when a severe drought occurred especially in the southern region of Brazil (Fig. 8 A, drought recorded in the S2iD database). In Fig. 8 B we present the output of the model in a simulation scenario (CHIRPS + NCEP), while in Figs. 8 C- 8 F we illustrate the ANN output in forecasts with lead times of 1 to 4-months (adding ECMWF-SEAS5). The maps indicate a drought index with continuous values in the interval [0–1] for all Brazilian municipalities. In general, the ANN model was successful in predicting drought with the application of seasonal forecasting. The forecast performance decreased according to the forecast lead time, as expected, but we can see that the ANN model satisfactorily predicted droughts up to 2 months in advance. We observed from the visual analysis of the results presented in Fig. 8 that the model was able to indicate drought in the municipalities of the Northeast and South of Brazil. On the other hand, we found that the predictive capacity of the model for forecasting droughts at the municipal level with a lead time of 3 or 4 months is lower. In the example in Fig. 8 , the drought in southern Brazil would have been underestimated with this anticipation of the forecast. However, we emphasize that the spatial pattern of the forecast is consistent with the observed map, and can regionally reproduce areas with a greater propensity for the occurrence of droughts. The results of the ANN model can also be presented using a binary classification, considering the threshold that presented the highest accuracy in the model training sample set (Fig. 9 ). We observed a high degree of agreement between the drought observed in January 2022 (Fig. 9 A) and the scenarios simulated (Fig. 9 B) and predicted 1 month in advance (Fig. 9 C). This indicates the high capability of correctly allocating municipalities with drought by the trained ANN model. The precise allocation of municipalities facing drought can also be considered good for a 2-month forecast (Fig. 9 D), significantly reducing it for 3 and 4 months, as previously mentioned. It is important to mention that this binary classification would not need to be used in future forecasting situations, as it is necessary only to provide model validation statistics, allowing for a quantitative assessment of the model in a forecast scenario. Once the model is validated, and its accuracy and precision are known, the ANN model can be used for forecasting considering the continuous interval [0–1], or another classification criterion, for example, in levels from low to high risk of drought occurrence in Brazilian municipalities. We show in Table 5 a summary of the performance metrics of the best ANN model for drought forecasting in Brazilian municipalities, considering only the testing period between January 2021 and December 2022. The overall accuracy of drought prediction ranged from 0.922 (lead time 1) to 0.757 (lead time 4). The ANN model consistently outperformed the persistent model, which would predict that current conditions will persist in the later time interval. Table 5 Performance metrics of the best ANN model for drought forecast in Brazilian municipalities, testing period between January 2021 and December 2022 ANN prediction and lead times Overall Accuracy (OA) Area Under the ROC Curve (AUC) Precision Recall Simulation t + 0 0.931 0.952 0.805 0.784 Forecasting t + 1 0.922 0.944 0.766 0.751 t + 2 0.868 0.883 0.612 0.550 t + 3 0.801 0.827 0.337 0.265 t + 4 0.757 0.785 0.209 0.146 The ANN model in the 1-month forecast scenario showed an accuracy greater than 0.86 in all months of the test period, with values always close to the simulated scenario (without the inclusion of seasonal forecast data from the ECMWF model). We observed that in forecasts with a lead time of 2 months, the model presented OA equal to 0.868 (the entire test period), with values always higher than 0.8 in all months of the analyzed period. This result reinforces what is visually observed in the maps, indicating that the ANN model, combined with ECMWF data, has a good forecasting capacity for this lead time. The AUC index of the drought forecast ranged from 0.944 (lead time 1) to 0.785 (lead time 4). In all months, the AUC was greater than 0.82 considering forecasts up to 2 months in advance. Although the OA and AUC values indicate that the ANN model makes correct predictions in lead times of 3 and 4 months, we observed a strong reduction in precision and recall indices, with values below 0.5 (Table 5 ). The OA and AUC values are strongly influenced by the ability of the ANN model to correctly indicate the municipalities where there will be no drought. This observation is important, as model allocation performance must be analyzed carefully, based on both visual analysis of the prediction and other performance metrics. In Fig. 10 we present the difference maps between observed and predicted droughts (January 2022), which exemplifies the reduction in precision and recall for forecasts with lead times of 3 and 4 months. The analysis of the difference map for the 1-month ahead prediction (Fig. 10 A) demonstrates the satisfactory allocation quality of the model. The model showed an accuracy of 0.928, precision of 0.803, and recall of 0.792, maintaining the observed performance metrics in the simulation. When analyzing the difference map for the 2-month lead time (Fig. 10 B), we observed an increase in errors, but still with satisfactory performance: accuracy of 0.87, precision of 0.67, and recall of 0.55. Discussion Precipitation and temperature anomalies are responsible for the occurrence of several types of natural disasters (Nandgude et al. 2023 ). In this context, Vicente-Serrano et al. ( 2020 ) addressed future climate change scenarios that suggest an increase in the severity of droughts across the world. As a result, it becomes even more important to monitor, model and predict droughts, seeking to mitigate the effects on societies and the environment (Tanguy et al. 2023 ). Predicting droughts with the potential to cause losses is extremely challenging due to the involvement of several hydrometeorological factors, in addition to local characteristics, such as crop seasonality, land use management, irrigation techniques, among others (Musolino et al. 2018 ). In our study, we trained artificial neural network models for this purpose, predicting droughts with the potential to cause losses as early as possible. In reviewing deep learning models for drought prediction, Gyaneshwar et al. ( 2023 ) summarize that the spatial and temporal variability of precipitation is seen as a factor that impairs the performance of models, due to its random components. In our study, in general, the precipitation variables calculated from the CHIRPS product showed a significant correlation with the occurrence of droughts in Brazil. These results are corroborated by Wu et al. ( 2019 ) and Sousa et al. ( 2023 ) in which CHIRPS data performed well for estimating monthly precipitation, and considered adequate to capture the spatial distribution of rainfall. Other researchers use CHIRPS data in specific studies and have concluded that this dataset can capture the characteristics of droughts. Pandey et al. (2022) performed a comparative analysis of three satellite precipitation products (TRMM-3B43 V7, PERSIANN-CDR and CHIRPS V2) and observed that CHIRPS data stands out compared to the others. Gao et al. ( 2018 ) investigated the applicability of the CHIRPS product for drought monitoring using the Standardized Precipitation Index (SPI). The results indicated good performance on multiple temporal scales (monthly, seasonal and annual). The researchers concluded that the product can be used to estimate precipitation and monitor droughts in the study area. Similarly, Guo et al. ( 2017 ) performed a meteorological analysis of droughts using the CHIRPS dataset and suggest that the product can adequately capture drought characteristics at multiple time scales, with the best performance at the three-month scale. The results we present indicate that precipitation variables are the most explanatory for the occurrence of droughts in Brazilian municipalities, especially the precipitation anomaly. In research related to droughts in California, Luo et al. ( 2017 ) investigated the influence of precipitation deficits and temperature anomalies on drought development. The study revealed that precipitation deficits have been largely responsible for extreme agricultural drought. Another result that we highlight refers to the importance of considering different time scales when calculating precipitation indices. In our study we used lag times of 1, 3, 6 and 12 months to calculate the variables of accumulated precipitation (P), precipitation anomaly (PA) and Standardized Precipitation Index (SPI). When removing any of the temporal scales we observe that there is degradation in forecast performance, especially when we remove the lag times of 3 and 6 months. Shahid ( 2010 ) and Yaseen et al. ( 2021 ) indicate that SPI-1, for example, reveals short-period precipitation deficits that directly affect ecology, air temperature and public health. The 3-month precipitation indices provide a characterization of water availability in the medium term, being appropriate in agricultural areas as an indicator of soil moisture (WMO 2008; Alamgir et al. 2019 ; Achour et al. 2020 ). The 6-month indices provide an indication of precipitation trends over a season (Khan et al. 2008), while the 12-month precipitation indices should be considered for long-term estimates for hydrological droughts and water resource management, being indicators of the decline of river flow and groundwater levels (Achour et al. 2020 ; Yaseen et al. 2021 ). Overall, the developed ANN model was successful in drought prediction. We observed that the model predicted droughts with very good accuracy in a 2-month lead time, which was possible using the seasonal prediction system of ECMWF-SEAS5. The quality of data from the ECMWF-SEAS5 model has already been demonstrated in some previous studies. Ferreira et al. ( 2022 ) evaluated the model's seasonal temperature and precipitation predictions for South America. The results showed that ECMWF-SEAS5 climate predictions are potentially useful and should be considered for planning activities that depend on hydrometeorological dynamics. Boas et al. ( 2023 ) developed a land surface modeling approach for seasonal soil moisture and crop yield prediction using SEAS5. The forecasting experiments were able to satisfactorily capture the interannual variations recorded in crop yields. Sutanto et al. ( 2020b ) evaluated the hydro-meteorological drought forecast skill for the pan-European region. Hydrological drought forecasts show good predictive power more than 2 months in advance. The authors suggest that the development of seasonal hydrological drought forecasting systems is beneficial, including the use of data from ECMWF-SEAS5. Busker et al. ( 2023 ) investigated the potential for early action for droughts by using seasonal forecasts from the SEAS5. The researchers conclude that early action using ECMWF-SEAS5 forecasts can reduce drought impacts. The proposed methodology proved to be adequate and our results indicate robust potential in the integration between satellite precipitation data, seasonal climate forecasting and a model based on artificial neural networks for drought forecasting and warning. According to Dikshit et al. ( 2022 ) as droughts are inherently non-linear and have many explanatory variables, ANN models have the ability to capture this dynamic relationship efficiently. In this study, the overall accuracy of the ANN model for drought simulation was 0.931. The forecast accuracy ranged from 0.922 for a 1-month lead time to 0.757 for 4 months. The AUC index of the drought forecast ranged from 0.944 (lead time 1) to 0.785 (lead time 4). In all months, the AUC was greater than 0.82 considering forecasts up to 2 months in advance. The performance metrics obtained by us are similar to previous scientific studies on the same topic. Mohamadi et al. ( 2020 ) developed a zoning map for drought prediction (SPI-3) using machine learning models and nomadic people optimization algorithm (NPA). The Adaptive Neuro-Fuzzy Interface System (ANFIS–NPA) model obtained an AUC value equal to 0.92 for extreme droughts. Sutanto et al. ( 2020a ) analyzed the skill of large-scale seasonal drought impact forecasts. The researchers developed drought impact functions using machine learning approaches to predict drought impacts with lead times of 1 to 7 months. The AUC values presented by the authors are compatible with those from our study. Dikshit et al. ( 2021 ) presented a drought prediction approach (SPEI index) using the long short-term memory neural network (LSTM). The authors obtained AUC values of 0.83 and 0.82 for SPEI-1 and SPEI-3, respectively. We observed a progressive decrease in precision and recall values as the forecast lead time increased. This reduction in model performance must be mainly associated with the limitations of the seasonal weather forecast model for specific applications at the local level. For Nandgude et al. ( 2023 ) it may be important to consider external driving forces for the occurrence of droughts, such as Sea Surface Temperature (SST) anomalies and region-specific climate indices, such as the Pacific Decadal Oscillation (PDO), the El Niño-Southern Oscillation (ENSO), and the Indian Ocean Dipole (IOD). The authors indicate that the use of these climate indices can improve drought forecasting results. Although the results presented by us are satisfactory and promising, we emphasize that caution should be taken when using the ANN model to forecast droughts using only the binary classification (0 = no drought; 1 = drought) for seasonal forecasts of 3 or 4 months ahead. On the other hand, the model reproduced the spatial pattern of droughts, especially when the output is interpreted as a continuous index of drought risk. We can affirm from the results obtained in this study that the trained model is efficient and the results indicate strong potential for drought forecasting and warning, using ANN, remote sensing data, hydrometeorological and climate indexes. Conclusions In this study, we developed a drought forecast model in Brazilian municipalities based on the integration between remote sensing data, hydro-meteorological and climate indexes, and ANN-based model. We used an inventory of droughts that caused losses by municipalities to generate the model's dependent variable. After model training, we tested the ANN for drought forecasts for lead times of 1–4 months, using seasonal forecast data from ECMWF-SEAS5. The overall accuracy of the ANN model for drought simulation was 0.931. In turn, the forecast accuracy ranged from 0.922 for a 1-month lead time to 0.757 for 4 months. We can affirm that the ANN model consistently outperformed the persistent model. Additionally, we observed a progressive decrease in precision and recall values as the forecast lead time increased. Remarkably, the model reproduced the spatial pattern of droughts, especially when the output is interpreted as a continuous index of drought risk. We conclude that the trained model is efficient and the results indicate strong potential for drought forecasting and warning, using ANN, remote sensing data, hydrometeorological and climate indexes. Finally, we conclude that the model trained in this study can be applied for forecasting and warning of droughts at any time of the year for the entire Brazilian territory. The methodological approach has the following main advantages: (i) it indicates with very good accuracy the municipalities with the potential to present losses due to droughts; (ii) does not require spatial interpolation to generate precipitation or temperature data; (iii) and the practicality of accessing data and processing the forecast, which can be carried out using computer codes, making processing more agile. Declarations Acknowledgements We are grateful for the support of the National Council for Scientific and Technological Development (Conselho Nacional de Desenvolvimento Científico e Tecnológico - CNPq), with financial support to the first author through CNPq Call No. 09/2022 - Research Productivity, process 311009/2022-0. Funding Partial financial support was received from National Council for Scientific and Technological Development (Conselho Nacional de Desenvolvimento Científico e Tecnológico - CNPq), with financial support to the first author through CNPq Call No. 09/2022 Research Productivity, process 311009/2022-0. Competing Interests The authors have no conflicts of interest to declare that are relevant to the content of this manuscript. Author Contributions Guilherme Garcia de Oliveira contributed to the study conception and design. Material preparation, data collection and analysis were performed by all authors. The first draft of the manuscript was written by Guilherme Garcia de Oliveira and Nicholas Becker Pires Pi. All authors commented on previous versions of the manuscript. All authors read and approved the final manuscript. References Achour K, Meddi M, Zeroual A, Bouabdelli S, Maccioni P, Moramarco T (2020) Spatio-temporal analysis and forecasting of drought in the plains of northwestern Algeria using the standardized precipitation index. J Earth Syst Sci 129(1). https://doi.org/10.1007/s12040-019-1306-3 Alamgir M, Mohsenipour M, Homsi R, Wang X, Shahid S, Shiru M, Alias N, Yuzir A (2019) Parametric assessment of seasonal drought risk to crop production in Bangladesh. Sustainability 11(5):1442. https://doi.org/10.3390/su11051442 Belayneh A, Adamowski J (2013) Drought forecasting using new machine learning methods. J Water Land Development 18(9):3-12. https://doi.org/10.2478/jwld-2013-0001 Blaikie P, Cannon T, Davis I, Wisner B (2014) At risk: Natural hazards, people’s vulnerability and disasters. Routledge. Boas T, Bogena HR, Ryu D, Vereecken H, Western A, Hendricks Franssen HJ (2023) Seasonal soil moisture and crop yield prediction with fifth-generation seasonal forecasting system (SEAS5) long-range meteorological forecasts in a land surface modelling approach. Hydrol Earth Syst Sci 27(16):3143-3167. https://doi.org/10.5194/hess-27-3143-2023 Boluwade A (2020) Spatial-Temporal assessment of satellite-based rainfall estimates in different precipitation regimes in water-scarce and data-sparse regions. Atmos 11(9):901. https://doi.org/10.3390/atmos11090901 Boult VL, Asfaw DT, Young M, Maidment R, Mwangi E, Ambani M, Waruru S, Otieno G, Todd MC, Black E (2020) Evaluation and validation of TAMSAT‐ALERT soil moisture and WRSI for use in drought anticipatory action. Meteorol Appl 27(5). https://doi.org/10.1002/met.1959 Busker T, de Moel H, van den Hurk B, Aerts J (2023) Impact-based seasonal rainfall forecasting to trigger early action for droughts. Sci Total Environ 898:165506. https://doi.org/10.1016/j.scitotenv.2023.165506 Canedo Rosso C, Hochrainer-Stigler S, Pflug G, Condori B, Berndtsson R (2018) Early warning and drought risk assessment for the Bolivian Altiplano agriculture using high resolution satellite imagery data. Nat Hazard Earth Syst Sci Discuss 1-23. http://dx.doi.org/10.5194/nhess-2018-133 Cheng S, Wang W, Yu Z (2021) Evaluating the drought-monitoring utility of GPM and TRMM precipitation products over mainland China. Remote Sens 13(20):4153. https://doi.org/10.3390/rs13204153 Dikshit A, Pradhan B, Alamri AM (2020) Temporal hydrological drought index forecasting for New South Wales, Australia using machine learning approaches. Atmos 11(6):585. https://doi.org/10.3390/atmos11060585 Dikshit A, Pradhan B, Huete A (2021) An improved SPEI drought forecasting approach using the long short-term memory neural network. J Environ Manag 283:111979. https://doi.org/10.1016/j.jenvman.2021.111979 Dikshit A, Pradhan B, Santosh M (2022) Artificial neural networks in drought prediction in the 21st century - A scientometric analysis. Appl Soft Comput 114:108080. https://doi.org/10.1016/j.asoc.2021.108080 ECMWF European Centre for Medium-Range Weather Forecasts (2023) Seasonal forecast anomalies on single levels. https://cds.climate.copernicus.eu/cdsapp#!/dataset/seasonal-postprocessed-single-levels?tab=form. Accessed 13 December 2023 Fawcett T (2006) An introduction to ROC analysis. Pattern Recognit Lett 27(8):861-874. https://doi.org/10.1016/j.patrec.2005.10.010 Ferreira GWS, Reboita MS (2022) A new look into the South America precipitation regimes: Observation and forecast. Atmos 13(6):873. https://doi.org/10.3390/atmos13060873 Ferreira GWS, Reboita MS, Drumond A (2022) Evaluation of ECMWF-SEAS5 Seasonal Temperature and Precipitation Predictions over South America. Clim 10(9):128. https://doi.org/10.3390/cli10090128 Gao F, Zhang Y, Ren X, Yao Y, Hao Z, Cai W (2018) Evaluation of CHIRPS and its application for drought monitoring over the Haihe River Basin, China. Nat Hazard 92(1):155-172. https://doi.org/10.1007/s11069-018-3196-0 Guo H, Bao A, Liu T, Ndayisaba F, He D, Kurban A, De Maeyer P (2017) Meteorological drought analysis in the Lower Mekong Basin using satellite-based long-term CHIRPS product. Sustainability 9(6):901. https://doi.org/10.3390/su9060901 Gyaneshwar A, Mishra A, Chadha U, Raj Vincent PMD, Rajinikanth V, Pattukandan Ganapathy G, Srinivasan K (2023) A contemporary review on deep learning models for drought prediction. Sustainability 15(7):6160. https://doi.org/10.3390/su15076160 Hagenlocher M, Meza I, Anderson CC, Min A, Renaud FG, Walz Y, Siebert S, Sebesvari Z (2019) Drought vulnerability and risk assessments: state of the art, persistent gaps, and research agenda. Environ Res Lett 14(8):083002. https://doi.org/10.1088/1748-9326/ab225d Hanley JA (1989) Receiver operating characteristic (ROC) methodology: the state of the art. Crit Rev Diagn Imaging, 29(3):307-335. He X, Feng K, Li X, Craft AB, Wada Y, Burek P, Wood EF, Sheffield J (2019) Solar and wind energy enhances drought resilience and groundwater sustainability. Nat Commun 10(1). https://doi.org/10.1038/s41467-019-12810-5 Held IM, Soden BJ (2006) Robust Responses of the Hydrological Cycle to Global Warming. J Clim 19(21):5686-5699. https://doi.org/10.1175/jcli3990.1 Howells M, Hermann S, Welsch M, Bazilian M, Segerström R, Alfstad T, Gielen D, Rogner H, Fischer G, van Velthuizen H, Wiberg D, Young C, Roehrl RA, Mueller A, Steduto P, Ramma I (2013) Integrated analysis of climate change, land-use, energy and water strategies. Nat Clim Chang 3(7):621-626. https://doi.org/10.1038/nclimate1789 Huete A, Didan K, Miura T, Rodriguez EP, Gao X, Ferreira LG (2002) Overview of the radiometric and biophysical performance of the MODIS vegetation indices. Remote Sens Environ 83(1-2):195-213. https://doi.org/10.1016/s0034-4257(02)00096-2 Huynh TD, Nguyen TH, Truong C (2020) Climate risk: The price of drought. J Corp Finance 65:101750. https://doi.org/10.1016/j.jcorpfin.2020.101750 IBGE Instituto Brasileiro de Geografia e Estatística (2019) Biomas e sistema costeiro-marinho do Brasil. https://www.ibge.gov.br/geociencias/informacoes-ambientais/vegetacao/15842-biomas.html. Accessed 22 January 2023 IBGE Instituto Brasileiro de Geografia e Estatística (2023) Malhas territoriais. https://www.ibge.gov.br/geociencias/downloads-geociencias.html. Accessed 22 January 2023 IPCC (2021) Climate Change 2021: The Physical Science Basis. Contribution of Working Group I to the Sixth Assessment Report of the Intergovernmental Panel on Climate Change [Masson-Delmotte, V, Zhai P, Pirani A, Connors SL, Péan C, Berger S, Caud N, Chen Y, Goldfarb L, Gomis MI, Huang M, Leitzell K, Lonnoy E, Matthews JBR, Maycock TK, Waterfield T, Yelekçi O, Yu R, Zhou B (eds.)]. Cambridge University Press. In Press. Johnson SJ, Stockdale TN, Ferranti L, Balmaseda MA, Molteni F, Magnusson L, Tietsche S, Decremer D, Weisheimer A, Balsamo G, Keeley SPE, Mogensen K, Zuo H, Monge-Sanz BM (2019) SEAS5: the new ECMWF seasonal forecast system. Geoscientific Model Dev 12(3):1087-1117. https://doi.org/10.5194/gmd-12-1087-2019 Kallis G (2008) Droughts. Annu Rev Environ Resour 33(1):85-118. https://doi.org/10.1146/annurev.environ.33.081307.123117 Kanamitsu M, Ebisuzaki W, Woollen J, Yang SK, Hnilo JJ, Fiorino M, Potter GL (2002) NCEP–DOE AMIP-II Reanalysis (R-2). Bull Am Meteorol Soc 83(11):1631-1643. https://doi.org/10.1175/bams-83-11-1631 Kim TW, Jehanzaib M (2020) Drought Risk Analysis, Forecasting and Assessment under Climate Change. Water 12(7):1862. https://doi.org/10.3390/w12071862 Kofidou M, Stathopoulos S, Gemitzi A (2023) Review on spatial downscaling of satellite derived precipitation estimates. Environ Earth Sci 82(18). https://doi.org/10.1007/s12665-023-11115-7 Liu Y, Chen J (2021) Future global socioeconomic risk to droughts based on estimates of hazard, exposure, and vulnerability in a changing climate. Sci Total Environ 751:142159. https://doi.org/10.1016/j.scitotenv.2020.142159 Luo L, Apps D, Arcand S, Xu H, Pan M, Hoerling M (2017) Contribution of temperature and precipitation anomalies to the California drought during 2012–2015. Geophys Res Lett 44(7):3184-3192. https://doi.org/10.1002/2016gl072027 Marj AF, Meijerink AMJ (2011) Agricultural drought forecasting using satellite images, climate indices and artificial neural network. Int J Remote Sens 32(24):9707-9719. https://doi.org/10.1080/01431161.2011.575896 McKee TB, Doesken NJ, Kleist J (1993) The relationship of drought frequency and duration to time scales. 8th Conference on Applied Climatology, Anaheim, 17-22, 179-184. (n.d.). Scientific Research Publishing. https://www.scirp.org/reference/ReferencesPapers?ReferenceID=2099290 Mishra AK, Desai VR (2006) Drought forecasting using feed-forward recursive neural network. Ecol Model 198(1-2):127-138. https://doi.org/10.1016/j.ecolmodel.2006.04.017 Mohamadi S, Sammen S, Panahi F, Ehteram M, Kisi O, Mosavi A, Ahmed AN, El-Shafie A, Al-Ansari N (2020) Zoning map for drought prediction using integrated machine learning models with a nomadic people optimization algorithm. Nat Hazard 104(1):537-579. https://doi.org/10.1007/s11069-020-04180-9 Mokhtari R, Akhoondzadeh M (2019) Neural network method for drought modeling using satellite data. The International Archives of the Photogrammetry, Remote Sensing and Spatial Information Sciences, XLII-4/W18, pp 749-753. https://doi.org/10.5194/isprs-archives-xlii-4-w18-749-2019 Mulualem GM, Liou YA (2020) Application of Artificial Neural Networks in Forecasting a Standardized Precipitation Evapotranspiration Index for the Upper Blue Nile Basin. Water 12(3):643. https://doi.org/10.3390/w12030643 Musolino DA, Massarutto A, de Carli A (2018) Does drought always cause economic losses in agriculture? An empirical investigation on the distributive effects of drought events in some areas of Southern Europe. Sci Total Environ 633:1560-1570. https://doi.org/10.1016/j.scitotenv.2018.02.308 Nandgude N, Singh TP, Nandgude S, Tiwari M (2023) Drought Prediction: A Comprehensive Review of Different Drought Prediction Models and Adopted Technologies. Sustainability 15(15):11684. https://doi.org/10.3390/su151511684 NASA (2023) Global Climate Change - Vital Signs of the Planet. Goddard Institute for Space Studies (GISS). https://climate.nasa.gov/. Accessed 28 November 2023 Oliveira GG, Ruiz LFC, Guasselli LA, Teixeira B, Klabin G (2023) Modelo baseado em redes neurais artificiais e dados de chuva por satélites para alerta de estiagens e secas. 