CoSal: A remote sensing and machine learning framework for mapping coastal soil salinity trends around aquaculture in South Asia

CoSal: A remote sensing and machine learning framework for mapping coastal soil salinity trends around aquaculture in South Asia | 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 Article CoSal: A remote sensing and machine learning framework for mapping coastal soil salinity trends around aquaculture in South Asia Garima Jain, Wenxin Yang, Showmitra Kumar Sarkar, Amy E. Frazier, and 1 more This is a preprint; it has not been peer reviewed by a journal. https://doi.org/ 10.21203/rs.3.rs-6173117/v1 This work is licensed under a CC BY 4.0 License Status: Posted Version 1 posted You are reading this latest preprint version Abstract Coastal salinity represents a critical global environmental crisis that threatens agricultural productivity and food security. Traditional remote sensing methods to measure soil salinity in coastal areas are confounded by the presence of soil moisture and ubiquitous water-based land uses. This study introduces CoSal, a remote sensing and machine learning framework for mapping long-term coastal soil salinity trends while accounting for soil moisture and aquaculture, a fast-growing land-based practice of fish farming. We apply CoSal in South Asia (CoSal-SA), where salinity and aquaculture acutely impact agriculture, where we integrate Landsat imagery with soil data from coastal India and Bangladesh. Using 28 metrics and a stacked ensemble of nine machine learning models, CoSal-SA identifies saline soils in waterlogged coastal areas with over 91% accuracy. Applying CoSal-SA to a coastal district in India reveals that 10 percent of the area in 2024 had salinity levels unsuitable for rice cultivation. While interior regions showed decreasing salinity between 1995–2024, the coastal belt experienced intensifying salinity alongside increased aquaculture adoption. CoSal can be adopted for diverse coastal contexts and time periods with additional soil data. CoSal enables crucial research on salinity dynamics at different geographical scales that can guide targeted interventions to ultimately address agricultural productivity losses, food insecurity, and poverty in vulnerable coastal regions. Earth and environmental sciences/Environmental sciences/Environmental chemistry Earth and environmental sciences/Environmental sciences/Environmental impact Earth and environmental sciences/Environmental social sciences/Climate change impacts Satellite data ensemble model salt-affected soil climate adaptation shrimp farming Odisha Figures Figure 1 Figure 2 Figure 3 Figure 4 Figure 5 Figure 6 Figure 7 Figure 8 Background and Summary Land change and degradation are second only to climate change in being the most influential drivers of ecosystem collapse 1 – 3 . Among degraded lands, salt-affected soils, characterized by the presence of soluble salts in quantities that inhibit plant growth, cover over 1,128 million ha of global arable lands resulting in lowered agricultural productivity costing the world US $ 12-27.3 billion annually 4 – 6 . Soil salinization demands urgent attention to curb food insecurity, social conflicts, distress migration, and deep poverty prevalence 7 – 10 . There is growing evidence of increasing salinity in coastal areas due to simultaneously occurring seawater intrusion, poor agricultural practices, and water-logging including land conversions to fish farms or aquaculture 8 , 11 – 13 . Aquaculture, the practice of farming fish on land, is found to be both a driver and an outcome of salinity in coastal areas 7 , 14 – 16 . It is the fastest-growing animal food production sector, and is expected to expand mainly on coastal lands in the coming decades 17 – 20 , making it an active area of research and policy concern. Our lack of understanding of the historical salinity dynamics at the local level limits us from identifying the specific causal pathways of salinity increase and taking targeted actions. Remote sensing is increasingly used for mapping soil salinity as it provides consistent data across time, spatial scales, and diverse climatic contexts 21 – 24 . However, a global review 21 has found that 65 percent of remote-sensing-based salinity identification methods have been developed in arid and semi-arid regions e.g. 25 – 28 , while coastal areas—home to 40 percent of the global population and significant agricultural lands—remain understudied. Coastal soils are a complex composite of minerals and saline deposits alongside high moisture and organic matter. These components simultaneously interfere with incoming light, making remote salinity detection complex by increasing spectral confusion, especially between salts and moisture 29 . Clear water scatters some blue and green light but absorbs most red, making it appear darker in true color composites, whereas dry saline or bare soils are highly reflective, particularly in the red, and NIR/SWIR spectrum. High salinity, often associated with low vegetation cover, further heightens red reflectance by reducing photosynthesis activity that tends to absorb red spectrum light 22 , 30 . Such spectral interference is particularly problematic in coarse-resolution imagery like 30m Landsat, where mixed pixels, containing several surface components including bare soil and surface water, are common 29 , 31 , 32 . While higher-resolution data from Sentinel-2 (10m) could reduce this error, such imagery lacks the long historical record needed to track salinity changes over time. Accurately mapping soil salinity in coastal environments thus requires adjusting for the presence of water-based land uses, such as aquaculture, and integrating multiple surface reflectance measures. Recent efforts have developed remote sensing-based salinity models for coastal areas 33 – 36 , but these models often do not distinguish between the signals created from soil moisture and those from water-based land uses such as aquaculture, which is found to bias their results 31 . Besides, traditional modeling approaches rely on univariate linear regressions between electrical conductivity and surface reflectance to predict salinity 33 , 37 – 41 , but have lower accuracy likely due to the non-linear relationships and soil-moisture confounding 24 . Modeling multiple surface reflectance measures and capturing non-linear relationships using machine learning has shown higher accuracy 24 , 28 , 36 , 42 , 43 . However, even machine learning approaches have so far failed to account for the confounding effects of surface water, which can still introduce bias into salinity predictions 31 . This study introduces CoSal, a remote-sensing and machine learning-based framework for mapping long-term soil salinity trends in coastal areas, accounting for the confounding effects from water-based land uses. Since coastal geographies, especially across Asia and Latin America, are increasingly characterized by aquaculture, our framework overcomes the challenge posed by the presence of surface water that is otherwise found to bias existing salinity models 31 . We use Landsat satellite imagery to leverage its long time span (available 1984 onwards). We identify and mask all pixels containing surface water using a supervised land use land cover classification, such that the salinity model developed is based on soil reflectance only. In the specific application of CoSal in South Asia, CoSal-SA, we use 28 variables (visible and infrared bands, popularly used salinity indices as well as all normalized band combinations) and apply 9 machine learning methods to build a stacked ensemble model for identifying coastal saline soils in a district of India. We train the model to identify soils that are too saline for growing rice (> 1900 µS/cm electrical conductivity) 44 . In this study, we apply the CoSal framework in South Asia (CoSal-SA). South Asia has the world's largest undernourished population, 14.4% of 1.3 billion in India alone 45 , with food security threatened by land degradation and climate change 46 , 47 . As per recent estimates 47 , about 7 million ha are salt affected in India, of which 20 per cent are in the coastal areas. This is estimated to expand to over 20 million ha or 50% of its arable land by 2050. Remote-sensing based salinity assessments conducted at the national scale are, however, too coarse to understand local level dynamics required for targeted actions 46 . For instance, a remote-sensing based national salinity mapping exercise in India uses a benchmark of 4000 µS/cm for saline soils, which is too high for most food crops 48 . Crops such as rice, however, are affected by salinity above 1900 µS/cm 44 and are not captured by this assessment. These national-level benchmarks highlight salinity in arid and semi-arid regions such as Gujarat, Rajasthan, and Uttar Pradesh, however, coastal salinity remains underrepresented 48 . CoSal-SA is developed by combining remotely sensed data with primary field data from India and Bangladesh. Soil data collected in 2024 from Jagatsinghpur district in India is used to train the model, and test the model’s internal validity. Soil data collected in 2016 from Satkhira district in Bangladesh 34 , 35 is used to test its external validity across time for South Asian coastal geographies. Jagatsinghpur is a district in the state of Odisha on the east coast of India and is home to 1.2 million people 49 . With a land area of 1668 sq. km and 55 km of coastline, it falls within the coastal plain agro-climatic zone 50 . Historically, the region has been agriculturally productive, with over 81 thousand ha largely monocropped with rice 51 . While 54% of the district’s workers depend on agriculture for their livelihoods, their contribution is less than 20% of its district domestic product (2% less than the national contribution from the sector) 50 . Coastal livelihoods in the district have been suffering both from climatic shocks, such as cyclones, and stresses, such as coastal erosion and salinization 52 – 54 . The district was the worst affected area after the 1999 Supercyclone (Cyclone 05B Paradeep), which resulted in 10,000 deaths and US $ 3.5 million in agricultural losses 50 , 55 . Over 8000 ha of land in the district are affected by salinization. These lands are largely concentrated in the district’s Ersama block, the one also most affected by the cyclonic storm surge, however, no causal connections are tested empirically 50 . Meanwhile, the district is also witnessing an expansion of aquaculture. In other contexts, aquaculture is found to be both an outcome and a dominant cause for salinization and agricultural productivity losses 7 – 10 , 47 , 56 . A historical salinity mapping at the local scale can give urgently needed insights into the potential spatial-temporal drivers of salinization, to develop targeted land management practices. Our application of CoSal-SA in Jagatsinghpur district reveals that 155.16 sq.km. or nearly 10 percent of its total area had salinity levels not conducive for rice cultivation in 2024. While salinity increased in the years immediately following the cyclonic storm surge in 1999, it reduced in the interiors of the study area over time, likely due to natural processes of leaching. Meanwhile, salinity has been increasing in the coastal belt of the study area in India in recent years, alongside increased land conversions to aquaculture in the same belt. This simultaneous increase in salinity and aquaculture points at a potential salinity lock-in and acceleration effect due to coastal land change practices. CoSal-SA has a training accuracy of 91.7% and internal testing accuracy of 85.7%. Its overall external testing accuracy in Bangladesh is 69.3%, and this drop is expected given different environmental factors, surface reflectance distributions between different images, and differences in soil sampling methodologies. However, a high recall score of 82.2% with this external data still makes it highly reliable for predicting high salinity values in a different geography and time period. This study establishes the unique characteristic of coastal soils that are simultaneously saline and have high moisture content. The CoSal framework demonstrates a comprehensive approach to accurately map historical coastal salinity in multiple geographical contexts, including near emerging coastal land uses, such as aquaculture. Fine grained salinity mapping can enable urgently needed research on local causal pathways of coastal salinity to inform targeted action and counter large losses in agricultural productivity, food insecurity, and related poverty prevalence. CoSal Framework: Method and Materials ColSal framework follows a six-step process to combine remotely sensed data with field data, to create, validate, and apply a machine learning based salinity model for mapping historical soil salinity levels (1995–2024) (Fig. 1 ). Step1 : Primary soil data collection - To build a relationship between remotely sensed data and field readings of salinity, we collected 113 soil samples distributed relatively evenly across the 200 sq. km. of a predetermined sampling area. We identified two neighboring areas in Jagatsinghpur district such that one area had widespread aquaculture while the second did not. We imposed a grid of 10km x 10km on each such that the areas were equally distant from the coast (Fig. 2 ). This distinction was made to help assess the difference in salinity specifically in relation to aquaculture. We divided these grids into 2km x 2km grid cells and systematically collected two geolocated soil samples from each grid cell. We collected these samples at 15 cm depth (topsoil) using a soil coring device. We collected these samples from the middle of the fields and away from any tree cover or buildings to avoid shade in surface reflectances. Critically, we ensured a minimum distance of 30m (i.e. Landsat pixel size) from any surface water or water-based land uses to prevent the sampling points from falling within a satellite image pixel with mixed surface components that could confound the surface reflectance results. The samples were collected in the dry month of February in 2024 to avoid cloud cover, a common deterrent in remote sensing studies especially in tropical contexts. This timing also helped avoid the presence of irrigation water logging for rice cultivation, a common practice in these geographies, that could also introduce a similar mixed pixel issue as aquaculture. We collected some additional samples (~ 13) from each area to potentially replace samples falling within mixed pixels despite our efforts on the field. These samples were used to measure the electrical conductivity (in µS/cm), pH, and soil moisture (in %) at one of India’s National Accreditation Board for Testing and Calibration Laboratories (NABL) certified labs. For methods of analysis, we followed IS: 14767:2000 for EC, IS: 2720 (Part 26)-1987 for pH, and IS: 2720 (Part 2) − 1973 for soil moisture, all as prescribed by NABL. Electrical conductivity (EC) is our key variable of interest for soil salinity measure, while moisture and pH were measured to study their role as potential confounders. Step 2: Supervised land use land cover classification and aquaculture masking - To reduce the confounding presence of surface water and aquaculture in the satellite reflectance data, we employed available perennial water data for