20th Simpósio Brasileiro de Sensoriamento Remoto. São José dos Campos, INPE. https://proceedings.science/sbsr-2023/trabalhos/modelo-baseado-em-redes-neurais-artificiais-e-dados-de-chuva-por-satelites-para?lang=pt-br. Accessed 28 July 2023 Palharini R, Vila D, Rodrigues D, Palharini R, Mattos E, Undurraga E (2022) Analysis of Extreme Rainfall and Natural Disasters Events Using Satellite Precipitation Products in Different Regions of Brazil. Atmos 13(10):1680. https://doi.org/10.3390/atmos13101680 Pandey V, Srivastava PK, Mall RK, Munoz-Arriola F, Han D (2020) Multi-satellite precipitation products for meteorological drought assessment and forecasting in Central India. Geocarto Int 37(7):1899-1918. https://doi.org/10.1080/10106049.2020.1801862 Panu US, Sharma TC (2002) Challenges in drought research: some perspectives and future directions. Hydrol Sci J 47:19-30. https://doi.org/10.1080/02626660209493019 Rumelhart DE, Hinton GE, Williams RJ (1986) Learning representations by back-propagating errors. Nature 323(6088):533-536. https://doi.org/10.1038/323533a0 Santos JF, Portela MM, Pulido-Calvo I (2012) Spring drought prediction based on winter NAO and global SST in Portugal. Hydrol Process 28(3):1009-1024. https://doi.org/10.1002/hyp.9641 Santos SRQ, Cunha APMA, Ribeiro-Neto GG (2019) Avaliação de dados de precipitação para o monitoramento do padrão espaço-temporal da seca no nordeste do Brasil. Rev Bras Climatol 25:80-100. https://doi.org/10.5380/abclima.v25i0.62018 SEDEC Secretaria Nacional de Proteção e Defesa Civil (2023) Sistema Integrado de Informações sobre Desastres - S2ID. https://s2id.mi.gov.br/paginas/relatorios/. Accessed 10 July 2023 Shahid S (2010) Recent trends in the climate of Bangladesh. Clim Res 42(3):185-193. https://doi.org/10.3354/cr00889 Shao C, Paynabar K, Kim TH, Jin J, Hu SJ, Spicer JP, Wang H, Abell JA (2013) Feature selection for manufacturing process monitoring using cross-validation. J Manuf Syst 32(4):550-555. https://doi.org/10.1016/j.jmsy.2013.05.006 Sousa LB, de Assunção Montenegro AA, da Silva MV, Almeida TAB, de Carvalho AA, da Silva TGF, de Lima JLMP (2023) Spatiotemporal Analysis of Rainfall and Droughts in a Semiarid Basin of Brazil: Land Use and Land Cover Dynamics. Remote Sens 15(10):2550. https://doi.org/10.3390/rs15102550 Stewart IT, Rogers J, Graham A (2020) Water security under severe drought and climate change: Disparate impacts of the recent severe drought on environmental flows and water supplies in Central California. J Hydrol X 7:100054. https://doi.org/10.1016/j.hydroa.2020.100054 Sundararajan K, Garg L, Srinivasan K, Kashif Bashir A, Kaliappan J, Pattukandan Ganapathy G, Kumaran Selvaraj S, Meena T (2021) A Contemporary Review on Drought Modeling Using Machine Learning Approaches. Comput Model Eng Sci 128(2):447-487. https://doi.org/10.32604/cmes.2021.015528 Sutanto SJ, van der Weert M, Blauhut V, Van Lanen HAJ (2020a) Skill of large-scale seasonal drought impact forecasts. Nat Hazard Earth Syst Sci 20(6):1595-1608. https://doi.org/10.5194/nhess-20-1595-2020 Sutanto SJ, Wetterhall F, Van Lanen HAJ (2020b) Hydrological drought forecasts outperform meteorological drought forecasts. Environ Res Lett 15(8):084010. https://doi.org/10.1088/1748-9326/ab8b13 Tanguy M, Chevuturi A, Marchant BP, Mackay JD, Parry S, Hannaford J (2023) How will climate change affect the spatial coherence of streamflow and groundwater droughts in Great Britain? Environ Res Lett 18(6):064048. https://doi.org/10.1088/1748-9326/acd655 Trnka M, Hlavinka P, Možný M, Semerádová D, Štěpánek P, Balek J, Bartošová L, Zahradníček P, Bláhová M, Skalák P, Farda A, Hayes M, Svoboda M, Wagner W, Eitzinger J, Fischer M, Žalud Z (2020) Czech Drought Monitor System for monitoring and forecasting agricultural drought and drought impacts. Int J Climatol 40(14):5941-5958. https://doi.org/10.1002/joc.6557 UCSB University of California at Santa Barbara (2023) Climate Hazards Group Infrared Precipitation with Stations (CHIRPS). https://data.chc.ucsb.edu/products/CHIRPS-2.0. Accessed 26 June 2023 Vicente-Serrano SM, Beguería S, López-Moreno JI (2010) A Multiscalar Drought Index Sensitive to Global Warming: The Standardized Precipitation Evapotranspiration Index. J Clim 23(7):1696-1718. https://doi.org/10.1175/2009jcli2909.1 Vicente-Serrano SM, Quiring SM, Peña-Gallardo M, Yuan S, Domínguez-Castro F (2020) A review of environmental droughts: Increased risk under global warming? Earth-Sci Rev 201:102953. https://doi.org/10.1016/j.earscirev.2019.102953 WMO World Meteorological Organization (2008) Guide to Hydrological Practice, Volume I, Hydrology – From Measurement to Hydrological Information. 6th ed, Geneva, Switzerland. Wu W, Li Y, Luo X, Zhang Y, Ji X, Li X (2019) Performance evaluation of the CHIRPS precipitation dataset and its utility in drought monitoring over Yunnan Province, China. Geomat Nat Hazard Risk 10(1):2145-2162. https://doi.org/10.1080/19475705.2019.1683082 Yaseen ZM, Ali M, Sharafati A, Al-Ansari N, Shahid S (2021) Forecasting standardized precipitation index using data intelligence models: Regional investigation of Bangladesh. Sci Rep 11(1):3435. https://doi.org/10.1038/s41598-021-82977-9 Zhang R, Chen ZY, Xu LJ, Ou CQ (2019) Meteorological drought forecasting based on a statistical model with machine learning techniques in Shaanxi province, China. Sci Total Environ 665: 338-346. https://doi.org/10.1016/j.scitotenv.2019.01.431 Zhang Y, Li W, Chen Q, Pu X, Xiang L (2017) Multi-models for SPI drought forecasting in the north of Haihe River Basin, China. Stoch Environ Res Risk Assess 31(10):2471-2481. https://doi.org/10.1007/s00477-017-1437-5 Cite Share Download PDF Status: Under Revision Version 1 posted Editorial decision: Major revisions 26 May, 2025 Reviewers agreed at journal 16 Aug, 2024 Reviewers invited by journal 29 Jul, 2024 Editor assigned by journal 23 Jul, 2024 First submitted to journal 22 Jul, 2024 You are reading this latest preprint version Research Square lets you share your work early, gain feedback from the community, and start making changes to your manuscript prior to peer review in a journal. As a division of Research Square Company, we’re committed to making research communication faster, fairer, and more useful. We do this by developing innovative software and high quality services for the global research community. Our growing team is made up of researchers and industry professionals working together to solve the most critical problems facing scientific publishing. Also discoverable on Platform About Our Team In Review Editorial Policies Advisory Board Help Center Resources Author Services Accessibility API Access RSS feed Manage Cookie Preferences © Research Square 2026 | ISSN 2693-5015 (online) Privacy Policy Terms of Service Do Not Sell My Personal Information {"props":{"pageProps":{"initialData":{"identity":"rs-4784321","acceptedTermsAndConditions":true,"allowDirectSubmit":false,"archivedVersions":[],"articleType":"Research Article","associatedPublications":[],"authors":[{"id":332999058,"identity":"dc2e14c1-aab8-4658-bd15-9cfd5f1290fc","order_by":0,"name":"Guilherme Garcia de Oliveira","email":"data:image/png;base64,iVBORw0KGgoAAAANSUhEUgAAAZAAAAAyAQMAAABI0h/eAAAABlBMVEX///8AAABVwtN+AAAACXBIWXMAAA7EAAAOxAGVKw4bAAABAUlEQVRIiWNgGAWjYNCCAgkGBmYgnWBgAyQZGw8Q1mIA1fKgIg2kpYEYLRCK8cGZw2AGXi3yDuwXGD4YWCT2s/M+/JDYdt5ubfthoC01NtG4tBge4ClgnGEgkTizmd1YIrHtdvK2M4lALcfSchtwaWngSWDmMZAwNjjMxgDWYnYAqIWx4TBhLfaH2Zh/JLadSzY7/xC/FnkG9gMgLXIGzGxsEglnDtiZ3SBgiwEzD8NBoF/kJA6zsVkkVCQnmN0A2pKAxy/y7e0PH3yoqOPh7z/GfPOHgZ292fl0oEiNDW5bDvMYHEAWSASrTMChHGxLA/sDFAF7PIpHwSgYBaNghAIAJTdbuf/p/X0AAAAASUVORK5CYII=","orcid":"https://orcid.org/0000-0003-4197-5704","institution":"Universidade Federal do Rio Grande do Sul","correspondingAuthor":true,"prefix":"","firstName":"Guilherme","middleName":"Garcia","lastName":"de Oliveira","suffix":""},{"id":332999059,"identity":"ac68622d-7981-4ee8-91bf-0c79f2495285","order_by":1,"name":"Nicholas Becker Pires Pi","email":"","orcid":"","institution":"Universidade Federal do Rio Grande do Sul","correspondingAuthor":false,"prefix":"","firstName":"Nicholas","middleName":"Becker Pires","lastName":"Pi","suffix":""},{"id":332999060,"identity":"b4aed6f2-8e23-40ab-839a-8db3b4635b8d","order_by":2,"name":"Laurindo Antonio Guasselli","email":"","orcid":"","institution":"Universidade Federal do Rio Grande do Sul","correspondingAuthor":false,"prefix":"","firstName":"Laurindo","middleName":"Antonio","lastName":"Guasselli","suffix":""},{"id":332999061,"identity":"9f2b219f-f8ee-4114-84e9-cab87185be55","order_by":3,"name":"Luis Fernando Chimelo Ruiz","email":"","orcid":"","institution":"Universidade Federal do Rio Grande do Sul","correspondingAuthor":false,"prefix":"","firstName":"Luis","middleName":"Fernando Chimelo","lastName":"Ruiz","suffix":""}],"badges":[],"createdAt":"2024-07-22 21:00:03","currentVersionCode":1,"declarations":"","doi":"10.21203/rs.3.rs-4784321/v1","doiUrl":"https://doi.org/10.21203/rs.3.rs-4784321/v1","draftVersion":[],"editorialEvents":[],"editorialNote":"","failedWorkflow":false,"files":[{"id":63110594,"identity":"8d83a50f-28a1-453b-96c4-b1f2a4759900","added_by":"auto","created_at":"2024-08-23 08:49:26","extension":"jpg","order_by":1,"title":"Figure 1","display":"","copyAsset":false,"role":"figure","size":2161447,"visible":true,"origin":"","legend":"\u003cp\u003eNumber of months with drought between 2013 and 2022 in Brazilian municipalities\u003c/p\u003e","description":"","filename":"FIG01.jpg","url":"https://assets-eu.researchsquare.com/files/rs-4784321/v1/7bef6e9666cc8bc9506330f2.jpg"},{"id":63109986,"identity":"f856cbcc-2602-441a-8d81-36debfd55fdf","added_by":"auto","created_at":"2024-08-23 08:41:25","extension":"jpg","order_by":2,"title":"Figure 2","display":"","copyAsset":false,"role":"figure","size":258889,"visible":true,"origin":"","legend":"\u003cp\u003eFlowchart of methods in the study\u003c/p\u003e","description":"","filename":"FIG02.jpg","url":"https://assets-eu.researchsquare.com/files/rs-4784321/v1/9b181a1a43aec878f9276119.jpg"},{"id":63110592,"identity":"063b8cde-d22d-472c-88a6-ab944ef83694","added_by":"auto","created_at":"2024-08-23 08:49:25","extension":"jpg","order_by":3,"title":"Figure 3","display":"","copyAsset":false,"role":"figure","size":2055036,"visible":true,"origin":"","legend":"\u003cp\u003ePrecipitation variables used in the modeling process. Example: January 2022.\u003c/p\u003e","description":"","filename":"FIG03.jpg","url":"https://assets-eu.researchsquare.com/files/rs-4784321/v1/e2e4ad505c3cb4badf887fc3.jpg"},{"id":63109989,"identity":"8e2cd955-c978-428b-97f2-e793e8aee8b8","added_by":"auto","created_at":"2024-08-23 08:41:25","extension":"jpg","order_by":4,"title":"Figure 4","display":"","copyAsset":false,"role":"figure","size":175353,"visible":true,"origin":"","legend":"\u003cp\u003eANN models and combinations of precipitation and temperature input variables: blue cells indicate that the variable was used\u003c/p\u003e","description":"","filename":"FIG04.jpg","url":"https://assets-eu.researchsquare.com/files/rs-4784321/v1/cba6bd9fd4c9ea19b7d2d90c.jpg"},{"id":63109991,"identity":"ee56e9ff-8f5a-41ec-b076-0fc81a4653ed","added_by":"auto","created_at":"2024-08-23 08:41:25","extension":"jpg","order_by":5,"title":"Figure 5","display":"","copyAsset":false,"role":"figure","size":329935,"visible":true,"origin":"","legend":"\u003cp\u003eVariable replacement process for precipitation (P), temperature (T), and initial condition (STA) to perform the forecast/warning, according to the established lead time\u003c/p\u003e","description":"","filename":"FIG05.jpg","url":"https://assets-eu.researchsquare.com/files/rs-4784321/v1/1b329d081ef7b40ac5dbefdc.jpg"},{"id":63109994,"identity":"d04f6dc2-9a38-499c-af3c-4db14801df4a","added_by":"auto","created_at":"2024-08-23 08:41:26","extension":"jpg","order_by":6,"title":"Figure 6","display":"","copyAsset":false,"role":"figure","size":521960,"visible":true,"origin":"","legend":"\u003cp\u003eAccumulated precipitation (P), precipitation anomaly (PA), Standardized Precipitation Index (SPI) and temperature (T) for lag-times of 1, 3, 6 and 12 months per biome in Brazil: mean values for drought and non-drought conditions\u003c/p\u003e","description":"","filename":"FIG06.jpg","url":"https://assets-eu.researchsquare.com/files/rs-4784321/v1/6a687790f2946be2d2846588.jpg"},{"id":63110593,"identity":"4c8eb856-0b3c-446b-b7d0-79a3ddb3fe86","added_by":"auto","created_at":"2024-08-23 08:49:25","extension":"jpg","order_by":7,"title":"Figure 7","display":"","copyAsset":false,"role":"figure","size":314971,"visible":true,"origin":"","legend":"\u003cp\u003eAccumulated precipitation in 12 months (P-12) for January 2022, calculated according to the forecast lead time: (A) observed, CHIRPS; (B) 1-month forecast, ECMWF-SEAS5; (C) 2-month forecast, ECMWF-SEAS5; (D) 3-month forecast, ECMWF-SEAS5; (E) 4-month forecast, ECMWF-SEAS5\u003c/p\u003e","description":"","filename":"FIG07.jpg","url":"https://assets-eu.researchsquare.com/files/rs-4784321/v1/a2cb9160b1962d0fc78acbc4.jpg"},{"id":63109992,"identity":"12deda1a-a1aa-4bd5-ab9d-a89cc36f054d","added_by":"auto","created_at":"2024-08-23 08:41:26","extension":"jpg","order_by":8,"title":"Figure 8","display":"","copyAsset":false,"role":"figure","size":507296,"visible":true,"origin":"","legend":"\u003cp\u003eANN outputs in continuous values in the interval [0-1] for January 2022. A) observed; B) ANN simulation, time t; C) ANN forecast, time t+1; D) forecast in t+2; E) forecast in t+3; F) forecast in t+4\u003c/p\u003e","description":"","filename":"FIG08.jpg","url":"https://assets-eu.researchsquare.com/files/rs-4784321/v1/80d5877419297a5909e2eb5d.jpg"},{"id":63110595,"identity":"469e9e40-0e19-4e4b-a742-f69a64818b4c","added_by":"auto","created_at":"2024-08-23 08:49:26","extension":"jpg","order_by":9,"title":"Figure 9","display":"","copyAsset":false,"role":"figure","size":437640,"visible":true,"origin":"","legend":"\u003cp\u003eANN model results using binary classification for January 2022. A) observed; B) ANN simulation, time t; C) ANN forecast, time t+1; D) forecast in t+2; E) forecast in t+3; F) forecast in t+4\u003c/p\u003e","description":"","filename":"FIG09.jpg","url":"https://assets-eu.researchsquare.com/files/rs-4784321/v1/6de6966b866e2101de3f5d39.jpg"},{"id":63109997,"identity":"80147a54-2c74-421c-86fa-ba1ad9b3230a","added_by":"auto","created_at":"2024-08-23 08:41:26","extension":"jpg","order_by":10,"title":"Figure 10","display":"","copyAsset":false,"role":"figure","size":781442,"visible":true,"origin":"","legend":"\u003cp\u003eMap of differences between the drought observed and predicted by the ANN model, lead times of 1 to 4 months, by municipalities in Brazil: January 2022\u003c/p\u003e","description":"","filename":"FIG10.jpg","url":"https://assets-eu.researchsquare.com/files/rs-4784321/v1/7a30fdfb7cfc0e9218f71a96.jpg"},{"id":63111340,"identity":"5892625a-6e03-458f-b44f-701a39b7985f","added_by":"auto","created_at":"2024-08-23 08:57:28","extension":"pdf","order_by":0,"title":"","display":"","copyAsset":false,"role":"manuscript-pdf","size":8221095,"visible":true,"origin":"","legend":"","description":"","filename":"manuscript.pdf","url":"https://assets-eu.researchsquare.com/files/rs-4784321/v1/da2be945-a967-4f54-8d9f-0e65dcfa6704.pdf"}],"financialInterests":"","formattedTitle":"Drought forecast model based on Artificial Neural Networks for Brazilian municipalities","fulltext":[{"header":"Introduction","content":"\u003cp\u003eThe spatial and temporal variability of precipitation has the potential to trigger extreme climatological processes, such as droughts. These events can cause serious economic losses and impacts on society, especially when they occur in regions where the population is more vulnerable to disasters (Blaikie et al. 2005; Hagenlocher et al. \u003cspan citationid=\"CR21\" class=\"CitationRef\"\u003e2019\u003c/span\u003e; Liu and Chen \u003cspan citationid=\"CR36\" class=\"CitationRef\"\u003e2021\u003c/span\u003e). In Brazil, droughts affect more than 40\u0026nbsp;million people in some years. In the 3-year period between 2020 and 2022, the accumulated loss was more than US\u003cspan\u003e$\u003c/span\u003e35\u0026nbsp;billion (SEDEC 2023).\u003c/p\u003e \u003cp\u003eThe frequency of droughts has intensified due to climate change and global warming (Kim and Jehanzaib \u003cspan citationid=\"CR34\" class=\"CitationRef\"\u003e2020\u003c/span\u003e). Recent data reveals that global surface temperatures have warmed by more than 1\u0026deg;C compared to the long-term average from 1951 to 1980 (NASA/GISS 2023). Although meteorological and climatic conditions are dynamic and extreme events are the results of adverse natural processes, the influence of human activities on global warming observed in recent decades is already considered unequivocal (IPCC \u003cspan citationid=\"CR30\" class=\"CitationRef\"\u003e2021\u003c/span\u003e). Changes in climate affect different natural systems and dynamics with severe environmental, social and economic impacts (Howells et al. \u003cspan citationid=\"CR25\" class=\"CitationRef\"\u003e2013\u003c/span\u003e; Shukla et al. 2019).\u003c/p\u003e \u003cp\u003eKnowledge about the spatial-temporal distribution of droughts and their trends is essential for effective risk management and the development of mitigation strategies (Mulualem and Liou \u003cspan citationid=\"CR43\" class=\"CitationRef\"\u003e2020\u003c/span\u003e). Developing a drought monitoring, early warning system or seasonal forecast can strengthen a country's resilience to losses caused by droughts (Tanguy et al. \u003cspan citationid=\"CR62\" class=\"CitationRef\"\u003e2023\u003c/span\u003e). Research and products related to drought diagnosis, analysis, forecasting and warning can mitigate the impacts caused by these extreme weather events (Panu and Sharma \u003cspan citationid=\"CR50\" class=\"CitationRef\"\u003e2002\u003c/span\u003e; Canedo Rosso et al. \u003cspan citationid=\"CR9\" class=\"CitationRef\"\u003e2018\u003c/span\u003e), as it allows the planning of different sectors of society, such as agriculture (Trnka et al. \u003cspan citationid=\"CR63\" class=\"CitationRef\"\u003e2020\u003c/span\u003e), energy (He et al. \u003cspan citationid=\"CR23\" class=\"CitationRef\"\u003e2019\u003c/span\u003e), public water supply (Stewart et al. \u003cspan citationid=\"CR58\" class=\"CitationRef\"\u003e2020\u003c/span\u003e) and banks and insurance companies (Huynh et al. \u003cspan citationid=\"CR27\" class=\"CitationRef\"\u003e2020\u003c/span\u003e).\u003c/p\u003e \u003cp\u003eThe decrease in accumulated precipitation is the main explanation for triggering droughts. Satellite-derived precipitation, including products that merge remote sensing data with measurements from rainfall stations, has the main advantage of extensive spatial coverage (Boluwade \u003cspan citationid=\"CR6\" class=\"CitationRef\"\u003e2020\u003c/span\u003e; Kofidou et al. \u003cspan citationid=\"CR35\" class=\"CitationRef\"\u003e2023\u003c/span\u003e). These datasets make it possible to analyze precipitation over large areas efficiently and without the need for interpolation. Climate disaster monitoring is one of the possible and currently expanding applications (Cheng et al. \u003cspan citationid=\"CR10\" class=\"CitationRef\"\u003e2021\u003c/span\u003e; Palharini et al. \u003cspan citationid=\"CR48\" class=\"CitationRef\"\u003e2022\u003c/span\u003e). Based on the temporal and spatial precipitation series, researchers can understand how previous accumulated precipitation relates to disasters, aiming to warn about extreme events. These studies are of interest to the scientific community and society, with some researchers dedicating themselves to this (Boult et al. \u003cspan citationid=\"CR7\" class=\"CitationRef\"\u003e2020\u003c/span\u003e).\u003c/p\u003e \u003cp\u003eAccording to Zhang et al. (\u003cspan citationid=\"CR70\" class=\"CitationRef\"\u003e2019\u003c/span\u003e) the inclusion of variables that refer to hydrometeorological dynamics such as rainfall, temperature, potential evapotranspiration, air pressure, wind speed and relative humidity, often results in improvements in drought prediction. Regarding the dependent variable of the forecast model, defining drought is an important step. Usually, researchers use one or more indexes to define the occurrence of drought (Kallis \u003cspan citationid=\"CR32\" class=\"CitationRef\"\u003e2008\u003c/span\u003e; Dikshit et al. \u003cspan citationid=\"CR13\" class=\"CitationRef\"\u003e2022\u003c/span\u003e), such as the Standard Precipitation Index - SPI (McKee et al. \u003cspan citationid=\"CR39\" class=\"CitationRef\"\u003e1993\u003c/span\u003e), Standard Precipitation Evaporation Index - SPEI (Vicente-Serrano et al. \u003cspan citationid=\"CR65\" class=\"CitationRef\"\u003e2010\u003c/span\u003e), and vegetation indices based on remote sensing such as the Normalized Difference Vegetation Index - NDVI or Enhanced Vegetation Index - EVI (Huete et al. \u003cspan citationid=\"CR26\" class=\"CitationRef\"\u003e2002\u003c/span\u003e). However, these indexes do not consider local economic and social dynamics, that is, a drought defined by these parameters does not always result in effective losses for the population.\u003c/p\u003e \u003cp\u003ePredicting droughts with the potential to cause losses, for example, in agricultural production, is a complex task as it depends on the location, the dry period, the seasonality of crops, land use management, irrigation techniques, among others (Musolino et al. \u003cspan citationid=\"CR44\" class=\"CitationRef\"\u003e2018\u003c/span\u003e). The losses are strongly linked to the spatial and temporal context. Therefore, an inventory of drought occurrences, with losses certified by authorities or government entities, is a powerful tool that could be used as a dependent variable in the forecast model. In Brazil, the Integrated Disaster Information System - S2iD (SEDEC 2023), is used for municipal authorities to report the occurrence of droughts that result in losses to various economic and social sectors. The processing of this data set and its use as a dependent variable is a distinguishing feature of this research, making it possible to train a model that predicts droughts with the potential to cause losses in each municipality.\u003c/p\u003e \u003cp\u003eDrought forecasting models represent a notable challenge due to the intricate interaction of diverse hydrometeorological factors, compounded by the effects of climate change. Among the various approaches, such as statistical, physical and data-based methods, machine learning methods are among the most common for predicting drought index (Belayneh and Adamowski \u003cspan citationid=\"CR3\" class=\"CitationRef\"\u003e2013\u003c/span\u003e; Sundararajan et al. \u003cspan citationid=\"CR59\" class=\"CitationRef\"\u003e2021\u003c/span\u003e). In this approach, artificial neural networks (ANN) often provide excellent results for modeling and predictions in the environmental area, using data obtained through remote sensing, meteorological and climate data as model input variables (Marj and Meijerink \u003cspan citationid=\"CR38\" class=\"CitationRef\"\u003e2011\u003c/span\u003e; Mokhtari and Akhoondzadeh \u003cspan citationid=\"CR42\" class=\"CitationRef\"\u003e2019\u003c/span\u003e). The spatial complexity, evidenced by the non-linear and multivariate property of droughts, highlights the ability of ANN-based models to quickly and effectively capture dynamic relationships, considerably boosting their application (Dikshit et al. \u003cspan citationid=\"CR13\" class=\"CitationRef\"\u003e2022\u003c/span\u003e).\u003c/p\u003e \u003cp\u003eSeveral studies use ANN for drought index forecasts (for examples: Mishra et al. 2006; Santos et al. 2014; Zhang et al. \u003cspan citationid=\"CR71\" class=\"CitationRef\"\u003e2017\u003c/span\u003e; Dikshit et al. \u003cspan citationid=\"CR11\" class=\"CitationRef\"\u003e2020\u003c/span\u003e; Mulualem and Liou \u003cspan citationid=\"CR43\" class=\"CitationRef\"\u003e2020\u003c/span\u003e). Mishra et al. (2006) forecasted SPI at multiple time scales for West Bengal, India, using three approaches (RBF, ANFIS and ANN). Their study revealed ANN produced the best result, with the key finding of using a direct multi-step approach to forecast at higher lead times, instead of a recursive multi-step approach. Santos et al. (2014) tested the ability of neural network approaches to hindcast the spring SPI on a 6-month time scale in Portugal, based on winter large-scale climatic indices. The authors indicate that the ANN showed good prediction capacity, although they identify limitations in relation to modeling extreme SPI values. Zhang et al. (\u003cspan citationid=\"CR71\" class=\"CitationRef\"\u003e2017\u003c/span\u003e) tested ARIMA, ANN and W-ANN\u0026rsquo;s applicability, using SPI as the drought index for Haihe River Basin, China. The results obtained by the authors indicated an advantage for the wavelet-based models at lead times of 3 and 6 months. Dikshit et al. (\u003cspan citationid=\"CR11\" class=\"CitationRef\"\u003e2020\u003c/span\u003e) present SPEI prediction results for a region of Australia, comparing ANN and Support Vector Regression (SVR) models. The drought index was calculated at various time scales (1, 3, 6 and 12 months) using a Climate Research Unit (CRU) dataset. The results indicate that ANN was better than SVR in predicting temporal drought trends. Mulualem and Liou (\u003cspan citationid=\"CR43\" class=\"CitationRef\"\u003e2020\u003c/span\u003e) developed ANN predictive models incorporating hydro-meteorological, climate, sea surface temperatures, and topographic attributes to SPEI forecast in the Upper Blue Nile basin. The authors showed that accurate results in predicting SPEI values could be achieved by including large-scale climate indexes.\u003c/p\u003e \u003cp\u003eIn recent years, seasonal forecast meteorological models have been developed. For example, Johnson et al. (\u003cspan citationid=\"CR31\" class=\"CitationRef\"\u003e2019\u003c/span\u003e) describe SEAS5, the fifth-generation seasonal forecast system of European Center for Medium-Range Weather Forecasts (ECMWF). In summary, the researchers indicate that SEAS5 is a state-of-the-art seasonal forecast system which continues to display a particular strength in the El Ni\u0026ntilde;o Southern Oscillation (ENSO) prediction. Ferreira and Reboita (\u003cspan citationid=\"CR16\" class=\"CitationRef\"\u003e2022\u003c/span\u003e) and Ferreira et al. (\u003cspan citationid=\"CR17\" class=\"CitationRef\"\u003e2022\u003c/span\u003e) evaluated the performance of ECMWF-SEAS5 in simulating South American precipitation. The authors conclude that ECMWF-SEAS5 has good performance in the climate characterization of South America. However, we did not identify any study that tested coupling an ANN model for drought prediction using a seasonal meteorological forecast dataset, such as ECMWF-SEAS5. This is another scientific gap that we intend to explore in this paper.\u003c/p\u003e \u003cp\u003eIn this context, and considering the need for advances in relation to forecasting and warning of climate disasters in Brazil, our objective was to develop a monthly ANN model with satellite-derived precipitation and seasonal weather forecast to predict droughts with the potential to cause losses in Brazilian municipalities.\u003c/p\u003e"},{"header":"Materials and Methods","content":"\u003cp\u003eIn this study, we used a dataset with coverage across Brazil, considering all 5,570 municipalities in the country as units of analysis and prediction. The time period used for training, cross-validation, and testing of the calibrated models was between 2013 and 2022, considering a monthly data interval. These spatial and temporal scales are associated with the modeled dependent variable, namely, the record in the database recognizing situations of emergency related to droughts in the Integrated Disaster Information System (S2iD) of the National Secretariat for Civil Protection and Defense (SEDEC 2023). In Fig.\u0026nbsp;\u003cspan refid=\"Fig1\" class=\"InternalRef\"\u003e1\u003c/span\u003e we illustrate the number of months in which each municipality in Brazil was affected by droughts that resulted in losses, including the water supply, agriculture, livestock and energy sectors. Although all regions present occurrences of droughts, the Northeast Region stands out with a significant number of municipalities being affected in more than 20% of the months of analysis.\u003c/p\u003e \u003cp\u003e \u003c/p\u003e \u003cp\u003eThe methodology of this study is based on the use of artificial neural networks for drought warning based on seasonal precipitation and temperature forecasts. This research presents advances in relation to the methodology presented in Oliveira et al. (\u003cspan citationid=\"CR47\" class=\"CitationRef\"\u003e2023\u003c/span\u003e), including new predictive variables, increased forecast lead time and expansion of the forecast to the entire territory of Brazil.\u003c/p\u003e \u003cp\u003eWe organized the study into six main steps (Fig.\u0026nbsp;\u003cspan refid=\"Fig2\" class=\"InternalRef\"\u003e2\u003c/span\u003e), containing: i) processing an inventory of droughts that caused losses; ii-iii) calculation of monthly precipitation and temperature indexes; iv) extraction of spatial descriptors for each municipality; v-vi) training, validation and testing of the ANN model for drought forecasting.\u003c/p\u003e \u003cp\u003eIn this research, we use the datasets from the list below:\u003c/p\u003e \u003cp\u003e \u003cul\u003e \u003cli\u003e \u003cp\u003eS2iD Database (SEDEC 2023);\u003c/p\u003e \u003c/li\u003e \u003cli\u003e \u003cp\u003eGeospatial base of the Brazilian Institute of Geography and Statistics, including vector files of the limits of Brazilian biomes (IBGE 2019), and the territorial grids of municipalities, microregions and regions of Brazil (IBGE 2023);\u003c/p\u003e \u003c/li\u003e \u003cli\u003e \u003cp\u003eClimate Hazards Group InfraRed Precipitation with Station Data (CHIRPS), University of California at Santa Barbara (UCSB 2023);\u003c/p\u003e \u003c/li\u003e \u003cli\u003e \u003cp\u003eNational Centers for Environmental Prediction (NCEP) database, NCEP/DOE Reanalysis II product (Kanamitsu et al. \u003cspan citationid=\"CR33\" class=\"CitationRef\"\u003e2002\u003c/span\u003e);\u003c/p\u003e \u003c/li\u003e \u003cli\u003e \u003cp\u003eECMWF-SEAS5 Seasonal Forecast (ECMWF 2023).\u003c/p\u003e \u003c/li\u003e \u003c/ul\u003e \u003c/p\u003e \u003cp\u003e \u003c/p\u003e \u003cp\u003eIn the first step, we compiled and processed data on drought occurrences from the S2iD Database, producing an inventory of drought occurrences. The database consists of the municipal code, date of the emergency protocol, updating the registration date as the drought prolongs, in addition to detailed information on losses in various economic sectors. The time series was discretized into 120 monthly time intervals, considering the period between January 2013 and December 2022.\u003c/p\u003e \u003cp\u003eAn algorithm was developed in MATLAB to organize the data in the form of a matrix, where each row corresponds to a municipality, and each column represents a time interval (month/year). In the matrix, each cell (municipality vs. time interval) was assigned the code 1 for the occurrence of drought. In the remaining cells, the code 0 was established (no occurrence).\u003c/p\u003e \u003cp\u003eAfter an exploratory analysis of the data in spatial and temporal contexts, and due to uncertainties in establishing the beginning and end of each drought event, we performed an analysis of consistency and transformation on the occurrence matrix, considering spatial-temporal dependence. The procedure was executed for each cell \u003cem\u003ex\u003c/em\u003e\u003csub\u003e\u003cem\u003em,t\u003c/em\u003e\u003c/sub\u003e, where \u003cem\u003em\u003c/em\u003e is a municipality and \u003cem\u003et\u003c/em\u003e is a time interval, and the following criteria and rules were subsequently applied (in sequential loops):\u003c/p\u003e \u003cp\u003e \u003col\u003e \u003cspan\u003e \u003cli\u003e \u003cp\u003eWhen an occurrence was identified in municipality \u003cem\u003em\u003c/em\u003e at time \u003cem\u003et\u003c/em\u003e (\u003cem\u003ex\u003c/em\u003e\u003csub\u003e\u003cem\u003em,t\u003c/em\u003e\u003c/sub\u003e = 1), the code 2 was assigned to the cells of the same municipality in the immediately preceding time interval (\u003cem\u003ex\u003c/em\u003e\u003csub\u003e\u003cem\u003em,t\u0026minus;1\u003c/em\u003e\u003c/sub\u003e = 2) and the subsequent time (\u003cem\u003ex\u003c/em\u003e\u003csub\u003e\u003cem\u003em,t+1\u003c/em\u003e\u003c/sub\u003e = 2). This code represents a possible temporal extension of the drought, and it can be interpreted as a doubtful record;\u003c/p\u003e \u003c/li\u003e \u003c/span\u003e \u003cspan\u003e \u003cli\u003e \u003cp\u003eWhen a doubtful occurrence was identified in municipality \u003cem\u003em\u003c/em\u003e at time \u003cem\u003et\u003c/em\u003e (\u003cem\u003ex\u003c/em\u003e\u003csub\u003e\u003cem\u003em,t\u003c/em\u003e\u003c/sub\u003e = 2), with more than 50% of the municipalities in its geographic microregion in a situation of confirmed occurrence in the same time interval, the record was changed to drought (\u003cem\u003ex\u003c/em\u003e\u003csub\u003e\u003cem\u003em,t\u003c/em\u003e\u003c/sub\u003e = 1);\u003c/p\u003e \u003c/li\u003e \u003c/span\u003e \u003cspan\u003e \u003cli\u003e \u003cp\u003eWhen a non-occurrence was identified in municipality \u003cem\u003em\u003c/em\u003e at time \u003cem\u003et\u003c/em\u003e (\u003cem\u003ex\u003c/em\u003e\u003csub\u003e\u003cem\u003em,t\u003c/em\u003e\u003c/sub\u003e = 0), but with a record of occurrence in the immediately preceding time interval (\u003cem\u003ex\u003c/em\u003e\u003csub\u003e\u003cem\u003em,t\u0026minus;1\u003c/em\u003e\u003c/sub\u003e = 1) and subsequent time (\u003cem\u003ex\u003c/em\u003e\u003csub\u003e\u003cem\u003em,t+1\u003c/em\u003e\u003c/sub\u003e = 1), the record was changed to drought (\u003cem\u003ex\u003c/em\u003e\u003csub\u003e\u003cem\u003em,t\u003c/em\u003e\u003c/sub\u003e = 1);\u003c/p\u003e \u003c/li\u003e \u003c/span\u003e \u003cspan\u003e \u003cli\u003e \u003cp\u003eWhen a record of non-occurrence was identified in municipality \u003cem\u003em\u003c/em\u003e at time \u003cem\u003et\u003c/em\u003e (\u003cem\u003ex\u003c/em\u003e\u003csub\u003e\u003cem\u003em,t\u003c/em\u003e\u003c/sub\u003e = 0), with more than 60% of the municipalities in its geographic microregion in a situation of confirmed occurrence between the time intervals \u003cem\u003et-1\u003c/em\u003e and \u003cem\u003et\u0026thinsp;+\u0026thinsp;1\u003c/em\u003e, the record was changed to drought (\u003cem\u003ex\u003c/em\u003e\u003csub\u003e\u003cem\u003em,t\u003c/em\u003e\u003c/sub\u003e = 1).\u003c/p\u003e \u003c/li\u003e \u003c/span\u003e \u003c/ol\u003e \u003c/p\u003e \u003cp\u003eThe sum of extreme drought events was calculated to establish the number of non-occurrence samples for training the ANN considering a 1:1 ratio, that is, with a balance between occurrence and non-occurrence samples chosen randomly.\u003c/p\u003e \u003cp\u003eThe antecedent accumulated precipitation is the most important hydrological process in defining drought, establishing a strong correlation with these extreme events. In this study, we selected the CHIRPS product, version 2.0, monthly precipitation with a spatial resolution of 0.05\u0026deg;. In this product, precipitation is estimated by integrating satellite rainfall data with measurements from meteorological stations (UCSB 2023).\u003c/p\u003e \u003cp\u003eWe developed a MATLAB script that: i) performs spatial clipping of monthly data for the Brazilian territory; ii) calculates the mean value of monthly accumulated precipitation per municipality; iii) exports the data in table format, where each row represents a municipality and each column represents a time interval.\u003c/p\u003e \u003cp\u003eThe climate normal of average monthly precipitation was calculated using a 30-year baseline period of data, from 1993 to 2022. This data was used to extract the municipal monthly precipitation anomaly time series, according to Eq.\u0026nbsp;1:\u003cdiv id=\"Equa\" class=\"Equation\"\u003e\u003cdiv format=\"TEX\" class=\"mathdisplay\" id=\"FileID_Equa\" name=\"EquationSource\"\u003e\n$$\\:{AP}_{m/a}={P}_{m/a}-{\\mu\\:}_{m}\\:\\left(1\\right)$$\u003c/div\u003e\u003c/div\u003e\u003c/p\u003e \u003cp\u003eWhere: \u003cem\u003eAP\u003c/em\u003e\u003csub\u003e\u003cem\u003em/a\u003c/em\u003e\u003c/sub\u003e is the precipitation anomaly in mm.month\u003csup\u003e\u0026minus;\u0026thinsp;1\u003c/sup\u003e, for month \u003cem\u003em\u003c/em\u003e, year \u003cem\u003ea\u003c/em\u003e; \u003cem\u003eP\u003c/em\u003e\u003csub\u003e\u003cem\u003em/a\u003c/em\u003e\u003c/sub\u003e is the accumulated precipitation in mm.month\u003csup\u003e\u0026minus;\u0026thinsp;1\u003c/sup\u003e, for month \u003cem\u003em\u003c/em\u003e, year \u003cem\u003ea\u003c/em\u003e; \u003cem\u003e\u0026micro;\u003c/em\u003e\u003csub\u003e\u003cem\u003em\u003c/em\u003e\u003c/sub\u003e is the average precipitation in mm.month\u003csup\u003e\u0026minus;\u0026thinsp;1\u003c/sup\u003e, for month \u003cem\u003em\u003c/em\u003e, baseline 1993\u0026ndash;2022.\u003c/p\u003e \u003cp\u003eIn addition to the precipitation anomaly, we calculated the SPI index (McKee et al. \u003cspan citationid=\"CR39\" class=\"CitationRef\"\u003e1993\u003c/span\u003e), widely used to characterize meteorological droughts on various time scales. The SPI is closely related to soil moisture on short time scales. On longer time scales, the SPI may be related to groundwater reserves and storage.\u003c/p\u003e \u003cp\u003eIn the calculation of the SPI, observed precipitation is quantified as a standardized deviation from a selected probability distribution to model the monthly accumulated precipitation time series. After exploratory data analysis, we fit the Gamma distribution to the monthly precipitation series of each municipality in MATLAB. Based on the calculated parameters of the distribution, for each precipitation value in the municipal time series, the cumulative distribution function (CDF) is fitted describing the probability of that value not being exceeded according to the Gamma distribution.\u003c/p\u003e \u003cp\u003eThe SPI is obtained for each month and each municipality by transforming the probability value obtained in the previous step using the inverse function of the Normal distribution, with a mean of 0 and a standard deviation of 1. SPI values can be interpreted as the number of standard deviations by which the observed anomaly deviates from the long-term average. It is expected that 95% of the time, the SPI will remain between \u0026minus;\u0026thinsp;2 and +\u0026thinsp;2.\u003c/p\u003e \u003cp\u003eWith the aim of incorporating larger time scales to represent not only soil moisture but also groundwater storage and reserves, the same precipitation indexes (accumulated precipitation, precipitation anomaly, and SPI) were calculated for accumulations of 3, 6, and 12 months, resulting in 12 explanatory and dynamic variables (Table\u0026nbsp;\u003cspan refid=\"Tab1\" class=\"InternalRef\"\u003e1\u003c/span\u003e). For example, in Fig.\u0026nbsp;\u003cspan refid=\"Fig3\" class=\"InternalRef\"\u003e3\u003c/span\u003e we illustrate the combination of precipitation variables for the month of January 2022.\u003c/p\u003e \u003cp\u003e \u003cdiv class=\"gridtable\"\u003e\u003ctable float=\"Yes\" id=\"Tab1\" border=\"1\"\u003e \u003ccaption language=\"En\"\u003e \u003cdiv class=\"CaptionNumber\"\u003eTable 1\u003c/div\u003e \u003cdiv class=\"CaptionContent\"\u003e \u003cp\u003ePrecipitation variables used in the modeling process\u003c/p\u003e \u003c/div\u003e \u003c/caption\u003e \u003ccolgroup cols=\"3\"\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c1\" colnum=\"1\"\u003e\u003c/div\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c2\" colnum=\"2\"\u003e\u003c/div\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c3\" colnum=\"3\"\u003e\u003c/div\u003e \u003cthead\u003e \u003ctr\u003e \u003cth align=\"left\" colname=\"c1\"\u003e \u003cp\u003eTypology\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c2\"\u003e \u003cp\u003eVariable Acronym\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c3\"\u003e \u003cp\u003eTime Intervals\u003c/p\u003e \u003c/th\u003e \u003c/tr\u003e \u003c/thead\u003e \u003ctbody\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\" morerows=\"3\" rowspan=\"4\"\u003e \u003cp\u003eAccumulated Precipitation Indexes\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eP-01\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003eP (\u003cem\u003et\u003c/em\u003e)\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eP-03\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003eP (\u003cem\u003et\u003c/em\u003e) + [...]\u0026thinsp;+\u0026thinsp;P (\u003cem\u003et\u003c/em\u003e-2)\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eP-06\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003eP (\u003cem\u003et\u003c/em\u003e) + [...]\u0026thinsp;+\u0026thinsp;P (\u003cem\u003et\u003c/em\u003e-5)\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eP-12\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003eP (\u003cem\u003et\u003c/em\u003e) + [...]\u0026thinsp;+\u0026thinsp;P (\u003cem\u003et\u003c/em\u003e-11)\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\" morerows=\"3\" rowspan=\"4\"\u003e \u003cp\u003eAccumulated Precipitation Anomaly Indexes\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003ePA-01\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003ePA (\u003cem\u003et\u003c/em\u003e)\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003ePA-03\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003ePA (\u003cem\u003et\u003c/em\u003e) + [...]\u0026thinsp;+\u0026thinsp;PA (\u003cem\u003et\u003c/em\u003e-2)\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003ePA-06\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003ePA (\u003cem\u003et\u003c/em\u003e) + [...]\u0026thinsp;+\u0026thinsp;PA (\u003cem\u003et\u003c/em\u003e-5)\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003ePA-12\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003ePA (\u003cem\u003et\u003c/em\u003e) + [...]\u0026thinsp;+\u0026thinsp;PA (\u003cem\u003et\u003c/em\u003e-11)\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\" morerows=\"3\" rowspan=\"4\"\u003e \u003cp\u003eStandardized Precipitation Index (SPI)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eSPI-01\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003eSPI (\u003cem\u003et\u003c/em\u003e)\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eSPI-03\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003eSPI (\u003cem\u003et\u003c/em\u003e to \u003cem\u003et\u003c/em\u003e-2)\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eSPI-06\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003eSPI (\u003cem\u003et\u003c/em\u003e to \u003cem\u003et\u003c/em\u003e-5)\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eSPI-12\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003eSPI (\u003cem\u003et\u003c/em\u003e to \u003cem\u003et\u003c/em\u003e-11)\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003c/tbody\u003e \u003c/colgroup\u003e \u003c/table\u003e\u003c/div\u003e \u003c/p\u003e \u003cp\u003e \u003c/p\u003e \u003cp\u003eTemperature was used as a predictive variable in the neural network model as it directly influences the evapotranspiration rate. Seasonal forecasts are usually more accurate for the temperature variable, given that the estimation of evapotranspiration results from a complex interaction of meteorological factors such as wind, relative humidity, atmospheric pressure, among others. Since the goal is to adjust a model to have the ability to drought seasonal forecasts, using temperature should be advantageous.\u003c/p\u003e \u003cp\u003eWe used the NCEP/DOE Reanalysis II product (Kanamitsu et al. \u003cspan citationid=\"CR33\" class=\"CitationRef\"\u003e2002\u003c/span\u003e) to generate the monthly temperature time series in municipalities. The monthly temperature data in raster format were processed in MATLAB using a script developed to: i) perform spatial clipping of monthly data for the Brazilian territory; ii) calculate the mean monthly temperature value per municipality; iii) export the data in table format. We calculated four temperature variables using lag times of 1, 3, 6 and 12 months (Table\u0026nbsp;\u003cspan refid=\"Tab2\" class=\"InternalRef\"\u003e2\u003c/span\u003e).\u003c/p\u003e \u003cp\u003eWe defined descriptive and spatial attributes that contribute to the modeling process for each Brazilian municipality, allowing the ANN to seek a customized threshold for precipitation and temperature needed to trigger drought in each territorial unit. In total, we established six static variables to describe each municipality, in addition to two dynamic variables related to the temporal/seasonal context (Table\u0026nbsp;\u003cspan refid=\"Tab3\" class=\"InternalRef\"\u003e3\u003c/span\u003e).\u003c/p\u003e \u003cp\u003e \u003cdiv class=\"gridtable\"\u003e\u003ctable float=\"Yes\" id=\"Tab2\" border=\"1\"\u003e \u003ccaption language=\"En\"\u003e \u003cdiv class=\"CaptionNumber\"\u003eTable 2\u003c/div\u003e \u003cdiv class=\"CaptionContent\"\u003e \u003cp\u003eTemperature variables used in the modeling process\u003c/p\u003e \u003c/div\u003e \u003c/caption\u003e \u003ccolgroup cols=\"2\"\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c1\" colnum=\"1\"\u003e\u003c/div\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c2\" colnum=\"2\"\u003e\u003c/div\u003e \u003cthead\u003e \u003ctr\u003e \u003cth align=\"left\" colname=\"c1\"\u003e \u003cp\u003eVariable\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c2\"\u003e \u003cp\u003eTime Intervals\u003c/p\u003e \u003c/th\u003e \u003c/tr\u003e \u003c/thead\u003e \u003ctbody\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eT-01\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eT(\u003cem\u003et\u003c/em\u003e)\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eT-03\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e[T(\u003cem\u003et\u003c/em\u003e) + [...]\u0026thinsp;+\u0026thinsp;T(\u003cem\u003et\u003c/em\u003e-2)] / 3\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eT-06\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e[T(\u003cem\u003et\u003c/em\u003e) + [...]\u0026thinsp;+\u0026thinsp;T(\u003cem\u003et\u003c/em\u003e-5)] / 6\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eT-12\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e[T(\u003cem\u003et\u003c/em\u003e) + [...]\u0026thinsp;+\u0026thinsp;T(\u003cem\u003et\u003c/em\u003e-11)] / 12\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003c/tbody\u003e \u003c/colgroup\u003e \u003c/table\u003e\u003c/div\u003e \u003c/p\u003e \u003cp\u003e \u003cdiv class=\"gridtable\"\u003e\u003ctable float=\"Yes\" id=\"Tab3\" border=\"1\"\u003e \u003ccaption language=\"En\"\u003e \u003cdiv class=\"CaptionNumber\"\u003eTable 3\u003c/div\u003e \u003cdiv class=\"CaptionContent\"\u003e \u003cp\u003eDescriptive, spatial and dynamic temporal variables by municipality\u003c/p\u003e \u003c/div\u003e \u003c/caption\u003e \u003ccolgroup cols=\"2\"\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c1\" colnum=\"1\"\u003e\u003c/div\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c2\" colnum=\"2\"\u003e\u003c/div\u003e \u003cthead\u003e \u003ctr\u003e \u003cth align=\"left\" colname=\"c1\"\u003e \u003cp\u003eDescriptive and spatial variables\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c2\"\u003e \u003cp\u003eDynamic temporal variables\u003c/p\u003e \u003c/th\u003e \u003c/tr\u003e \u003c/thead\u003e \u003ctbody\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eBiome identifier (BIO)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e\u0026nbsp;\u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eRegion identifier (REG)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e\u0026nbsp;\u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eFrequency of occurrences (FRO)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eMonth of the year (MON)\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eLatitude (LAT)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eStatus - previous situation (STA)\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eDistance from oceans (DOC)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e\u0026nbsp;\u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eDistance from the Amazon (DAM)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e\u0026nbsp;\u003c/td\u003e \u003c/tr\u003e \u003c/tbody\u003e \u003c/colgroup\u003e \u003c/table\u003e\u003c/div\u003e \u003c/p\u003e \u003cp\u003eThe Brazilian biomes (BIO) (IBGE 2019) encompass types of vegetation that share similar geographical and climatic conditions, and they can aid in the process of regionalizing precipitation and temperature thresholds in the neural network to identify occurrences of droughts. The regions (REG) (IBGE 2023) incorporate, in addition to the aspect of location in the Brazilian territory, the political-administrative dimension inherent to the emergency decrees of municipalities and federative units. According to the biome and region of the country, we expect the ANN model to identify the appropriate thresholds for precipitation indexes to trigger droughts.\u003c/p\u003e \u003cp\u003eThe frequency of occurrences (FRO) in each municipality rescues the historical dimension of this type of climatic disaster and its impacts on society. Latitude (LAT) can assist in the modeling process, as this component directly influences global climate patterns and seasonality. The distance from oceans (DOC) and the Amazon (DAM) incorporates the importance of the proximity of areas that contribute to the distribution of moisture to continental areas, directly influencing meteorological and climatic conditions. Monthly identification (MON) allows the neural network to seek customized thresholds according to the time of the year, incorporating the seasonal dimension, while the previous situation (STA) reveals the initial condition of the municipality regarding droughts.\u003c/p\u003e \u003cp\u003eWe trained an ANN model for drought forecast and warning, specifically a Multi-Layered Perceptron (MLP) with a three-layer architecture, implemented through code developed in the MATLAB script editor. The assumption of a single intermediate layer is based on the existence theorem (Hecht-Nielsen 1990), which demonstrates that an ANN with a single hidden layer approximates any continuous numerical relationship. Training was conducted using the backpropagation algorithm (Rumelhart et al. \u003cspan citationid=\"CR51\" class=\"CitationRef\"\u003e1986\u003c/span\u003e) with the Delta Rule for updating synaptic weights (Widrow and Hoff 1960). We chose the unipolar sigmoid activation function for all neurons.\u003c/p\u003e \u003cp\u003eSeveral ANN models were developed to predict drought from the reported input variables. We tested 27 different combinations of precipitation and temperature predictor variables (Fig.\u0026nbsp;\u003cspan refid=\"Fig4\" class=\"InternalRef\"\u003e4\u003c/span\u003e). The objective of the tests was to verify how the indexes and lag-time windows affect the model's performance.\u003c/p\u003e \u003cp\u003eWe prepare a set of 129,609 samples, temporally and spatially distributed. The samples were divided: i) training (60%); ii) cross-validation (20%); iii) testing (20%). The training set is used to calibrate synaptic weights. The cross-validation set is used in parallel with training to prevent overfitting of the ANN (Hecht-Nielsen 1990). The testing set is used to assess the model's performance statistics.\u003c/p\u003e \u003cp\u003eFrom the 10 years of data (2013\u0026ndash;2022), we define the period from January 2021 to December 2022 as the test set (24 months). The training of the ANN models was performed using a k-fold cross-validation approach (Shao et al. \u003cspan citationid=\"CR56\" class=\"CitationRef\"\u003e2013\u003c/span\u003e; Jiang and Chen 2016). In this study, we set k\u0026thinsp;=\u0026thinsp;4, that is, the remaining dataset (from January 2013 to December 2020) was divided into four subsets (or folds), each with 24 months. In this manner, four ANN models were developed using different subsets of data samples. In each model round, three subsets were used for training and one for cross-validation. In the end, the ensemble result was considered, with the calculation of the average of the four adjusted ANN\u0026rsquo;s.