masking the former and conducted a supervised land use land cover classification to identify and mask the latter. First, we masked out perennial surface water pixels in the Landsat composite using the JRC Global Surface Water Mapping Layers, v1.4 57 . At this stage, we used a threshold of 90 percent occurrence of water, to not mix perennial surface water with aquaculture ponds that do not operate throughout the year. This would ensure isolating the patterns of change in aquaculture alongside salinity. Second, to conduct the supervised classification, we acquired atmospherically corrected US Geological Survey Landsat 8 OLI (Level 2 Collection 2 Tier 1) and Harmonized Sentinel-2 MSI satellite images for February 2024 to coincide with field data collection. We limited images to no more than 50 percent cloud cover to maximise the number of images available but reducing reflectance errors due to potential presence of clouds. We created image composites using the 30th percentile, since that helps filter out clouds, haze, and high reflectance noise while retaining meaningful surface reflectances without overemphasizing dark shadows. We tested alternate reducers, including geomedian, median, and 70th percentile (See supplementary 1 for accuracy comparisons). Since the 30th percentile gave the highest accuracy results amongst these, we use it as the reducer for our current purpose. As is required by the Landsat dataset, we scaled the Landsat pixel values to floating points (0–1) using published gain and offset values 58 , 59 . Some pixels (0.4 percent) had negative surface reflectance after conversion, which is likely due to cloud cover or presence of water 60 . These values were adjusted to 0 so as to not get included in the analysis further. We used primary observation data points as well as Sentinel-2 high-resolution composite to identify 274 active aquaculture pond locations (Fig. 3 ). We classified each Landsat composite for three classes - active aquaculture, dry aquaculture and other land uses. To do so, we used a random forest classifier, with 40 trees (see Supplementary 1 for hyperparameter tuning results) and employed 6 spectral bands (Blue, Green, Red, Near Infrared or NIR, and the two Short Wave Infrared bands SWIR1 and SWIR2) and three indices (Normalized Difference Vegetation Index, Normalized Difference Water Index, and Normalized Difference Salinity Index). Finally, we applied this aquaculture LULC classification to the entire Jagatsinghpur district and used the classification output to mask active aquaculture pixels in the Landsat composite to be used for salinity modeling in subsequent steps. All processing was performed in Google Earth Engine. Step 3 : Surface reflectance of pure soil and correlating electrical conductivity with key variables - To derive the surface reflectance values corresponding to the field locations, we used the masked Landsat composite from Step 2, and superimposed the field locations on it. 7 of the 113 soil sampling locations fell within a pixel masked for aquaculture, suggesting a mixed pixel scenario (Fig. 4 ). Including data from such mixed pixels containing both water and soil in the modelling has been found to bias salinity results (see Silvestri et al. 31 ). We, therefore, dropped these samples from the analysis. We calculated 11 indices widely used for measuring salinity, and 11 additional normalized band combinations (Supplementary 2). Including the six VIS-NIR-SWIR spectral bands, these 28 variables are used as the key predictor variables in the models. (Note: coastal blue band was not included since it is not available in older Landsat images). We assessed the linear, quadratic, and logarithmic relationships between EC and the 28 variables. We also tested the relationship using a principal component analysis (PCA) to help reduce the dimensionality of the model. See Supplementary 2 for details on each of these metrics, correlation matrix, and their regression model fit results with EC. The variables most correlated with EC were NDWI, NDSI2, NRSWIR1, NBNIR, NIR, and SAVI. Of these, two indices, NDWI and NDSI2, achieved the highest R2 of 0.39 and 0.37 in a second-degree polynomial regression model. Since this is not an adequate explanation, a machine learning-based approach was adopted to derive a better explanatory combination of the input variables. Step 4 : Ensemble model - We used the 106 remaining EC measurements after aquaculture masking as the key dependent variables, and their corresponding 28 surface reflectance measurements as the key predicting variables. Of these, we used 80 percent of the samples to train machine learning models, leaving 20% for testing. We used EC as a binary variable, and applied stratified sampling for the training and testing sets to ensure both categories are adequately present in each subset. Since rice is the main crop in the study region 50 , we use 1900 µS/cm of EC as the threshold for high salinity. This is based on the understanding that rice production is hampered above this level under all conditions of acidity and soil nutrients 44 . We first conducted 100 iterations for select machine learning methods most widely used in salinity studies including random forest, random forest with bagging, and artificial neural network 24 , 43 , and tested the average training and testing accuracies for each (Supplementary 6). We also reduced the dimensionality of the predictor variables using a lasso regression and repeated the 100-iteration test which improved test results and reduced overfitting (Supplementary 3). Random forest with bagging 61 resulted in the highest accuracy amongst these select machine learning models, however, the accuracy remained lower than that of the ensemble. Ensemble models combine outputs from multiple machine learning models and find an optimum result that balances between bias and variance (or noise). This reduces errors and improves overall accuracy, while also addressing over or underfitting the training data. This makes ensemble models particularly efficient, especially when data is limited. We included the following nine machine learning models in the ensemble model: (1) Random Forest (“rf”), which builds multiple decision trees using random subsets of data, is robust to outliers, and good at handling non-linear relationships. (2) Recursive Partitioning Decision Tree (“rpart”), which creates a single decision-tree through recursive partitioning of variables and is easier to interpret. (3) Neural Network (“nnet”), which is a basic feed-forward neural network that helps capture complex patterns. (4) Support Vector Machine with Radial Kernel (“svmRadial”), which creates non-linear decision boundaries in high-dimension data. (5) Gradient Boosting Machine (“gbm”), which builds trees sequentially, each correcting errors in the previous tree and improving accuracy. (6) XGBoost (“xgbTree”), which is an advanced implementation of gradient boosting. (7) Naive Bayes (“naive_bayes”), which is a probabilistic classifier based on Bayes Theorem useful for high-dimensional data. (8) K-nearest Neighbors (“knn”), which makes predictions on most similar training examples. (9) Elastic Net (“glmnet”), which helps with feature selection (using ridge and lasso) and helps handle correlated predictors. We used the R-package caretEnsemble 62 to enable fitting multiple linear and non-linear models to the same dataset. We stacked the models using the function caretStack(). This approach uses the predictions from all the 9 models as features, and combines them using a meta-model, in this case a random forest, to make the final ensemble predictions. The model returns probabilities or the likelihood that a given sample belongs to a specific class (high or low salinity). We used Youden’s J Statistic (Eq. 1) for deriving the most effective threshold to convert these probabilities back into binary classification predictions 63 , 64 . Eq. 1: Youden's J statistics: J = Sensitivity + Specificity − 1 While ensemble models inherently address overfitting, we took additional measures to reduce any remaining issues. We used a 5-fold cross-validation method such that the data is split into 5 equal parts. The model is trained 5 times, each time using 4 parts for training and 1 part for testing. We also reduced the dimensionality of the data by removing near-variance predictors and lasso variable selection (Supplementary 3). For the threshold for converting probabilities into class predictions, we tested all values between 0.1 and 0.9 with the increments of 0.01, to minimize the difference between training and testing accuracies (Supplementary 7). The difference was minimal across the range, suggesting low possibility of overfitting. To help interpret the machine learning model, we used the SHapley Additive exPlanations (SHAP) methodology 65 , 66 to calculate Shapley values as a measure of the contribution of each individual parameter in the final model 65 , 66 . These values improve our judgement about the model by theoretically testing and explaining specific predictor contributions. Step5: Model Validation - We conducted two separate analyses to test the internal and external validity of the model. Internal validity helps assess the predictive consistency of the model and potential overfitting. External validity on the other hand, gives insights into the generalizability of the model to different spatial contexts and time periods. Both validities give insights into the robustness of the fundamental physical relationship between salinity measures and surface reflectances across the training and testing data, rather than spurious correlations. For internal validation, we used the remaining 20 percent of the Jagatsinghpur field data to test how well the model performs on unseen data from the same geography and time period. For external validity of the coastal salinity model on a different geography and time period, we use georeferenced soil data (n = 92) collected from aquaculture practicing regions of Satkhira District in coastal Bangladesh in 2016 34,35 . These topsoil samples were also collected in the dry period of February; however, they were collected using a composite sampling method by mixing multiple smaller samples from an agricultural field. This is a standard practice in agricultural soil testing, and while this approach is useful for determining the average salinity levels across a large field, the corresponding surface reflectance may not capture the precise relationship with soil salinity. For each model and its prediction application on the various subsets of data (training and internal and external testing data), we calculated a confusion matrix of true positives (TP), true negatives (TN), false positives (FP), and false negatives (FN), and derived four metrics to evaluate the model comprehensively: Accuracy, precision, recall, and F1-score (Supplementary 4). Accuracy involves all components of the confusion matrix and usually serves as a coarse measure of overall model quality. In case of imbalanced data, such as ours where there are fewer high salinity points (30 out of 106) than there are for lower salinity (70 out of 106), there is a possibility of getting a relatively high accuracy even if all the high salinity points are incorrectly classified as low (accuracy would still be 70% because TN = 70 even when FN = 30). We, therefore, also consider the other three metrics in tandem to gain better insights into the robustness of the model predictions. More details on each metric is presented in Supplementary 4. Step 6: Historical coastal salinity mapping - To apply the CoSal-SA model in the current context of Jagatsinghpur, we extracted composites for the district for February for the years 1995, 2001, 2005, and 2010 from Landsat 5 Thematic Mapper (TM) and for years 2014, 2017, and 2021 from Landsat 8 Operational Land Imager (OLI). It is important to consider and address the sensor calibration and radiometric resolutions differences of the two sensors (TM and OLI) in a time series analysis. To align the spectral distributions across the two time periods, we conducted a mean-variance normalization for all bands in Landsat 5 images with corresponding bands in Landsat 8 (2024) (normalization results are in Supplementary 5) 67 . We then followed Steps 2 and 3 (above) on these images (i.e. applied the LULC classifier to these historical composites to create the aquaculture masks and extracted the values for the 28 predictors). We applied the coastal salinity ensemble model to this data to get the final historical coastal salinity maps for the district of Jagatsinghpur. Findings Soil salinity increases with soil moisture in coastal contexts Soil samples collected from coastal India show a strong positive correlation (correlation coefficient of 0.48) between electrical conductivity and soil moisture (Supplementary 2). This finding suggests that top-soil salinity increases with increasing moisture in these coastal areas. This result is further corroborated by indices such as NDWI, NBNIR and NGSWIR1, where they are found to be higher for both higher levels of soil moisture as well as higher levels of EC (Fig. 5 and Supplementary 8). While these indices behave as expected with the presence of moisture, their relationship with salinity is divergent. Clear water reflects some green and blue light but absorbs NIR and SWIR1 almost completely, increasing the numerator and decreasing the denominator of these three indices (Eq. 2), thereby showing a positive association with more moisture. Saline soils, however, are known to reflect more infrared light 25 , and therefore these three indices should decrease with increasing salinity. In our samples, NIR reflectance shows a negative relationship with salinity while SWIR reflectance shows nearly no correlation with salinity (Fig. 5 and Supplementary 8). These divergent relationships between salinity and these indices and the NIR and SWIR bands suggest that salinity in these coastal soils present themselves differently in surface reflectance as compared to how saline soils are known to behave when there is a lack of moisture, such as in arid and semi-arid contexts 25 , 29 . This positive correlation also confirms the presence of surface water land uses, such as aquaculture, is likely to cause a positive bias in salinity results if not masked. Previous studies, such as Nguyen et al. 33 in coastal Vietnam found NIR reflectance and the Vegetation Soil Salinity Index (VSSI) 13 respectively explaining 89 percent and 77 percent of the variation in electrical conductivity. Both of these predictors are highly sensitive to water, such that as water increases, NIR decreases and VSSI increases. After accounting for surface water land uses, we also find a positive relationship between electrical conductivity and both NIR and VSSI, however, these relationships are significantly weaker as compared to that found by Nguyen et al. 33 . NIR and VSSI only explained 20 percent and 8 percent of electrical conductivity variation, respectively, when tested on pure soil data (Supplementary 2). These findings collectively attest to the necessity of developing alternate measures of salinity in coastal areas than those used in dryer contexts, and the need to address the presence of water-based land uses in these approaches. Stacked ensemble method is an effective approach for modeling coastal salinity