\u003c/p\u003e \u003cp\u003e \u003c/p\u003e \u003cp\u003eEach round of modeling was performed with twenty repetitions using random initiations of the synaptic weights. This procedure was performed to reduce the effect of random initialization on the results. Additionally, we also tested the ideal number of neurons in the intermediate layer and the number of iterations required in each repetition of the model for the neural network to converge.\u003c/p\u003e \u003cp\u003eThe performance of the ANN was measured using four metrics: i) AUC index, area under the Receiver Operator Characteristic curve (Hanley \u003cspan citationid=\"CR22\" class=\"CitationRef\"\u003e1989\u003c/span\u003e; Fawcett \u003cspan citationid=\"CR15\" class=\"CitationRef\"\u003e2006\u003c/span\u003e); ii) overall accuracy (OA), where the number of correctly classified samples is divided by the total number of samples, obtained from the confusion matrix, which discriminates the output of the ANN into two classes (occurrence or non-occurrence of droughts); precision or positive predictive value, which indicates the fraction of positive predictions that were correct, obtained by dividing the number of True Positive (TP) samples by Predicted Positive (PP); recall or sensitivity, which indicates the proportion of observed droughts that were predicted, obtained by dividing TP by the total number of observed droughts.\u003c/p\u003e \u003cp\u003eIn the last stage of the study, we tested the best ANN model for droughts forecast. We progressively replaced the precipitation (CHIRPS dataset) and temperature (NCEP dataset) with the ECMWF-SEAS5 Seasonal Forecast (ECMWF 2023), using for lead time prediction of 1, 2, 3, and 4 months. The STA variable (occurrence or non-occurrence of drought in the previous month), when necessary, was replaced by the ANN's forecast for the immediately previous lead time (Fig.\u0026nbsp;\u003cspan refid=\"Fig5\" class=\"InternalRef\"\u003e5\u003c/span\u003e). We tested the performance of the forecast for the period between January 2021 and December 2022, considering all prediction lead times.\u003c/p\u003e \u003cp\u003e \u003c/p\u003e"},{"header":"Results","content":"\u003cp\u003eThe exploratory analysis of the sample data showed us relevant results on the explanatory variables of precipitation and temperature. In Fig.\u0026nbsp;\u003cspan\u003e6\u003c/span\u003e, we show the average values of the accumulated precipitation (P), precipitation anomaly (PA), Standardized Precipitation Index (SPI) and temperature (T), in drought and non-drought conditions. We extracted the values for lag-times of 1, 3, 6 and 12 months per biome in Brazilian territory.\u003c/p\u003e\n\u003cp\u003eThe temperature tends to be higher in drought conditions, except in the Caatinga biome, which has a semi-arid tropical climate. In this region, the temperature has low seasonal variation and droughts occur almost every year, depending in particular on precipitation dynamics. Opposite to this, in the Pampa biome (extreme south of Brazil), the temperature presents much higher values during droughts, in particular, the variable T-01, 3.8 K of average difference in drought and non-drought situations. We found that the variables T-01 and T-03 are the most explanatory to define droughts. In Brazil, the temperature is 2 K higher during droughts considering the average of the variables T-01 and T-03.\u003c/p\u003e\n\u003cp\u003eIn Brazil, when droughts occur, precipitation decreases by 49.5% on average, ranging from \u0026minus;\u0026thinsp;53.9% (P-01) to -45.7% (P-12). Proportionally, variables P-01, P-03 and P-06 are the most explanatory. Once again, in the Caatinga biome we observed the smallest difference in drought and non-drought conditions, with an average decrease of 22.5%. The greatest relative differences in precipitation (P) were observed in the Cerrado and Pantanal biomes. In absolute values, we also highlight the Amazon, with an average decrease of 76 mm (P-01), 265 mm (P-03) and 389 mm (P-06). The biggest difference was observed in the Cerrado, with a decrease of 491 mm in the P-12 variable in droughts (Fig.\u0026nbsp;\u003cspan\u003e6\u003c/span\u003e).\u003c/p\u003e\n\u003cp\u003eWe observed that precipitation anomaly (PA) indexes are strongly related to drought and non-drought conditions in the Pampa biome. For example, the PA-03 variable has an average value of -181 mm in droughts, while in other periods it is +\u0026thinsp;30 mm. The largest precipitation anomaly was observed in the Amazon, with an average of -332 mm (PA-12). In the Caatinga, the precipitation anomaly presents lower values. For example, the average PA-03 in droughts is only \u0026minus;\u0026thinsp;24 mm.\u003c/p\u003e\n\u003cp\u003eRegarding the SPI index, we observed that it is more explanatory for the droughts that occurred in the Pampa, Pantanal and Amazon biomes. For example, the SPI-03 variable in Pampa presented an average value of -1.3 in droughts. In the Caatinga, the SPI shows little change in drought and non-drought conditions, except for the SPI-12 variable, with an average of -0.42 in droughts.\u003c/p\u003e\n\u003cp\u003eIn general, we observe that precipitation indexes calculated from the CHIRPS product show a significant correlation with the occurrence of drought in Brazil. The P-12 variable was the one with the strongest linear correlation, with a correlation coefficient (r) of -0.592, indicating a significant dependence between the occurrence of drought and the accumulated precipitation in 12-months interval. The weakest linear correlation was observed for the SPI, with values between \u0026minus;\u0026thinsp;0.153 (SPI-01) and \u0026minus;\u0026thinsp;0.271 (SPI-12). Temperature variables obtained from the NCEP database showed linear correlations between 0.306 (T-06) and 0.33 (T-01), values that are also statistically significant.\u003c/p\u003e\n\u003cp\u003eIn Table\u0026nbsp;\u003cspan\u003e4\u003c/span\u003e we present the performance metrics of ANN models for drought simulation in Brazilian municipalities. ANN-1 was the best model, with OA\u0026thinsp;=\u0026thinsp;0.931, AUC\u0026thinsp;=\u0026thinsp;0.952 and precision of 0.805. This model has 24 input variables, 47 neurons in the hidden layer and one output neuron (ANN 24-47-1). Regarding recall, we highlight the ANN-13 model, which obtained a value of 0.79 in this performance metric. The worst performing model was the ANN-20, in which we excluded all P, PA and SPI variables. This model obtained OA\u0026thinsp;=\u0026thinsp;0.764, AUC\u0026thinsp;=\u0026thinsp;0.84, precision of 0.383 and recall of 0.639.\u003c/p\u003e\n\u003cp\u003eIn general, the ANN models that exclude the precipitation anomaly (PA) variables present the lowest values of OA (average\u0026thinsp;=\u0026thinsp;0.858), AUC (0.903), precision (0.576) and recall (0.72). Opposite to this, the exclusion of temperature variables (T) from ANN models slightly weakens the performance for drought simulation, with average values of OA\u0026thinsp;=\u0026thinsp;0.899, AUC\u0026thinsp;=\u0026thinsp;0.931, precision of 0.689 and recall of 0.756. Regarding lag times, we observed that the ANN models without the variables at lag times of 3 and 6 months show slightly lower performance.\u003c/p\u003e\n\u003cp\u003eThese results indicate that the trained ANN\u0026apos;s (especially ANN-1) was able to simulate the droughts that cause damage in Brazilian municipalities, using various explanatory variables such as precipitation indexes (accumulated precipitation, precipitation anomaly and SPI) extracted from the CHIRPS product and temperature from the NCEP product, considering lag times from 1 to 12 months.\u003c/p\u003e\n\u003cp\u003eHowever, only by subjecting the ANN to a situation where rainfall/temperature was not observed in the prediction time interval can we verify if the model has the ability to forecast drought. For this purpose, we tested the combined use of CHIRPS and NCEP products, along with the precipitation and temperature forecast from the ECMWF-SEAS5 Seasonal Forecast, for prediction lead times from 1 to 4 months. We carried out forecast tests for the time interval between July 2021 and June 2022, due to the numerous occurrences of drought, mainly in the Southern Region of Brazil.\u003c/p\u003e\n\u003cdiv\u003e\n \u003cdiv\u003e\u003cstrong\u003eTable 4\u003c/strong\u003e Performance metrics of ANN models for drought simulation in Brazilian municipalities\u003c/div\u003e\n \u003cdiv style='margin-top:0in;margin-right:0in;margin-bottom:8.0pt;margin-left:0in;font-size:11.0pt;font-family:\"Calibri\",sans-serif;'\u003e\n \u003ctable style=\"width: 3.5e+2pt;border-collapse:collapse;border:none;\"\u003e\n \u003ctbody\u003e\n \u003ctr\u003e\n \u003ctd rowspan=\"2\" style=\"width:45.75pt;border-top:solid black 1.0pt;border-left:none;border-bottom:solid black 1.0pt;border-right:none;padding:2.15pt 2.15pt 2.15pt 2.15pt;height:10.35pt;\"\u003e\n \u003cp style='margin-top:0in;margin-right:0in;margin-bottom:0in;margin-left:0in;font-size:11.0pt;font-family:\"Calibri\",sans-serif;text-align:right;line-height:normal;'\u003e\u003cspan style='font-size:12px;font-family:\"Times New Roman\",serif;'\u003eANN Model\u003c/span\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd rowspan=\"2\" style=\"width:79.5pt;border-top:solid black 1.0pt;border-left:none;border-bottom:solid black 1.0pt;border-right:none;padding:2.15pt 2.15pt 2.15pt 2.15pt;height:10.35pt;\"\u003e\n \u003cp style='margin-top:0in;margin-right:0in;margin-bottom:0in;margin-left:0in;font-size:11.0pt;font-family:\"Calibri\",sans-serif;text-align:right;line-height:normal;'\u003e\u003cspan style='font-size:12px;font-family:\"Times New Roman\",serif;'\u003eExcluded Variable\u003c/span\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd rowspan=\"2\" style=\"width:54.95pt;border-top:solid black 1.0pt;border-left:none;border-bottom:solid black 1.0pt;border-right:none;padding:2.15pt 2.15pt 2.15pt 2.15pt;height:10.35pt;\"\u003e\n \u003cp style='margin-top:0in;margin-right:0in;margin-bottom:0in;margin-left:0in;font-size:11.0pt;font-family:\"Calibri\",sans-serif;text-align:right;line-height:normal;'\u003e\u003cspan style='font-size:12px;font-family:\"Times New Roman\",serif;'\u003eOverall Accuracy (OA)\u003c/span\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd rowspan=\"2\" style=\"width:54.95pt;border-top:solid black 1.0pt;border-left:none;border-bottom:solid black 1.0pt;border-right:none;padding:2.15pt 2.15pt 2.15pt 2.15pt;height:10.35pt;\"\u003e\n \u003cp style='margin-top:0in;margin-right:0in;margin-bottom:0in;margin-left:0in;font-size:11.0pt;font-family:\"Calibri\",sans-serif;text-align:right;line-height:normal;'\u003e\u003cspan style='font-size:12px;font-family:\"Times New Roman\",serif;'\u003eArea Under the ROC Curve (AUC)\u003c/span\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd rowspan=\"2\" style=\"width:54.95pt;border-top:solid black 1.0pt;border-left:none;border-bottom:solid black 1.0pt;border-right:none;padding:2.15pt 2.15pt 2.15pt 2.15pt;height:10.35pt;\"\u003e\n \u003cp style='margin-top:0in;margin-right:0in;margin-bottom:0in;margin-left:0in;font-size:11.0pt;font-family:\"Calibri\",sans-serif;text-align:right;line-height:normal;'\u003e\u003cspan style='font-size:12px;font-family:\"Times New Roman\",serif;'\u003ePrecision\u003c/span\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd rowspan=\"2\" style=\"width:54.95pt;border-top:solid black 1.0pt;border-left:none;border-bottom:solid black 1.0pt;border-right:none;padding:2.15pt 2.15pt 2.15pt 2.15pt;height:10.35pt;\"\u003e\n \u003cp style='margin-top:0in;margin-right:0in;margin-bottom:0in;margin-left:0in;font-size:11.0pt;font-family:\"Calibri\",sans-serif;text-align:right;line-height:normal;'\u003e\u003cspan style='font-size:12px;font-family:\"Times New Roman\",serif;'\u003eRecall\u003c/span\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"height:10.35pt;border:none;\"\u003e\u003cbr\u003e\u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd style=\"height:25.95pt;border:none;\"\u003e\u003cbr\u003e\u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd style=\"width:45.75pt;border:none;padding:2.15pt 2.15pt 2.15pt 2.15pt;height:3.3pt;\"\u003e\n \u003cp style='margin-top:0in;margin-right:0in;margin-bottom:0in;margin-left:0in;font-size:11.0pt;font-family:\"Calibri\",sans-serif;text-align:right;line-height:normal;'\u003e\u003cspan style='font-size:12px;font-family:\"Times New Roman\",serif;'\u003e1\u003c/span\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width:79.5pt;border:none;padding:2.15pt 2.15pt 2.15pt 2.15pt;height:3.3pt;\"\u003e\n \u003cp style='margin-top:0in;margin-right:0in;margin-bottom:0in;margin-left:0in;font-size:11.0pt;font-family:\"Calibri\",sans-serif;text-align:right;line-height:normal;'\u003e\u003cspan style='font-size:12px;font-family:\"Times New Roman\",serif;'\u003e-\u003c/span\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width:54.95pt;border:none;background:#5A8AC6;padding:2.15pt 2.15pt 2.15pt 2.15pt;height:3.3pt;\"\u003e\n \u003cp style='margin-top:0in;margin-right:0in;margin-bottom:0in;margin-left:0in;font-size:11.0pt;font-family:\"Calibri\",sans-serif;text-align:right;line-height:normal;'\u003e\u003cspan style='font-size:12px;font-family:\"Times New Roman\",serif;color:black;'\u003e0.931\u003c/span\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width:54.95pt;border:none;background:#5A8AC6;padding:2.15pt 2.15pt 2.15pt 2.15pt;height:3.3pt;\"\u003e\n \u003cp style='margin-top:0in;margin-right:0in;margin-bottom:0in;margin-left:0in;font-size:11.0pt;font-family:\"Calibri\",sans-serif;text-align:right;line-height:normal;'\u003e\u003cspan style='font-size:12px;font-family:\"Times New Roman\",serif;color:black;'\u003e0.952\u003c/span\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width:54.95pt;border:none;background:#5A8AC6;padding:2.15pt 2.15pt 2.15pt 2.15pt;height:3.3pt;\"\u003e\n \u003cp style='margin-top:0in;margin-right:0in;margin-bottom:0in;margin-left:0in;font-size:11.0pt;font-family:\"Calibri\",sans-serif;text-align:right;line-height:normal;'\u003e\u003cspan style='font-size:12px;font-family:\"Times New Roman\",serif;color:black;'\u003e0.805\u003c/span\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width:54.95pt;border:none;background:#719ACE;padding:2.15pt 2.15pt 2.15pt 2.15pt;height:3.3pt;\"\u003e\n \u003cp style='margin-top:0in;margin-right:0in;margin-bottom:0in;margin-left:0in;font-size:11.0pt;font-family:\"Calibri\",sans-serif;text-align:right;line-height:normal;'\u003e\u003cspan style='font-size:12px;font-family:\"Times New Roman\",serif;color:black;'\u003e0.784\u003c/span\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"height:3.3pt;border:none;\"\u003e\u003cbr\u003e\u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd style=\"width:45.75pt;border:none;padding:2.15pt 2.15pt 2.15pt 2.15pt;height:3.3pt;\"\u003e\n \u003cp style='margin-top:0in;margin-right:0in;margin-bottom:0in;margin-left:0in;font-size:11.0pt;font-family:\"Calibri\",sans-serif;text-align:right;line-height:normal;'\u003e\u003cspan style='font-size:12px;font-family:\"Times New Roman\",serif;'\u003e2\u003c/span\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width:79.5pt;border:none;padding:2.15pt 2.15pt 2.15pt 2.15pt;height:3.3pt;\"\u003e\n \u003cp style='margin-top:0in;margin-right:0in;margin-bottom:0in;margin-left:0in;font-size:11.0pt;font-family:\"Calibri\",sans-serif;text-align:right;line-height:normal;'\u003e\u003cspan style='font-size:12px;font-family:\"Times New Roman\",serif;'\u003eSPI\u003c/span\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width:54.95pt;border:none;background:#7099CE;padding: 2.15pt 2.15pt 2.15pt 2.15pt;height:3.3pt;\"\u003e\n \u003cp style='margin-top:0in;margin-right:0in;margin-bottom:0in;margin-left:0in;font-size:11.0pt;font-family:\"Calibri\",sans-serif;text-align:right;line-height:normal;'\u003e\u003cspan style='font-size:12px;font-family:\"Times New Roman\",serif;color:black;'\u003e0.925\u003c/span\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width:54.95pt;border:none;background:#729BCF;padding: 2.15pt 2.15pt 2.15pt 2.15pt;height:3.3pt;\"\u003e\n \u003cp style='margin-top:0in;margin-right:0in;margin-bottom:0in;margin-left:0in;font-size:11.0pt;font-family:\"Calibri\",sans-serif;text-align:right;line-height:normal;'\u003e\u003cspan style='font-size:12px;font-family:\"Times New Roman\",serif;color:black;'\u003e0.947\u003c/span\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width:54.95pt;border:none;background:#6D98CD;padding: 2.15pt 2.15pt 2.15pt 2.15pt;height:3.3pt;\"\u003e\n \u003cp style='margin-top:0in;margin-right:0in;margin-bottom:0in;margin-left:0in;font-size:11.0pt;font-family:\"Calibri\",sans-serif;text-align:right;line-height:normal;'\u003e\u003cspan style='font-size:12px;font-family:\"Times New Roman\",serif;color:black;'\u003e0.786\u003c/span\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width:54.95pt;border:none;background:#B1C7E5;padding: 2.15pt 2.15pt 2.15pt 2.15pt;height:3.3pt;\"\u003e\n \u003cp style='margin-top:0in;margin-right:0in;margin-bottom:0in;margin-left:0in;font-size:11.0pt;font-family:\"Calibri\",sans-serif;text-align:right;line-height:normal;'\u003e\u003cspan style='font-size:12px;font-family:\"Times New Roman\",serif;color:black;'\u003e0.766\u003c/span\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"height:3.3pt;border:none;\"\u003e\u003cbr\u003e\u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd style=\"width:45.75pt;border:none;padding:2.15pt 2.15pt 2.15pt 2.15pt;height:3.3pt;\"\u003e\n \u003cp style='margin-top:0in;margin-right:0in;margin-bottom:0in;margin-left:0in;font-size:11.0pt;font-family:\"Calibri\",sans-serif;text-align:right;line-height:normal;'\u003e\u003cspan style='font-size:12px;font-family:\"Times New Roman\",serif;'\u003e3\u003c/span\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width:79.5pt;border:none;padding:2.15pt 2.15pt 2.15pt 2.15pt;height:3.3pt;\"\u003e\n \u003cp style='margin-top:0in;margin-right:0in;margin-bottom:0in;margin-left:0in;font-size:11.0pt;font-family:\"Calibri\",sans-serif;text-align:right;line-height:normal;'\u003e\u003cspan style='font-size:12px;font-family:\"Times New Roman\",serif;'\u003ePA\u003c/span\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width:54.95pt;border:none;background:#CBDAEE;padding: 2.15pt 2.15pt 2.15pt 2.15pt;height:3.3pt;\"\u003e\n \u003cp style='margin-top:0in;margin-right:0in;margin-bottom:0in;margin-left:0in;font-size:11.0pt;font-family:\"Calibri\",sans-serif;text-align:right;line-height:normal;'\u003e\u003cspan style='font-size:12px;font-family:\"Times New Roman\",serif;color:black;'\u003e0.899\u003c/span\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width:54.95pt;border:none;background:#CCDBEF;padding: 2.15pt 2.15pt 2.15pt 2.15pt;height:3.3pt;\"\u003e\n \u003cp style='margin-top:0in;margin-right:0in;margin-bottom:0in;margin-left:0in;font-size:11.0pt;font-family:\"Calibri\",sans-serif;text-align:right;line-height:normal;'\u003e\u003cspan style='font-size:12px;font-family:\"Times New Roman\",serif;color:black;'\u003e0.929\u003c/span\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width:54.95pt;border:none;background:#D4E0F1;padding: 2.15pt 2.15pt 2.15pt 2.15pt;height:3.3pt;\"\u003e\n \u003cp style='margin-top:0in;margin-right:0in;margin-bottom:0in;margin-left:0in;font-size:11.0pt;font-family:\"Calibri\",sans-serif;text-align:right;line-height:normal;'\u003e\u003cspan style='font-size:12px;font-family:\"Times New Roman\",serif;color:black;'\u003e0.681\u003c/span\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width:54.95pt;border:none;background:#C0D2EA;padding: 2.15pt 2.15pt 2.15pt 2.15pt;height:3.3pt;\"\u003e\n \u003cp style='margin-top:0in;margin-right:0in;margin-bottom:0in;margin-left:0in;font-size:11.0pt;font-family:\"Calibri\",sans-serif;text-align:right;line-height:normal;'\u003e\u003cspan style='font-size:12px;font-family:\"Times New Roman\",serif;color:black;'\u003e0.762\u003c/span\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"height:3.3pt;border:none;\"\u003e\u003cbr\u003e\u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd style=\"width:45.75pt;border:none;padding:2.15pt 2.15pt 2.15pt 2.15pt;height:3.3pt;\"\u003e\n \u003cp style='margin-top:0in;margin-right:0in;margin-bottom:0in;margin-left:0in;font-size:11.0pt;font-family:\"Calibri\",sans-serif;text-align:right;line-height:normal;'\u003e\u003cspan style='font-size:12px;font-family:\"Times New Roman\",serif;'\u003e4\u003c/span\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width:79.5pt;border:none;padding:2.15pt 2.15pt 2.15pt 2.15pt;height:3.3pt;\"\u003e\n \u003cp style='margin-top:0in;margin-right:0in;margin-bottom:0in;margin-left:0in;font-size:11.0pt;font-family:\"Calibri\",sans-serif;text-align:right;line-height:normal;'\u003e\u003cspan style='font-size:12px;font-family:\"Times New Roman\",serif;'\u003eP\u003c/span\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width:54.95pt;border:none;background:#DDE6F4;padding: 2.15pt 2.15pt 2.15pt 2.15pt;height:3.3pt;\"\u003e\n \u003cp style='margin-top:0in;margin-right:0in;margin-bottom:0in;margin-left:0in;font-size:11.0pt;font-family:\"Calibri\",sans-serif;text-align:right;line-height:normal;'\u003e\u003cspan style='font-size:12px;font-family:\"Times New Roman\",serif;color:black;'\u003e0.894\u003c/span\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width:54.95pt;border:none;background:#D2DFF1;padding: 2.15pt 2.15pt 2.15pt 2.15pt;height:3.3pt;\"\u003e\n \u003cp style='margin-top:0in;margin-right:0in;margin-bottom:0in;margin-left:0in;font-size:11.0pt;font-family:\"Calibri\",sans-serif;text-align:right;line-height:normal;'\u003e\u003cspan style='font-size:12px;font-family:\"Times New Roman\",serif;color:black;'\u003e0.928\u003c/span\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width:54.95pt;border:none;background:#E5ECF7;padding: 2.15pt 2.15pt 2.15pt 2.15pt;height:3.3pt;\"\u003e\n \u003cp style='margin-top:0in;margin-right:0in;margin-bottom:0in;margin-left:0in;font-size:11.0pt;font-family:\"Calibri\",sans-serif;text-align:right;line-height:normal;'\u003e\u003cspan style='font-size:12px;font-family:\"Times New Roman\",serif;color:black;'\u003e0.663\u003c/span\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width:54.95pt;border:none;background:#C0D2EA;padding: 2.15pt 2.15pt 2.15pt 2.15pt;height:3.3pt;\"\u003e\n \u003cp style='margin-top:0in;margin-right:0in;margin-bottom:0in;margin-left:0in;font-size:11.0pt;font-family:\"Calibri\",sans-serif;text-align:right;line-height:normal;'\u003e\u003cspan style='font-size:12px;font-family:\"Times New Roman\",serif;color:black;'\u003e0.762\u003c/span\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"height:3.3pt;border:none;\"\u003e\u003cbr\u003e\u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd style=\"width:45.75pt;border:none;padding:2.15pt 2.15pt 2.15pt 2.15pt;height:3.3pt;\"\u003e\n \u003cp style='margin-top:0in;margin-right:0in;margin-bottom:0in;margin-left:0in;font-size:11.0pt;font-family:\"Calibri\",sans-serif;text-align:right;line-height:normal;'\u003e\u003cspan style='font-size:12px;font-family:\"Times New Roman\",serif;'\u003e5\u003c/span\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width:79.5pt;border:none;padding:2.15pt 2.15pt 2.15pt 2.15pt;height:3.3pt;\"\u003e\n \u003cp style='margin-top:0in;margin-right:0in;margin-bottom:0in;margin-left:0in;font-size:11.0pt;font-family:\"Calibri\",sans-serif;text-align:right;line-height:normal;'\u003e\u003cspan style='font-size:12px;font-family:\"Times New Roman\",serif;'\u003eT\u003c/span\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width:54.95pt;border:none;background:#7AA1D2;padding: 2.15pt 2.15pt 2.15pt 2.15pt;height:3.3pt;\"\u003e\n \u003cp style='margin-top:0in;margin-right:0in;margin-bottom:0in;margin-left:0in;font-size:11.0pt;font-family:\"Calibri\",sans-serif;text-align:right;line-height:normal;'\u003e\u003cspan style='font-size:12px;font-family:\"Times New Roman\",serif;color:black;'\u003e0.922\u003c/span\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width:54.95pt;border:none;background:#779ED0;padding: 2.15pt 2.15pt 2.15pt 2.15pt;height:3.3pt;\"\u003e\n \u003cp style='margin-top:0in;margin-right:0in;margin-bottom:0in;margin-left:0in;font-size:11.0pt;font-family:\"Calibri\",sans-serif;text-align:right;line-height:normal;'\u003e\u003cspan style='font-size:12px;font-family:\"Times New Roman\",serif;color:black;'\u003e0.946\u003c/span\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width:54.95pt;border:none;background:#7CA2D2;padding: 2.15pt 2.15pt 2.15pt 2.15pt;height:3.3pt;\"\u003e\n \u003cp style='margin-top:0in;margin-right:0in;margin-bottom:0in;margin-left:0in;font-size:11.0pt;font-family:\"Calibri\",sans-serif;text-align:right;line-height:normal;'\u003e\u003cspan style='font-size:12px;font-family:\"Times New Roman\",serif;color:black;'\u003e0.771\u003c/span\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width:54.95pt;border:none;background:#A6C0E1;padding: 2.15pt 2.15pt 2.15pt 2.15pt;height:3.3pt;\"\u003e\n \u003cp style='margin-top:0in;margin-right:0in;margin-bottom:0in;margin-left:0in;font-size:11.0pt;font-family:\"Calibri\",sans-serif;text-align:right;line-height:normal;'\u003e\u003cspan style='font-size:12px;font-family:\"Times New Roman\",serif;color:black;'\u003e0.769\u003c/span\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"height:3.3pt;border:none;\"\u003e\u003cbr\u003e\u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd style=\"width:45.75pt;border:none;padding:2.15pt 2.15pt 2.15pt 2.15pt;height:3.3pt;\"\u003e\n \u003cp style='margin-top:0in;margin-right:0in;margin-bottom:0in;margin-left:0in;font-size:11.0pt;font-family:\"Calibri\",sans-serif;text-align:right;line-height:normal;'\u003e\u003cspan style='font-size:12px;font-family:\"Times New Roman\",serif;'\u003e6\u003c/span\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width:79.5pt;border:none;padding:2.15pt 2.15pt 2.15pt 2.15pt;height:3.3pt;\"\u003e\n \u003cp style='margin-top:0in;margin-right:0in;margin-bottom:0in;margin-left:0in;font-size:11.0pt;font-family:\"Calibri\",sans-serif;text-align:right;line-height:normal;'\u003e\u003cspan style='font-size:12px;font-family:\"Times New Roman\",serif;'\u003eLag-12\u003c/span\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width:54.95pt;border:none;background:#C4D5EC;padding: 2.15pt 2.15pt 2.15pt 2.15pt;height:3.3pt;\"\u003e\n \u003cp style='margin-top:0in;margin-right:0in;margin-bottom:0in;margin-left:0in;font-size:11.0pt;font-family:\"Calibri\",sans-serif;text-align:right;line-height:normal;'\u003e\u003cspan style='font-size:12px;font-family:\"Times New Roman\",serif;color:black;'\u003e0.901\u003c/span\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width:54.95pt;border:none;background:#CBDAEE;padding: 