Linear, log-linear, and polynomial regression models with single predictors for electrical conductivity found the best fit with NDWI, NDSI2, NIR, NRSWIR2, and NBNIR (Supplementary 2). The explanatory powers of all these predictors individually, however, was low (R-squared less than 0.40). These explanatory powers were also lower compared to what previous studies have presented, a bias potentially owing to unaccounted presence of surface water 31 , 33 . Adding all variables in a single polynomial regression model increased the overall adjusted R-squared as compared to that of individual variables, but the explanatory power of the model remains moderate at 0.57 percent. In this model, most bands remained statistically insignificant (except NRSWIR2 and NBNIR at 95% confidence). Results did not significantly improve with a principal component analysis (PCA) either, although it reduced overfitting with reduced dimensionality. The training and testing accuracies remained low since a PCA still treated the relationships between EC and surface reflectance variables as linear (Fig. 5 shows many non-linear relationships including NDWI). We applied a Lasso regression to remove multicollinear variables. The predictors selected based on the coefficients of the Lasso model included Green, SWIR2, NDWI, NDSI1, VSSI, NBNIR, NBSWIR2, NRSWIR1, NGSWIR1, and NNIRSWIR1 (Supplementary 3). We used these predictors to train 9 machine-learning models. Applying individual machine learning models to this data improved the overall accuracy. Mean accuracy of the models was 0.8496, and the models found to be above this threshold were nnet, svmRadial, knn, glmnet (Fig. 6 ). This accuracy was similar to what was achieved in previous studies, such as Sarkar et al. 43 , but could be improved further. We therefore used a more advanced machine learning algorithm, ensemble stacking, to arrive at the final model for CoSal-SA. Apart from the two surface reflectance bands (SWIR2 and Green), other variables that contribute the most to the CoSal-SA model in predicting salinity include Normalized Difference Water Index (NDWI), Normalized Blue and NIR (NBNIR), Normalized Red and SWIR1 (NRSWIR1), Vegetation Soil Salinity Index (VSSI) 13 , Normalized NIR and SWIR1 (NNIRSWIR1), Normalized Difference Salinity Index (NDSI1) 68 , Normalized Green and SWIR1 (NGSWIR1), and Normalized Blue and SWIR2 (NBSWIR2) (Fig. 7 and Eq. 2). Our correlation analysis (Supplementary 8) shows varying direction of relationship between predictor variables with EC and moisture. NDWI, NBNIR, Green, and VSSI have a positive relationship with EC; NNIRSWIR1, and NDSI1 have a negative relationship with EC; while NRSWIR1, NGSWIR1, and NBSWIR2 have a non-linear relationship with EC. Simultaneously, NRSWIR1, VSSI, NNIRSWIR1, NDSI1, NGSWIR1, and NBSWIR2 have a positive relationship with moisture, SWIR2 and Green have a negative relationship with moisture, while NDWI and NBNIR have a non-linear relationship with moisture. Overall, a combination of these different indices in the CoSal-SA model appears to help identify saline soil in the presence of water. The model is validated by the quality metrics. The training accuracy of the CoSal-SA model is 91.8%, with high precision of 100% and a reasonable recall value of 71%. The high F1-score of 83% suggests it is a balanced model. We get a test accuracy of 85.7%, precision or proportion of accurately classified positive predictions of 80%, and recall or true positive rate of 67%. The model has an F1 score of 72.7% suggesting a balanced performance between true positives and false negatives on unseen data. With the Bangladesh sample, we found a testing accuracy of 69.3%, precision of 71.2%, recall of 82.2% and F1 Score of 76.3%. The accuracy drop on external data is common and expected since there are variations in data collection methods and other differences in the geography. The model maintains high precision and recall, suggesting a reasonable reliability in positive predictions of salinity and the maintained F1-score suggests our model is robust. Long term salinity trends suggest overall reduction across the region except in the coastal belt alongside aquaculture ponds To explore the applicability of CoSal-SA, we applied it to Jagatsinghpur district at two scales - one at the district level, and another about 200 sq.km. study area in the Ersama block in the district’s coastal belt. The study area was identified such that the (lower) half of it was around aquaculture ponds, while the (upper) half was without much aquaculture conversions (Fig. 2 ). Both areas, however, were equally flooded by the 1999 storm surge, and are equally distant from the coast with similar elevations from the sea (less than 5m). This selection was done to account for potential coastal influence on the observed salinity, with the key differentiator primarily being aquaculture. Applying the CoSal-SA model to Jagatsinghpur district’s historical satellite imagery reveals insightful patterns of salinity change (Fig. 8 top panel). In 1995, about 138.2 sq. km of the district’s total area (1668 sq. km) was too saline for rice cultivation. In the years after the 1999 storm surge, salinity increased towards the interiors and became more concentrated over time. It increased from 160.26 sq. km in 2001, to 230.05 sq. km in 2005. This salinity trend, however, declined in the years after 2005, especially in the interior regions, potentially owing to natural leaching processes. By 2010, saline areas had reduced to 199.05 sq. km. and remained somewhat constant up until 2014, however salinity had started to become more concentrated in the coastal belt again. It was also in this period when the district experienced over 70 percent increase in aquaculture activity in the coastal belt, expanding from 66.7 sq.km in 2014 to over 112 sq. km in 2024. By 2024, about 10 percent of the Jagatsinghpur district had salinity levels that were not conducive for rice cultivation, and over 6 percent of area under active aquaculture. Compared to 1995, when about 6 percent of the district had high salinity areas spread across the district, salinity seems to have almost doubled in the district and become more concentrated in the coastal belt (~ 10km from the coastline) over these three decades. Salinity changes in the study area reflect similar trends as were for the district, however, with more closer insights into the salinity dynamics with aquaculture land change in the area. Prior to the 1999 storm surge, salinity was concentrated around the southern part of the study area, and alongside the perennial rivers and deltas containing brackish water from the sea. Salinity became more widespread across both the north and south parts after the surge and became more intense at least until 2005. Over the next few years, the northern parts experienced a reduction in salinity, from 2010 to about 2014, before briefly seeing another period of increased salinity around 2017. Throughout this period, however, the southern part of the study area has sustained high salinity. It was also the same southern part that witnessed a change in land use to aquaculture in the more recent times. Aquaculture land use increased from 32 sq km. in 2010 to about 50 sq. km in 2024, or about half of the lower study region. Given that the two parts of the study area are similar in all ways except the one attribute of aquaculture and prevalence of salinity, there could be empirical exercises undertaken to test the causal pathways for salinity and aquaculture in coastal areas. Discussion and conclusion Soil salinization is an urgent global problem in the face of changing climate and increasing food insecurity 1 – 5 . Coastal soil salinity is dynamic with multiple interconnected natural and anthropogenic drivers, including sea level rise, coastal erosion, rainfall patterns, land use changes, and agricultural practices 8 , 11 . Estimating coastal soil salinity with remote sensing, however, has remained challenging owing to the confounding presence of water 22 , 29 , 31 , 69 , 70 . Insight into the specific channels of salinity increase at a local level, however, requires information on the historical salinity patterns, which has been limited. Important attempts at consolidating national-level data on soil quality and salinity, have been made but are dependent on a country’s participation and data collection costs (e.g. GSASMap) 71 . These attempts, however, have not addressed tracing historical salinity changes at an actionable sub-national or local scale. This study offers a framework to estimate long-term coastal soil salinity in the context of soil moisture and confounding water-based land uses. This approach offers a means to study salinity changes over time at varying geographical scales, to enable future causal inferential research. We find that soil salinity in coastal India presents itself differently in surface reflectance as compared to how it is observed in arid and semi-arid contexts 25 , 29 . This justifies our undertaking to develop a soil salinity model more suitable for coastal areas with high moisture than the models developed for arid and semi-arid regions. We also find that univariate models based on indices sensitive to the presence of water cannot be used to estimate soil salinity, due to the growing presence of water-based land uses. Our findings hold up against evidence presented by Silvestri et al. 31 that the presence of water in aquaculture ponds causes a bias in the explanatory power of certain bands and indices (e.g. NIR and VSSI). We argue that water-based land uses must be accounted for before modeling salinity in coastal areas, and over reliance on a single band or parameter must be avoided. Despite their coarser resolution, we demonstrate the utility of Landsat imagery in conducting soil salinity analysis, by accounting for mixed pixels with partial proportion of water before correlating with soil salinity values 31 . The additional step of performing a supervised land use land cover classification in the framework in order to mask water-based land uses such as aquaculture, is key for reducing bias while unlocking the advantage of long-term satellite data. While machine learning based models help estimate non-linear relationships between multiple variables and salinity, their “black box” tendencies keep the direction and magnitude of influence of different variables opaque 65 . We use Shapley feature values to assess the specific contributions of all features behind the model, enabling us to make a better judgement of the model. For instance, we find a strong contribution of NDWI in the CoSal-SA model, suggesting its ability to detect salinity in the context of high soil moisture. This holds up against the results from previous soil science studies conducted in coastal and wetland contexts with varying levels of moisture 29 , 69 . Most other indices in the model employ Red, SWIR and NIR bands, which are widely used for soil mineral and moisture detection 72 . Meanwhile, the ultimate challenge related to soil salinity facing science and policy is to identify and address its causes and consequences. This requires better understanding of the various causal pathways in a context where multiple drivers, such as coastal erosion, agricultural practices, and sea-water intrusion, are at play. While we do not claim any causal inferences in this study, our findings give some insights in that direction. Previous studies have found that soil moisture loss increases top-soil salinity due to capillary action that moves salts to the surface 73 . The positive correlation between the presence of moisture and higher levels of salinity in this coastal context is perhaps due to the mediating presence of saline or brackish water that is both associated with high salinity and high moisture. Further to this end, the spatial-temporal findings from Jagatsinghpur also highlight the increasing salinity in the interiors after the storm surge, and in the coastal belt alongside growing aquaculture, both related to saline water intrusion. One could argue that salinity in the coastal belt is due to the proximity to the sea, however, our comparative study area results show that parts of the coast with no aquaculture activity have lower salinity compared to those with aquaculture transitions, despite being equally distant, with equal elevation from the coast and having similar history of storm surge impact. A farm-level temporal analysis of salinity and land change can give specific insights into the causal relationships between aquaculture and soil salinity. The CoSal-SA model is, however, not without its limitations. It currently does not factor soil type and color, which may influence the visible bands of the spectrum. Other factors, such as pH were not included because of the low correlation between EC and pH in the field samples. In other contexts where acidity and EC are more closely related, pH could be included in the assessment, similar to moisture. More soil samples could perhaps improve the model further, however, our numbers are similar to other comparable studies 25 , 33 , 74 . We substantiate the results by adding Bangladesh soil data, to gain better insights into the external generalizability of the model, which to the best of our knowledge is done for the first time in a salinity modelling setting. This research identifies the distinctive property of coastal soils that exhibit both high salinity and moisture levels. Through the innovative CoSal framework, the study offers a thorough methodology for precisely tracking historical patterns of coastal salinity across various geographical settings, including areas where new coastal land uses like aquaculture are emerging. To the best of our knowledge, this is the first framework that comprehensively addresses the presence of water-based land uses in coastal areas into a soil salinity assessment. Additional coastal soil data, with similar salinity and moisture measurement approaches, could help adapt the framework to other geographical and historical contexts in the future. The ability to map salinity at a detailed level opens the door to critically needed research on the specific local mechanisms driving coastal salinity, which can guide focused interventions to address significant agricultural productivity losses, food insecurity, and the resulting poverty in affected regions. Declarations Competing Interest Statement The author(s) declare no competing interests. Author Contribution G.J., W.Y., A.E.F., and D.S.C. conceptualized the study; G.J., W.Y., and S.K.S. performed the methodology; G.J., W.Y., and S.K.S. supported data collection and visuals; all authors reviewed and edited the manuscript. Acknowledgement The authors would like to thank Germán Silva, Krishnachandran Balakrishnan, Teja Malladi, Pratyush Tripathy, Jiwon Jang, Nicole Corcoran, and B L Turner II for their inputs at various stages of the research. We would also like to thank Greenforce Labs and Deepthi Nagappa for their support with the field data collection. This study is supported by the US National Science Foundation (NSF Award Number 2409616), Intergovernmental Panel for Climate Change (IPCC), Cuomo Foundation, and Horowitz Foundation for Social Policy. This study would not have been possible without the support of Arizona State University and its resources. Data Availability All data, codes, and supplementary material from this study are available on the GitHub repository: https://github.com/garimajain2002/coastal_salinity_index.git. Further information may be available from the corresponding author based on request. References Díaz, S. et al. Summary for Policymakers of the IPBES Global Assessment Report on Biodiversity and Ecosystem Services . 