2.15pt 2.15pt 2.15pt 2.15pt;height:3.3pt;\"\u003e\n \u003cp style='margin-top:0in;margin-right:0in;margin-bottom:0in;margin-left:0in;font-size:11.0pt;font-family:\"Calibri\",sans-serif;text-align:right;line-height:normal;'\u003e\u003cspan style='font-size:12px;font-family:\"Times New Roman\",serif;color:black;'\u003e0.929\u003c/span\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width:54.95pt;border:none;background:#CBD9EE;padding: 2.15pt 2.15pt 2.15pt 2.15pt;height:3.3pt;\"\u003e\n \u003cp style='margin-top:0in;margin-right:0in;margin-bottom:0in;margin-left:0in;font-size:11.0pt;font-family:\"Calibri\",sans-serif;text-align:right;line-height:normal;'\u003e\u003cspan style='font-size:12px;font-family:\"Times New Roman\",serif;color:black;'\u003e0.690\u003c/span\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width:54.95pt;border:none;background:#D7E2F2;padding: 2.15pt 2.15pt 2.15pt 2.15pt;height:3.3pt;\"\u003e\n \u003cp style='margin-top:0in;margin-right:0in;margin-bottom:0in;margin-left:0in;font-size:11.0pt;font-family:\"Calibri\",sans-serif;text-align:right;line-height:normal;'\u003e\u003cspan style='font-size:12px;font-family:\"Times New Roman\",serif;color:black;'\u003e0.755\u003c/span\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"height:3.3pt;border:none;\"\u003e\u003cbr\u003e\u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd style=\"width:45.75pt;border:none;padding:2.15pt 2.15pt 2.15pt 2.15pt;height:3.3pt;\"\u003e\n \u003cp style='margin-top:0in;margin-right:0in;margin-bottom:0in;margin-left:0in;font-size:11.0pt;font-family:\"Calibri\",sans-serif;text-align:right;line-height:normal;'\u003e\u003cspan style='font-size:12px;font-family:\"Times New Roman\",serif;'\u003e7\u003c/span\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width:79.5pt;border:none;padding:2.15pt 2.15pt 2.15pt 2.15pt;height:3.3pt;\"\u003e\n \u003cp style='margin-top:0in;margin-right:0in;margin-bottom:0in;margin-left:0in;font-size:11.0pt;font-family:\"Calibri\",sans-serif;text-align:right;line-height:normal;'\u003e\u003cspan style='font-size:12px;font-family:\"Times New Roman\",serif;'\u003eLag-06\u003c/span\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width:54.95pt;border:none;background:#8FB0D9;padding: 2.15pt 2.15pt 2.15pt 2.15pt;height:3.3pt;\"\u003e\n \u003cp style='margin-top:0in;margin-right:0in;margin-bottom:0in;margin-left:0in;font-size:11.0pt;font-family:\"Calibri\",sans-serif;text-align:right;line-height:normal;'\u003e\u003cspan style='font-size:12px;font-family:\"Times New Roman\",serif;color:black;'\u003e0.916\u003c/span\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width:54.95pt;border:none;background:#90B0D9;padding: 2.15pt 2.15pt 2.15pt 2.15pt;height:3.3pt;\"\u003e\n \u003cp style='margin-top:0in;margin-right:0in;margin-bottom:0in;margin-left:0in;font-size:11.0pt;font-family:\"Calibri\",sans-serif;text-align:right;line-height:normal;'\u003e\u003cspan style='font-size:12px;font-family:\"Times New Roman\",serif;color:black;'\u003e0.941\u003c/span\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width:54.95pt;border:none;background:#93B2DA;padding: 2.15pt 2.15pt 2.15pt 2.15pt;height:3.3pt;\"\u003e\n \u003cp style='margin-top:0in;margin-right:0in;margin-bottom:0in;margin-left:0in;font-size:11.0pt;font-family:\"Calibri\",sans-serif;text-align:right;line-height:normal;'\u003e\u003cspan style='font-size:12px;font-family:\"Times New Roman\",serif;color:black;'\u003e0.747\u003c/span\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width:54.95pt;border:none;background:#BCCFE9;padding: 2.15pt 2.15pt 2.15pt 2.15pt;height:3.3pt;\"\u003e\n \u003cp style='margin-top:0in;margin-right:0in;margin-bottom:0in;margin-left:0in;font-size:11.0pt;font-family:\"Calibri\",sans-serif;text-align:right;line-height:normal;'\u003e\u003cspan style='font-size:12px;font-family:\"Times New Roman\",serif;color:black;'\u003e0.763\u003c/span\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"height:3.3pt;border:none;\"\u003e\u003cbr\u003e\u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd style=\"width:45.75pt;border:none;padding:2.15pt 2.15pt 2.15pt 2.15pt;height:3.3pt;\"\u003e\n \u003cp style='margin-top:0in;margin-right:0in;margin-bottom:0in;margin-left:0in;font-size:11.0pt;font-family:\"Calibri\",sans-serif;text-align:right;line-height:normal;'\u003e\u003cspan style='font-size:12px;font-family:\"Times New Roman\",serif;'\u003e8\u003c/span\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width:79.5pt;border:none;padding:2.15pt 2.15pt 2.15pt 2.15pt;height:3.3pt;\"\u003e\n \u003cp style='margin-top:0in;margin-right:0in;margin-bottom:0in;margin-left:0in;font-size:11.0pt;font-family:\"Calibri\",sans-serif;text-align:right;line-height:normal;'\u003e\u003cspan style='font-size:12px;font-family:\"Times New Roman\",serif;'\u003eLag-03\u003c/span\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width:54.95pt;border:none;background:#BACDE8;padding: 2.15pt 2.15pt 2.15pt 2.15pt;height:3.3pt;\"\u003e\n \u003cp style='margin-top:0in;margin-right:0in;margin-bottom:0in;margin-left:0in;font-size:11.0pt;font-family:\"Calibri\",sans-serif;text-align:right;line-height:normal;'\u003e\u003cspan style='font-size:12px;font-family:\"Times New Roman\",serif;color:black;'\u003e0.904\u003c/span\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width:54.95pt;border:none;background:#B9CDE8;padding: 2.15pt 2.15pt 2.15pt 2.15pt;height:3.3pt;\"\u003e\n \u003cp style='margin-top:0in;margin-right:0in;margin-bottom:0in;margin-left:0in;font-size:11.0pt;font-family:\"Calibri\",sans-serif;text-align:right;line-height:normal;'\u003e\u003cspan style='font-size:12px;font-family:\"Times New Roman\",serif;color:black;'\u003e0.933\u003c/span\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width:54.95pt;border:none;background:#C3D4EB;padding: 2.15pt 2.15pt 2.15pt 2.15pt;height:3.3pt;\"\u003e\n \u003cp style='margin-top:0in;margin-right:0in;margin-bottom:0in;margin-left:0in;font-size:11.0pt;font-family:\"Calibri\",sans-serif;text-align:right;line-height:normal;'\u003e\u003cspan style='font-size:12px;font-family:\"Times New Roman\",serif;color:black;'\u003e0.698\u003c/span\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width:54.95pt;border:none;background:#B5CAE6;padding: 2.15pt 2.15pt 2.15pt 2.15pt;height:3.3pt;\"\u003e\n \u003cp style='margin-top:0in;margin-right:0in;margin-bottom:0in;margin-left:0in;font-size:11.0pt;font-family:\"Calibri\",sans-serif;text-align:right;line-height:normal;'\u003e\u003cspan style='font-size:12px;font-family:\"Times New Roman\",serif;color:black;'\u003e0.765\u003c/span\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"height:3.3pt;border:none;\"\u003e\u003cbr\u003e\u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd style=\"width:45.75pt;border:none;padding:2.15pt 2.15pt 2.15pt 2.15pt;height:3.3pt;\"\u003e\n \u003cp style='margin-top:0in;margin-right:0in;margin-bottom:0in;margin-left:0in;font-size:11.0pt;font-family:\"Calibri\",sans-serif;text-align:right;line-height:normal;'\u003e\u003cspan style='font-size:12px;font-family:\"Times New Roman\",serif;'\u003e9\u003c/span\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width:79.5pt;border:none;padding:2.15pt 2.15pt 2.15pt 2.15pt;height:3.3pt;\"\u003e\n \u003cp style='margin-top:0in;margin-right:0in;margin-bottom:0in;margin-left:0in;font-size:11.0pt;font-family:\"Calibri\",sans-serif;text-align:right;line-height:normal;'\u003e\u003cspan style='font-size:12px;font-family:\"Times New Roman\",serif;'\u003eLag-01\u003c/span\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width:54.95pt;border:none;background:#E7EEF8;padding: 2.15pt 2.15pt 2.15pt 2.15pt;height:3.3pt;\"\u003e\n \u003cp style='margin-top:0in;margin-right:0in;margin-bottom:0in;margin-left:0in;font-size:11.0pt;font-family:\"Calibri\",sans-serif;text-align:right;line-height:normal;'\u003e\u003cspan style='font-size:12px;font-family:\"Times New Roman\",serif;color:black;'\u003e0.891\u003c/span\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width:54.95pt;border:none;background:#E4EBF7;padding: 2.15pt 2.15pt 2.15pt 2.15pt;height:3.3pt;\"\u003e\n \u003cp style='margin-top:0in;margin-right:0in;margin-bottom:0in;margin-left:0in;font-size:11.0pt;font-family:\"Calibri\",sans-serif;text-align:right;line-height:normal;'\u003e\u003cspan style='font-size:12px;font-family:\"Times New Roman\",serif;color:black;'\u003e0.925\u003c/span\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width:54.95pt;border:none;background:#EDF2FA;padding: 2.15pt 2.15pt 2.15pt 2.15pt;height:3.3pt;\"\u003e\n \u003cp style='margin-top:0in;margin-right:0in;margin-bottom:0in;margin-left:0in;font-size:11.0pt;font-family:\"Calibri\",sans-serif;text-align:right;line-height:normal;'\u003e\u003cspan style='font-size:12px;font-family:\"Times New Roman\",serif;color:black;'\u003e0.655\u003c/span\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width:54.95pt;border:none;background:#D7E2F2;padding: 2.15pt 2.15pt 2.15pt 2.15pt;height:3.3pt;\"\u003e\n \u003cp style='margin-top:0in;margin-right:0in;margin-bottom:0in;margin-left:0in;font-size:11.0pt;font-family:\"Calibri\",sans-serif;text-align:right;line-height:normal;'\u003e\u003cspan style='font-size:12px;font-family:\"Times New Roman\",serif;color:black;'\u003e0.755\u003c/span\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"height:3.3pt;border:none;\"\u003e\u003cbr\u003e\u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd style=\"width:45.75pt;border:none;padding:2.15pt 2.15pt 2.15pt 2.15pt;height:3.3pt;\"\u003e\n \u003cp style='margin-top:0in;margin-right:0in;margin-bottom:0in;margin-left:0in;font-size:11.0pt;font-family:\"Calibri\",sans-serif;text-align:right;line-height:normal;'\u003e\u003cspan style='font-size:12px;font-family:\"Times New Roman\",serif;'\u003e10\u003c/span\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width:79.5pt;border:none;padding:2.15pt 2.15pt 2.15pt 2.15pt;height:3.3pt;\"\u003e\n \u003cp style='margin-top:0in;margin-right:0in;margin-bottom:0in;margin-left:0in;font-size:11.0pt;font-family:\"Calibri\",sans-serif;text-align:right;line-height:normal;'\u003e\u003cspan style='font-size:12px;font-family:\"Times New Roman\",serif;'\u003ePA and SPI\u003c/span\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width:54.95pt;border:none;background:#FAD0D2;padding: 2.15pt 2.15pt 2.15pt 2.15pt;height:3.3pt;\"\u003e\n \u003cp style='margin-top:0in;margin-right:0in;margin-bottom:0in;margin-left:0in;font-size:11.0pt;font-family:\"Calibri\",sans-serif;text-align:right;line-height:normal;'\u003e\u003cspan style='font-size:12px;font-family:\"Times New Roman\",serif;color:black;'\u003e0.849\u003c/span\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width:54.95pt;border:none;background:#FBD7DA;padding: 2.15pt 2.15pt 2.15pt 2.15pt;height:3.3pt;\"\u003e\n \u003cp style='margin-top:0in;margin-right:0in;margin-bottom:0in;margin-left:0in;font-size:11.0pt;font-family:\"Calibri\",sans-serif;text-align:right;line-height:normal;'\u003e\u003cspan style='font-size:12px;font-family:\"Times New Roman\",serif;color:black;'\u003e0.900\u003c/span\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width:54.95pt;border:none;background:#FAC4C6;padding: 2.15pt 2.15pt 2.15pt 2.15pt;height:3.3pt;\"\u003e\n \u003cp style='margin-top:0in;margin-right:0in;margin-bottom:0in;margin-left:0in;font-size:11.0pt;font-family:\"Calibri\",sans-serif;text-align:right;line-height:normal;'\u003e\u003cspan style='font-size:12px;font-family:\"Times New Roman\",serif;color:black;'\u003e0.542\u003c/span\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width:54.95pt;border:none;background:#FAC8CB;padding: 2.15pt 2.15pt 2.15pt 2.15pt;height:3.3pt;\"\u003e\n \u003cp style='margin-top:0in;margin-right:0in;margin-bottom:0in;margin-left:0in;font-size:11.0pt;font-family:\"Calibri\",sans-serif;text-align:right;line-height:normal;'\u003e\u003cspan style='font-size:12px;font-family:\"Times New Roman\",serif;color:black;'\u003e0.708\u003c/span\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"height:3.3pt;border:none;\"\u003e\u003cbr\u003e\u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd style=\"width:45.75pt;border:none;padding:2.15pt 2.15pt 2.15pt 2.15pt;height:3.3pt;\"\u003e\n \u003cp style='margin-top:0in;margin-right:0in;margin-bottom:0in;margin-left:0in;font-size:11.0pt;font-family:\"Calibri\",sans-serif;text-align:right;line-height:normal;'\u003e\u003cspan style='font-size:12px;font-family:\"Times New Roman\",serif;'\u003e11\u003c/span\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width:79.5pt;border:none;padding:2.15pt 2.15pt 2.15pt 2.15pt;height:3.3pt;\"\u003e\n \u003cp style='margin-top:0in;margin-right:0in;margin-bottom:0in;margin-left:0in;font-size:11.0pt;font-family:\"Calibri\",sans-serif;text-align:right;line-height:normal;'\u003e\u003cspan style='font-size:12px;font-family:\"Times New Roman\",serif;'\u003eP and SPI\u003c/span\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width:54.95pt;border:none;background:#FBF8FB;padding: 2.15pt 2.15pt 2.15pt 2.15pt;height:3.3pt;\"\u003e\n \u003cp style='margin-top:0in;margin-right:0in;margin-bottom:0in;margin-left:0in;font-size:11.0pt;font-family:\"Calibri\",sans-serif;text-align:right;line-height:normal;'\u003e\u003cspan style='font-size:12px;font-family:\"Times New Roman\",serif;color:black;'\u003e0.882\u003c/span\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width:54.95pt;border:none;background:#FBFBFE;padding: 2.15pt 2.15pt 2.15pt 2.15pt;height:3.3pt;\"\u003e\n \u003cp style='margin-top:0in;margin-right:0in;margin-bottom:0in;margin-left:0in;font-size:11.0pt;font-family:\"Calibri\",sans-serif;text-align:right;line-height:normal;'\u003e\u003cspan style='font-size:12px;font-family:\"Times New Roman\",serif;color:black;'\u003e0.919\u003c/span\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width:54.95pt;border:none;background:#FBF6F9;padding: 2.15pt 2.15pt 2.15pt 2.15pt;height:3.3pt;\"\u003e\n \u003cp style='margin-top:0in;margin-right:0in;margin-bottom:0in;margin-left:0in;font-size:11.0pt;font-family:\"Calibri\",sans-serif;text-align:right;line-height:normal;'\u003e\u003cspan style='font-size:12px;font-family:\"Times New Roman\",serif;color:black;'\u003e0.630\u003c/span\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width:54.95pt;border:none;background:#FBF1F4;padding: 2.15pt 2.15pt 2.15pt 2.15pt;height:3.3pt;\"\u003e\n \u003cp style='margin-top:0in;margin-right:0in;margin-bottom:0in;margin-left:0in;font-size:11.0pt;font-family:\"Calibri\",sans-serif;text-align:right;line-height:normal;'\u003e\u003cspan style='font-size:12px;font-family:\"Times New Roman\",serif;color:black;'\u003e0.737\u003c/span\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"height:3.3pt;border:none;\"\u003e\u003cbr\u003e\u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd style=\"width:45.75pt;border:none;padding:2.15pt 2.15pt 2.15pt 2.15pt;height:3.3pt;\"\u003e\n \u003cp style='margin-top:0in;margin-right:0in;margin-bottom:0in;margin-left:0in;font-size:11.0pt;font-family:\"Calibri\",sans-serif;text-align:right;line-height:normal;'\u003e\u003cspan style='font-size:12px;font-family:\"Times New Roman\",serif;'\u003e12\u003c/span\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width:79.5pt;border:none;padding:2.15pt 2.15pt 2.15pt 2.15pt;height:3.3pt;\"\u003e\n \u003cp style='margin-top:0in;margin-right:0in;margin-bottom:0in;margin-left:0in;font-size:11.0pt;font-family:\"Calibri\",sans-serif;text-align:right;line-height:normal;'\u003e\u003cspan style='font-size:12px;font-family:\"Times New Roman\",serif;'\u003eP and PA\u003c/span\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width:54.95pt;border:none;background:#FBE7EA;padding: 2.15pt 2.15pt 2.15pt 2.15pt;height:3.3pt;\"\u003e\n \u003cp style='margin-top:0in;margin-right:0in;margin-bottom:0in;margin-left:0in;font-size:11.0pt;font-family:\"Calibri\",sans-serif;text-align:right;line-height:normal;'\u003e\u003cspan style='font-size:12px;font-family:\"Times New Roman\",serif;color:black;'\u003e0.868\u003c/span\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width:54.95pt;border:none;background:#FBE6E8;padding: 2.15pt 2.15pt 2.15pt 2.15pt;height:3.3pt;\"\u003e\n \u003cp style='margin-top:0in;margin-right:0in;margin-bottom:0in;margin-left:0in;font-size:11.0pt;font-family:\"Calibri\",sans-serif;text-align:right;line-height:normal;'\u003e\u003cspan style='font-size:12px;font-family:\"Times New Roman\",serif;color:black;'\u003e0.908\u003c/span\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width:54.95pt;border:none;background:#FBDEE1;padding: 2.15pt 2.15pt 2.15pt 2.15pt;height:3.3pt;\"\u003e\n \u003cp style='margin-top:0in;margin-right:0in;margin-bottom:0in;margin-left:0in;font-size:11.0pt;font-family:\"Calibri\",sans-serif;text-align:right;line-height:normal;'\u003e\u003cspan style='font-size:12px;font-family:\"Times New Roman\",serif;color:black;'\u003e0.587\u003c/span\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width:54.95pt;border:none;background:#FCFCFF;padding: 2.15pt 2.15pt 2.15pt 2.15pt;height:3.3pt;\"\u003e\n \u003cp style='margin-top:0in;margin-right:0in;margin-bottom:0in;margin-left:0in;font-size:11.0pt;font-family:\"Calibri\",sans-serif;text-align:right;line-height:normal;'\u003e\u003cspan style='font-size:12px;font-family:\"Times New Roman\",serif;color:black;'\u003e0.745\u003c/span\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"height:3.3pt;border:none;\"\u003e\u003cbr\u003e\u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd style=\"width:45.75pt;border:none;padding:2.15pt 2.15pt 2.15pt 2.15pt;height:3.3pt;\"\u003e\n \u003cp style='margin-top:0in;margin-right:0in;margin-bottom:0in;margin-left:0in;font-size:11.0pt;font-family:\"Calibri\",sans-serif;text-align:right;line-height:normal;'\u003e\u003cspan style='font-size:12px;font-family:\"Times New Roman\",serif;'\u003e13\u003c/span\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width:79.5pt;border:none;padding:2.15pt 2.15pt 2.15pt 2.15pt;height:3.3pt;\"\u003e\n \u003cp style='margin-top:0in;margin-right:0in;margin-bottom:0in;margin-left:0in;font-size:11.0pt;font-family:\"Calibri\",sans-serif;text-align:right;line-height:normal;'\u003e\u003cspan style='font-size:12px;font-family:\"Times New Roman\",serif;'\u003eT and SPI\u003c/span\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width:54.95pt;border:none;background:#81A6D4;padding: 2.15pt 2.15pt 2.15pt 2.15pt;height:3.3pt;\"\u003e\n \u003cp style='margin-top:0in;margin-right:0in;margin-bottom:0in;margin-left:0in;font-size:11.0pt;font-family:\"Calibri\",sans-serif;text-align:right;line-height:normal;'\u003e\u003cspan style='font-size:12px;font-family:\"Times New Roman\",serif;color:black;'\u003e0.920\u003c/span\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width:54.95pt;border:none;background:#80A5D4;padding: 2.15pt 2.15pt 2.15pt 2.15pt;height:3.3pt;\"\u003e\n \u003cp style='margin-top:0in;margin-right:0in;margin-bottom:0in;margin-left:0in;font-size:11.0pt;font-family:\"Calibri\",sans-serif;text-align:right;line-height:normal;'\u003e\u003cspan style='font-size:12px;font-family:\"Times New Roman\",serif;color:black;'\u003e0.945\u003c/span\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width:54.95pt;border:none;background:#90B0D9;padding: 2.15pt 2.15pt 2.15pt 2.15pt;height:3.3pt;\"\u003e\n \u003cp style='margin-top:0in;margin-right:0in;margin-bottom:0in;margin-left:0in;font-size:11.0pt;font-family:\"Calibri\",sans-serif;text-align:right;line-height:normal;'\u003e\u003cspan style='font-size:12px;font-family:\"Times New Roman\",serif;color:black;'\u003e0.751\u003c/span\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width:54.95pt;border:none;background:#5A8AC6;padding: 2.15pt 2.15pt 2.15pt 2.15pt;height:3.3pt;\"\u003e\n \u003cp style='margin-top:0in;margin-right:0in;margin-bottom:0in;margin-left:0in;font-size:11.0pt;font-family:\"Calibri\",sans-serif;text-align:right;line-height:normal;'\u003e\u003cspan style='font-size:12px;font-family:\"Times New Roman\",serif;color:black;'\u003e0.790\u003c/span\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"height:3.3pt;border:none;\"\u003e\u003cbr\u003e\u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd style=\"width:45.75pt;border:none;padding:2.15pt 2.15pt 2.15pt 2.15pt;height:3.3pt;\"\u003e\n \u003cp style='margin-top:0in;margin-right:0in;margin-bottom:0in;margin-left:0in;font-size:11.0pt;font-family:\"Calibri\",sans-serif;text-align:right;line-height:normal;'\u003e\u003cspan style='font-size:12px;font-family:\"Times New Roman\",serif;'\u003e14\u003c/span\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width:79.5pt;border:none;padding:2.15pt 2.15pt 2.15pt 2.15pt;height:3.3pt;\"\u003e\n \u003cp style='margin-top:0in;margin-right:0in;margin-bottom:0in;margin-left:0in;font-size:11.0pt;font-family:\"Calibri\",sans-serif;text-align:right;line-height:normal;'\u003e\u003cspan style='font-size:12px;font-family:\"Times New Roman\",serif;'\u003eT and PA\u003c/span\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width:54.95pt;border:none;background:#D6E1F2;padding: 2.15pt 2.15pt 2.15pt 2.15pt;height:3.3pt;\"\u003e\n \u003cp style='margin-top:0in;margin-right:0in;margin-bottom:0in;margin-left:0in;font-size:11.0pt;font-family:\"Calibri\",sans-serif;text-align:right;line-height:normal;'\u003e\u003cspan style='font-size:12px;font-family:\"Times New Roman\",serif;color:black;'\u003e0.896\u003c/span\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width:54.95pt;border:none;background:#C2D3EB;padding: 2.15pt 2.15pt 2.15pt 2.15pt;height:3.3pt;\"\u003e\n \u003cp style='margin-top:0in;margin-right:0in;margin-bottom:0in;margin-left:0in;font-size:11.0pt;font-family:\"Calibri\",sans-serif;text-align:right;line-height:normal;'\u003e\u003cspan style='font-size:12px;font-family:\"Times New Roman\",serif;color:black;'\u003e0.931\u003c/span\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width:54.95pt;border:none;background:#DDE6F4;padding: 2.15pt 2.15pt 2.15pt 2.15pt;height:3.3pt;\"\u003e\n \u003cp style='margin-top:0in;margin-right:0in;margin-bottom:0in;margin-left:0in;font-size:11.0pt;font-family:\"Calibri\",sans-serif;text-align:right;line-height:normal;'\u003e\u003cspan style='font-size:12px;font-family:\"Times New Roman\",serif;color:black;'\u003e0.672\u003c/span\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width:54.95pt;border:none;background:#D3DFF1;padding: 2.15pt 2.15pt 2.15pt 2.15pt;height:3.3pt;\"\u003e\n \u003cp style='margin-top:0in;margin-right:0in;margin-bottom:0in;margin-left:0in;font-size:11.0pt;font-family:\"Calibri\",sans-serif;text-align:right;line-height:normal;'\u003e\u003cspan style='font-size:12px;font-family:\"Times New Roman\",serif;color:black;'\u003e0.756\u003c/span\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"height:3.3pt;border:none;\"\u003e\u003cbr\u003e\u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd style=\"width:45.75pt;border:none;padding:2.15pt 2.15pt 2.15pt 2.15pt;height:3.3pt;\"\u003e\n \u003cp style='margin-top:0in;margin-right:0in;margin-bottom:0in;margin-left:0in;font-size:11.0pt;font-family:\"Calibri\",sans-serif;text-align:right;line-height:normal;'\u003e\u003cspan style='font-size:12px;font-family:\"Times New Roman\",serif;'\u003e15\u003c/span\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width:79.5pt;border:none;padding:2.15pt 2.15pt 2.15pt 2.15pt;height:3.3pt;\"\u003e\n \u003cp style='margin-top:0in;margin-right:0in;margin-bottom:0in;margin-left:0in;font-size:11.0pt;font-family:\"Calibri\",sans-serif;text-align:right;line-height:normal;'\u003e\u003cspan style='font-size:12px;font-family:\"Times New Roman\",serif;'\u003eT and P\u003c/span\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width:54.95pt;border:none;background:#88ABD7;padding: 2.15pt 2.15pt 2.15pt 2.15pt;height:3.3pt;\"\u003e\n \u003cp style='margin-top:0in;margin-right:0in;margin-bottom:0in;margin-left:0in;font-size:11.0pt;font-family:\"Calibri\",sans-serif;text-align:right;line-height:normal;'\u003e\u003cspan style='font-size:12px;font-family:\"Times New Roman\",serif;color:black;'\u003e0.918\u003c/span\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width:54.95pt;border:none;background:#8AACD7;padding: 2.15pt 2.15pt 2.15pt 2.15pt;height:3.3pt;\"\u003e\n \u003cp style='margin-top:0in;margin-right:0in;margin-bottom:0in;margin-left:0in;font-size:11.0pt;font-family:\"Calibri\",sans-serif;text-align:right;line-height:normal;'\u003e\u003cspan style='font-size:12px;font-family:\"Times New Roman\",serif;color:black;'\u003e0.943\u003c/span\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width:54.95pt;border:none;background:#92B1DA;padding: 2.15pt 2.15pt 2.15pt 2.15pt;height:3.3pt;\"\u003e\n \u003cp style='margin-top:0in;margin-right:0in;margin-bottom:0in;margin-left:0in;font-size:11.0pt;font-family:\"Calibri\",sans-serif;text-align:right;line-height:normal;'\u003e\u003cspan style='font-size:12px;font-family:\"Times New Roman\",serif;color:black;'\u003e0.749\u003c/span\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width:54.95pt;border:none;background:#8BADD8;padding: 2.15pt 2.15pt 2.15pt 2.15pt;height:3.3pt;\"\u003e\n \u003cp style='margin-top:0in;margin-right:0in;margin-bottom:0in;margin-left:0in;font-size:11.0pt;font-family:\"Calibri\",sans-serif;text-align:right;line-height:normal;'\u003e\u003cspan style='font-size:12px;font-family:\"Times New Roman\",serif;color:black;'\u003e0.776\u003c/span\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"height:3.3pt;border:none;\"\u003e\u003cbr\u003e\u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd style=\"width:45.75pt;border:none;padding:2.15pt 2.15pt 2.15pt 2.15pt;height:3.3pt;\"\u003e\n \u003cp style='margin-top:0in;margin-right:0in;margin-bottom:0in;margin-left:0in;font-size:11.0pt;font-family:\"Calibri\",sans-serif;text-align:right;line-height:normal;'\u003e\u003cspan style='font-size:12px;font-family:\"Times New Roman\",serif;'\u003e16\u003c/span\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width:79.5pt;border:none;padding:2.15pt 2.15pt 2.15pt 2.15pt;height:3.3pt;\"\u003e\n \u003cp style='margin-top:0in;margin-right:0in;margin-bottom:0in;margin-left:0in;font-size:11.0pt;font-family:\"Calibri\",sans-serif;text-align:right;line-height:normal;'\u003e\u003cspan style='font-size:12px;font-family:\"Times New Roman\",serif;'\u003eLag-06 and 12\u003c/span\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width:54.95pt;border:none;background:#FABABC;padding: 2.15pt 2.15pt 2.15pt 2.15pt;height:3.3pt;\"\u003e\n \u003cp style='margin-top:0in;margin-right:0in;margin-bottom:0in;margin-left:0in;font-size:11.0pt;font-family:\"Calibri\",sans-serif;text-align:right;line-height:normal;'\u003e\u003cspan style='font-size:12px;font-family:\"Times New