56 (2019). https://ipbes.net/system/tdf/ipbes_global_assessment_report_summary_for_policymakers.pdf?file=1&type=node&id=35329 IPCC. Summary for Policymakers. In: Climate Change and Land: An IPCC Special Report on Climate Change, Desertification, Land Degradation, Sustainable Land Management, Food Security, and Greenhouse Gas Fluxes in Terrestrial Ecosystems (Cambridge University Press, 2019). Olsson, L. et al. Land degradation. in Climate Change and Land: an IPCC special report on climate change, desertification, land degradation, sustainable land management, food security, and greenhouse gas fluxes in terrestrial ecosystems 345–436 (Cambridge University Press, (2022). Qadir, M. et al. Economics of salt-induced land degradation and restoration. Nat. Resour. Forum . 38 , 282–295 (2014). Wicke, B. et al. The global technical and economic potential of bioenergy from salt-affected soils. Energy Environ. Sci. 4 , 2669–2681 (2011). FAO. Global Map of salt-affected soils. GSASMap v1.0 (2021). https://openknowledge.fao.org/server/api/core/bitstreams/31be1fac-a057-4b6b-80ea-a4554910368c/content Amoako Johnson, F., Hutton, C. W., Hornby, D., Lázár, A. N. & Mukhopadhyay, A. Is shrimp farming a successful adaptation to salinity intrusion? A geospatial associative analysis of poverty in the populous Ganges-Brahmaputra-Meghna Delta of Bangladesh. Sustain. Sci. 11 , 423–439 (2016). Hossain, M. S., Uddin, M. J. & Fakhruddin, A. N. M. Impacts of shrimp farming on the coastal environment of Bangladesh and approach for management. Rev. Environ. Sci. Biotechnol. 12 , 313–332 (2013). van Schie, D. et al. Addressing non-economic loss and damage: learning from autonomous responses in Bangladesh. Clim. Change . 177 , 1–22 (2024). Chen, J. & Mueller, V. Salt of the earth: Migration, adaptation, and soil salinity in coastal Bangladesh. Nat. Clim. Change (2018). Eswar, D., Karuppusamy, R. & Chellamuthu, S. Drivers of soil salinity and their correlation with climate change. Curr. Opin. Environ. Sustain. 50 , 310–318 (2021). Sen, R. Salt in the wound: embodied everyday adaptations to salinity intrusion in the Sundarbans. Ecol. Soc. 28 , (2023). Dehni, A. & Lounis, M. Remote sensing techniques for salt affected soil mapping: Application to the Oran region of Algeria. Procedia Eng. 33 , 188–198 (2012). Sá, T. D., de Sousa, R. R., Rocha, Í. R. C., de Lima, G. C. & Costa, F. H. F. Brackish shrimp farming in Northeastern Brazil: The environmental and Socio-economic impacts and sustainability. Natural Resources 538–550 (2013). (2013). Ali, A. M. S. Rice to shrimp: Land use/land cover changes and soil degradation in Southwestern Bangladesh. Land. Use Policy . 23 , 421–435 (2006). Belton, B. Shrimp, prawn and the political economy of social wellbeing in rural Bangladesh. J. Rural Stud. 45 , 230–242 (2016). Zhang, W. et al. Aquaculture will continue to depend more on land than sea. Nature 603 , E2–E4 (2022). Ottinger, M., Clauss, K., Kuenzer, C. & Aquaculture Relevance, distribution, impacts and spatial assessments – A review. Ocean. Coast Manag . 119 , 244–266 (2016). Azad, A. K., Jensen, K. R. & Lin, C. K. Coastal aquaculture development in Bangladesh: unsustainable and sustainable experiences. Environ. Manage. 44 , 800–809 (2009). Primavera, J. H. Overcoming the impacts of aquaculture on the coastal zone. Ocean. Coast Manag . 49 , 531–545 (2006). Gorji, T., Yildirim, A., Sertel, E. & Tanik, A. Remote sensing approaches and mapping methods for monitoring soil salinity under different climate regimes. Int. J. Environ. Geoinformatics . 10.30897/IJEGEO.500452 (2019). Metternicht, G. I. & Zinck, J. A. Remote sensing of soil salinity: potentials and constraints. Remote Sens. Environ. 85 , 1–20 (2003). Morshed, M. M., Islam, M. T. & Jamil, R. Soil salinity detection from satellite image analysis: an integrated approach of salinity indices and field data. Environ. Monit. Assess. 188 , 119 (2016). Farifteh, J., Van der Meer, F., Atzberger, C. & Carranza, E. J. M. Quantitative analysis of salt-affected soil reflectance spectra: A comparison of two adaptive methods (PLSR and ANN). Remote Sens. Environ. 110 , 59–78 (2007). Abuelgasim, A. & Ammad, R. Mapping soil salinity in arid and semi-arid regions using Landsat 8 OLI satellite data. Remote Sens. Applications: Soc. Environ. 13 , 415–425 (2019). Allbed, A. & Kumar, L. Soil Salinity Mapping and Monitoring in Arid and Semi-Arid Regions Using Remote Sensing Technology: A Review. Advances in Remote Sensing (2013). (2013). Bannari, A. & Al-Ali, Z. M. Assessing climate change impact on soil salinity dynamics between 1987–2017 in arid landscape using Landsat TM, ETM + and OLI data. Remote Sens. (Basel) . 12 , 2794 (2020). Elshewy, M. A., Mohamed, M. H. A. & Refaat, M. Developing a soil salinity model from Landsat 8 satellite bands based on advanced machine learning algorithms. J. Ind. Soc. Remote Sens. 52 , 617–632 (2024). Yang, X. & Yu, Y. Estimating soil salinity under various moisture conditions: An experimental study. IEEE Trans. Geosci. Remote Sens. 55 , 2525–2533 (2017). Allbed, A., Kumar, L. & Aldakheel, Y. Y. Assessing soil salinity using soil salinity and vegetation indices derived from IKONOS high-spatial resolution imageries: Applications in a date palm dominated region. Geoderma 230–231 , 1–8 (2014). Silvestri, S., Nguyen, D. N. & Chiapponi, E. Comment on ‘Soil salinity assessment by using near-infrared channel and Vegetation Soil Salinity Index derived from Landsat 8 OLI data: a case study in the Tra Vinh Province, Mekong Delta, Vietnam’ by Kim-Anh Nguyen, Yuei-An Liou, Ha-Phuong Tran, Phi-Phung Hoang and Thanh-Hung Nguyen. Progress Earth Planet. Sci. 9 , 1–8 (2022). Lee, S. & Lathrop, R. G. Sub-pixel estimation of urban land cover components with linear mixture model analysis and Landsat Thematic Mapper imagery. Int. J. Remote Sens. 26 , 4885–4905 (2005). Nguyen, K. A., Liou, Y. A., Tran, H. P., Hoang, P. P. & Nguyen, T. H. Soil salinity assessment by using near-infrared channel and Vegetation Soil Salinity Index derived from Landsat 8 OLI data: a case study in the Tra Vinh Province, Mekong Delta, Vietnam. Progress Earth Planet. Sci. 7 , 1–16 (2020). Morshed, M. M., Sarkar, S. K., Zzaman, M. R. U. & Islam, M. M. Application of remote sensing for salinity based coastal land use zoning in Bangladesh. Spat. Inf. Res. 29 , 353–364 (2021). Sarkar, S. K., Rudra, R. R., Nur, M. S. & Das, P. C. Partial least-squares regression for soil salinity mapping in Bangladesh. Ecol. Indic. 154 , 110825 (2023). Mahajan, G. R. et al. Monitoring properties of the salt-affected soils by multivariate analysis of the visible and near-infrared hyperspectral data. Catena 198 , 105041 (2021). An, D. et al. Hyperspectral field estimation and remote-sensing inversion of salt content in coastal saline soils of the Yellow River Delta. Int. J. Remote Sens. 37 , 455–470 (2016). Davis, E., Wang, C. & Dow, K. Comparing Sentinel-2 MSI and Landsat 8 OLI in soil salinity detection: a case study of agricultural lands in coastal North Carolina. Int. J. Remote Sens. 40 , 6134–6153 (2019). Eldeiry, A. A. & Garcia, L. A. Detecting soil salinity in alfalfa fields using spatial modeling and remote sensing. Soil. Sci. Soc. Am. J. 72 , 201–211 (2008). Pang, G., Wang, T., Liao, J. & Li, S. Quantitative model based on field-derived spectral characteristics to estimate soil salinity in minqin county, China. Soil. Sci. Soc. Am. J. 78 , 546–555 (2014). Fadl, M. E. et al. Soil salinity assessing and mapping using several statistical and distribution techniques in arid and semi-arid ecosystems, Egypt. Agron. (Basel) . 13 , 583 (2023). Gómez, A. M. R. et al. Digital mapping of the soil available water capacity: tool for the resilience of agricultural systems to climate change. Sci. Total Environ. 882 , (2023). Sarkar, S. K. et al. Coupling of machine learning and remote sensing for soil salinity mapping in coastal area of Bangladesh. Sci. Rep. 13 , 17056 (2023). Grattan, S. R., Zeng, L., Shannon, M. C. & Roberts, S. R. Rice is more sensitive to salinity than previously thought. Calif. Agric. (Berkeley) . 56 , 189–195 (2002). FAO. The state of food security and nutrition in the world 2024. Preprint at (2024). https://doi.org/10.4060/cd1254en Sharma, D. K. & Singh, A. Salinity research in India - achievements, challenges and future prospects. Water Energy Int. 35–45 (2015). Kumar, P. & Sharma, P. K. Soil Salinity and Food Security in India. Front. Sustainable Food Syst. 4 , (2020). NRSC-ISRO. Status of Land Degradation in India: 2015-16 . (2019). https://bhuvan-app1.nrsc.gov.in/2dresources/thematic/ld0506/ATLASLD.pdf Census of India. District Statistical Handbook. Preprint at (2011). https://jagatsinghpur.odisha.gov.in/sites/default/files/2023-05/2018051173.pdf (2011). Government of Odisha. Odisha District Gazetteers - Jagatsinghpur. Preprint at (2018). https://jagatsinghpur.odisha.gov.in/sites/default/files/2023-05/2018051050.pdf Mahata, K. R., Singh, D. P., Saha, S., Ismail, A. M. & Haefele, S. M. Improving rice productivity in the coastal saline soils of the Mahanadi Delta of India through integrated nutrient management. in Tropical deltas and coastal zones: Food production, communities and environment at the land and water interface 239–248 (CABI, UK, (2010). Rafi, S., Mourya, N. K. & Balasani, R. Evaluation of shoreline alteration along theJagatsinghpur district coast, India (1990–2020) using DSAS. Ocean. Coast Manag . 253 , 107132 (2024). Mohapatra, M., Mandal, G. S., Bandyopadhyay, B. K., Tyagi, A. & Mohanty, U. C. Classification of cyclone hazard prone districts of India. Nat. Hazards . 63 , 1601–1620 (2012). Mandal, U. K. et al. Delineation of saline soils in coastal India using satellite remote sensing. Curr. Sci. 125 , 1339–1353 (2023). DDMA. District Disaster Management Plan: Jagatsinghpur. Preprint at (2019). https://www.osdma.org/districtplan/jagatsinghpur/#gsc.tab=0 Chhotray, V. A supercyclone, landscapes of ‘emptiness’ and shrimp aquaculture: The lesser-known trajectories of disaster recovery in coastal Odisha, India. World Dev. 153 , 105823 (2022). Pekel, J. F., Cottam, A., Gorelick, N. & Belward, A. S. High-resolution mapping of global surface water and its long-term changes. Nature 540 , 418–422 (2016). USGS. How do I use a scale factor with Landsat Level-2 science products? USGS (2023). https://www.usgs.gov/faqs/how-do-i-use-a-scale-factor-landsat-level-2-science-products USGS. Landsat Project Documents. USGS https://www.usgs.gov/landsat-missions/landsat-project-documents USGS. Why are negative values observed over water in some Landsat Surface Reflectance products? USGS (2022). https://www.usgs.gov/faqs/why-are-negative-values-observed-over-water-some-landsat-surface-reflectance-products Breiman, L. Bagging predictors. Mach. Learn. 24 , 123–140 (1996). Deane-Mayer, Z. A., Knowles, J. E. & López, A. caretEnsemble: Ensembles of Caret Models . https://cran.r-project.org/web/packages/caretEnsemble/caretEnsemble.pdf (2024). 10.32614/CRAN.package.caretEnsemble Xu, Y. et al. Classification using ensemble learning under weighted misclassification loss: Ensemble Learning under Weighted Misclassification Loss. Stat. Med. 38 , 2002–2012 (2019). Aznar-Gimeno, R., Esteban, L. & Sanz, G. del-Hoyo-Alonso, R. Comparing the min-max-Median/IQR approach with the min-max approach, logistic regression and XGBoost, maximising the Youden index. Symmetry (Basel) . 15 , 756 (2023). Rodríguez-Pérez, R. & Bajorath, J. Interpretation of machine learning models using shapley values: application to compound potency and multi-target activity predictions. J. Comput. Aided Mol. Des. 34 , 1013–1026 (2020). Shapley, L. S. In Contributions to the Theory of Games. Annals of Mathematical Studies. (1953). Richards, J. A. & Xiuping, J. Remote Sensing Digital Image Analysis: An Introduction (Springer, 2005). Henrich, V., Krauss, G., Götze, C. & Sandow, C. IDB - Index DataBase. Index. DataBase (2012). https://www.indexdatabase.de/ Shahmoradi, S., Malmiri, G., Sharifi Pichoon, M. & H. R. & Modeling and mapping of soil salinity and moisture using spectral and radar remote sensing. Appl. Soil. Res. 10 , 43–65 (2022). Guo, Y. et al. Integrating remote sensing and proximal sensors for the detection of soil moisture and salinity variability in coastal areas. J. Integr. Agric. 12 , 723–731 (2013). FAO. GSASmap. (2021). https://www.fao.org/global-soil-partnership/gsasmap/en USGS. What are the best Landsat spectral bands for use in my research? USGS (2022). https://www.usgs.gov/faqs/what-are-best-landsat-spectral-bands-use-my-research Fu, Z. et al. Composition, seasonal variation, and salinization characteristics of soil salinity in the Chenier Island of the Yellow River Delta. Glob Ecol. Conserv. 24 , e01318 (2020). Abbas, A., Khan, S., Hussain, N., Hanjra, M. A. & Akbar, S. Characterizing soil salinity in irrigated agriculture using a remote sensing approach. Phys. Chem. Earth (2002) 55–57, 43–52 (2013). Additional Declarations No competing interests reported. Supplementary Files 20250306NSRCoSalJainSUPPLEMENTARY.docx Cite Share Download PDF Status: Posted Version 1 posted 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-6173117","acceptedTermsAndConditions":true,"allowDirectSubmit":true,"archivedVersions":[],"articleType":"Article","associatedPublications":[],"authors":[{"id":431110947,"identity":"dc71c11a-6c0f-418f-b766-5af06b701358","order_by":0,"name":"Garima Jain","email":"data:image/png;base64,iVBORw0KGgoAAAANSUhEUgAAAZAAAAAyAQMAAABI0h/eAAAABlBMVEX///8AAABVwtN+AAAACXBIWXMAAA7EAAAOxAGVKw4bAAAAt0lEQVRIiWNgGAWjYFCCA0BYYQNh8xCv5UwaSVqAgLHtMAla5B0PPzxc2HZeXn5GAuODt21EaDE8cMzg8Ixztw033EhgNpxLlJaGAwaHecpuM26QSGCT5iVOy/EPh3nYztnPn5HA/psoLfIMZ4C2tB1IbLiRwMZMlBYDhjMFh3nOJCdvOPOwWXLOOWJsmXF882eeCjvb+e3JBz+8KSPGlhsHYEzGBiLUg2zpJ1LhKBgFo2AUjGAAANeaPiY75rKuAAAAAElFTkSuQmCC","orcid":"","institution":"Arizona State University","correspondingAuthor":true,"prefix":"","firstName":"Garima","middleName":"","lastName":"Jain","suffix":""},{"id":431110949,"identity":"c2afd36f-26f2-4ec1-928e-aa80fdb26c1f","order_by":1,"name":"Wenxin Yang","email":"","orcid":"","institution":"University of California, Santa Barbara","correspondingAuthor":false,"prefix":"","firstName":"Wenxin","middleName":"","lastName":"Yang","suffix":""},{"id":431110950,"identity":"6eb971af-415c-49d4-892c-319ecdc10cfb","order_by":2,"name":"Showmitra Kumar Sarkar","email":"","orcid":"","institution":"Khulna University of Engineering and Technology","correspondingAuthor":false,"prefix":"","firstName":"Showmitra","middleName":"Kumar","lastName":"Sarkar","suffix":""},{"id":431110951,"identity":"8e0dc5b0-08b7-48be-b6e5-4db05858e3f5","order_by":3,"name":"Amy E. Frazier","email":"","orcid":"","institution":"University of California, Santa Barbara","correspondingAuthor":false,"prefix":"","firstName":"Amy","middleName":"E.","lastName":"Frazier","suffix":""},{"id":431110952,"identity":"b4966ca2-609b-4d30-8373-6ce6b37c47f8","order_by":4,"name":"Dylan S. Connor","email":"","orcid":"","institution":"Arizona State University","correspondingAuthor":false,"prefix":"","firstName":"Dylan","middleName":"S.","lastName":"Connor","suffix":""}],"badges":[],"createdAt":"2025-03-06 19:23:14","currentVersionCode":1,"declarations":"","doi":"10.21203/rs.3.rs-6173117/v1","doiUrl":"https://doi.org/10.21203/rs.3.rs-6173117/v1","draftVersion":[],"editorialEvents":[],"editorialNote":"","failedWorkflow":false,"files":[{"id":79176777,"identity":"59eafdbd-56b3-4947-b852-88ae24418c5e","added_by":"auto","created_at":"2025-03-25 10:02:33","extension":"jpg","order_by":1,"title":"Figure 1","display":"","copyAsset":false,"role":"figure","size":117229,"visible":true,"origin":"","legend":"\u003cp\u003eColSal Modelling Framework for mapping historical coastal salinity in South Asia\u003c/p\u003e","description":"","filename":"Picture1.jpg","url":"https://assets-eu.researchsquare.com/files/rs-6173117/v1/61def35d0bfd3ec2d462ff62.jpg"},{"id":79178921,"identity":"29e2c7a7-3e33-496a-a7e2-057b2a228655","added_by":"auto","created_at":"2025-03-25 10:10:33","extension":"jpg","order_by":2,"title":"Figure 2","display":"","copyAsset":false,"role":"figure","size":191922,"visible":true,"origin":"","legend":"\u003cp\u003eSoil Sampling Area: Two adjacent areas of 100 sq.km. each were predetermined such that both were equally distant from the coast, had similar elevation from the sea (max. 5m), and had a similar history of being affected by storm surges in the past. The only differentiating aspect is the presence of aquaculture (more in the lower grid than in the upper grid). The points here show the sample locations and their electrical conductivity (EC) levels. The figure shows higher EC in the grid with higher aquaculture prevalence compared to the area with lower aquaculture prevalence.\u003c/p\u003e","description":"","filename":"Picture2.jpg","url":"https://assets-eu.researchsquare.com/files/rs-6173117/v1/d3f8ab0657af4f2d6ae26b4b.jpg"},{"id":79176779,"identity":"e560e319-fa60-4098-b2fa-80150e5d1e6a","added_by":"auto","created_at":"2025-03-25 10:02:33","extension":"jpg","order_by":3,"title":"Figure 3","display":"","copyAsset":false,"role":"figure","size":225272,"visible":true,"origin":"","legend":"\u003cp\u003eSupervised Land Use Land Cover \u0026nbsp;\u0026nbsp;classification. Using a Sentinel-2 composite (top) as a visual reference and \u0026nbsp;\u0026nbsp;field observations for 274 active aquaculture points, 220 dry/abandoned \u0026nbsp;\u0026nbsp;aquaculture points, and 391 non-aquaculture or other uses, a classifier is \u0026nbsp;\u0026nbsp;trained on the Landsat 8 composite (middle). The trained classifier is \u0026nbsp;\u0026nbsp;applied to Feb 2024 (bottom) and historical Landsat composites.\u003c/p\u003e","description":"","filename":"Picture3.jpg","url":"https://assets-eu.researchsquare.com/files/rs-6173117/v1/66397e9503ce0615500b8843.jpg"},{"id":79176785,"identity":"7c3724d7-0c24-4dfb-8ae1-c6ab0858c64f","added_by":"auto","created_at":"2025-03-25 10:02:34","extension":"jpg","order_by":4,"title":"Figure 4","display":"","copyAsset":false,"role":"figure","size":31145,"visible":true,"origin":"","legend":"\u003cp\u003eMasking Landsat \u0026nbsp;\u0026nbsp;composite to extract pixel values at Field Sample Points. Seven field samples \u0026nbsp;\u0026nbsp;were eliminated as a result of masking (in red).\u003c/p\u003e","description":"","filename":"4.jpg","url":"https://assets-eu.researchsquare.com/files/rs-6173117/v1/91100c5cc0c8d041045721b6.jpg"},{"id":79176783,"identity":"be7b7833-0eb7-453c-bcad-198e97c93ffc","added_by":"auto","created_at":"2025-03-25 10:02:33","extension":"jpg","order_by":5,"title":"Figure 5","display":"","copyAsset":false,"role":"figure","size":155651,"visible":true,"origin":"","legend":"\u003cp\u003eRelationships between indices featuring as high contributors in the CoSal-SA model, with EC and moisture. Different indices detect salinity and moisture in varying ways: Indices, such as NDWI, increase for higher levels of EC as well as moisture. Other indices, such as SWIR2 and Green only detect water or salinity respectively; Indices such as NRSWIR1, have a non-linear relationship with salinity but a positive relationship with moisture. See Supplementary 8 for all index comparisons.\u003c/p\u003e","description":"","filename":"5.jpg","url":"https://assets-eu.researchsquare.com/files/rs-6173117/v1/8e38968bea192cc1ce80d8d4.jpg"},{"id":79178922,"identity":"9e0233ab-0966-4f32-9f9e-f0da53915036","added_by":"auto","created_at":"2025-03-25 10:10:33","extension":"jpg","order_by":6,"title":"Figure 6","display":"","copyAsset":false,"role":"figure","size":27255,"visible":true,"origin":"","legend":"\u003cp\u003eComparing model overall accuracies and \u0026nbsp;\u0026nbsp;Kappa coefficients for the nine machine learning models\u003c/p\u003e","description":"","filename":"6.jpg","url":"https://assets-eu.researchsquare.com/files/rs-6173117/v1/c495a1e1be9dd53e1e20983e.jpg"},{"id":79178924,"identity":"f4f1b2f3-4079-48a4-aa4f-2a4ab605ce97","added_by":"auto","created_at":"2025-03-25 10:10:33","extension":"jpg","order_by":7,"title":"Figure 7","display":"","copyAsset":false,"role":"figure","size":42650,"visible":true,"origin":"","legend":"\u003cp\u003eShapley Mean Importance Values of predictive power in the Ensemble Model\u003c/p\u003e","description":"","filename":"7.jpg","url":"https://assets-eu.researchsquare.com/files/rs-6173117/v1/9df219ab1e41913ae7d3779f.jpg"},{"id":79176792,"identity":"f9845edf-9159-4d37-9bcf-5c1e7c297efc","added_by":"auto","created_at":"2025-03-25 10:02:34","extension":"jpg","order_by":8,"title":"Figure 8","display":"","copyAsset":false,"role":"figure","size":401248,"visible":true,"origin":"","legend":"\u003cp\u003eHistorical predictions of salinity and aquaculture in Jagatsinghpur District (top half) and the study area (bottom half) within that.\u003c/p\u003e","description":"","filename":"8.jpg","url":"https://assets-eu.researchsquare.com/files/rs-6173117/v1/a6aec939243bab489cd6e02a.jpg"},{"id":90588655,"identity":"be6f6c45-bf3d-49f9-881d-3c240eee0bea","added_by":"auto","created_at":"2025-09-04 12:02:13","extension":"pdf","order_by":0,"title":"","display":"","copyAsset":false,"role":"manuscript-pdf","size":2095363,"visible":true,"origin":"","legend":"","description":"","filename":"manuscript.pdf","url":"https://assets-eu.researchsquare.com/files/rs-6173117/v1/bcd347db-dc24-409b-b146-6fbff9091b79.pdf"},{"id":79178930,"identity":"3ea43f3a-b851-4070-98d6-0c7e717a678c","added_by":"auto","created_at":"2025-03-25 10:10:34","extension":"docx","order_by":0,"title":"","display":"","copyAsset":false,"role":"supplement","size":3065634,"visible":true,"origin":"","legend":"","description":"","filename":"20250306NSRCoSalJainSUPPLEMENTARY.docx","url":"https://assets-eu.researchsquare.com/files/rs-6173117/v1/f4816fd9fadde88463b7200c.docx"}],"financialInterests":"No competing interests reported.","formattedTitle":"CoSal: A remote sensing and machine learning framework for mapping coastal soil salinity trends around aquaculture in South Asia","fulltext":[{"header":"Background and Summary","content":"\u003cp\u003eLand change and degradation are second only to climate change in being the most influential drivers of ecosystem collapse \u003csup\u003e \u003cspan additionalcitationids=\"CR2\" citationid=\"CR1\" class=\"CitationRef\"\u003e1\u003c/span\u003e\u0026ndash;\u003cspan citationid=\"CR3\" class=\"CitationRef\"\u003e3\u003c/span\u003e \u003c/sup\u003e. Among degraded lands, salt-affected soils, characterized by the presence of soluble salts in quantities that inhibit plant growth, cover over 1,128\u0026nbsp;million ha of global arable lands resulting in lowered agricultural productivity costing the world US\u003cspan\u003e$\u003c/span\u003e12-27.3\u0026nbsp;billion annually \u003csup\u003e \u003cspan additionalcitationids=\"CR5\" citationid=\"CR4\" class=\"CitationRef\"\u003e4\u003c/span\u003e\u0026ndash;\u003cspan citationid=\"CR6\" class=\"CitationRef\"\u003e6\u003c/span\u003e \u003c/sup\u003e. Soil salinization demands urgent attention to curb food insecurity, social conflicts, distress migration, and deep poverty prevalence \u003csup\u003e \u003cspan additionalcitationids=\"CR8 CR9\" citationid=\"CR7\" class=\"CitationRef\"\u003e7\u003c/span\u003e\u0026ndash;\u003cspan citationid=\"CR10\" class=\"CitationRef\"\u003e10\u003c/span\u003e \u003c/sup\u003e. There is growing evidence of increasing salinity in coastal areas due to simultaneously occurring seawater intrusion, poor agricultural practices, and water-logging including land conversions to fish farms or aquaculture \u003csup\u003e \u003cspan citationid=\"CR8\" class=\"CitationRef\"\u003e8\u003c/span\u003e,\u003cspan additionalcitationids=\"CR12\" citationid=\"CR11\" class=\"CitationRef\"\u003e11\u003c/span\u003e\u0026ndash;\u003cspan citationid=\"CR13\" class=\"CitationRef\"\u003e13\u003c/span\u003e \u003c/sup\u003e. Aquaculture, the practice of farming fish on land, is found to be both a driver and an outcome of salinity in coastal areas \u003csup\u003e \u003cspan citationid=\"CR7\" class=\"CitationRef\"\u003e7\u003c/span\u003e,\u003cspan additionalcitationids=\"CR15\" citationid=\"CR14\" class=\"CitationRef\"\u003e14\u003c/span\u003e\u0026ndash;\u003cspan citationid=\"CR16\" class=\"CitationRef\"\u003e16\u003c/span\u003e \u003c/sup\u003e. It is the fastest-growing animal food production sector, and is expected to expand mainly on coastal lands in the coming decades \u003csup\u003e \u003cspan additionalcitationids=\"CR18 CR19\" citationid=\"CR17\" class=\"CitationRef\"\u003e17\u003c/span\u003e\u0026ndash;\u003cspan citationid=\"CR20\" class=\"CitationRef\"\u003e20\u003c/span\u003e \u003c/sup\u003e, making it an active area of research and policy concern. Our lack of understanding of the historical salinity dynamics at the local level limits us from identifying the specific causal pathways of salinity increase and taking targeted actions.\u003c/p\u003e \u003cp\u003eRemote sensing is increasingly used for mapping soil salinity as it provides consistent data across time, spatial scales, and diverse climatic contexts \u003csup\u003e\u003cspan additionalcitationids=\"CR22 CR23\" citationid=\"CR21\" class=\"CitationRef\"\u003e21\u003c/span\u003e\u0026ndash;\u003cspan citationid=\"CR24\" class=\"CitationRef\"\u003e24\u003c/span\u003e\u003c/sup\u003e. However, a global review \u003csup\u003e\u003cspan citationid=\"CR21\" class=\"CitationRef\"\u003e21\u003c/span\u003e\u003c/sup\u003e has found that 65 percent of remote-sensing-based salinity identification methods have been developed in arid and semi-arid regions \u003csup\u003ee.g. \u003cspan additionalcitationids=\"CR26 CR27\" citationid=\"CR25\" class=\"CitationRef\"\u003e25\u003c/span\u003e\u0026ndash;\u003cspan citationid=\"CR28\" class=\"CitationRef\"\u003e28\u003c/span\u003e\u003c/sup\u003e, while coastal areas\u0026mdash;home to 40 percent of the global population and significant agricultural lands\u0026mdash;remain understudied. Coastal soils are a complex composite of minerals and saline deposits alongside high moisture and organic matter. These components simultaneously interfere with incoming light, making remote salinity detection complex by increasing spectral confusion, especially between salts and moisture \u003csup\u003e\u003cspan citationid=\"CR29\" class=\"CitationRef\"\u003e29\u003c/span\u003e\u003c/sup\u003e. Clear water scatters some blue and green light but absorbs most red, making it appear darker in true color composites, whereas dry saline or bare soils are highly reflective, particularly in the red, and NIR/SWIR spectrum. High salinity, often associated with low vegetation cover, further heightens red reflectance by reducing photosynthesis activity that tends to absorb red spectrum light \u003csup\u003e\u003cspan citationid=\"CR22\" class=\"CitationRef\"\u003e22\u003c/span\u003e,\u003cspan citationid=\"CR30\" class=\"CitationRef\"\u003e30\u003c/span\u003e\u003c/sup\u003e. Such spectral interference is particularly problematic in coarse-resolution imagery like 30m Landsat, where mixed pixels, containing several surface components including bare soil and surface water, are common \u003csup\u003e\u003cspan citationid=\"CR29\" class=\"CitationRef\"\u003e29\u003c/span\u003e,\u003cspan citationid=\"CR31\" class=\"CitationRef\"\u003e31\u003c/span\u003e,\u003cspan citationid=\"CR32\" class=\"CitationRef\"\u003e32\u003c/span\u003e\u003c/sup\u003e. While higher-resolution data from Sentinel-2 (10m) could reduce this error, such imagery lacks the long historical record needed to track salinity changes over time.\u003c/p\u003e \u003cp\u003eAccurately mapping soil salinity in coastal environments thus requires adjusting for the presence of water-based land uses, such as aquaculture, and integrating multiple surface reflectance measures. Recent efforts have developed remote sensing-based salinity models for coastal areas \u003csup\u003e\u003cspan additionalcitationids=\"CR34 CR35\" citationid=\"CR33\" class=\"CitationRef\"\u003e33\u003c/span\u003e\u0026ndash;\u003cspan citationid=\"CR36\" class=\"CitationRef\"\u003e36\u003c/span\u003e\u003c/sup\u003e, but these models often do not distinguish between the signals created from soil moisture and those from water-based land uses such as aquaculture, which is found to bias their results \u003csup\u003e\u003cspan citationid=\"CR31\" class=\"CitationRef\"\u003e31\u003c/span\u003e\u003c/sup\u003e. Besides, traditional modeling approaches rely on univariate linear regressions between electrical conductivity and surface reflectance to predict salinity \u003csup\u003e\u003cspan citationid=\"CR33\" class=\"CitationRef\"\u003e33\u003c/span\u003e,\u003cspan additionalcitationids=\"CR38 CR39 CR40\" citationid=\"CR37\" class=\"CitationRef\"\u003e37\u003c/span\u003e\u0026ndash;\u003cspan citationid=\"CR41\" class=\"CitationRef\"\u003e41\u003c/span\u003e\u003c/sup\u003e, but have lower accuracy likely due to the non-linear relationships and soil-moisture confounding\u003csup\u003e\u003cspan citationid=\"CR24\" class=\"CitationRef\"\u003e24\u003c/span\u003e\u003c/sup\u003e. Modeling multiple surface reflectance measures and capturing non-linear relationships using machine learning has shown higher accuracy \u003csup\u003e\u003cspan citationid=\"CR24\" class=\"CitationRef\"\u003e24\u003c/span\u003e,\u003cspan citationid=\"CR28\" class=\"CitationRef\"\u003e28\u003c/span\u003e,\u003cspan citationid=\"CR36\" class=\"CitationRef\"\u003e36\u003c/span\u003e,\u003cspan citationid=\"CR42\" class=\"CitationRef\"\u003e42\u003c/span\u003e,\u003cspan citationid=\"CR43\" class=\"CitationRef\"\u003e43\u003c/span\u003e\u003c/sup\u003e. However, even machine learning approaches have so far failed to account for the confounding effects of surface water, which can still introduce bias into salinity predictions \u003csup\u003e\u003cspan citationid=\"CR31\" class=\"CitationRef\"\u003e31\u003c/span\u003e\u003c/sup\u003e.