Roman\",serif;color:black;'\u003e0.831\u003c/span\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width:54.95pt;border:none;background:#FAB6B9;padding: 2.15pt 2.15pt 2.15pt 2.15pt;height:3.3pt;\"\u003e\n \u003cp style='margin-top:0in;margin-right:0in;margin-bottom:0in;margin-left:0in;font-size:11.0pt;font-family:\"Calibri\",sans-serif;text-align:right;line-height:normal;'\u003e\u003cspan style='font-size:12px;font-family:\"Times New Roman\",serif;color:black;'\u003e0.882\u003c/span\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width:54.95pt;border:none;background:#F9ACAF;padding: 2.15pt 2.15pt 2.15pt 2.15pt;height:3.3pt;\"\u003e\n \u003cp style='margin-top:0in;margin-right:0in;margin-bottom:0in;margin-left:0in;font-size:11.0pt;font-family:\"Calibri\",sans-serif;text-align:right;line-height:normal;'\u003e\u003cspan style='font-size:12px;font-family:\"Times New Roman\",serif;color:black;'\u003e0.501\u003c/span\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width:54.95pt;border:none;background:#FABBBD;padding: 2.15pt 2.15pt 2.15pt 2.15pt;height:3.3pt;\"\u003e\n \u003cp style='margin-top:0in;margin-right:0in;margin-bottom:0in;margin-left:0in;font-size:11.0pt;font-family:\"Calibri\",sans-serif;text-align:right;line-height:normal;'\u003e\u003cspan style='font-size:12px;font-family:\"Times New Roman\",serif;color:black;'\u003e0.698\u003c/span\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"height:3.3pt;border:none;\"\u003e\u003cbr\u003e\u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd style=\"width:45.75pt;border:none;padding:2.15pt 2.15pt 2.15pt 2.15pt;height:3.3pt;\"\u003e\n \u003cp style='margin-top:0in;margin-right:0in;margin-bottom:0in;margin-left:0in;font-size:11.0pt;font-family:\"Calibri\",sans-serif;text-align:right;line-height:normal;'\u003e\u003cspan style='font-size:12px;font-family:\"Times New Roman\",serif;'\u003e17\u003c/span\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width:79.5pt;border:none;padding:2.15pt 2.15pt 2.15pt 2.15pt;height:3.3pt;\"\u003e\n \u003cp style='margin-top:0in;margin-right:0in;margin-bottom:0in;margin-left:0in;font-size:11.0pt;font-family:\"Calibri\",sans-serif;text-align:right;line-height:normal;'\u003e\u003cspan style='font-size:12px;font-family:\"Times New Roman\",serif;'\u003eLag-03 and 12\u003c/span\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width:54.95pt;border:none;background:#FCFCFF;padding: 2.15pt 2.15pt 2.15pt 2.15pt;height:3.3pt;\"\u003e\n \u003cp style='margin-top:0in;margin-right:0in;margin-bottom:0in;margin-left:0in;font-size:11.0pt;font-family:\"Calibri\",sans-serif;text-align:right;line-height:normal;'\u003e\u003cspan style='font-size:12px;font-family:\"Times New Roman\",serif;color:black;'\u003e0.885\u003c/span\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width:54.95pt;border:none;background:#FCFCFF;padding: 2.15pt 2.15pt 2.15pt 2.15pt;height:3.3pt;\"\u003e\n \u003cp style='margin-top:0in;margin-right:0in;margin-bottom:0in;margin-left:0in;font-size:11.0pt;font-family:\"Calibri\",sans-serif;text-align:right;line-height:normal;'\u003e\u003cspan style='font-size:12px;font-family:\"Times New Roman\",serif;color:black;'\u003e0.920\u003c/span\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width:54.95pt;border:none;background:#FCFCFF;padding: 2.15pt 2.15pt 2.15pt 2.15pt;height:3.3pt;\"\u003e\n \u003cp style='margin-top:0in;margin-right:0in;margin-bottom:0in;margin-left:0in;font-size:11.0pt;font-family:\"Calibri\",sans-serif;text-align:right;line-height:normal;'\u003e\u003cspan style='font-size:12px;font-family:\"Times New Roman\",serif;color:black;'\u003e0.639\u003c/span\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width:54.95pt;border:none;background:#FBF6F9;padding: 2.15pt 2.15pt 2.15pt 2.15pt;height:3.3pt;\"\u003e\n \u003cp style='margin-top:0in;margin-right:0in;margin-bottom:0in;margin-left:0in;font-size:11.0pt;font-family:\"Calibri\",sans-serif;text-align:right;line-height:normal;'\u003e\u003cspan style='font-size:12px;font-family:\"Times New Roman\",serif;color:black;'\u003e0.740\u003c/span\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"height:3.3pt;border:none;\"\u003e\u003cbr\u003e\u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd style=\"width:45.75pt;border:none;padding:2.15pt 2.15pt 2.15pt 2.15pt;height:3.3pt;\"\u003e\n \u003cp style='margin-top:0in;margin-right:0in;margin-bottom:0in;margin-left:0in;font-size:11.0pt;font-family:\"Calibri\",sans-serif;text-align:right;line-height:normal;'\u003e\u003cspan style='font-size:12px;font-family:\"Times New Roman\",serif;'\u003e18\u003c/span\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width:79.5pt;border:none;padding:2.15pt 2.15pt 2.15pt 2.15pt;height:3.3pt;\"\u003e\n \u003cp style='margin-top:0in;margin-right:0in;margin-bottom:0in;margin-left:0in;font-size:11.0pt;font-family:\"Calibri\",sans-serif;text-align:right;line-height:normal;'\u003e\u003cspan style='font-size:12px;font-family:\"Times New Roman\",serif;'\u003eLag-01 and 12\u003c/span\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width:54.95pt;border:none;background:#FBE1E4;padding: 2.15pt 2.15pt 2.15pt 2.15pt;height:3.3pt;\"\u003e\n \u003cp style='margin-top:0in;margin-right:0in;margin-bottom:0in;margin-left:0in;font-size:11.0pt;font-family:\"Calibri\",sans-serif;text-align:right;line-height:normal;'\u003e\u003cspan style='font-size:12px;font-family:\"Times New Roman\",serif;color:black;'\u003e0.863\u003c/span\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width:54.95pt;border:none;background:#FBE4E6;padding: 2.15pt 2.15pt 2.15pt 2.15pt;height:3.3pt;\"\u003e\n \u003cp style='margin-top:0in;margin-right:0in;margin-bottom:0in;margin-left:0in;font-size:11.0pt;font-family:\"Calibri\",sans-serif;text-align:right;line-height:normal;'\u003e\u003cspan style='font-size:12px;font-family:\"Times New Roman\",serif;color:black;'\u003e0.907\u003c/span\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width:54.95pt;border:none;background:#FBD7DA;padding: 2.15pt 2.15pt 2.15pt 2.15pt;height:3.3pt;\"\u003e\n \u003cp style='margin-top:0in;margin-right:0in;margin-bottom:0in;margin-left:0in;font-size:11.0pt;font-family:\"Calibri\",sans-serif;text-align:right;line-height:normal;'\u003e\u003cspan style='font-size:12px;font-family:\"Times New Roman\",serif;color:black;'\u003e0.575\u003c/span\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width:54.95pt;border:none;background:#FBEEF1;padding: 2.15pt 2.15pt 2.15pt 2.15pt;height:3.3pt;\"\u003e\n \u003cp style='margin-top:0in;margin-right:0in;margin-bottom:0in;margin-left:0in;font-size:11.0pt;font-family:\"Calibri\",sans-serif;text-align:right;line-height:normal;'\u003e\u003cspan style='font-size:12px;font-family:\"Times New Roman\",serif;color:black;'\u003e0.735\u003c/span\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"height:3.3pt;border:none;\"\u003e\u003cbr\u003e\u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd style=\"width:45.75pt;border:none;padding:2.15pt 2.15pt 2.15pt 2.15pt;height:3.3pt;\"\u003e\n \u003cp style='margin-top:0in;margin-right:0in;margin-bottom:0in;margin-left:0in;font-size:11.0pt;font-family:\"Calibri\",sans-serif;text-align:right;line-height:normal;'\u003e\u003cspan style='font-size:12px;font-family:\"Times New Roman\",serif;'\u003e19\u003c/span\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width:79.5pt;border:none;padding:2.15pt 2.15pt 2.15pt 2.15pt;height:3.3pt;\"\u003e\n \u003cp style='margin-top:0in;margin-right:0in;margin-bottom:0in;margin-left:0in;font-size:11.0pt;font-family:\"Calibri\",sans-serif;text-align:right;line-height:normal;'\u003e\u003cspan style='font-size:12px;font-family:\"Times New Roman\",serif;'\u003eLag-01 and 06\u003c/span\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width:54.95pt;border:none;background:#FBF2F5;padding: 2.15pt 2.15pt 2.15pt 2.15pt;height:3.3pt;\"\u003e\n \u003cp style='margin-top:0in;margin-right:0in;margin-bottom:0in;margin-left:0in;font-size:11.0pt;font-family:\"Calibri\",sans-serif;text-align:right;line-height:normal;'\u003e\u003cspan style='font-size:12px;font-family:\"Times New Roman\",serif;color:black;'\u003e0.877\u003c/span\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width:54.95pt;border:none;background:#FBF4F7;padding: 2.15pt 2.15pt 2.15pt 2.15pt;height:3.3pt;\"\u003e\n \u003cp style='margin-top:0in;margin-right:0in;margin-bottom:0in;margin-left:0in;font-size:11.0pt;font-family:\"Calibri\",sans-serif;text-align:right;line-height:normal;'\u003e\u003cspan style='font-size:12px;font-family:\"Times New Roman\",serif;color:black;'\u003e0.915\u003c/span\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width:54.95pt;border:none;background:#FBEDF0;padding: 2.15pt 2.15pt 2.15pt 2.15pt;height:3.3pt;\"\u003e\n \u003cp style='margin-top:0in;margin-right:0in;margin-bottom:0in;margin-left:0in;font-size:11.0pt;font-family:\"Calibri\",sans-serif;text-align:right;line-height:normal;'\u003e\u003cspan style='font-size:12px;font-family:\"Times New Roman\",serif;color:black;'\u003e0.613\u003c/span\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width:54.95pt;border:none;background:#FBF7FA;padding: 2.15pt 2.15pt 2.15pt 2.15pt;height:3.3pt;\"\u003e\n \u003cp style='margin-top:0in;margin-right:0in;margin-bottom:0in;margin-left:0in;font-size:11.0pt;font-family:\"Calibri\",sans-serif;text-align:right;line-height:normal;'\u003e\u003cspan style='font-size:12px;font-family:\"Times New Roman\",serif;color:black;'\u003e0.742\u003c/span\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"height:3.3pt;border:none;\"\u003e\u003cbr\u003e\u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd style=\"width:45.75pt;border:none;padding:2.15pt 2.15pt 2.15pt 2.15pt;height:3.3pt;\"\u003e\n \u003cp style='margin-top:0in;margin-right:0in;margin-bottom:0in;margin-left:0in;font-size:11.0pt;font-family:\"Calibri\",sans-serif;text-align:right;line-height:normal;'\u003e\u003cspan style='font-size:12px;font-family:\"Times New Roman\",serif;'\u003e20\u003c/span\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width:79.5pt;border:none;padding:2.15pt 2.15pt 2.15pt 2.15pt;height:3.3pt;\"\u003e\n \u003cp style='margin-top:0in;margin-right:0in;margin-bottom:0in;margin-left:0in;font-size:11.0pt;font-family:\"Calibri\",sans-serif;text-align:right;line-height:normal;'\u003e\u003cspan style='font-size:12px;font-family:\"Times New Roman\",serif;'\u003eP, PA and SPI\u003c/span\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width:54.95pt;border:none;background:#F8696B;padding: 2.15pt 2.15pt 2.15pt 2.15pt;height:3.3pt;\"\u003e\n \u003cp style='margin-top:0in;margin-right:0in;margin-bottom:0in;margin-left:0in;font-size:11.0pt;font-family:\"Calibri\",sans-serif;text-align:right;line-height:normal;'\u003e\u003cspan style='font-size:12px;font-family:\"Times New Roman\",serif;color:black;'\u003e0.764\u003c/span\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width:54.95pt;border:none;background:#F8696B;padding: 2.15pt 2.15pt 2.15pt 2.15pt;height:3.3pt;\"\u003e\n \u003cp style='margin-top:0in;margin-right:0in;margin-bottom:0in;margin-left:0in;font-size:11.0pt;font-family:\"Calibri\",sans-serif;text-align:right;line-height:normal;'\u003e\u003cspan style='font-size:12px;font-family:\"Times New Roman\",serif;color:black;'\u003e0.840\u003c/span\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width:54.95pt;border:none;background:#F8696B;padding: 2.15pt 2.15pt 2.15pt 2.15pt;height:3.3pt;\"\u003e\n \u003cp style='margin-top:0in;margin-right:0in;margin-bottom:0in;margin-left:0in;font-size:11.0pt;font-family:\"Calibri\",sans-serif;text-align:right;line-height:normal;'\u003e\u003cspan style='font-size:12px;font-family:\"Times New Roman\",serif;color:black;'\u003e0.383\u003c/span\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width:54.95pt;border:none;background:#F8696B;padding: 2.15pt 2.15pt 2.15pt 2.15pt;height:3.3pt;\"\u003e\n \u003cp style='margin-top:0in;margin-right:0in;margin-bottom:0in;margin-left:0in;font-size:11.0pt;font-family:\"Calibri\",sans-serif;text-align:right;line-height:normal;'\u003e\u003cspan style='font-size:12px;font-family:\"Times New Roman\",serif;color:black;'\u003e0.639\u003c/span\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"height:3.3pt;border:none;\"\u003e\u003cbr\u003e\u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd style=\"width:45.75pt;border:none;padding:2.15pt 2.15pt 2.15pt 2.15pt;height:3.3pt;\"\u003e\n \u003cp style='margin-top:0in;margin-right:0in;margin-bottom:0in;margin-left:0in;font-size:11.0pt;font-family:\"Calibri\",sans-serif;text-align:right;line-height:normal;'\u003e\u003cspan style='font-size:12px;font-family:\"Times New Roman\",serif;'\u003e21\u003c/span\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width:79.5pt;border:none;padding:2.15pt 2.15pt 2.15pt 2.15pt;height:3.3pt;\"\u003e\n \u003cp style='margin-top:0in;margin-right:0in;margin-bottom:0in;margin-left:0in;font-size:11.0pt;font-family:\"Calibri\",sans-serif;text-align:right;line-height:normal;'\u003e\u003cspan style='font-size:12px;font-family:\"Times New Roman\",serif;'\u003eT, PA and SPI\u003c/span\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width:54.95pt;border:none;background:#FBD8DB;padding: 2.15pt 2.15pt 2.15pt 2.15pt;height:3.3pt;\"\u003e\n \u003cp style='margin-top:0in;margin-right:0in;margin-bottom:0in;margin-left:0in;font-size:11.0pt;font-family:\"Calibri\",sans-serif;text-align:right;line-height:normal;'\u003e\u003cspan style='font-size:12px;font-family:\"Times New Roman\",serif;color:black;'\u003e0.856\u003c/span\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width:54.95pt;border:none;background:#FAD7D9;padding: 2.15pt 2.15pt 2.15pt 2.15pt;height:3.3pt;\"\u003e\n \u003cp style='margin-top:0in;margin-right:0in;margin-bottom:0in;margin-left:0in;font-size:11.0pt;font-family:\"Calibri\",sans-serif;text-align:right;line-height:normal;'\u003e\u003cspan style='font-size:12px;font-family:\"Times New Roman\",serif;color:black;'\u003e0.900\u003c/span\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width:54.95pt;border:none;background:#FACED1;padding: 2.15pt 2.15pt 2.15pt 2.15pt;height:3.3pt;\"\u003e\n \u003cp style='margin-top:0in;margin-right:0in;margin-bottom:0in;margin-left:0in;font-size:11.0pt;font-family:\"Calibri\",sans-serif;text-align:right;line-height:normal;'\u003e\u003cspan style='font-size:12px;font-family:\"Times New Roman\",serif;color:black;'\u003e0.560\u003c/span\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width:54.95pt;border:none;background:#FABEC0;padding: 2.15pt 2.15pt 2.15pt 2.15pt;height:3.3pt;\"\u003e\n \u003cp style='margin-top:0in;margin-right:0in;margin-bottom:0in;margin-left:0in;font-size:11.0pt;font-family:\"Calibri\",sans-serif;text-align:right;line-height:normal;'\u003e\u003cspan style='font-size:12px;font-family:\"Times New Roman\",serif;color:black;'\u003e0.700\u003c/span\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"height:3.3pt;border:none;\"\u003e\u003cbr\u003e\u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd style=\"width:45.75pt;border:none;padding:2.15pt 2.15pt 2.15pt 2.15pt;height:3.3pt;\"\u003e\n \u003cp style='margin-top:0in;margin-right:0in;margin-bottom:0in;margin-left:0in;font-size:11.0pt;font-family:\"Calibri\",sans-serif;text-align:right;line-height:normal;'\u003e\u003cspan style='font-size:12px;font-family:\"Times New Roman\",serif;'\u003e22\u003c/span\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width:79.5pt;border:none;padding:2.15pt 2.15pt 2.15pt 2.15pt;height:3.3pt;\"\u003e\n \u003cp style='margin-top:0in;margin-right:0in;margin-bottom:0in;margin-left:0in;font-size:11.0pt;font-family:\"Calibri\",sans-serif;text-align:right;line-height:normal;'\u003e\u003cspan style='font-size:12px;font-family:\"Times New Roman\",serif;'\u003eT, P and SPI\u003c/span\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width:54.95pt;border:none;background:#A8C1E2;padding: 2.15pt 2.15pt 2.15pt 2.15pt;height:3.3pt;\"\u003e\n \u003cp style='margin-top:0in;margin-right:0in;margin-bottom:0in;margin-left:0in;font-size:11.0pt;font-family:\"Calibri\",sans-serif;text-align:right;line-height:normal;'\u003e\u003cspan style='font-size:12px;font-family:\"Times New Roman\",serif;color:black;'\u003e0.909\u003c/span\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width:54.95pt;border:none;background:#A1BCDF;padding: 2.15pt 2.15pt 2.15pt 2.15pt;height:3.3pt;\"\u003e\n \u003cp style='margin-top:0in;margin-right:0in;margin-bottom:0in;margin-left:0in;font-size:11.0pt;font-family:\"Calibri\",sans-serif;text-align:right;line-height:normal;'\u003e\u003cspan style='font-size:12px;font-family:\"Times New Roman\",serif;color:black;'\u003e0.938\u003c/span\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width:54.95pt;border:none;background:#B3C8E5;padding: 2.15pt 2.15pt 2.15pt 2.15pt;height:3.3pt;\"\u003e\n \u003cp style='margin-top:0in;margin-right:0in;margin-bottom:0in;margin-left:0in;font-size:11.0pt;font-family:\"Calibri\",sans-serif;text-align:right;line-height:normal;'\u003e\u003cspan style='font-size:12px;font-family:\"Times New Roman\",serif;color:black;'\u003e0.715\u003c/span\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width:54.95pt;border:none;background:#A2BDE0;padding: 2.15pt 2.15pt 2.15pt 2.15pt;height:3.3pt;\"\u003e\n \u003cp style='margin-top:0in;margin-right:0in;margin-bottom:0in;margin-left:0in;font-size:11.0pt;font-family:\"Calibri\",sans-serif;text-align:right;line-height:normal;'\u003e\u003cspan style='font-size:12px;font-family:\"Times New Roman\",serif;color:black;'\u003e0.770\u003c/span\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"height:3.3pt;border:none;\"\u003e\u003cbr\u003e\u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd style=\"width:45.75pt;border:none;padding:2.15pt 2.15pt 2.15pt 2.15pt;height:3.3pt;\"\u003e\n \u003cp style='margin-top:0in;margin-right:0in;margin-bottom:0in;margin-left:0in;font-size:11.0pt;font-family:\"Calibri\",sans-serif;text-align:right;line-height:normal;'\u003e\u003cspan style='font-size:12px;font-family:\"Times New Roman\",serif;'\u003e23\u003c/span\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width:79.5pt;border:none;padding:2.15pt 2.15pt 2.15pt 2.15pt;height:3.3pt;\"\u003e\n \u003cp style='margin-top:0in;margin-right:0in;margin-bottom:0in;margin-left:0in;font-size:11.0pt;font-family:\"Calibri\",sans-serif;text-align:right;line-height:normal;'\u003e\u003cspan style='font-size:12px;font-family:\"Times New Roman\",serif;'\u003eT, P and PA\u003c/span\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width:54.95pt;border:none;background:#FBEDF0;padding: 2.15pt 2.15pt 2.15pt 2.15pt;height:3.3pt;\"\u003e\n \u003cp style='margin-top:0in;margin-right:0in;margin-bottom:0in;margin-left:0in;font-size:11.0pt;font-family:\"Calibri\",sans-serif;text-align:right;line-height:normal;'\u003e\u003cspan style='font-size:12px;font-family:\"Times New Roman\",serif;color:black;'\u003e0.873\u003c/span\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width:54.95pt;border:none;background:#FBF0F3;padding: 2.15pt 2.15pt 2.15pt 2.15pt;height:3.3pt;\"\u003e\n \u003cp style='margin-top:0in;margin-right:0in;margin-bottom:0in;margin-left:0in;font-size:11.0pt;font-family:\"Calibri\",sans-serif;text-align:right;line-height:normal;'\u003e\u003cspan style='font-size:12px;font-family:\"Times New Roman\",serif;color:black;'\u003e0.913\u003c/span\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width:54.95pt;border:none;background:#FBE8EA;padding: 2.15pt 2.15pt 2.15pt 2.15pt;height:3.3pt;\"\u003e\n \u003cp style='margin-top:0in;margin-right:0in;margin-bottom:0in;margin-left:0in;font-size:11.0pt;font-family:\"Calibri\",sans-serif;text-align:right;line-height:normal;'\u003e\u003cspan style='font-size:12px;font-family:\"Times New Roman\",serif;color:black;'\u003e0.604\u003c/span\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width:54.95pt;border:none;background:#FBE4E7;padding: 2.15pt 2.15pt 2.15pt 2.15pt;height:3.3pt;\"\u003e\n \u003cp style='margin-top:0in;margin-right:0in;margin-bottom:0in;margin-left:0in;font-size:11.0pt;font-family:\"Calibri\",sans-serif;text-align:right;line-height:normal;'\u003e\u003cspan style='font-size:12px;font-family:\"Times New Roman\",serif;color:black;'\u003e0.728\u003c/span\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"height:3.3pt;border:none;\"\u003e\u003cbr\u003e\u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd style=\"width:45.75pt;border:none;padding:2.15pt 2.15pt 2.15pt 2.15pt;height:3.3pt;\"\u003e\n \u003cp style='margin-top:0in;margin-right:0in;margin-bottom:0in;margin-left:0in;font-size:11.0pt;font-family:\"Calibri\",sans-serif;text-align:right;line-height:normal;'\u003e\u003cspan style='font-size:12px;font-family:\"Times New Roman\",serif;'\u003e24\u003c/span\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width:79.5pt;border:none;padding:2.15pt 2.15pt 2.15pt 2.15pt;height:3.3pt;\"\u003e\n \u003cp style='margin-top:0in;margin-right:0in;margin-bottom:0in;margin-left:0in;font-size:11.0pt;font-family:\"Calibri\",sans-serif;text-align:right;line-height:normal;'\u003e\u003cspan style='font-size:12px;font-family:\"Times New Roman\",serif;'\u003eLag-03, 06 and 12\u003c/span\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width:54.95pt;border:none;background:#F87D7F;padding: 2.15pt 2.15pt 2.15pt 2.15pt;height:3.3pt;\"\u003e\n \u003cp style='margin-top:0in;margin-right:0in;margin-bottom:0in;margin-left:0in;font-size:11.0pt;font-family:\"Calibri\",sans-serif;text-align:right;line-height:normal;'\u003e\u003cspan style='font-size:12px;font-family:\"Times New Roman\",serif;color:black;'\u003e0.781\u003c/span\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width:54.95pt;border:none;background:#F87D7F;padding: 2.15pt 2.15pt 2.15pt 2.15pt;height:3.3pt;\"\u003e\n \u003cp style='margin-top:0in;margin-right:0in;margin-bottom:0in;margin-left:0in;font-size:11.0pt;font-family:\"Calibri\",sans-serif;text-align:right;line-height:normal;'\u003e\u003cspan style='font-size:12px;font-family:\"Times New Roman\",serif;color:black;'\u003e0.852\u003c/span\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width:54.95pt;border:none;background:#F8787A;padding: 2.15pt 2.15pt 2.15pt 2.15pt;height:3.3pt;\"\u003e\n \u003cp style='margin-top:0in;margin-right:0in;margin-bottom:0in;margin-left:0in;font-size:11.0pt;font-family:\"Calibri\",sans-serif;text-align:right;line-height:normal;'\u003e\u003cspan style='font-size:12px;font-family:\"Times New Roman\",serif;color:black;'\u003e0.410\u003c/span\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width:54.95pt;border:none;background:#F88A8D;padding: 2.15pt 2.15pt 2.15pt 2.15pt;height:3.3pt;\"\u003e\n \u003cp style='margin-top:0in;margin-right:0in;margin-bottom:0in;margin-left:0in;font-size:11.0pt;font-family:\"Calibri\",sans-serif;text-align:right;line-height:normal;'\u003e\u003cspan style='font-size:12px;font-family:\"Times New Roman\",serif;color:black;'\u003e0.663\u003c/span\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"height:3.3pt;border:none;\"\u003e\u003cbr\u003e\u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd style=\"width:45.75pt;border:none;padding:2.15pt 2.15pt 2.15pt 2.15pt;height:3.3pt;\"\u003e\n \u003cp style='margin-top:0in;margin-right:0in;margin-bottom:0in;margin-left:0in;font-size:11.0pt;font-family:\"Calibri\",sans-serif;text-align:right;line-height:normal;'\u003e\u003cspan style='font-size:12px;font-family:\"Times New Roman\",serif;'\u003e25\u003c/span\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width:79.5pt;border:none;padding:2.15pt 2.15pt 2.15pt 2.15pt;height:3.3pt;\"\u003e\n \u003cp style='margin-top:0in;margin-right:0in;margin-bottom:0in;margin-left:0in;font-size:11.0pt;font-family:\"Calibri\",sans-serif;text-align:right;line-height:normal;'\u003e\u003cspan style='font-size:12px;font-family:\"Times New Roman\",serif;'\u003eLag-01, 06 and 12\u003c/span\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width:54.95pt;border:none;background:#F99497;padding: 2.15pt 2.15pt 2.15pt 2.15pt;height:3.3pt;\"\u003e\n \u003cp style='margin-top:0in;margin-right:0in;margin-bottom:0in;margin-left:0in;font-size:11.0pt;font-family:\"Calibri\",sans-serif;text-align:right;line-height:normal;'\u003e\u003cspan style='font-size:12px;font-family:\"Times New Roman\",serif;color:black;'\u003e0.800\u003c/span\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width:54.95pt;border:none;background:#F99395;padding: 2.15pt 2.15pt 2.15pt 2.15pt;height:3.3pt;\"\u003e\n \u003cp style='margin-top:0in;margin-right:0in;margin-bottom:0in;margin-left:0in;font-size:11.0pt;font-family:\"Calibri\",sans-serif;text-align:right;line-height:normal;'\u003e\u003cspan style='font-size:12px;font-family:\"Times New Roman\",serif;color:black;'\u003e0.863\u003c/span\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width:54.95pt;border:none;background:#F88A8C;padding: 2.15pt 2.15pt 2.15pt 2.15pt;height:3.3pt;\"\u003e\n \u003cp style='margin-top:0in;margin-right:0in;margin-bottom:0in;margin-left:0in;font-size:11.0pt;font-family:\"Calibri\",sans-serif;text-align:right;line-height:normal;'\u003e\u003cspan style='font-size:12px;font-family:\"Times New Roman\",serif;color:black;'\u003e0.441\u003c/span\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width:54.95pt;border:none;background:#F99DA0;padding: 2.15pt 2.15pt 2.15pt 2.15pt;height:3.3pt;\"\u003e\n \u003cp style='margin-top:0in;margin-right:0in;margin-bottom:0in;margin-left:0in;font-size:11.0pt;font-family:\"Calibri\",sans-serif;text-align:right;line-height:normal;'\u003e\u003cspan style='font-size:12px;font-family:\"Times New Roman\",serif;color:black;'\u003e0.677\u003c/span\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"height:3.3pt;border:none;\"\u003e\u003cbr\u003e\u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd style=\"width:45.75pt;border:none;padding:2.15pt 2.15pt 2.15pt 2.15pt;height:3.3pt;\"\u003e\n \u003cp style='margin-top:0in;margin-right:0in;margin-bottom:0in;margin-left:0in;font-size:11.0pt;font-family:\"Calibri\",sans-serif;text-align:right;line-height:normal;'\u003e\u003cspan style='font-size:12px;font-family:\"Times New Roman\",serif;'\u003e26\u003c/span\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width:79.5pt;border:none;padding:2.15pt 2.15pt 2.15pt 2.15pt;height:3.3pt;\"\u003e\n \u003cp style='margin-top:0in;margin-right:0in;margin-bottom:0in;margin-left:0in;font-size:11.0pt;font-family:\"Calibri\",sans-serif;text-align:right;line-height:normal;'\u003e\u003cspan style='font-size:12px;font-family:\"Times New Roman\",serif;'\u003eLag-01, 03 and 12\u003c/span\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width:54.95pt;border:none;background:#F99EA0;padding: 2.15pt 2.15pt 2.15pt 2.15pt;height:3.3pt;\"\u003e\n \u003cp style='margin-top:0in;margin-right:0in;margin-bottom:0in;margin-left:0in;font-size:11.0pt;font-family:\"Calibri\",sans-serif;text-align:right;line-height:normal;'\u003e\u003cspan