\u003c/p\u003e \u003cp\u003eThis study introduces CoSal, a remote-sensing and machine learning-based framework for mapping long-term soil salinity trends in coastal areas, accounting for the confounding effects from water-based land uses. Since coastal geographies, especially across Asia and Latin America, are increasingly characterized by aquaculture, our framework overcomes the challenge posed by the presence of surface water that is otherwise found to bias existing salinity models \u003csup\u003e\u003cspan citationid=\"CR31\" class=\"CitationRef\"\u003e31\u003c/span\u003e\u003c/sup\u003e. We use Landsat satellite imagery to leverage its long time span (available 1984 onwards). We identify and mask all pixels containing surface water using a supervised land use land cover classification, such that the salinity model developed is based on soil reflectance only. In the specific application of CoSal in South Asia, CoSal-SA, we use 28 variables (visible and infrared bands, popularly used salinity indices as well as all normalized band combinations) and apply 9 machine learning methods to build a stacked ensemble model for identifying coastal saline soils in a district of India. We train the model to identify soils that are too saline for growing rice (\u0026gt;\u0026thinsp;1900 \u0026micro;S/cm electrical conductivity)\u003csup\u003e\u003cspan citationid=\"CR44\" class=\"CitationRef\"\u003e44\u003c/span\u003e\u003c/sup\u003e.\u003c/p\u003e \u003cp\u003eIn this study, we apply the CoSal framework in South Asia (CoSal-SA). South Asia has the world's largest undernourished population, 14.4% of 1.3\u0026nbsp;billion in India alone \u003csup\u003e\u003cspan citationid=\"CR45\" class=\"CitationRef\"\u003e45\u003c/span\u003e\u003c/sup\u003e, with food security threatened by land degradation and climate change \u003csup\u003e\u003cspan citationid=\"CR46\" class=\"CitationRef\"\u003e46\u003c/span\u003e,\u003cspan citationid=\"CR47\" class=\"CitationRef\"\u003e47\u003c/span\u003e\u003c/sup\u003e. As per recent estimates \u003csup\u003e\u003cspan citationid=\"CR47\" class=\"CitationRef\"\u003e47\u003c/span\u003e\u003c/sup\u003e, about 7\u0026nbsp;million ha are salt affected in India, of which 20 per cent are in the coastal areas. This is estimated to expand to over 20\u0026nbsp;million ha or 50% of its arable land by 2050.\u003c/p\u003e \u003cp\u003eRemote-sensing based salinity assessments conducted at the national scale are, however, too coarse to understand local level dynamics required for targeted actions\u003csup\u003e\u003cspan citationid=\"CR46\" class=\"CitationRef\"\u003e46\u003c/span\u003e\u003c/sup\u003e. For instance, a remote-sensing based national salinity mapping exercise in India uses a benchmark of 4000 \u0026micro;S/cm for saline soils, which is too high for most food crops \u003csup\u003e\u003cspan citationid=\"CR48\" class=\"CitationRef\"\u003e48\u003c/span\u003e\u003c/sup\u003e. Crops such as rice, however, are affected by salinity above 1900 \u0026micro;S/cm \u003csup\u003e\u003cspan citationid=\"CR44\" class=\"CitationRef\"\u003e44\u003c/span\u003e\u003c/sup\u003e and are not captured by this assessment. These national-level benchmarks highlight salinity in arid and semi-arid regions such as Gujarat, Rajasthan, and Uttar Pradesh, however, coastal salinity remains underrepresented \u003csup\u003e\u003cspan citationid=\"CR48\" class=\"CitationRef\"\u003e48\u003c/span\u003e\u003c/sup\u003e.\u003c/p\u003e \u003cp\u003eCoSal-SA is developed by combining remotely sensed data with primary field data from India and Bangladesh. Soil data collected in 2024 from Jagatsinghpur district in India is used to train the model, and test the model\u0026rsquo;s internal validity. Soil data collected in 2016 from Satkhira district in Bangladesh \u003csup\u003e\u003cspan citationid=\"CR34\" class=\"CitationRef\"\u003e34\u003c/span\u003e,\u003cspan citationid=\"CR35\" class=\"CitationRef\"\u003e35\u003c/span\u003e\u003c/sup\u003e is used to test its external validity across time for South Asian coastal geographies.\u003c/p\u003e \u003cp\u003eJagatsinghpur is a district in the state of Odisha on the east coast of India and is home to 1.2\u0026nbsp;million people \u003csup\u003e \u003cspan citationid=\"CR49\" class=\"CitationRef\"\u003e49\u003c/span\u003e \u003c/sup\u003e. With a land area of 1668 sq. km and 55 km of coastline, it falls within the coastal plain agro-climatic zone \u003csup\u003e \u003cspan citationid=\"CR50\" class=\"CitationRef\"\u003e50\u003c/span\u003e \u003c/sup\u003e. Historically, the region has been agriculturally productive, with over 81 thousand ha largely monocropped with rice \u003csup\u003e \u003cspan citationid=\"CR51\" class=\"CitationRef\"\u003e51\u003c/span\u003e \u003c/sup\u003e. While 54% of the district\u0026rsquo;s workers depend on agriculture for their livelihoods, their contribution is less than 20% of its district domestic product (2% less than the national contribution from the sector)\u003csup\u003e \u003cspan citationid=\"CR50\" class=\"CitationRef\"\u003e50\u003c/span\u003e \u003c/sup\u003e. Coastal livelihoods in the district have been suffering both from climatic shocks, such as cyclones, and stresses, such as coastal erosion and salinization \u003csup\u003e \u003cspan additionalcitationids=\"CR53\" citationid=\"CR52\" class=\"CitationRef\"\u003e52\u003c/span\u003e\u0026ndash;\u003cspan citationid=\"CR54\" class=\"CitationRef\"\u003e54\u003c/span\u003e \u003c/sup\u003e. The district was the worst affected area after the 1999 Supercyclone (Cyclone 05B Paradeep), which resulted in 10,000 deaths and US\u003cspan\u003e$\u003c/span\u003e 3.5\u0026nbsp;million in agricultural losses \u003csup\u003e \u003cspan citationid=\"CR50\" class=\"CitationRef\"\u003e50\u003c/span\u003e,\u003cspan citationid=\"CR55\" class=\"CitationRef\"\u003e55\u003c/span\u003e \u003c/sup\u003e.\u003c/p\u003e \u003cp\u003eOver 8000 ha of land in the district are affected by salinization. These lands are largely concentrated in the district\u0026rsquo;s Ersama block, the one also most affected by the cyclonic storm surge, however, no causal connections are tested empirically \u003csup\u003e\u003cspan citationid=\"CR50\" class=\"CitationRef\"\u003e50\u003c/span\u003e\u003c/sup\u003e. Meanwhile, the district is also witnessing an expansion of aquaculture. In other contexts, aquaculture is found to be both an outcome and a dominant cause for salinization and agricultural productivity losses \u003csup\u003e\u003cspan additionalcitationids=\"CR8 CR9\" citationid=\"CR7\" class=\"CitationRef\"\u003e7\u003c/span\u003e\u0026ndash;\u003cspan citationid=\"CR10\" class=\"CitationRef\"\u003e10\u003c/span\u003e,\u003cspan citationid=\"CR47\" class=\"CitationRef\"\u003e47\u003c/span\u003e,\u003cspan citationid=\"CR56\" class=\"CitationRef\"\u003e56\u003c/span\u003e\u003c/sup\u003e. A historical salinity mapping at the local scale can give urgently needed insights into the potential spatial-temporal drivers of salinization, to develop targeted land management practices.\u003c/p\u003e \u003cp\u003eOur application of CoSal-SA in Jagatsinghpur district reveals that 155.16 sq.km. or nearly 10 percent of its total area had salinity levels not conducive for rice cultivation in 2024. While salinity increased in the years immediately following the cyclonic storm surge in 1999, it reduced in the interiors of the study area over time, likely due to natural processes of leaching. Meanwhile, salinity has been increasing in the coastal belt of the study area in India in recent years, alongside increased land conversions to aquaculture in the same belt. This simultaneous increase in salinity and aquaculture points at a potential salinity lock-in and acceleration effect due to coastal land change practices.\u003c/p\u003e \u003cp\u003eCoSal-SA has a training accuracy of 91.7% and internal testing accuracy of 85.7%. Its overall external testing accuracy in Bangladesh is 69.3%, and this drop is expected given different environmental factors, surface reflectance distributions between different images, and differences in soil sampling methodologies. However, a high recall score of 82.2% with this external data still makes it highly reliable for predicting high salinity values in a different geography and time period.\u003c/p\u003e \u003cp\u003eThis study establishes the unique characteristic of coastal soils that are simultaneously saline and have high moisture content. The CoSal framework demonstrates a comprehensive approach to accurately map historical coastal salinity in multiple geographical contexts, including near emerging coastal land uses, such as aquaculture. Fine grained salinity mapping can enable urgently needed research on local causal pathways of coastal salinity to inform targeted action and counter large losses in agricultural productivity, food insecurity, and related poverty prevalence.\u003c/p\u003e"},{"header":"CoSal Framework: Method and Materials","content":"\u003cp\u003eColSal framework follows a six-step process to combine remotely sensed data with field data, to create, validate, and apply a machine learning based salinity model for mapping historical soil salinity levels (1995\u0026ndash;2024) (Fig.\u0026nbsp;\u003cspan refid=\"Fig1\" class=\"InternalRef\"\u003e1\u003c/span\u003e).\u003c/p\u003e \u003cp\u003e \u003c/p\u003e \u003cp\u003e \u003cb\u003eStep1\u003c/b\u003e: \u003cb\u003ePrimary soil data collection\u003c/b\u003e - To build a relationship between remotely sensed data and field readings of salinity, we collected 113 soil samples distributed relatively evenly across the 200 sq. km. of a predetermined sampling area. We identified two neighboring areas in Jagatsinghpur district such that one area had widespread aquaculture while the second did not. We imposed a grid of 10km x 10km on each such that the areas were equally distant from the coast (Fig.\u0026nbsp;\u003cspan refid=\"Fig2\" class=\"InternalRef\"\u003e2\u003c/span\u003e). This distinction was made to help assess the difference in salinity specifically in relation to aquaculture. We divided these grids into 2km x 2km grid cells and systematically collected two geolocated soil samples from each grid cell.\u003c/p\u003e \u003cp\u003e \u003c/p\u003e \u003cp\u003eWe collected these samples at 15 cm depth (topsoil) using a soil coring device. We collected these samples from the middle of the fields and away from any tree cover or buildings to avoid shade in surface reflectances. Critically, we ensured a minimum distance of 30m (i.e. Landsat pixel size) from any surface water or water-based land uses to prevent the sampling points from falling within a satellite image pixel with mixed surface components that could confound the surface reflectance results. The samples were collected in the dry month of February in 2024 to avoid cloud cover, a common deterrent in remote sensing studies especially in tropical contexts. This timing also helped avoid the presence of irrigation water logging for rice cultivation, a common practice in these geographies, that could also introduce a similar mixed pixel issue as aquaculture. We collected some additional samples (~\u0026thinsp;13) from each area to potentially replace samples falling within mixed pixels despite our efforts on the field.\u003c/p\u003e \u003cp\u003eThese samples were used to measure the electrical conductivity (in \u0026micro;S/cm), pH, and soil moisture (in %) at one of India\u0026rsquo;s National Accreditation Board for Testing and Calibration Laboratories (NABL) certified labs. For methods of analysis, we followed IS: 14767:2000 for EC, IS: 2720 (Part 26)-1987 for pH, and IS: 2720 (Part 2) \u0026minus;\u0026thinsp;1973 for soil moisture, all as prescribed by NABL. Electrical conductivity (EC) is our key variable of interest for soil salinity measure, while moisture and pH were measured to study their role as potential confounders.\u003c/p\u003e \u003cp\u003e \u003cb\u003eStep 2: Supervised land use land cover classification and aquaculture masking -\u003c/b\u003e To reduce the confounding presence of surface water and aquaculture in the satellite reflectance data, we employed available perennial water data for masking the former and conducted a supervised land use land cover classification to identify and mask the latter.\u003c/p\u003e \u003cp\u003eFirst, we masked out perennial surface water pixels in the Landsat composite using the JRC Global Surface Water Mapping Layers, v1.4 \u003csup\u003e57\u003c/sup\u003e. At this stage, we used a threshold of 90 percent occurrence of water, to not mix perennial surface water with aquaculture ponds that do not operate throughout the year. This would ensure isolating the patterns of change in aquaculture alongside salinity.\u003c/p\u003e \u003cp\u003eSecond, to conduct the supervised classification, we acquired atmospherically corrected US Geological Survey Landsat 8 OLI (Level 2 Collection 2 Tier 1) and Harmonized Sentinel-2 MSI satellite images for February 2024 to coincide with field data collection. We limited images to no more than 50 percent cloud cover to maximise the number of images available but reducing reflectance errors due to potential presence of clouds. We created image composites using the 30th percentile, since that helps filter out clouds, haze, and high reflectance noise while retaining meaningful surface reflectances without overemphasizing dark shadows. We tested alternate reducers, including geomedian, median, and 70th percentile (See supplementary 1 for accuracy comparisons). Since the 30th percentile gave the highest accuracy results amongst these, we use it as the reducer for our current purpose.\u003c/p\u003e \u003cp\u003eAs is required by the Landsat dataset, we scaled the Landsat pixel values to floating points (0\u0026ndash;1) using published gain and offset values \u003csup\u003e\u003cspan citationid=\"CR58\" class=\"CitationRef\"\u003e58\u003c/span\u003e,\u003cspan citationid=\"CR59\" class=\"CitationRef\"\u003e59\u003c/span\u003e\u003c/sup\u003e. Some pixels (0.4 percent) had negative surface reflectance after conversion, which is likely due to cloud cover or presence of water \u003csup\u003e\u003cspan citationid=\"CR60\" class=\"CitationRef\"\u003e60\u003c/span\u003e\u003c/sup\u003e. These values were adjusted to 0 so as to not get included in the analysis further.\u003c/p\u003e \u003cp\u003eWe used primary observation data points as well as Sentinel-2 high-resolution composite to identify 274 active aquaculture pond locations (Fig.\u0026nbsp;\u003cspan refid=\"Fig3\" class=\"InternalRef\"\u003e3\u003c/span\u003e). We classified each Landsat composite for three classes - active aquaculture, dry aquaculture and other land uses. To do so, we used a random forest classifier, with 40 trees (see Supplementary 1 for hyperparameter tuning results) and employed 6 spectral bands (Blue, Green, Red, Near Infrared or NIR, and the two Short Wave Infrared bands SWIR1 and SWIR2) and three indices (Normalized Difference Vegetation Index, Normalized Difference Water Index, and Normalized Difference Salinity Index).\u003c/p\u003e \u003cp\u003e \u003c/p\u003e \u003cp\u003eFinally, we applied this aquaculture LULC classification to the entire Jagatsinghpur district and used the classification output to mask active aquaculture pixels in the Landsat composite to be used for salinity modeling in subsequent steps. All processing was performed in Google Earth Engine.\u003c/p\u003e \u003cp\u003e \u003cb\u003eStep 3\u003c/b\u003e: \u003cb\u003eSurface reflectance of pure soil and correlating electrical conductivity with key variables\u003c/b\u003e - To derive the surface reflectance values corresponding to the field locations, we used the masked Landsat composite from Step 2, and superimposed the field locations on it. 7 of the 113 soil sampling locations fell within a pixel masked for aquaculture, suggesting a mixed pixel scenario (Fig.