style='font-size:12px;font-family:\"Times New Roman\",serif;color:black;'\u003e0.808\u003c/span\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width:54.95pt;border:none;background:#F9A0A3;padding: 2.15pt 2.15pt 2.15pt 2.15pt;height:3.3pt;\"\u003e\n \u003cp style='margin-top:0in;margin-right:0in;margin-bottom:0in;margin-left:0in;font-size:11.0pt;font-family:\"Calibri\",sans-serif;text-align:right;line-height:normal;'\u003e\u003cspan style='font-size:12px;font-family:\"Times New Roman\",serif;color:black;'\u003e0.870\u003c/span\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width:54.95pt;border:none;background:#F99294;padding: 2.15pt 2.15pt 2.15pt 2.15pt;height:3.3pt;\"\u003e\n \u003cp style='margin-top:0in;margin-right:0in;margin-bottom:0in;margin-left:0in;font-size:11.0pt;font-family:\"Calibri\",sans-serif;text-align:right;line-height:normal;'\u003e\u003cspan style='font-size:12px;font-family:\"Times New Roman\",serif;color:black;'\u003e0.455\u003c/span\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width:54.95pt;border:none;background:#F99698;padding: 2.15pt 2.15pt 2.15pt 2.15pt;height:3.3pt;\"\u003e\n \u003cp style='margin-top:0in;margin-right:0in;margin-bottom:0in;margin-left:0in;font-size:11.0pt;font-family:\"Calibri\",sans-serif;text-align:right;line-height:normal;'\u003e\u003cspan style='font-size:12px;font-family:\"Times New Roman\",serif;color:black;'\u003e0.672\u003c/span\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"height:3.3pt;border:none;\"\u003e\u003cbr\u003e\u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd style=\"width:45.75pt;border:none;border-bottom:solid black 1.0pt;padding:2.15pt 2.15pt 2.15pt 2.15pt;height:3.3pt;\"\u003e\n \u003cp style='margin-top:0in;margin-right:0in;margin-bottom:0in;margin-left:0in;font-size:11.0pt;font-family:\"Calibri\",sans-serif;text-align:right;line-height:normal;'\u003e\u003cspan style='font-size:12px;font-family:\"Times New Roman\",serif;'\u003e27\u003c/span\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width:79.5pt;border:none;border-bottom:solid black 1.0pt;padding:2.15pt 2.15pt 2.15pt 2.15pt;height:3.3pt;\"\u003e\n \u003cp style='margin-top:0in;margin-right:0in;margin-bottom:0in;margin-left:0in;font-size:11.0pt;font-family:\"Calibri\",sans-serif;text-align:right;line-height:normal;'\u003e\u003cspan style='font-size:12px;font-family:\"Times New Roman\",serif;'\u003eLag-01, 03 and 06\u003c/span\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width:54.95pt;border:none;border-bottom:solid black 1.0pt;background:#F88C8E;padding:2.15pt 2.15pt 2.15pt 2.15pt;height:3.3pt;\"\u003e\n \u003cp style='margin-top:0in;margin-right:0in;margin-bottom:0in;margin-left:0in;font-size:11.0pt;font-family:\"Calibri\",sans-serif;text-align:right;line-height:normal;'\u003e\u003cspan style='font-size:12px;font-family:\"Times New Roman\",serif;color:black;'\u003e0.793\u003c/span\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width:54.95pt;border:none;border-bottom:solid black 1.0pt;background:#F88A8C;padding:2.15pt 2.15pt 2.15pt 2.15pt;height:3.3pt;\"\u003e\n \u003cp style='margin-top:0in;margin-right:0in;margin-bottom:0in;margin-left:0in;font-size:11.0pt;font-family:\"Calibri\",sans-serif;text-align:right;line-height:normal;'\u003e\u003cspan style='font-size:12px;font-family:\"Times New Roman\",serif;color:black;'\u003e0.858\u003c/span\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width:54.95pt;border:none;border-bottom:solid black 1.0pt;background:#F88385;padding:2.15pt 2.15pt 2.15pt 2.15pt;height:3.3pt;\"\u003e\n \u003cp style='margin-top:0in;margin-right:0in;margin-bottom:0in;margin-left:0in;font-size:11.0pt;font-family:\"Calibri\",sans-serif;text-align:right;line-height:normal;'\u003e\u003cspan style='font-size:12px;font-family:\"Times New Roman\",serif;color:black;'\u003e0.428\u003c/span\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width:54.95pt;border:none;border-bottom:solid black 1.0pt;background:#F88487;padding:2.15pt 2.15pt 2.15pt 2.15pt;height:3.3pt;\"\u003e\n \u003cp style='margin-top:0in;margin-right:0in;margin-bottom:0in;margin-left:0in;font-size:11.0pt;font-family:\"Calibri\",sans-serif;text-align:right;line-height:normal;'\u003e\u003cspan style='font-size:12px;font-family:\"Times New Roman\",serif;color:black;'\u003e0.659\u003c/span\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"height:3.3pt;border:none;\"\u003e\u003cbr\u003e\u003c/td\u003e\n \u003c/tr\u003e\n \u003c/tbody\u003e\n \u003c/table\u003e\n \u003c/div\u003e\n\u003c/div\u003e\n\u003cp\u003eFor example, Fig.\u0026nbsp;\u003cspan\u003e7\u003c/span\u003e illustrates the precipitation variable P-12 (accumulated precipitation in 12 months) for January 2022, calculated from the integration of CHIRPS and ECMWF data, according to the forecast lead time. We emphasize the visual similarity between the map of 12-month accumulated precipitation (CHIRPS dataset only), Fig.\u0026nbsp;\u003cspan\u003e7\u003c/span\u003eA, and the other four maps, including predicted precipitation from the ECMWF-SEAS5 model (Figs.\u0026nbsp;\u003cspan\u003e7\u003c/span\u003eB-\u003cspan\u003e7\u003c/span\u003eE). Overall, combining the two datasets did not significantly change long-term precipitation accumulation. For January 2022, for example, the linear correlation was 0.956 between the 12-month observed accumulated precipitation (CHIRPS: 02/2021-01/2022), Fig.\u0026nbsp;\u003cspan\u003e7\u003c/span\u003eA, and the version of the same variable with 4-month forecasted precipitation (CHIRPS: 02/2021-09/2021; ECMWF: 10/2021-01/2022), Fig.\u0026nbsp;\u003cspan\u003e7\u003c/span\u003eE. The linear correlation coefficient was 0.993 for the 1-month forecasted precipitation, Fig.\u0026nbsp;\u003cspan\u003e7\u003c/span\u003eB.\u003c/p\u003e\n\u003cp\u003eThe mean absolute difference in Brazilian municipalities varied between 49.3 mm (1-month forecast) and 151.7 mm (4-months forecast), considering the 12-month accumulated precipitation (P-12). While the 90th percentile error of the P12 variable calculation is 114.9 mm for a 1-month forecast, the index increases to 297.3 mm when we include the 4-month seasonal forecast from the ECMWF-SEAS5. Overall, as expected, as forecasted precipitation data is added to the 12-month accumulated total, the uncertainty increases, which should have an impact on the drought forecast made by the ANN model.\u003c/p\u003e\n\u003cp\u003eWe present in Fig.\u0026nbsp;\u003cspan\u003e8\u003c/span\u003e some examples of the application of the ANN model for January 2022, when a severe drought occurred especially in the southern region of Brazil (Fig.\u0026nbsp;\u003cspan\u003e8\u003c/span\u003eA, drought recorded in the S2iD database). In Fig.\u0026nbsp;\u003cspan\u003e8\u003c/span\u003eB we present the output of the model in a simulation scenario (CHIRPS\u0026thinsp;+\u0026thinsp;NCEP), while in Figs.\u0026nbsp;\u003cspan\u003e8\u003c/span\u003eC-\u003cspan\u003e8\u003c/span\u003eF we illustrate the ANN output in forecasts with lead times of 1 to 4-months (adding ECMWF-SEAS5). The maps indicate a drought index with continuous values in the interval [0\u0026ndash;1] for all Brazilian municipalities. In general, the ANN model was successful in predicting drought with the application of seasonal forecasting. The forecast performance decreased according to the forecast lead time, as expected, but we can see that the ANN model satisfactorily predicted droughts up to 2 months in advance. We observed from the visual analysis of the results presented in Fig.\u0026nbsp;\u003cspan\u003e8\u003c/span\u003e that the model was able to indicate drought in the municipalities of the Northeast and South of Brazil.\u003c/p\u003e\n\u003cp\u003eOn the other hand, we found that the predictive capacity of the model for forecasting droughts at the municipal level with a lead time of 3 or 4 months is lower. In the example in Fig.\u0026nbsp;\u003cspan\u003e8\u003c/span\u003e, the drought in southern Brazil would have been underestimated with this anticipation of the forecast. However, we emphasize that the spatial pattern of the forecast is consistent with the observed map, and can regionally reproduce areas with a greater propensity for the occurrence of droughts.\u003c/p\u003e\n\u003cp\u003eThe results of the ANN model can also be presented using a binary classification, considering the threshold that presented the highest accuracy in the model training sample set (Fig.\u0026nbsp;\u003cspan\u003e9\u003c/span\u003e). We observed a high degree of agreement between the drought observed in January 2022 (Fig.\u0026nbsp;\u003cspan\u003e9\u003c/span\u003eA) and the scenarios simulated (Fig.\u0026nbsp;\u003cspan\u003e9\u003c/span\u003eB) and predicted 1 month in advance (Fig.\u0026nbsp;\u003cspan\u003e9\u003c/span\u003eC). This indicates the high capability of correctly allocating municipalities with drought by the trained ANN model. The precise allocation of municipalities facing drought can also be considered good for a 2-month forecast (Fig.\u0026nbsp;\u003cspan\u003e9\u003c/span\u003eD), significantly reducing it for 3 and 4 months, as previously mentioned.\u003c/p\u003e\n\u003cp\u003eIt is important to mention that this binary classification would not need to be used in future forecasting situations, as it is necessary only to provide model validation statistics, allowing for a quantitative assessment of the model in a forecast scenario. Once the model is validated, and its accuracy and precision are known, the ANN model can be used for forecasting considering the continuous interval [0\u0026ndash;1], or another classification criterion, for example, in levels from low to high risk of drought occurrence in Brazilian municipalities.\u003c/p\u003e\n\u003cp\u003eWe show in Table\u0026nbsp;\u003cspan\u003e5\u003c/span\u003ea summary of the performance metrics of the best ANN model for drought forecasting in Brazilian municipalities, considering only the testing period between January 2021 and December 2022. The overall accuracy of drought prediction ranged from 0.922 (lead time 1) to 0.757 (lead time 4). The ANN model consistently outperformed the persistent model, which would predict that current conditions will persist in the later time interval.\u003c/p\u003e\n\u003cdiv\u003e\n \u003ctable id=\"Tab5\" border=\"1\"\u003e\n \u003ccaption language=\"En\"\u003e\n \u003cdiv\u003eTable 5\u003c/div\u003e\n \u003cdiv\u003e\n \u003cp\u003ePerformance metrics of the best ANN model for drought forecast in Brazilian municipalities, testing period between January 2021 and December 2022\u003c/p\u003e\n \u003c/div\u003e\n \u003c/caption\u003e\n \u003cthead\u003e\n \u003ctr\u003e\n \u003cth align=\"left\" colspan=\"2\"\u003e\n \u003cp\u003eANN prediction and lead times\u003c/p\u003e\n \u003c/th\u003e\n \u003cth align=\"left\"\u003e\n \u003cp\u003eOverall Accuracy (OA)\u003c/p\u003e\n \u003c/th\u003e\n \u003cth align=\"left\"\u003e\n \u003cp\u003eArea Under the ROC Curve (AUC)\u003c/p\u003e\n \u003c/th\u003e\n \u003cth align=\"left\"\u003e\n \u003cp\u003ePrecision\u003c/p\u003e\n \u003c/th\u003e\n \u003cth align=\"left\"\u003e\n \u003cp\u003eRecall\u003c/p\u003e\n \u003c/th\u003e\n \u003c/tr\u003e\n \u003c/thead\u003e\n \u003ctbody\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003eSimulation\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003et\u0026thinsp;+\u0026thinsp;0\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e0.931\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e0.952\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e0.805\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e0.784\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\" rowspan=\"4\"\u003e\n \u003cp\u003eForecasting\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003et\u0026thinsp;+\u0026thinsp;1\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e0.922\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e0.944\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e0.766\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e0.751\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003et\u0026thinsp;+\u0026thinsp;2\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e0.868\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e0.883\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e0.612\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e0.550\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003et\u0026thinsp;+\u0026thinsp;3\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e0.801\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e0.827\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e0.337\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e0.265\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003et\u0026thinsp;+\u0026thinsp;4\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e0.757\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e0.785\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e0.209\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e0.146\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003c/tbody\u003e\n \u003c/table\u003e\n\u003c/div\u003e\n\u003cp\u003eThe ANN model in the 1-month forecast scenario showed an accuracy greater than 0.86 in all months of the test period, with values always close to the simulated scenario (without the inclusion of seasonal forecast data from the ECMWF model). We observed that in forecasts with a lead time of 2 months, the model presented OA equal to 0.868 (the entire test period), with values always higher than 0.8 in all months of the analyzed period. This result reinforces what is visually observed in the maps, indicating that the ANN model, combined with ECMWF data, has a good forecasting capacity for this lead time.\u003c/p\u003e\n\u003cp\u003eThe AUC index of the drought forecast ranged from 0.944 (lead time 1) to 0.785 (lead time 4). In all months, the AUC was greater than 0.82 considering forecasts up to 2 months in advance. Although the OA and AUC values indicate that the ANN model makes correct predictions in lead times of 3 and 4 months, we observed a strong reduction in precision and recall indices, with values below 0.5 (Table\u0026nbsp;\u003cspan\u003e5\u003c/span\u003e). The OA and AUC values are strongly influenced by the ability of the ANN model to correctly indicate the municipalities where there will be no drought. This observation is important, as model allocation performance must be analyzed carefully, based on both visual analysis of the prediction and other performance metrics. In Fig.\u0026nbsp;\u003cspan\u003e10\u003c/span\u003e we present the difference maps between observed and predicted droughts (January 2022), which exemplifies the reduction in precision and recall for forecasts with lead times of 3 and 4 months.\u003c/p\u003e\n\u003cp\u003eThe analysis of the difference map for the 1-month ahead prediction (Fig.\u0026nbsp;\u003cspan\u003e10\u003c/span\u003eA) demonstrates the satisfactory allocation quality of the model. The model showed an accuracy of 0.928, precision of 0.803, and recall of 0.792, maintaining the observed performance metrics in the simulation. When analyzing the difference map for the 2-month lead time (Fig.\u0026nbsp;\u003cspan\u003e10\u003c/span\u003eB), we observed an increase in errors, but still with satisfactory performance: accuracy of 0.87, precision of 0.67, and recall of 0.55.\u003c/p\u003e"},{"header":"Discussion","content":"\u003cp\u003ePrecipitation and temperature anomalies are responsible for the occurrence of several types of natural disasters (Nandgude et al. \u003cspan citationid=\"CR45\" class=\"CitationRef\"\u003e2023\u003c/span\u003e). In this context, Vicente-Serrano et al. (\u003cspan citationid=\"CR66\" class=\"CitationRef\"\u003e2020\u003c/span\u003e) addressed future climate change scenarios that suggest an increase in the severity of droughts across the world. As a result, it becomes even more important to monitor, model and predict droughts, seeking to mitigate the effects on societies and the environment (Tanguy et al. \u003cspan citationid=\"CR62\" class=\"CitationRef\"\u003e2023\u003c/span\u003e). Predicting droughts with the potential to cause losses is extremely challenging due to the involvement of several hydrometeorological factors, in addition to local characteristics, such as crop seasonality, land use management, irrigation techniques, among others (Musolino et al. \u003cspan citationid=\"CR44\" class=\"CitationRef\"\u003e2018\u003c/span\u003e). In our study, we trained artificial neural network models for this purpose, predicting droughts with the potential to cause losses as early as possible.\u003c/p\u003e \u003cp\u003eIn reviewing deep learning models for drought prediction, Gyaneshwar et al. (\u003cspan citationid=\"CR20\" class=\"CitationRef\"\u003e2023\u003c/span\u003e) summarize that the spatial and temporal variability of precipitation is seen as a factor that impairs the performance of models, due to its random components. In our study, in general, the precipitation variables calculated from the CHIRPS product showed a significant correlation with the occurrence of droughts in Brazil. These results are corroborated by Wu et al. (\u003cspan citationid=\"CR68\" class=\"CitationRef\"\u003e2019\u003c/span\u003e) and Sousa et al. (\u003cspan citationid=\"CR57\" class=\"CitationRef\"\u003e2023\u003c/span\u003e) in which CHIRPS data performed well for estimating monthly precipitation, and considered adequate to capture the spatial distribution of rainfall.\u003c/p\u003e \u003cp\u003eOther researchers use CHIRPS data in specific studies and have concluded that this dataset can capture the characteristics of droughts. Pandey et al. (2022) performed a comparative analysis of three satellite precipitation products (TRMM-3B43 V7, PERSIANN-CDR and CHIRPS V2) and observed that CHIRPS data stands out compared to the others. Gao et al. (\u003cspan citationid=\"CR18\" class=\"CitationRef\"\u003e2018\u003c/span\u003e) investigated the applicability of the CHIRPS product for drought monitoring using the Standardized Precipitation Index (SPI). The results indicated good performance on multiple temporal scales (monthly, seasonal and annual). The researchers concluded that the product can be used to estimate precipitation and monitor droughts in the study area. Similarly, Guo et al. (\u003cspan citationid=\"CR19\" class=\"CitationRef\"\u003e2017\u003c/span\u003e) performed a meteorological analysis of droughts using the CHIRPS dataset and suggest that the product can adequately capture drought characteristics at multiple time scales, with the best performance at the three-month scale.\u003c/p\u003e \u003cp\u003eThe results we present indicate that precipitation variables are the most explanatory for the occurrence of droughts in Brazilian municipalities, especially the precipitation anomaly. In research related to droughts in California, Luo et al. (\u003cspan citationid=\"CR37\" class=\"CitationRef\"\u003e2017\u003c/span\u003e) investigated the influence of precipitation deficits and temperature anomalies on drought development. The study revealed that precipitation deficits have been largely responsible for extreme agricultural drought.\u003c/p\u003e \u003cp\u003eAnother result that we highlight refers to the importance of considering different time scales when calculating precipitation indices. In our study we used lag times of 1, 3, 6 and 12 months to calculate the variables of accumulated precipitation (P), precipitation anomaly (PA) and Standardized Precipitation Index (SPI). When removing any of the temporal scales we observe that there is degradation in forecast performance, especially when we remove the lag times of 3 and 6 months. Shahid (\u003cspan citationid=\"CR55\" class=\"CitationRef\"\u003e2010\u003c/span\u003e) and Yaseen et al. (\u003cspan citationid=\"CR69\" class=\"CitationRef\"\u003e2021\u003c/span\u003e) indicate that SPI-1, for example, reveals short-period precipitation deficits that directly affect ecology, air temperature and public health. The 3-month precipitation indices provide a characterization of water availability in the medium term, being appropriate in agricultural areas as an indicator of soil moisture (WMO 2008; Alamgir et al. \u003cspan citationid=\"CR2\" class=\"CitationRef\"\u003e2019\u003c/span\u003e; Achour et al. \u003cspan citationid=\"CR1\" class=\"CitationRef\"\u003e2020\u003c/span\u003e). The 6-month indices provide an indication of precipitation trends over a season (Khan et al. 2008), while the 12-month precipitation indices should be considered for long-term estimates for hydrological droughts and water resource management, being indicators of the decline of river flow and groundwater levels (Achour et al. \u003cspan citationid=\"CR1\" class=\"CitationRef\"\u003e2020\u003c/span\u003e; Yaseen et al. \u003cspan citationid=\"CR69\" class=\"CitationRef\"\u003e2021\u003c/span\u003e).\u003c/p\u003e \u003cp\u003eOverall, the developed ANN model was successful in drought prediction. We observed that the model predicted droughts with very good accuracy in a 2-month lead time, which was possible using the seasonal prediction system of ECMWF-SEAS5. The quality of data from the ECMWF-SEAS5 model has already been demonstrated in some previous studies. Ferreira et al. (\u003cspan citationid=\"CR17\" class=\"CitationRef\"\u003e2022\u003c/span\u003e) evaluated the model's seasonal temperature and precipitation predictions for South America. The results showed that ECMWF-SEAS5 climate predictions are potentially useful and should be considered for planning activities that depend on hydrometeorological dynamics. Boas et al. (\u003cspan citationid=\"CR5\" class=\"CitationRef\"\u003e2023\u003c/span\u003e) developed a land surface modeling approach for seasonal soil moisture and crop yield prediction using SEAS5. The forecasting experiments were able to satisfactorily capture the interannual variations recorded in crop yields. Sutanto et al. (\u003cspan citationid=\"CR61\" class=\"CitationRef\"\u003e2020b\u003c/span\u003e) evaluated the hydro-meteorological drought forecast skill for the pan-European region. Hydrological drought forecasts show good predictive power more than 2 months in advance. The authors suggest that the development of seasonal hydrological drought forecasting systems is beneficial, including the use of data from ECMWF-SEAS5. Busker et al. (\u003cspan citationid=\"CR8\" class=\"CitationRef\"\u003e2023\u003c/span\u003e) investigated the potential for early action for droughts by using seasonal forecasts from the SEAS5. The researchers conclude that early action using ECMWF-SEAS5 forecasts can reduce drought impacts.\u003c/p\u003e \u003cp\u003eThe proposed methodology proved to be adequate and our results indicate robust potential in the integration between satellite precipitation data, seasonal climate forecasting and a model based on artificial neural networks for drought forecasting and warning. According to Dikshit et al. (\u003cspan citationid=\"CR13\" class=\"CitationRef\"\u003e2022\u003c/span\u003e) as droughts are inherently non-linear and have many explanatory variables, ANN models have the ability to capture this dynamic relationship efficiently.\u003c/p\u003e \u003cp\u003eIn this study, the overall accuracy of the ANN model for drought simulation was 0.931. The forecast accuracy ranged from 0.922 for a 1-month lead time to 0.757 for 4 months. The AUC index of the drought forecast ranged from 0.944 (lead time 1) to 0.785 (lead time 4). In all months, the AUC was greater than 0.82 considering forecasts up to 2 months in advance. The performance metrics obtained by us are similar to previous scientific studies on the same topic. Mohamadi et al. (\u003cspan citationid=\"CR41\" class=\"CitationRef\"\u003e2020\u003c/span\u003e) developed a zoning map for drought prediction (SPI-3) using machine learning models and nomadic people optimization algorithm (NPA). The Adaptive Neuro-Fuzzy Interface System (ANFIS\u0026ndash;NPA) model obtained an AUC value equal to 0.92 for extreme droughts. Sutanto et al. (\u003cspan citationid=\"CR60\" class=\"CitationRef\"\u003e2020a\u003c/span\u003e) analyzed the skill of large-scale seasonal drought impact forecasts. The researchers developed drought impact functions using machine learning approaches to predict drought impacts with lead times of 1 to 7 months. The AUC values presented by the authors are compatible with those from our study. Dikshit et al. (\u003cspan citationid=\"CR12\" class=\"CitationRef\"\u003e2021\u003c/span\u003e) presented a drought prediction approach (SPEI index) using the long short-term memory neural network (LSTM). The authors obtained AUC values of 0.83 and 0.82 for SPEI-1 and SPEI-3, respectively.\u003c/p\u003e \u003cp\u003eWe observed a progressive decrease in precision and recall values as the forecast lead time increased. This reduction in model performance must be mainly associated with the limitations of the seasonal weather forecast model for specific applications at the local level. For Nandgude et al. (\u003cspan citationid=\"CR45\" class=\"CitationRef\"\u003e2023\u003c/span\u003e) it may be important to consider external driving forces for the occurrence of droughts, such as Sea Surface Temperature (SST) anomalies and region-specific climate indices, such as the Pacific Decadal Oscillation (PDO), the El Ni\u0026ntilde;o-Southern Oscillation (ENSO), and the Indian Ocean Dipole (IOD). The authors indicate that the use of these climate indices can improve drought forecasting results.\u003c/p\u003e \u003cp\u003eAlthough the results presented by us are satisfactory and promising, we emphasize that caution should be taken when using the ANN model to forecast droughts using only the binary classification (0\u0026thinsp;=\u0026thinsp;no drought; 1\u0026thinsp;=\u0026thinsp;drought) for seasonal forecasts of 3 or 4 months ahead. On the other hand, the model reproduced the spatial pattern of droughts, especially when the output is interpreted as a continuous index of drought risk. We can affirm from the results obtained in this study that the trained model is efficient and the results indicate strong potential for drought forecasting and warning, using ANN, remote sensing data, hydrometeorological and climate indexes.\u003c/p\u003e"},{"header":"Conclusions","content":"\u003cp\u003eIn this study, we developed a drought forecast model in Brazilian municipalities based on the integration between remote sensing data, hydro-meteorological and climate indexes, and ANN-based model. We used an inventory of droughts that caused losses by municipalities to generate the model's dependent variable. After model training, we tested the ANN for drought forecasts for lead times of 1\u0026ndash;4 months, using seasonal forecast data from ECMWF-SEAS5.