\u0026nbsp;\u003cspan refid=\"Fig4\" class=\"InternalRef\"\u003e4\u003c/span\u003e). Including data from such mixed pixels containing both water and soil in the modelling has been found to bias salinity results (see Silvestri et al.\u003csup\u003e\u003cspan citationid=\"CR31\" class=\"CitationRef\"\u003e31\u003c/span\u003e\u003c/sup\u003e). We, therefore, dropped these samples from the analysis.\u003c/p\u003e \u003cp\u003e \u003c/p\u003e \u003cp\u003eWe calculated 11 indices widely used for measuring salinity, and 11 additional normalized band combinations (Supplementary 2). Including the six VIS-NIR-SWIR spectral bands, these 28 variables are used as the key predictor variables in the models. (Note: coastal blue band was not included since it is not available in older Landsat images). We assessed the linear, quadratic, and logarithmic relationships between EC and the 28 variables. We also tested the relationship using a principal component analysis (PCA) to help reduce the dimensionality of the model. See Supplementary 2 for details on each of these metrics, correlation matrix, and their regression model fit results with EC. The variables most correlated with EC were NDWI, NDSI2, NRSWIR1, NBNIR, NIR, and SAVI. Of these, two indices, NDWI and NDSI2, achieved the highest R2 of 0.39 and 0.37 in a second-degree polynomial regression model.\u003c/p\u003e \u003cp\u003eSince this is not an adequate explanation, a machine learning-based approach was adopted to derive a better explanatory combination of the input variables.\u003c/p\u003e \u003cp\u003e \u003cb\u003eStep 4\u003c/b\u003e: \u003cb\u003eEnsemble model\u003c/b\u003e - We used the 106 remaining EC measurements after aquaculture masking as the key dependent variables, and their corresponding 28 surface reflectance measurements as the key predicting variables. Of these, we used 80 percent of the samples to train machine learning models, leaving 20% for testing. We used EC as a binary variable, and applied stratified sampling for the training and testing sets to ensure both categories are adequately present in each subset. Since rice is the main crop in the study region \u003csup\u003e\u003cspan citationid=\"CR50\" class=\"CitationRef\"\u003e50\u003c/span\u003e\u003c/sup\u003e, we use 1900 \u0026micro;S/cm of EC as the threshold for high salinity. This is based on the understanding that rice production is hampered above this level under all conditions of acidity and soil nutrients \u003csup\u003e\u003cspan citationid=\"CR44\" class=\"CitationRef\"\u003e44\u003c/span\u003e\u003c/sup\u003e.\u003c/p\u003e \u003cp\u003eWe first conducted 100 iterations for select machine learning methods most widely used in salinity studies including random forest, random forest with bagging, and artificial neural network \u003csup\u003e\u003cspan citationid=\"CR24\" class=\"CitationRef\"\u003e24\u003c/span\u003e,\u003cspan citationid=\"CR43\" class=\"CitationRef\"\u003e43\u003c/span\u003e\u003c/sup\u003e, and tested the average training and testing accuracies for each (Supplementary 6). We also reduced the dimensionality of the predictor variables using a lasso regression and repeated the 100-iteration test which improved test results and reduced overfitting (Supplementary 3). Random forest with bagging \u003csup\u003e\u003cspan citationid=\"CR61\" class=\"CitationRef\"\u003e61\u003c/span\u003e\u003c/sup\u003e resulted in the highest accuracy amongst these select machine learning models, however, the accuracy remained lower than that of the ensemble.\u003c/p\u003e \u003cp\u003eEnsemble models combine outputs from multiple machine learning models and find an optimum result that balances between bias and variance (or noise). This reduces errors and improves overall accuracy, while also addressing over or underfitting the training data. This makes ensemble models particularly efficient, especially when data is limited.\u003c/p\u003e \u003cp\u003eWe included the following nine machine learning models in the ensemble model: (1) Random Forest (\u0026ldquo;rf\u0026rdquo;), which builds multiple decision trees using random subsets of data, is robust to outliers, and good at handling non-linear relationships. (2) Recursive Partitioning Decision Tree (\u0026ldquo;rpart\u0026rdquo;), which creates a single decision-tree through recursive partitioning of variables and is easier to interpret. (3) Neural Network (\u0026ldquo;nnet\u0026rdquo;), which is a basic feed-forward neural network that helps capture complex patterns. (4) Support Vector Machine with Radial Kernel (\u0026ldquo;svmRadial\u0026rdquo;), which creates non-linear decision boundaries in high-dimension data. (5) Gradient Boosting Machine (\u0026ldquo;gbm\u0026rdquo;), which builds trees sequentially, each correcting errors in the previous tree and improving accuracy. (6) XGBoost (\u0026ldquo;xgbTree\u0026rdquo;), which is an advanced implementation of gradient boosting. (7) Naive Bayes (\u0026ldquo;naive_bayes\u0026rdquo;), which is a probabilistic classifier based on Bayes Theorem useful for high-dimensional data. (8) K-nearest Neighbors (\u0026ldquo;knn\u0026rdquo;), which makes predictions on most similar training examples. (9) Elastic Net (\u0026ldquo;glmnet\u0026rdquo;), which helps with feature selection (using ridge and lasso) and helps handle correlated predictors.\u003c/p\u003e \u003cp\u003eWe used the R-package caretEnsemble \u003csup\u003e\u003cspan citationid=\"CR62\" class=\"CitationRef\"\u003e62\u003c/span\u003e\u003c/sup\u003e to enable fitting multiple linear and non-linear models to the same dataset. We stacked the models using the function caretStack(). This approach uses the predictions from all the 9 models as features, and combines them using a meta-model, in this case a random forest, to make the final ensemble predictions.\u003c/p\u003e \u003cp\u003eThe model returns probabilities or the likelihood that a given sample belongs to a specific class (high or low salinity). We used Youden\u0026rsquo;s J Statistic (Eq.\u0026nbsp;1) for deriving the most effective threshold to convert these probabilities back into binary classification predictions \u003csup\u003e\u003cspan citationid=\"CR63\" class=\"CitationRef\"\u003e63\u003c/span\u003e,\u003cspan citationid=\"CR64\" class=\"CitationRef\"\u003e64\u003c/span\u003e\u003c/sup\u003e.\u003c/p\u003e \u003cp\u003eEq.\u0026nbsp;1: Youden's J statistics: J\u0026thinsp;=\u0026thinsp;Sensitivity\u0026thinsp;+\u0026thinsp;Specificity \u0026minus;\u0026thinsp;1\u003c/p\u003e \u003cp\u003eWhile ensemble models inherently address overfitting, we took additional measures to reduce any remaining issues. We used a 5-fold cross-validation method such that the data is split into 5 equal parts. The model is trained 5 times, each time using 4 parts for training and 1 part for testing. We also reduced the dimensionality of the data by removing near-variance predictors and lasso variable selection (Supplementary 3). For the threshold for converting probabilities into class predictions, we tested all values between 0.1 and 0.9 with the increments of 0.01, to minimize the difference between training and testing accuracies (Supplementary 7). The difference was minimal across the range, suggesting low possibility of overfitting.\u003c/p\u003e \u003cp\u003eTo help interpret the machine learning model, we used the SHapley Additive exPlanations (SHAP) methodology \u003csup\u003e\u003cspan citationid=\"CR65\" class=\"CitationRef\"\u003e65\u003c/span\u003e,\u003cspan citationid=\"CR66\" class=\"CitationRef\"\u003e66\u003c/span\u003e\u003c/sup\u003e to calculate Shapley values as a measure of the contribution of each individual parameter in the final model \u003csup\u003e\u003cspan citationid=\"CR65\" class=\"CitationRef\"\u003e65\u003c/span\u003e,\u003cspan citationid=\"CR66\" class=\"CitationRef\"\u003e66\u003c/span\u003e\u003c/sup\u003e. These values improve our judgement about the model by theoretically testing and explaining specific predictor contributions.\u003c/p\u003e \u003cp\u003e \u003cb\u003eStep5: Model Validation\u003c/b\u003e - We conducted two separate analyses to test the internal and external validity of the model. Internal validity helps assess the predictive consistency of the model and potential overfitting. External validity on the other hand, gives insights into the generalizability of the model to different spatial contexts and time periods. Both validities give insights into the robustness of the fundamental physical relationship between salinity measures and surface reflectances across the training and testing data, rather than spurious correlations.\u003c/p\u003e \u003cp\u003eFor internal validation, we used the remaining 20 percent of the Jagatsinghpur field data to test how well the model performs on unseen data from the same geography and time period. For external validity of the coastal salinity model on a different geography and time period, we use georeferenced soil data (n\u0026thinsp;=\u0026thinsp;92) collected from aquaculture practicing regions of Satkhira District in coastal Bangladesh in 2016 \u003csup\u003e34,35\u003c/sup\u003e. These topsoil samples were also collected in the dry period of February; however, they were collected using a composite sampling method by mixing multiple smaller samples from an agricultural field. This is a standard practice in agricultural soil testing, and while this approach is useful for determining the average salinity levels across a large field, the corresponding surface reflectance may not capture the precise relationship with soil salinity.\u003c/p\u003e \u003cp\u003eFor each model and its prediction application on the various subsets of data (training and internal and external testing data), we calculated a confusion matrix of true positives (TP), true negatives (TN), false positives (FP), and false negatives (FN), and derived four metrics to evaluate the model comprehensively: Accuracy, precision, recall, and F1-score (Supplementary 4).\u003c/p\u003e \u003cp\u003eAccuracy involves all components of the confusion matrix and usually serves as a coarse measure of overall model quality. In case of imbalanced data, such as ours where there are fewer high salinity points (30 out of 106) than there are for lower salinity (70 out of 106), there is a possibility of getting a relatively high accuracy even if all the high salinity points are incorrectly classified as low (accuracy would still be 70% because TN\u0026thinsp;=\u0026thinsp;70 even when FN\u0026thinsp;=\u0026thinsp;30). We, therefore, also consider the other three metrics in tandem to gain better insights into the robustness of the model predictions. More details on each metric is presented in Supplementary 4.\u003c/p\u003e \u003cp\u003e \u003cb\u003eStep 6: Historical coastal salinity mapping\u003c/b\u003e - To apply the CoSal-SA model in the current context of Jagatsinghpur, we extracted composites for the district for February for the years 1995, 2001, 2005, and 2010 from Landsat 5 Thematic Mapper (TM) and for years 2014, 2017, and 2021 from Landsat 8 Operational Land Imager (OLI). It is important to consider and address the sensor calibration and radiometric resolutions differences of the two sensors (TM and OLI) in a time series analysis. To align the spectral distributions across the two time periods, we conducted a mean-variance normalization for all bands in Landsat 5 images with corresponding bands in Landsat 8 (2024) (normalization results are in Supplementary 5)\u003csup\u003e\u003cspan citationid=\"CR67\" class=\"CitationRef\"\u003e67\u003c/span\u003e\u003c/sup\u003e. We then followed Steps 2 and 3 (above) on these images (i.e. applied the LULC classifier to these historical composites to create the aquaculture masks and extracted the values for the 28 predictors). We applied the coastal salinity ensemble model to this data to get the final historical coastal salinity maps for the district of Jagatsinghpur.\u003c/p\u003e "},{"header":"Findings","content":"\u003cdiv id=\"Sec3\" class=\"Section2\"\u003e\n \u003cdiv id=\"Sec4\" class=\"Section3\"\u003e\n \u003ch2\u003eSoil salinity increases with soil moisture in coastal contexts\u003c/h2\u003e\n \u003cp\u003eSoil samples collected from coastal India show a strong positive correlation (correlation coefficient of 0.48) between electrical conductivity and soil moisture (Supplementary 2). This finding suggests that top-soil salinity increases with increasing moisture in these coastal areas. This result is further corroborated by indices such as NDWI, NBNIR and NGSWIR1, where they are found to be higher for both higher levels of soil moisture as well as higher levels of EC (Fig.\u0026nbsp;5 and Supplementary 8).\u003c/p\u003e\n \u003cp\u003eWhile these indices behave as expected with the presence of moisture, their relationship with salinity is divergent. Clear water reflects some green and blue light but absorbs NIR and SWIR1 almost completely, increasing the numerator and decreasing the denominator of these three indices (Eq. 2), thereby showing a positive association with more moisture. Saline soils, however, are known to reflect more infrared light \u003csup\u003e\u003cspan class=\"CitationRef\"\u003e25\u003c/span\u003e\u003c/sup\u003e, and therefore these three indices should decrease with increasing salinity. In our samples, NIR reflectance shows a negative relationship with salinity while SWIR reflectance shows nearly no correlation with salinity (Fig. 5 and Supplementary 8). These divergent relationships between salinity and these indices and the NIR and SWIR bands suggest that salinity in these coastal soils present themselves differently in surface reflectance as compared to how saline soils are known to behave when there is a lack of moisture, such as in arid and semi-arid contexts \u003csup\u003e\u003cspan class=\"CitationRef\"\u003e25\u003c/span\u003e,\u003cspan class=\"CitationRef\"\u003e29\u003c/span\u003e\u003c/sup\u003e.