\u003c/p\u003e \u003cp\u003eThe overall accuracy of the ANN model for drought simulation was 0.931. In turn, the forecast accuracy ranged from 0.922 for a 1-month lead time to 0.757 for 4 months. We can affirm that the ANN model consistently outperformed the persistent model. Additionally, we observed a progressive decrease in precision and recall values as the forecast lead time increased.\u003c/p\u003e \u003cp\u003eRemarkably, the model reproduced the spatial pattern of droughts, especially when the output is interpreted as a continuous index of drought risk. We conclude that the trained model is efficient and the results indicate strong potential for drought forecasting and warning, using ANN, remote sensing data, hydrometeorological and climate indexes.\u003c/p\u003e \u003cp\u003eFinally, we conclude that the model trained in this study can be applied for forecasting and warning of droughts at any time of the year for the entire Brazilian territory. The methodological approach has the following main advantages: (i) it indicates with very good accuracy the municipalities with the potential to present losses due to droughts; (ii) does not require spatial interpolation to generate precipitation or temperature data; (iii) and the practicality of accessing data and processing the forecast, which can be carried out using computer codes, making processing more agile.\u003c/p\u003e"},{"header":"Declarations","content":"\u003cp\u003e\u0026nbsp;\u003cstrong\u003eAcknowledgements\u003c/strong\u003e\u003cstrong\u003e\u0026nbsp;\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eWe are grateful for the support of the National Council for Scientific and Technological Development (Conselho Nacional de Desenvolvimento Cient\u0026iacute;fico e Tecnol\u0026oacute;gico - CNPq), with financial support to the first author through CNPq Call No. 09/2022 - Research Productivity, process 311009/2022-0.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eFunding\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003ePartial financial support was received from National Council for Scientific and Technological Development (Conselho Nacional de Desenvolvimento Cient\u0026iacute;fico e Tecnol\u0026oacute;gico - CNPq), with financial support to the first author through CNPq Call No. 09/2022 Research Productivity, process 311009/2022-0.\u003c/p\u003e\n\u003cp\u003e\u0026nbsp;\u003cstrong\u003eCompeting Interests\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eThe authors have no conflicts of interest to declare that are relevant to the content of this manuscript.\u003c/p\u003e\n\u003cp\u003e\u0026nbsp;\u003cstrong\u003eAuthor Contributions\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eGuilherme Garcia de Oliveira contributed to the study conception and design. Material preparation, data collection and analysis were performed by all authors. The first draft of the manuscript was written by Guilherme Garcia de Oliveira and Nicholas Becker Pires Pi. All authors commented on previous versions of the manuscript. All authors read and approved the final manuscript.\u003c/p\u003e"},{"header":"References","content":"\u003col\u003e\n\u003cli\u003eAchour K, Meddi M, Zeroual A, Bouabdelli S, Maccioni P, Moramarco T (2020) Spatio-temporal analysis and forecasting of drought in the plains of northwestern Algeria using the standardized precipitation index. J Earth Syst Sci 129(1). https://doi.org/10.1007/s12040-019-1306-3\u003c/li\u003e\n\u003cli\u003eAlamgir M, Mohsenipour M, Homsi R, Wang X, Shahid S, Shiru M, Alias N, Yuzir A (2019) Parametric assessment of seasonal drought risk to crop production in Bangladesh. Sustainability 11(5):1442. https://doi.org/10.3390/su11051442 \u003c/li\u003e\n\u003cli\u003eBelayneh A, Adamowski J (2013) Drought forecasting using new machine learning methods. J Water Land Development 18(9):3-12. https://doi.org/10.2478/jwld-2013-0001\u003c/li\u003e\n\u003cli\u003eBlaikie P, Cannon T, Davis I, Wisner B (2014) At risk: Natural hazards, people\u0026rsquo;s vulnerability and disasters. Routledge.\u003c/li\u003e\n\u003cli\u003eBoas T, Bogena HR, Ryu D, Vereecken H, Western A, Hendricks Franssen HJ (2023) Seasonal soil moisture and crop yield prediction with fifth-generation seasonal forecasting system (SEAS5) long-range meteorological forecasts in a land surface modelling approach. Hydrol Earth Syst Sci 27(16):3143-3167. https://doi.org/10.5194/hess-27-3143-2023 \u003c/li\u003e\n\u003cli\u003eBoluwade A (2020) Spatial-Temporal assessment of satellite-based rainfall estimates in different precipitation regimes in water-scarce and data-sparse regions. Atmos 11(9):901. https://doi.org/10.3390/atmos11090901\u003c/li\u003e\n\u003cli\u003eBoult VL, Asfaw DT, Young M, Maidment R, Mwangi E, Ambani M, Waruru S, Otieno G, Todd MC, Black E (2020) Evaluation and validation of TAMSAT‐ALERT soil moisture and WRSI for use in drought anticipatory action. Meteorol Appl 27(5). https://doi.org/10.1002/met.1959\u003c/li\u003e\n\u003cli\u003eBusker T, de Moel H, van den Hurk B, Aerts J (2023) Impact-based seasonal rainfall forecasting to trigger early action for droughts. Sci Total Environ 898:165506. https://doi.org/10.1016/j.scitotenv.2023.165506\u003c/li\u003e\n\u003cli\u003eCanedo Rosso C, Hochrainer-Stigler S, Pflug G, Condori B, Berndtsson R (2018) Early warning and drought risk assessment for the Bolivian Altiplano agriculture using high resolution satellite imagery data. Nat Hazard Earth Syst Sci Discuss 1-23. http://dx.doi.org/10.5194/nhess-2018-133\u003c/li\u003e\n\u003cli\u003eCheng S, Wang W, Yu Z (2021) Evaluating the drought-monitoring utility of GPM and TRMM precipitation products over mainland China. Remote Sens 13(20):4153. https://doi.org/10.3390/rs13204153\u003c/li\u003e\n\u003cli\u003eDikshit A, Pradhan B, Alamri AM (2020) Temporal hydrological drought index forecasting for New South Wales, Australia using machine learning approaches. Atmos 11(6):585. https://doi.org/10.3390/atmos11060585\u003c/li\u003e\n\u003cli\u003eDikshit A, Pradhan B, Huete A (2021) An improved SPEI drought forecasting approach using the long short-term memory neural network. J Environ Manag 283:111979. https://doi.org/10.1016/j.jenvman.2021.111979 \u003c/li\u003e\n\u003cli\u003eDikshit A, Pradhan B, Santosh M (2022) Artificial neural networks in drought prediction in the 21st century - A scientometric analysis. Appl Soft Comput 114:108080. https://doi.org/10.1016/j.asoc.2021.108080\u003c/li\u003e\n\u003cli\u003eECMWF European Centre for Medium-Range Weather Forecasts (2023) Seasonal forecast anomalies on single levels. https://cds.climate.copernicus.eu/cdsapp#!/dataset/seasonal-postprocessed-single-levels?tab=form. Accessed 13 December 2023\u003c/li\u003e\n\u003cli\u003eFawcett T (2006) An introduction to ROC analysis. Pattern Recognit Lett 27(8):861-874. https://doi.org/10.1016/j.patrec.2005.10.010 \u003c/li\u003e\n\u003cli\u003eFerreira GWS, Reboita MS (2022) A new look into the South America precipitation regimes: Observation and forecast. Atmos 13(6):873. https://doi.org/10.3390/atmos13060873\u003c/li\u003e\n\u003cli\u003eFerreira GWS, Reboita MS, Drumond A (2022) Evaluation of ECMWF-SEAS5 Seasonal Temperature and Precipitation Predictions over South America. Clim 10(9):128. https://doi.org/10.3390/cli10090128\u003c/li\u003e\n\u003cli\u003eGao F, Zhang Y, Ren X, Yao Y, Hao Z, Cai W (2018) Evaluation of CHIRPS and its application for drought monitoring over the Haihe River Basin, China. Nat Hazard 92(1):155-172. https://doi.org/10.1007/s11069-018-3196-0\u003c/li\u003e\n\u003cli\u003eGuo H, Bao A, Liu T, Ndayisaba F, He D, Kurban A, De Maeyer P (2017) Meteorological drought analysis in the Lower Mekong Basin using satellite-based long-term CHIRPS product. Sustainability 9(6):901. https://doi.org/10.3390/su9060901\u003c/li\u003e\n\u003cli\u003eGyaneshwar A, Mishra A, Chadha U, Raj Vincent PMD, Rajinikanth V, Pattukandan Ganapathy G, Srinivasan K (2023) A contemporary review on deep learning models for drought prediction. Sustainability 15(7):6160. https://doi.org/10.3390/su15076160 \u003c/li\u003e\n\u003cli\u003eHagenlocher M, Meza I, Anderson CC, Min A, Renaud FG, Walz Y, Siebert S, Sebesvari Z (2019) Drought vulnerability and risk assessments: state of the art, persistent gaps, and research agenda. Environ Res Lett 14(8):083002. https://doi.org/10.1088/1748-9326/ab225d\u003c/li\u003e\n\u003cli\u003eHanley JA (1989) Receiver operating characteristic (ROC) methodology: the state of the art. Crit Rev Diagn Imaging, 29(3):307-335.\u003c/li\u003e\n\u003cli\u003eHe X, Feng K, Li X, Craft AB, Wada Y, Burek P, Wood EF, Sheffield J (2019) Solar and wind energy enhances drought resilience and groundwater sustainability. Nat Commun 10(1). https://doi.org/10.1038/s41467-019-12810-5\u003c/li\u003e\n\u003cli\u003eHeld IM, Soden BJ (2006) Robust Responses of the Hydrological Cycle to Global Warming. J Clim 19(21):5686-5699. https://doi.org/10.1175/jcli3990.1\u003c/li\u003e\n\u003cli\u003eHowells M, Hermann S, Welsch M, Bazilian M, Segerstr\u0026ouml;m R, Alfstad T, Gielen D, Rogner H, Fischer G, van Velthuizen H, Wiberg D, Young C, Roehrl RA, Mueller A, Steduto P, Ramma I (2013) Integrated analysis of climate change, land-use, energy and water strategies. Nat Clim Chang 3(7):621-626. https://doi.org/10.1038/nclimate1789\u003c/li\u003e\n\u003cli\u003eHuete A, Didan K, Miura T, Rodriguez EP, Gao X, Ferreira LG (2002) Overview of the radiometric and biophysical performance of the MODIS vegetation indices. Remote Sens Environ 83(1-2):195-213. https://doi.org/10.1016/s0034-4257(02)00096-2\u003c/li\u003e\n\u003cli\u003eHuynh TD, Nguyen TH, Truong C (2020) Climate risk: The price of drought. J Corp Finance 65:101750. https://doi.org/10.1016/j.jcorpfin.2020.101750\u003c/li\u003e\n\u003cli\u003eIBGE Instituto Brasileiro de Geografia e Estat\u0026iacute;stica (2019) Biomas e sistema costeiro-marinho do Brasil. https://www.ibge.gov.br/geociencias/informacoes-ambientais/vegetacao/15842-biomas.html. Accessed 22 January 2023\u003c/li\u003e\n\u003cli\u003eIBGE Instituto Brasileiro de Geografia e Estat\u0026iacute;stica (2023) Malhas territoriais. https://www.ibge.gov.br/geociencias/downloads-geociencias.html. Accessed 22 January 2023\u003c/li\u003e\n\u003cli\u003eIPCC (2021) Climate Change 2021: The Physical Science Basis. Contribution of Working Group I to the Sixth Assessment Report of the Intergovernmental Panel on Climate Change [Masson-Delmotte, V, Zhai P, Pirani A, Connors SL, P\u0026eacute;an C, Berger S, Caud N, Chen Y, Goldfarb L, Gomis MI, Huang M, Leitzell K, Lonnoy E, Matthews JBR, Maycock TK, Waterfield T, Yelek\u0026ccedil;i O, Yu R, Zhou B (eds.)]. Cambridge University Press. In Press.\u003c/li\u003e\n\u003cli\u003eJohnson SJ, Stockdale TN, Ferranti L, Balmaseda MA, Molteni F, Magnusson L, Tietsche S, Decremer D, Weisheimer A, Balsamo G, Keeley SPE, Mogensen K, Zuo H, Monge-Sanz BM (2019) SEAS5: the new ECMWF seasonal forecast system. Geoscientific Model Dev 12(3):1087-1117. https://doi.org/10.5194/gmd-12-1087-2019\u003c/li\u003e\n\u003cli\u003eKallis G (2008) Droughts. Annu Rev Environ Resour 33(1):85-118. https://doi.org/10.1146/annurev.environ.33.081307.123117\u003c/li\u003e\n\u003cli\u003eKanamitsu M, Ebisuzaki W, Woollen J, Yang SK, Hnilo JJ, Fiorino M, Potter GL (2002) NCEP\u0026ndash;DOE AMIP-II Reanalysis (R-2). Bull Am Meteorol Soc 83(11):1631-1643. https://doi.org/10.1175/bams-83-11-1631\u003c/li\u003e\n\u003cli\u003eKim TW, Jehanzaib M (2020) Drought Risk Analysis, Forecasting and Assessment under Climate Change. Water 12(7):1862. https://doi.org/10.3390/w12071862\u003c/li\u003e\n\u003cli\u003eKofidou M, Stathopoulos S, Gemitzi A (2023) Review on spatial downscaling of satellite derived precipitation estimates. Environ Earth Sci 82(18). https://doi.org/10.1007/s12665-023-11115-7\u003c/li\u003e\n\u003cli\u003eLiu Y, Chen J (2021) Future global socioeconomic risk to droughts based on estimates of hazard, exposure, and vulnerability in a changing climate. Sci Total Environ 751:142159. https://doi.org/10.1016/j.scitotenv.2020.142159\u003c/li\u003e\n\u003cli\u003eLuo L, Apps D, Arcand S, Xu H, Pan M, Hoerling M (2017) Contribution of temperature and precipitation anomalies to the California drought during 2012\u0026ndash;2015. Geophys Res Lett 44(7):3184-3192. https://doi.org/10.1002/2016gl072027 \u003c/li\u003e\n\u003cli\u003eMarj AF, Meijerink AMJ (2011) Agricultural drought forecasting using satellite images, climate indices and artificial neural network. Int J Remote Sens 32(24):9707-9719. https://doi.org/10.1080/01431161.2011.575896\u003c/li\u003e\n\u003cli\u003eMcKee TB, Doesken NJ, Kleist J (1993) The relationship of drought frequency and duration to time scales. 8th Conference on Applied Climatology, Anaheim, 17-22, 179-184. (n.d.). Scientific Research Publishing. https://www.scirp.org/reference/ReferencesPapers?ReferenceID=2099290 \u003c/li\u003e\n\u003cli\u003eMishra AK, Desai VR (2006) Drought forecasting using feed-forward recursive neural network. Ecol Model 198(1-2):127-138. https://doi.org/10.1016/j.ecolmodel.2006.04.017\u003c/li\u003e\n\u003cli\u003eMohamadi S, Sammen S, Panahi F, Ehteram M, Kisi O, Mosavi A, Ahmed AN, El-Shafie A, Al-Ansari N (2020) Zoning map for drought prediction using integrated machine learning models with a nomadic people optimization algorithm. Nat Hazard 104(1):537-579. https://doi.org/10.1007/s11069-020-04180-9 \u003c/li\u003e\n\u003cli\u003eMokhtari R, Akhoondzadeh M (2019) Neural network method for drought modeling using satellite data. The International Archives of the Photogrammetry, Remote Sensing and Spatial Information Sciences, XLII-4/W18, pp 749-753. https://doi.org/10.5194/isprs-archives-xlii-4-w18-749-2019\u003c/li\u003e\n\u003cli\u003eMulualem GM, Liou YA (2020) Application of Artificial Neural Networks in Forecasting a Standardized Precipitation Evapotranspiration Index for the Upper Blue Nile Basin. Water 12(3):643. https://doi.org/10.3390/w12030643\u003c/li\u003e\n\u003cli\u003eMusolino DA, Massarutto A, de Carli A (2018) Does drought always cause economic losses in agriculture? An empirical investigation on the distributive effects of drought events in some areas of Southern Europe. Sci Total Environ 633:1560-1570. https://doi.org/10.1016/j.scitotenv.2018.02.308\u003c/li\u003e\n\u003cli\u003eNandgude N, Singh TP, Nandgude S, Tiwari M (2023) Drought Prediction: A Comprehensive Review of Different Drought Prediction Models and Adopted Technologies. Sustainability 15(15):11684. https://doi.org/10.3390/su151511684\u003c/li\u003e\n\u003cli\u003eNASA (2023) Global Climate Change - Vital Signs of the Planet. Goddard Institute for Space Studies (GISS). https://climate.nasa.gov/. Accessed 28 November 2023\u003c/li\u003e\n\u003cli\u003eOliveira GG, Ruiz LFC, Guasselli LA, Teixeira B, Klabin G (2023) Modelo baseado em redes neurais artificiais e dados de chuva por sat\u0026eacute;lites para alerta de estiagens e secas. 20th Simp\u0026oacute;sio Brasileiro de Sensoriamento Remoto. S\u0026atilde;o Jos\u0026eacute; dos Campos, INPE. https://proceedings.science/sbsr-2023/trabalhos/modelo-baseado-em-redes-neurais-artificiais-e-dados-de-chuva-por-satelites-para?lang=pt-br. Accessed 28 July 2023\u003c/li\u003e\n\u003cli\u003ePalharini R, Vila D, Rodrigues D, Palharini R, Mattos E, Undurraga E (2022) Analysis of Extreme Rainfall and Natural Disasters Events Using Satellite Precipitation Products in Different Regions of Brazil. Atmos 13(10):1680. https://doi.org/10.3390/atmos13101680\u003c/li\u003e\n\u003cli\u003ePandey V, Srivastava PK, Mall RK, Munoz-Arriola F, Han D (2020) Multi-satellite precipitation products for meteorological drought assessment and forecasting in Central India. Geocarto Int 37(7):1899-1918. https://doi.org/10.1080/10106049.2020.1801862 \u003c/li\u003e\n\u003cli\u003ePanu US, Sharma TC (2002) Challenges in drought research: some perspectives and future directions. Hydrol Sci J 47:19-30. https://doi.org/10.1080/02626660209493019\u003c/li\u003e\n\u003cli\u003eRumelhart DE, Hinton GE, Williams RJ (1986) Learning representations by back-propagating errors. Nature 323(6088):533-536. https://doi.org/10.1038/323533a0\u003c/li\u003e\n\u003cli\u003eSantos JF, Portela MM, Pulido-Calvo I (2012) Spring drought prediction based on winter NAO and global SST in Portugal. Hydrol Process 28(3):1009-1024. https://doi.org/10.1002/hyp.9641\u003c/li\u003e\n\u003cli\u003eSantos SRQ, Cunha APMA, Ribeiro-Neto GG (2019) Avalia\u0026ccedil;\u0026atilde;o de dados de precipita\u0026ccedil;\u0026atilde;o para o monitoramento do padr\u0026atilde;o espa\u0026ccedil;o-temporal da seca no nordeste do Brasil. Rev Bras Climatol 25:80-100. https://doi.org/10.5380/abclima.v25i0.62018\u003c/li\u003e\n\u003cli\u003eSEDEC Secretaria Nacional de Prote\u0026ccedil;\u0026atilde;o e Defesa Civil (2023) Sistema Integrado de Informa\u0026ccedil;\u0026otilde;es sobre Desastres - S2ID. https://s2id.mi.gov.br/paginas/relatorios/. Accessed 10 July 2023\u003c/li\u003e\n\u003cli\u003eShahid S (2010) Recent trends in the climate of Bangladesh. Clim Res 42(3):185-193. https://doi.org/10.3354/cr00889 \u003c/li\u003e\n\u003cli\u003eShao C, Paynabar K, Kim TH, Jin J, Hu SJ, Spicer JP, Wang H, Abell JA (2013) Feature selection for manufacturing process monitoring using cross-validation. J Manuf Syst 32(4):550-555. https://doi.org/10.1016/j.jmsy.2013.05.006 \u003c/li\u003e\n\u003cli\u003eSousa LB, de Assun\u0026ccedil;\u0026atilde;o Montenegro AA, da Silva MV, Almeida TAB, de Carvalho AA, da Silva TGF, de Lima JLMP (2023) Spatiotemporal Analysis of Rainfall and Droughts in a Semiarid Basin of Brazil: Land Use and Land Cover Dynamics. Remote Sens 15(10):2550. https://doi.org/10.3390/rs15102550\u003c/li\u003e\n\u003cli\u003eStewart IT, Rogers J, Graham A (2020) Water security under severe drought and climate change: Disparate impacts of the recent severe drought on environmental flows and water supplies in Central California. J Hydrol X 7:100054. https://doi.org/10.1016/j.hydroa.2020.100054\u003c/li\u003e\n\u003cli\u003eSundararajan K, Garg L, Srinivasan K, Kashif Bashir A, Kaliappan J, Pattukandan Ganapathy G, Kumaran Selvaraj S, Meena T (2021) A Contemporary Review on Drought Modeling Using Machine Learning Approaches. Comput Model Eng Sci 128(2):447-487. https://doi.org/10.32604/cmes.2021.015528\u003c/li\u003e\n\u003cli\u003eSutanto SJ, van der Weert M, Blauhut V, Van Lanen HAJ (2020a) Skill of large-scale seasonal drought impact forecasts. Nat Hazard Earth Syst Sci 20(6):1595-1608. https://doi.org/10.5194/nhess-20-1595-2020 \u003c/li\u003e\n\u003cli\u003eSutanto SJ, Wetterhall F, Van Lanen HAJ (2020b) Hydrological drought forecasts outperform meteorological drought forecasts. Environ Res Lett 15(8):084010. https://doi.org/10.1088/1748-9326/ab8b13 \u003c/li\u003e\n\u003cli\u003eTanguy M, Chevuturi A, Marchant BP, Mackay JD, Parry S, Hannaford J (2023) How will climate change affect the spatial coherence of streamflow and groundwater droughts in Great Britain? Environ Res Lett 18(6):064048. https://doi.org/10.1088/1748-9326/acd655\u003c/li\u003e\n\u003cli\u003eTrnka M, Hlavinka P, Možn\u0026yacute; M, Semer\u0026aacute;dov\u0026aacute; D, \u0026Scaron;těp\u0026aacute;nek P, Balek J, Barto\u0026scaron;ov\u0026aacute; L, Zahradn\u0026iacute;ček P, Bl\u0026aacute;hov\u0026aacute; M, Skal\u0026aacute;k P, Farda A, Hayes M, Svoboda M, Wagner W, Eitzinger J, Fischer M, Žalud Z (2020) Czech Drought Monitor System for monitoring and forecasting agricultural drought and drought impacts. Int J Climatol 40(14):5941-5958. https://doi.org/10.1002/joc.6557\u003c/li\u003e\n\u003cli\u003eUCSB University of California at Santa Barbara (2023) Climate Hazards Group Infrared Precipitation with Stations (CHIRPS). https://data.chc.ucsb.edu/products/CHIRPS-2.0. Accessed 26 June 2023\u003c/li\u003e\n\u003cli\u003eVicente-Serrano SM, Beguer\u0026iacute;a S, L\u0026oacute;pez-Moreno JI (2010) A Multiscalar Drought Index Sensitive to Global Warming: The Standardized Precipitation Evapotranspiration Index. J Clim 23(7):1696-1718. https://doi.org/10.1175/2009jcli2909.1\u003c/li\u003e\n\u003cli\u003eVicente-Serrano SM, Quiring SM, Pe\u0026ntilde;a-Gallardo M, Yuan S, Dom\u0026iacute;nguez-Castro F (2020) A review of environmental droughts: Increased risk under global warming? Earth-Sci Rev 201:102953. https://doi.org/10.1016/j.earscirev.2019.102953\u003c/li\u003e\n\u003cli\u003eWMO World Meteorological Organization (2008) Guide to Hydrological Practice, Volume I, Hydrology \u0026ndash; From Measurement to Hydrological Information. 6th ed, Geneva, Switzerland.\u003c/li\u003e\n\u003cli\u003eWu W, Li Y, Luo X, Zhang Y, Ji X, Li X (2019) Performance evaluation of the CHIRPS precipitation dataset and its utility in drought monitoring over Yunnan Province, China. Geomat Nat Hazard Risk 10(1):2145-2162. https://doi.org/10.1080/19475705.2019.1683082\u003c/li\u003e\n\u003cli\u003eYaseen ZM, Ali M, Sharafati A, Al-Ansari N, Shahid S (2021) Forecasting standardized precipitation index using data intelligence models: Regional investigation of Bangladesh. Sci Rep 11(1):3435. https://doi.org/10.1038/s41598-021-82977-9 \u003c/li\u003e\n\u003cli\u003eZhang R, Chen ZY, Xu LJ, Ou CQ (2019) Meteorological drought forecasting based on a statistical model with machine learning techniques in Shaanxi province, China. Sci Total Environ 665: 338-346. https://doi.org/10.1016/j.scitotenv.2019.01.431\u003c/li\u003e\n\u003cli\u003eZhang Y, Li W, Chen Q, Pu X, Xiang L (2017) Multi-models for SPI drought forecasting in the north of Haihe River Basin, China. Stoch Environ Res Risk Assess 31(10):2471-2481. https://doi.org/10.1007/s00477-017-1437-5\u003c/li\u003e\n\u003c/ol\u003e"}],"fulltextSource":"","fullText":"","funders":[],"hasAdminPriorityOnWorkflow":false,"hasManuscriptDocX":true,"hasOptedInToPreprint":true,"hasPassedJournalQc":"","hasAnyPriority":false,"hideJournal":false,"highlight":"","institution":"","isAcceptedByJournal":false,"isAuthorSuppliedPdf":false,"isDeskRejected":"","isHiddenFromSearch":false,"isInQc":false,"isInWorkflow":false,"isPdf":false,"isPdfUpToDate":true,"isWithdrawnOrRetracted":false,"journal":{"display":true,"email":"[email protected]","identity":"natural-hazards","isNatureJournal":false,"hasQc":true,"allowDirectSubmit":false,"externalIdentity":"nhaz","sideBox":"Learn more about [Natural Hazards](https://www.springer.com/journal/11069)","snPcode":"11069","submissionUrl":"https://submission.nature.com/new-submission/11069/3","title":"Natural Hazards","twitterHandle":"","acdcEnabled":true,"dfaEnabled":true,"editorialSystem":"em","reportingPortfolio":"Springer Hybrid","inReviewEnabled":true,"inReviewRevisionsEnabled":false},"keywords":"Satellite-Derived Precipitation, Seasonal Weather Forecast, Natural Disaster, Drought Warning","lastPublishedDoi":"10.21203/rs.3.rs-4784321/v1","lastPublishedDoiUrl":"https://doi.org/10.21203/rs.3.rs-4784321/v1","license":{"name":"CC BY 4.0","url":"https://creativecommons.org/licenses/by/4.0/"},"manuscriptAbstract":"\u003cp\u003eThe increase in the frequency of droughts, driven by climate change, implies the need to understand and mitigate these extreme events. In Brazil, there are technical-scientific gaps in relation to climate disaster warnings. The integration of an inventory of droughts that caused losses with remote sensing data, hydrometeorological and climate indexes, using artificial neural networks (ANN) can contribute to a drought forecast. In this study, we developed a monthly forecast model for droughts in Brazilian municipalities using ANN. Precipitation and temperature indexes, in addition to municipal descriptors, for example, the region of the country, the biome, and distance from the oceans and the Amazon, were used as predictor variables in the model. We used an inventory of droughts that caused losses by municipalities (2013\u0026ndash;2022) from the Brazilian Integrated Disaster Information System. After model training, we tested the ANN for drought forecasts for lead times of 1\u0026ndash;4 months, using seasonal forecast data from the European Center for Medium-Range Weather Forecasts (ECMWF). The overall accuracy of the ANN model for drought simulation was 0.931. The forecast accuracy ranged from 0.922 for a 1-month lead time to 0.757 for 4 months. Remarkably, the model reproduced the spatial pattern of droughts, especially when the output is interpreted as a continuous index of drought risk. We conclude that the trained model is efficient and the results indicate strong potential for drought forecasting and warning, using ANN, remote sensing data, hydrometeorological and climate indexes.\u003c/p\u003e","manuscriptTitle":"Drought forecast model based on Artificial Neural Networks for Brazilian municipalities","msid":"","msnumber":"","nonDraftVersions":[{"code":1,"date":"2024-08-23 08:41:20","doi":"10.21203/rs.3.rs-4784321/v1","editorialEvents":[{"type":"communityComments","content":0},{"type":"decision","content":"Major revisions","date":"2025-05-26T08:28:46+00:00","index":"","fulltext":""},{"type":"reviewerAgreed","content":"","date":"2024-08-16T12:58:16+00:00","index":0,"fulltext":""},{"type":"reviewersInvited","content":"","date":"2024-07-29T06:58:55+00:00","index":"","fulltext":""},{"type":"editorAssigned","content":"","date":"2024-07-23T11:17:04+00:00","index":"","fulltext":""},{"type":"submitted","content":"Natural Hazards","date":"2024-07-22T16:59:21+00:00","index":"","fulltext":""}],"status":"published","journal":{"display":true,"email":"[email protected]","identity":"natural-hazards","isNatureJournal":false,"hasQc":true,"allowDirectSubmit":false,"externalIdentity":"nhaz","sideBox":"Learn more about [Natural Hazards](https://www.springer.com/journal/11069)","snPcode":"11069","submissionUrl":"https://submission.nature.com/new-submission/11069/3","title":"Natural Hazards","twitterHandle":"","acdcEnabled":true,"dfaEnabled":true,"editorialSystem":"em","reportingPortfolio":"Springer Hybrid","inReviewEnabled":true,"inReviewRevisionsEnabled":false}}],"origin":"","ownerIdentity":"54f0520b-b3f0-4e46-a612-c726004315b1","owner":[],"postedDate":"August 23rd, 2024","published":true,"recentEditorialEvents":[],"rejectedJournal":[],"revision":"","amendment":"","status":"in-revision","subjectAreas":[],"tags":[],"updatedAt":"2025-05-26T12:28:59+00:00","versionOfRecord":[],"versionCreatedAt":"2024-08-23 08:41:20","video":"","vorDoi":"","vorDoiUrl":"","workflowStages":[]},"version":"v1","identity":"rs-4784321","journalConfig":"researchsquare"},"__N_SSP":true},"page":"/article/[identity]/[[...version]]","query":{"redirect":"/article/rs-4784321","identity":"rs-4784321","version":["v1"]},"buildId":"qtupq5eGEP_6zYnWcrvyt","isFallback":false,"isExperimentalCompile":false,"dynamicIds":[84888],"gssp":true,"scriptLoader":[]}