\u003c/p\u003e\n \u003cp\u003e\u003cimg src=\"data:image/png;base64,iVBORw0KGgoAAAANSUhEUgAAAnYAAAEsCAYAAABOqf71AAAAAXNSR0IArs4c6QAAAARnQU1BAACxjwv8YQUAAAAJcEhZcwAADsMAAA7DAcdvqGQAAFhcSURBVHhe7Z3/i5Vl/v9vP7+njvqTmIijsIvvxbDRllaFAh1zl2VjtbFYQlBytSW2L6s1Km2Y2lgmROkYBRLvd2NYFLF+DRQak/IbSoYLqUhEP/kl7Q+Yz3lcc79Or7m8z5eZc5w5c8/zAbf3dV339eV1fTnnes3ret3HUT0FEiGEEEIIMez5f+ldCCGEEEIMc6TYCSGEEELkhOJR7KhRo0KCEEIIIYQYPnivuj6KnX8ghBBCCCEam1h/01GsEEIIIUROkGInhBBCCJETpNgJIYQQQuQEKXZCCCGEEDlBip0QQgghRE6QYieEEEIIkROk2AkhhBBC5AQpdkIIIYQQOUGKnRBCCCFETpBiJ4QQQgiRE6TYiRHB+fPnw3+7wnX8+PE0NUn27t0b0saNG9cn7q9p06Yl27ZtS65fvx7y+LrWrFkT0oDnpO3evTtN6Vsf5WiHMHXWi/Xr14f6Fi1aFOL0b/bs2cV2ly1blly5ciU8s/btIk4frG+NALLQJz9WNqbcvfxc9PXAgQPhea3E7VULstoYx9Af5oA8rCODtWHzBEPZ77vBUK1DxjhuT4gRBf9XLLigELmkvb29p6mpqaelpSVN6aW1tbVn//79aaw3H2lGV1dXsdy1a9dCWkdHxx2fGfKRtnr16jSlF+riGdAOebq7u0O8VpC1ubm55/LlyyGOfMja2dkZ4ufOnQtxay9unzvP29raQrwRYGwZa+sT4+nng2cWp788p0/0tR7YPNqcVYJ8Xr4Y5GV8be1A1joY6n7Xk6Feh4whY8VnQ4i8w2fJI4udGDGMGTMm/BVf+NK/wyIzevToNNSbz0OZzz//PDl9+nTy8ccfh7QlS5aEu7f+ffnll0lhs02OHDmSpvRaYQ4fPpwsWLAgxH07tULdW7ZsSV599dVk6tSpIe3ixYvJzZs3k1WrVoX4zJkzg0xG3P7cuXPvkHmoOXr0aLJ06dJinzZt2hTuxvjx49NQb9ief/311+FeK8w3Y7J9+/Y0pTzkW758eRrrCxY11g3z5OV++eWXQxuMvzHU/e4PWMT82o8Z6nXIGO7cuTN81hvZqinE3UCKnRhRTJkyJXnxxReTl156qV/HPmw8LS0tyaeffhribBzNzc3JiRMnQhzYlNjI2EzsCOibb74JG5bflOsFmyfMmTMn3GHixInh7o/8Nm/e3EeBiGED9nUMNYztu+++WxxDxu7QoUMhnMWtW7fCfcaMGeFeD+bPnx8UskpwvM58owxmceHChbBOTFkzqJs2PI3Q73rRCOvQPqPMgRAjCSl2YsSxYsWK4NOzcePGNKU6YuVs4cKFwcoCbPDEsUw0NTUl//3vf0P6f/7zn+RPf/pTCFcCC4j3O/KX3yCNH3/8Mdy90kC4q6srWbduXfDRqmStoE18vZ555pk0pbcvlKVd7oMNlijGkE25kt8Vz954442gdJvSgMzm+4hllrmmT+bv6MPWR9I8kyZNCvdKPlrU7y1RMVevXr1jDK1Oa8Ootd+k4WdJn6zPQ0Wt6xDZ6QNpgOJs48jdFGnCNo/4J8bwnDkQYiQhxU6MOFDQ3nrrrWTXrl1lj5MqgcLGMStwNwsMCt5XX30VwljxFi9eHMKVYIPu6enJvNauXZvm+pUffvghDfWFTQ8rEnL88Y9/LCo5nnnz5oXN8M9//nNy7NixPjL+9a9/Tfbs2RPatc10MGF+Tp06FRQDxm/69Ol3KCmMN/JPmDAhxMlvPPXUU+GOAoUFFSsQCjfj0tra2idMHznKtnmM+emnn9LQnaBIoXjwh0IpshTDUnXW2m/+UHn44YdDn7BK//LLL+mT+mAvJXCBrSGLx9SyDpkj+sA6hH/84x/FeeVO3MLd3d3JtWvXSh7jVlLOhcgbUuzEiIQNhI2dTX3s2LFpannYyL117IEHHgh3lMN9+/YV/egeeuihsMmQzqY2FCAnPkbnzp0LCmxsMbHNEOsQVhUDRQJlzqxAdhSIZQprkF2WD6uKbZxYTNikzXJiZXgzknzAmJgywFVOsUYxuHTpUhjDlStXpqm9mFKGEpSloKN0Y9Gqltivshq++OKLoDT6NVEPBtpvlEz7A4A7c8g8Mc6mlEE8T/FcWhnLRxyok7a5gDXk41kMdB3C2bNng++iWe0qEVvUhRipSLETIxY2HCwg1fhSsYGSzx+rspFwFHbw4MFiHH7/+9+HvKRXewwLsdLjr6yj2CzI5zdCLB+Q5WeEvDt27AhjYGVKWXmwfKFsMGawdevW5JNPPgkvlfBCCZs/1hk2+TNnzoQ8VgblECWF/sVWSVMgPbGVcNasWSXnyCxv8csL5pxfzlpj1q/PPvus6OTfH8q9NDEQau03PmoxzD/5uFirWfMUz6WVeeyxx5LOzs6S1sxy1LoO4eeffw79NKtdKbD6Yd3csGFDmiLEyEaKnRgxmLO5gTWho6MjbHSeOB9WBo6K8KWKj1VRWHjjkbcZDTYxfKWwUMT5b9++nYbupL9HsZMnT05DfWGDM4XGLDps1BC3T5ttbW2ZPl1mubHNFsWDMUNRY9O97777wqaKnyEK4UcffRTys0Fbu5Rh40ZJqRbmwxRZZMJyg4xGLKfNYaz8vvDCC2WtdowJ1iLKllIA7SWAGPqHIoXiUY4sa16pOuvVb8D65o8+WZMc22bNUzyXBmVqeTGjXusQ5fXDDz9MY3eC1Q/l1P7Aiqm3RVWIhqewaQRcUIjcwe9osca5/O9+8Xtbzc3Nxd/T6kp/w8xfLS0t4TfGsqBcXCfwe1wFRTCN/Qq/1UV+2qwVa7uwuacpvf2kXd8OfTKy2qc86VyUtzvQd6OwKaeh3v75PhMmLcbKMH42xpXgt89oFzm5qJd5smeW7vvF7/mRRt99W8S93Flh6qG8x9ZBKaydSpAna65j+aHWfpOffJZuY+D7nDVPxP1cgpWhjmr6GUN9ta5DLzf5TA4/vz5MHhsvw5cTIq/wWfJIsRNimMImxuc2VhBqhfrYXNkU2ZwBhcE2XGAztU3YNlYUBOQxxcGX4R4rFHcLLzfKglfgkA95fBhQqLxyR3mv1Hps3L1CXQr7Id44L3WbjPWCvtB3xtoUO9Jo36+ReJ7iufRlyEP6YIMMtGtyczcFjT6a7D5MHsbV5pQxpx/MgRB5hnXuGcU/hcRgmk+DQohhAkdXvKiBH1KWv5roPxwbcixZUAjuOEoHXgrhDU7vD1YOjqO58E+0Y0HKPv744+EliErHuaL/cJz79NNPh6Nq/+awEHkk1t/kYyfEMIbfPmtraws+gLyBKmqDL8hnn302KFxZSh28/vrryZNPPpnGKoPizZvXzc3NRX84lLnOzs7gh0abon4wxrxMAeV+4FmIvCKLnRBCCCHEMEUWOyGEEEKInCLFTgghhBAiJ0ixE0IIIYTICVLshBBCCCFyghQ7IYQQQoicIMVOCCGEECInSLETQgghhMgJUuyEEEIIIXKCFDshhBBCiJwgxU4IIYQQIidIsRNCCCGEyAlS7IQQQgghcoIUOyGEEEKInCDFTgghhBAiJ0ixE0IIIYTICVLsxIhk2bJlyahRo5JFixalKb1MmzYtpO/evTvELZ+/yLN+/frwHPpb17hx45Lz58+HNMJWrz1bs2ZNcv369fC8HuzduzeZPXt2qB+oG/mtbeQ0Gbl7ebgoe+DAgfB8sEEeG0cbG+Nujt22bdtCnczZlStX0tTKIK+XMSaeC0Be2kF+2oWh6nctDOW6itsTYkTTk+KCQuSea9eu9TQ1NYV1v3///jS1p6e7u7unpaUljfXmIw/pRkdHR59y/a2L8gb5ff3cqautrS3EayWuH2gfuS5fvhziq1ev7mltbQ1h4JnFkZnn1HHu3LmQNljYWJjsXV1dQQ7jbo8d48M4NDc3pymVIa8fa0/WXADjjcyMNVg/LN9g93ugDPW6snHiLsRIwn8/gCx2YkQyfvz45ObNm0l7e3vyzDPP9LF28MzwYWPFihXhfvv27XCvpa7Ro0enoV7mzp2bFDa85MiRI2lKbbz88suhPuo1jh49mixdujSZOnVqiG/atCncjVhme/7111+H+2Bx4sSJZM6cOUXZsWoVlIMQhrs9dozPzp07k4KiUpVlyfL4sfZkzQVlTp8+nWzZsqU47kPd7yywJMYW6ZihXleME+Owffv2NEWIkYkUOzGiee6555IbN24k77//fppSHpS2jRs3ho12wYIFaWov/a2rFCiJbOz1AKVh/vz5aawXNt533323eMTIJnvo0KEQzuLWrVvhPmPGjHAfLCZPnpwcPnw4OX78eJqSJKdOnUpD2dRz7ICxam5uTi5cuJCmlOaDDz5I/vWvf6WxO8maC+qlflOGoBH6PRAaYV0xvoyzECMZKXZiRMPms3Xr1mTdunVlfanmzZsX/HgmTJgQLCMfffRRHwsEVFtXOdjM8cPC8peF91HyV5Y1xWSYNGlSuBtYSpqamoJCUck3i2dvvPFGUGTNgmS+X9YueZCZeD0xCwxjTzvml1iKeOxMJhuHcakvGvUQxh/LwuTjTpkY+nv16tU0lg1tsCaWLFmSpvSl1FxQL/V76tXvuE+0Yz58d4Na1xVlbCzoM/Iju/WHNN838sZjY+M70M+fELkgPZK944xWiLzj13xhgwn+P/greb8gIJ/5M5lvED5N5ksE/anL+9iRhzS7qLdePkdWt8kegy8SPmFxm8jsZaK/HuS3OnlmYfLeDajfZPL+U9Y/u7LGDvlMfi+3nxcLm09dDGlZ6R7qjsfJY7Ja+0a5umvpt9Xp+4SMfu1Vg2/HX+XqGei6ol+kWV/9uGSFyRfLUWqchcgzrHmPLHZCFNixY0c4/vruu+/SlGywyuF3xdHXxx9/nKb2pdq6PIWNKCkojcHagcVvMMAydOnSpWThwoXJypUr09ReCpsn3xRJYfNMdu3a1edYMGbixIlpqBcsKWZFwrICWGiwPhE3XzSLcxEuBRYdjvQ6OzuTxx9//A5LULmx42gcqrXgjB07Ng31D44gn3jiiTRWH2rpt/HLL78k999/fxrrhXmx8fZWPNKx8PG2qsEa4CooUMU1wbV27do0x53Usq5mzpyZ7NmzJ41VZsyYMWlICGFIsROigDmgv/7662nKwBloXSiNphSyyWZhilB8lVOMYuy4y5g1a1ZJvyQ2aTbj5cuXpym92NF0U1NTH/8weOqpp8KdsgZK8KOPPhoUEXzRAKXFFIUsXyz65BUy88W6ePFiuHtKjd3PP/+cvPDCC+HYrxz0BUXknXfeSVOqp9JLE/2lHv0mTp8eeuih4ss+hp8XmyvgpYPvv/8+vABRTpEvRT3WFX53pJVa/2B9++yzz5JVq1alqUIIQ4qdGJFk+f/gI2QWHiPORxxfIHjwwQfDfaB1gb1Za6ActLW1lfRRMkUovrIUo9iSZvCWp1lpaOPs2bOhTSNuF2uNLwNYi2gXy2UlHzBgw/773/8efBTxRasW3jQ2eXhbFEXSFKhqx84Uz3J+cvQFf69vvvkmTelLrLx63nrrreTVV19NY9mUmotS9dbab7OMoVhRVzWQl/nB+keb/aUe6wpQ+spZ7egbfyBQtpQlttR4CzESkGInRiRPP/10uHvrhb384LF8/uUJ3lDcv39/caPtb12vvfZaURn629/+Fu7ecsFPX8D06dOrUppKYUrDjz/+GO4GR3v79u0r9gfMUsULBWzw3gLE8RjKARu+KbUejvtiUAyQHYUEODKjXVNEq8FeBkBGZEXmY8eOhTToz9hhtePorxyU5SdJYjhWnDJlShrrC4rFyZMn73hDOqbUXFAv9Xvq2W/AahmDzChaXtlFsUVhYn7io1biWX88eOq5rugbeUvB5+v555+/4w10G99yirgQuafwIQ64oBAiJ/AiRzmn/oGAYzzfF1ytra3hhRIc2Ylzx3Edx3na5Tlp5LFyBQUgrenuQrvIYQ78yINsxElHDh+2PMhsXL58OcjMjwJnUVBMqh7frLmwHxumnXphc8FF34gD488FbW1txX5bGmHKEGe+BhvGxmQBC1t/mCsfBsaUOTCogzQhRhJ8Jjyj+KeQGP7KSoNCiJyAdQTH+8KG2MeiKCqDRQsrK0d+pX5HjpcNsKZhfapEqbnAN42Ll3JkaRo4+AViWceavnjx4jRViPwT6286ihUix6BAdHZ2Jhs2bAgfflEd+H1xrAmljiBR1Jqbm6tS6qDUXPC7iLyNS12xv5moDsbz2WefDUqzlDox0pHFTgghhBBimCKLnRBCCCFETpFiJ4QQQgiRE6TYCSGEEELkBCl2QgghhBA5QYqdEEIIIUROkGInhBBCCJETpNgJIYQQQuQEKXZCCCGEEDlBip0QQgghRE6QYieEEEIIkROk2AkhhBBC5AQpdkIIIYQQOUGKnRBCCCFETpBiJ4QQQgiRE6TYCSGEEELkBCl2QgghhBA5QYqdGBbs3r07GTVqVDJu3Ljk+vXraWqSrFmzJqQvW7YsxC2fv2bPnp0cOHAgPDeOHz8e0i0P5a9cuRLuxGnn/Pnzae4kWb9+fUjzTJs2rVjWt7t3797w3OryF2m+XoM+8Qy56onVi+zbtm0LaaX6DuTz8hJnjP2YDxXIYPOAbIw/4w7VzvtgEctXCtYC+TysH5sfGE79Lke5dUfY98GeZ31W7jbISfvIWu/PoxCDQk+KCwrRkLS0tPQ0NTX1tLe3pym9sHavXbuWxnrztba2hjDpq1evDnnOnTtXTKOezs7OECedeHd3d/FZR0dHeGY0NzeHOi5fvpym9IT8pBu+XaAuylhdlLU+eKiHNPISrie019bWVhyfcn2H/fv395HDZKOOoYZxpD82B8yrH+9K8z7YdHV1hfa5lwIZ/VqLxx+GW7+zqLTueI7MlT4rgwXy8D2DTP4zL0Qjwjr1yGInhg3jx49PCptWsmXLljv+kueZEYc3bdoUwl9//XW4X7x4Mbl582ayatWqEJ85c2aoF8iPpeDo0aMhDlgVCptMUthkkpMnT6apSXLw4MGkoPCksb7tQhyfOnVq8sorr4S2TX4sKydOnEhu3LgR4vWEuk+fPh3Gy2Qp13cYPXp0Gupl7ty54fmRI0fSlKGDOVm6dGkYR7B5Nfx4E47nfbBhHTF227dvT1P6giVu165dyZIlS9KUJHn55ZdDGcbdaMR+m1WrWiqtO98HyPqsDCbIs3nz5qTwh1vy/vvvp6lCDA+k2IlhxSOPPJK0trYm69atS1Mqc+vWrXCfMWNGuE+cODHc7WgS+BK3zXT+/PnJ4cOHQxi++OKL5KGHHkoWLlyYfPnll2lqEpSdFStWpLHquH37drizscHixYuTtWvXhnC9uXDhQtiYTCGASn3Pgs11zpw5aWzooB/vvvtu8fiOzffQoUMhnEU870MBawnlOouPP/44/GHg54e8lPEMx37HDGTdxZ+VoYDP/JkzZ9KYEMMDKXZi2NHR0REUL/NlKwdWkTfeeCNY3GwTYaPs6uoKyiH+SrE/kikx5l9z7NixZMGCBUG5O3XqVEizTdZvypWgDNab9vb2NKV/sCl6HyR/ZfkCXb169Q7/rUp9j6FexvmZZ54JccLWJuWrsabgo7do0aI0NnCwRGE1RVmt5PeXNe/Ibn5qhKmDuPWBMBAn7PMAYes7/fHtkzfLn27SpEnhbuvF8/rrrydPPvlkGvs1j5Uxau0382Ry46tXCZvjetLfdZf1WUF25GI+9u3bF+6MBzD2pjSSZuueO5ZTC5caB+JZn6EpU6Ykly5dSmNCDBPSI9k7zmiFaDTwIzKfHPxf8L8x3xwP+Uiza/Xq1emTvuA7w7OsPPj3mL+P+dHhF0Re2sRXyPyFDNo1XyfDy8FVzt+K5963qlay5DFK9Z32vbyMceyrZXXSF+8fVo5ScgwE2mVOYtlow8ueNafMHX00uUkD8+my8beyPq8P89zy4hdHetZYkAdZLK9B3NaVUSqvMdB+Ixd10nfrbyVKzVfcll3Vzm+5z1xcp/+s0F/z87T+8PmzPIRNBuaDCyyvD8fjQFk/nx7KIIsQjUy8RmWxE8OS5557Ltzxf8GaEVP4kmelBysBfkxZf41jRdi5c2dS2DRCHm9F4AgG3ybSzI+OIyHa+uabb5JPP/00WPGqobA5BFkKm0dFi8tgUa7vUNjkksIGGKxE5Y69x4wZE+5YubCIYEUxyxN1EuftwixiC2S1YIHBisIcrVy5Mk3tpdy805cff/wxWLLs+JtjTeYD38nnn38++e6770IcS00l7rnnnnDnOP3BBx8M4Wr58MMPk6eeeiqNVcdA++3xvmxmATPrFfNmc1jKSsURMO2wPoAwV7mjYU+ldVfqs4IV8+effw5x5o455Kj522+/Dc89X331VZiTcvhxwO+vmvkWYrggxU4MS/hiZoNA6WAjKgWbIZve8uXL05RehYLjJsN8ePBJM/7whz+E4142CcIGmyppbDBsUv3BnNo3btwY7v2lv0exWVTTd4Mx3rFjxx3H3sRp87PPPis6w2/dujX55JNPks8//zz4jgEvAnz//ffJe++9F+IxbNCmGHBVwo7XjFmzZpX0X8uad14iYb34Y7iHH344OPbjl8YRPAoqinspRY3y9J1xG6jvF2sHpca/NFGOWvsN8+bNS6ZPn55s2LAhxOnn5cuXw7ibDxlHuHv27AlzVm/6s+4g/qywFpGdI3CrB+UO2VFI77333vB5tD8qShGPgxB5RIqdGDbYX++GbWJsUJ44H1YA8rC5GHyx2yZgShF1GQ888EC4owz4v/7xsyMttphA3G4cZ3NCAcqyqFheLEqliBUhf5k/laeU4lmu7+awblAvFktvPSEv1jzG1OrBmnLfffeFjdPeKKa/XANVgGL8HCLL2bNn+7yVHI93PO+MB5alyZMnF5UDFDjeSkaxM8UAZSNrPIE6sTaVUlZLYS8PAFZmLFLx/Pg8nlr7DVjYWHu8yQ2//PJL8tFHHwUlFUWddUDf6TdzFiuT9aDcuov7kPVZQTZ8XPmDwuoBrK2/+c1vwjojzPyWIh6HasDSK8SworApBFxQiIYDPxjWKP5FHnxv/Nq1fFzeR8d+kwpfGspwpy7S8FvyeY3CpnPH77dZe4WNM03pJatdyhKPfaLw7/H5aMdk4SKeJU9/KWzwd/hxVeq7TzfoK+nWD+QDyjGuQF99H4E6zKfNytQCY2xjx0Wb1G/PLN33x8+7zSV5GRugvB8D8tgzQHaL+zD1mR8X+Gce6qV9D+2RP4tYfqi138hl7dmcMFc2HgZx8jHffv7rQaV1R9uk2xozrN///ve/Q1/B5AT6aWuLcuT35X3fs8bB8M88yFyPtSvE3YTPiEeKnRA5xX7sNlZCa8EUFds82UjZXNkUTfmzDdKUDsrEG/ZQYEoCm7ofE9ItzgbvFQ5TRuKwKT/k5SLMM1N0DcpQv8GckLcU5KVMPaE9U+KQlTaYC1Om7JnNoSkzfhyGGuSlH8jrFS3G049XPLa+76XGgTmj3zyP+0xaPKdCNBp8Ljyj+KeQGEzyaVAIkRM4UuPCH7HU0ay4O3CEyNF0QfkoHufjPsBxvvkmxnBE/Pjjj4eXIMgrhg6Osl977bXgz6jPjmhkYv1NPnZC5Bh+RHns2LHBT8j7XIm7C1+0zz77bFDQTKnDLwy/tnIvTaDMdXZ2Bn806hCDDwo5Y4+vKL9hKaVODDdksRNCCCGEGKbIYieEEEIIkVOk2AkhhBBC5AQpdkIIIYQQOUGKnRBCCCFETpBiJ4QQQgiRE6TYCSGEEELkBCl2QgghhBA5QYqdEEIIIUROkGInhBBCCJETpNgJIYQQQuQEKXZCCCGEEDlBip0QQgghRE6QYieEEEIIkROk2AkhhBBC5AQpdkIIUYHZs2enISGEaGyk2AlRYNq0acmoUaPC/fz58yGN+7hx40L6mjVrQtru3buLeXlm6XD8+PGgAPCMa9myZcmVK1fCM6uHsneDvXv3FtuG69evJ9u2bevTL/LUCv2hX9ZH2qTf4NPtIs3G825Df61NG/eBYPX4q14wFosWLQp1Mm7M0/r16/usD9aY5fUy2HPkawTKrfe7uRZol/r82hNCOHpSXFCIEcfly5fDZ6CjoyNN6YV4a2trCHd3d/c0NTWFO3R1dRU/N9euXQvPOjs7Q/zcuXN98u7fvz/ktXg9yaobudva2oJcsHr16pCHftYCY0Fd1MvV0tJSHDPifgxpi+eMw2BBm8jY3NycpvQfP+f1BNkYCxsfIMwY2bwwtr7t9vb24hoD1hdxW2dDRaX1frfXAvXb2NjYCTFS8d8RIIudEAWmTp2aFDae5OrVq2lKL8SXL18ewidOnEjmzJmTzJ07N8SxQFAGLl68mNy8eTNZtWpViM+cOTMpbNIhDKNHj05D9efll18ObZlcsHbt2mChGz9+fIg/8cQT4f7TTz+F+0A5fPhwqIt6uV555ZX0SVJsy2BMec64DJbVjjZ37tyZFDb75MCBA2lqY/D+++8HyxxzYxw9ejRZunRpkBs2bdoU7saYMWPSUC+sL9bcsWPH0pS7AxaxctawSuv9bq8F6t+8eXNSUODDuAohfkWKnRApK1eu7HNcyTEZ8QULFoT45MmTg2LjN7xTp06F+8SJE8PdH5Ox8Xhl625x+vTpZP78+Wksm9u3b4e7yWmwydoxoB0r2/GtHQl62EjffvvtMDawePHiPopKjLXLxj9YoEQg54ULF9KUu0N/x+7MmTPJwoUL01gvyPruu+8WjzBRWA4dOhTCpUA5mjVrVhobGgay3u/GWmA8GVchxK9IsRMiBQXOWxS++OKLYJUz6wNhrBLz5s0LflLe8sAG3dXVlaxbty5s7rVYi1AMsi7ajDGFYNKkSeFeirfeeitpb28vWoYMNlmsP01NTUWrHpslfTFrjOeTTz5Jjhw5kkyfPr2irxeybd++PbRrMGam/HC/W1B3bH2tFhT4kydPFuXEl8vPtdHfsbt06VIyZcqUNNYLFjrKo4iiHJrCXAoUxhs3biRLliwJccr49VGpvCmjtfqm9Xe9x2vB5LA/pPhs2XrgTtzC1j98EWMYT8ZVCOFIj2TvOKMVYiSCHxC+O4CPGr5DMfgR4QfFZwY/Ow/+PvhJ8Yy7QRnS6u1jV029yOFlyYJ+4P+ET1Tcpxj8m8hHfsaLuIEs/orrwvfNZL0bfmwGddejfvrGOqCvpah27BgP8zmLoRxjQz1+zZHfj2c83oyl1ckcV7O+yJ+VL27LX6Uotd4hriMeG9qzOfL98PJZmD7T9xiTWYiRTPwZkMVOCAcWFyxSWD444sw6NuK4ieOyzs7O5PHHH+9jJcGSgY9XYXNOdu3aNeR+XnYkiEzlwELCxbGgHT2XAgsmx6/ff/99sHC++eab6ZNeCpst3zLBuumtUGats+M6O3JERqxNdlk+LDpmkcSyQxzrGVgZ4qRXi71RaVclyxV9/cc//hH6WSpvf8auFJTH8sT6wyUghvFkTeE7WM6n7J577gn3eLyA8aLP+/btS1P6wpzSDhcUFKo+8SwqrfdSawHOnj0b/Fe9+0M5zHIuhCiPFDshHI899lhQ6Ng8//Wvf6WpvbAxmqIBM2bMCHccyTmW9BuUKYQD8fPyioe/aL8/IM+nn35aUakDkx3H/KeffjqEY1BszJcM2GhRwEr5ONmLABs3bgz3X375JdxjOLZEqTE5t27dGo58P//88+Tjjz8OaRzjoUjyogFyWBmUQ5ShSgqagVJpygpXf3wgTWmKqWbsSsH4efCdY/1lwZp68cUXw/GnX4fEWR88t3UXjxcyPvzww6HPHPvWSn/Xe7wW4Oeffw4K7Z49e9KUbHB94Oh/w4YNaYoQohxS7IRwsEGx8bFZZllf8BEyqwNvyTY1NRWVAzYe23BN0WhtbQ13cxyvBq94+CvLqT5+GcIwn6b/+7//S1N6rVxZFkRk/fLLL4Ni9c477wTFwitwHqwy1jfaQLl69NFHQ9xbYwDFDyXNlzGwyqGMmHKAgoP1hz6y4d93331hQ+etUUCmCRMmhHlh3IEytFHqRYLYn7BakMssnfSJl0VQ2kx58fRn7LIUKixw5qtIW1ix2traQhxu3bqVhnpZsWJF0R/PwCqGxey9995LU+4crx9++CF58MEHw7N6vXhRbr33Zy1gtfvwww/T2J1gOaTswYMH05S+1ENRFSJXFDaMgAsKMaLBVwi/qhh+L450Pitc+PyYPxR3yuEjxbPCZtPHp8in1xvqjf2XkMXk9Jf5LhnITTryAfVY3vi30vBzwv+QPlgZ80cEGxvSbVyAcSKduv0z0g3ztQLq8eWBvN63DKyM+WF5kJH0gcDYWR+5kAdfspj+jB0gL3V7yGfjY21ZP61+Ll/OfrcQGem39ZM8PIN4vGiHfKTRRjxe/QXZaI++myx+DVazFvyc+/ny8+nD5InXADL4eoQYifCZ8kixE2KYw2YZKwyNiil3bNImM0qHVwDYyIlz2aZOHlMg2Nx9Ge4oEgZKGHlNyWkUTCmuJ6ZcAf2mfsY4Hi8uwqwV5Bjq9WLzZsogd1PskNPm04fJg/xeUeS5/+NCiJFIrL+N4p9CYjh+SINCiGEEx5m8xFHY+ILP0kiGI0D83DjitN8YbBQ4tiwoIuE4/7nnntPLADXCEfZrr70Wjp0HeuwuRB6I9Tf52AkxzEGZ6+zsDD5PfMBHKmz0ONlDlj/iUIPy0d3dHd66xv8t9jUT1cG4sc7xv+R3BKXUCdEXWeyEEEIIIYYpstgJIYQQQuQUKXZCCCGEEDlBip0QQgghRE6QYieEEEIIkROk2AkhhBBC5AQpdkIIIYQQOUGKnRBCCCFETpBiJ4QQQgiRE6TYCSGEEELkBCl2QgghhBA5QYqdEEIIIUROkGInhBBCCJETpNgJIYQQQuQEKXZCCCGEEDlBip0QQgghRE6QYicaht27dyejRo1Kxo0bl1y/fj1NTZI1a9aE9GXLloV4f/Nx7d27N6TxzNLsIu38+fPhuUHZadOmhee0Q91AmDSeeWbPnl1sF6jP8lKXtUuatWXP7bJ2fJ8MytBGvaHeRYsWhfaPHz8e0kr1nbxeXi7ybdu2LTwfaoaL3Iw38q1fvz5zrj2sG1u7QH7SKG/yD9f58rD2WN8mJ328cuVKeGafHX+RRv8GG+SkfWS1z4sQDUdPigsKMWS0tLT0NDU19bS3t6cpvbA+r127lsb6l6+1tTWN9YRn5Ono6Ajxy5cvF+syuru7Q5w7dHV1FT8fpBG2Z0AdpDU3N6cpvdDG6tWrQzhuF/bv39+nLmu3ra0txI3Ozs6Qj6ueIDfteZlMBpPJ9x0Ybx832bgPJcNNbuRkvdj6yMLmJ17PrA9LG67z5aEv9MFkOnfuXJ8+VfOZHUyQx8YVWYQYavxnHGSxEw3F+PHjk8Jml2zZsuWOv8h5ZvQnnyeOT506NXnllVeSmzdvFus5ceJEMmfOnGTu3LkhjnWgsJGEMGmFDTnkMU6ePBlkKXzJF60MsG/fvuRPf/pTCMftwujRo9NQL9RNPUeOHElTkmBduffee5PCJpem1I/3338/WHjWrl2bppTvO4wZMyYN9bJq1arw/NixY2nK0DDc5EbOV199Ndm1a1dJqx3zQz9s7Rw4cCA5ffp0WPOW1oj9NqtWtVy8eDF8/pANZs6cGT4HRjWf2cEEeTZv3hy+B5gjIRoNKXai4XjkkUeS1tbWZN26dWlKNtXmq8Tt27fDnQ0FJk+enBw+fLjPUcupU6fSUJIsXLgwOXr0aBpLks8++yxsSk1NTUHJAzZrNp7FixeHeLVQho3aQOnqbx3VcubMmdAXT6W+Z4HMs2bNSmNDw3CU2+YZxSYLlD5TduDChQtBmUCxMYbrfHkmTpwY7v6IGMXJlNUs4s/sUMBnh8+QEI2GFDvRkHR0dIQNy/sXZVFtvlJgYdu+fXvS3t6epvRaPbAYzJs3L/hDxVaB+fPnhzYNLHVsMHzRf/vttyHtiy++SNra2kK4Wtic6cczzzyTpvQP85XLurK4dOlSMmXKlDTWS6W+x+DfdePGjWTJkiUhjn8XflxAWdqmT1yESePCUkic+0DnzlOr3CjiNn7Ib/KabNRvyhPP6We1fbK6Y0xB+/HHH8PdQ3mUOK+4XL16tTi2Rq39tj6YpRnZ6Zv1h/zl+lYPGIeurq7wBxr9wzJZjqzPLP6KJh/yUo+NFXEbf2T3eRg/IEx5LuryEPeKs8Fnh8+QEA1HeiR7xxmtEEMB/nDmW4MfC3405mPj6U8+Lg95/IVfUhbUT9k4j7XDc67CxhrS8RFqaWkJYfyg8BXyUCb2ZzMZuOhDXMawvPUklsdTqu/kN3m56C/jYdh4WBk/9llhfJR8eq0MVG7mzny8TB7m0HyoCNtYsd6sbKU+kc+vixhkyZoDysfrkjTfnmeg/Qbmy9Yw+akLfFsWLjdf1n58lcofQ93IQRmTx4jr9H3kM8P8gMlPmtVBXht/+s78ge+rhXnu54q5ox7L57GxFWKoidehLHaiYXnuuefCHT8WjjlLUW0+T+FLmU9CsHZgocjyc+Io6NChQ0nhyz15/PHHi3nwsSl8+Qf/poMHDxb96H7/+98HHyjyca/2mKiwaSSFDSVYaGo9Vq4XpfpuMHaFzTNYK2M/I/q9Z8+eNFaZsWPHpqG+mAWFq9o3OQcq94wZM4IVBygPHFf+97//DRaiv/zlL8nZs2dDOsR+XzHWJ/JxnFopv4f2ONJfsGBBmlKZWuYLCx6Y1a4SpeaL9mmH9QyEuWw8K4HlbufOnUFOjqFjy12pz+ykSZOSn3/+OcRxXWAsWIN2JH3r1q1gTWd+v/nmm+DCUQ4/V8xdbNUWotGRYicaFr5g+aJH2eH4pBTV5sti06ZN4b5x48ZwB45t/CbHpg/eF8r87HjR4YEHHghpbCYolm+++Wa/j2Hpw44dO2o6VrajxKyrWqrpu0F/X3zxxTDuvgwb6fLlyyv2A7kYx3feeSdN6YspBlz+BY8sapXblAF+xsIUvAcffDD4tZmSZcpD/EKCp1KfqgHFiyPCapTBeswX/XrhhReSN954I03Jph59ywKl3a8V5ATGPov4M8s4bdiwIYyFr8eP3+9+97swJl999VWY6yw4zp4+fXqoS4jhjBQ70VDElgY2uNbW1mBp8Aw0Xxzny3/r1q3BQuD9aPDfsbxY5lDY/Ibwhz/8IShhlPcbCBsfby0+9thjaUovcbtgDuAG9aMQZlkQLa/fkGPMYpJ1ZYGFMItyfUdp86xYsSLUg8we5qOS1Q65sHxiRakHtcqNdYYxRPEByqK8//DDD2GOsVThO8k6K8VA+sQLEAbysxZRtGL8SxOeesyX1Y0fXynqPV8elClb2/Y5tHGOPwtZn1n6i4WOF5msnocffjgoeijiv/3tb4P/azmlHEsj9WKFr5ZSnyEhhpTChzXggkIMCfizsA4LG1Oa0gv+Mn599jcfl/nk4ItjZclv4Fdj+fh9OcvHxTOf1+AZbXiIF77s09ivZLVLmDSfv6CYhnTLx4WPkslCOvEsefoL9cS+TOX6zt3SfTn7PT76gS8SvkfAWJJu0J5hYfpLG7VSq9zITD58rPx8UI/1h7n1fYBq+xSXA/LRvvffYsyy8gJyxGur1n6Dl5t0k6favtUKciKf/zzY5xWyPjuAHKT/+9//Ln4OyWvy00/qQmagD/7zynhaXh+mjPdD9M88yFxqroQYTPgceIqx+IEQIt/gRB4rCrXCZufrtLApeWzMXGzStsk2wgaJfKZYmCIHjJEpGWzuXkGqtk/cyeeVLrDyXokgj7UXYwqZKSr1wPptciE3/fT9Kde3RgD5WGeMjZcrVtJ55seOZyiCcZgxsbli/uk7z+N5IY3nQgw1rH3PKP4pJAb/iTQohBgBcGRV2JzCUR4voPgjZXF3Yew5ti8oE+E32wDfvoceeqj4MkMW9jMe+JSWOpoVdx/8Al977bXwkpTmQQw1sf4mHzshRihsSN3d3eEFkAkTJmT+VpeoPzj5x0od8Jtxq1f/+j8uZMFc4euHQl7tm8KifvAZYRPF95L/vUNKnWhEZLETQgghhBimyGInhBBCCJFTpNgJIYQQQuQEKXZCCCGEEDlBip0QQgghRE6QYieEEEIIkROk2AkhhBBC5AQpdkIIIYQQOUGKnRBCCCFETpBiJ4QQQgiRE6TYCSGEEELkBCl2QgghhBA5QYqdEEIIIUROkGInhBBCCJETpNgJIYQQQuQEKXZCCCGEEDlBip3IJcuWLUtGjRqVLFq0KE3pZdq0aSF99+7dfeL+ouzx48fDc+PKlSvFOrlmz55dzGPp48aNS86fPx/SDOLkrTfUS99oFzmuX7+erF+/PshAGv2yPnI3ue1CpgMHDoTnQw3yI4/Jxngy3mD9sYv4mjVrQn8Hm23btt0hXykYc+Q0kJdyyE89MFz6XYrhtOZi+YTINT0pLijEsOfatWs9TU1NYV3v378/Te3p6e7u7mlpaUljvfnIQzpcvny5p62tLaR1dXWFNGhtbe1ZvXp1yM9FHR0dHeGZ1WFxo7OzM6Rz1RNkpG++PcLIxDNAVmQ2eGZx5OU5cp07dy6kDRXIQl8YK0Ae4jYfzB1yWpw7z5mjoYDxZRybm5vTlGx4bjID44/M9BeGW7+zGG5rjs8z7fvPtRB5gHXtkcVO5JLx48cnN2/eTNrb25Nnnnmmj6WDZ4YPw9SpU5O9e/cmhQ2pj8Xl8OHDyRNPPBHyc73yyivpkzvrAKwy9957b1LYkNOU+vH+++8HK8natWvTlCQ5evRosnTp0iA/bNq0KdyNuM/2/Ouvvw73oeLixYthnlatWhXiM2fOTAoKQAjD6NGj01Avc+fODc+PHDmSpgwujO/OnTuTgjJT0vpk6cgKxE+fPp1s2bKlOA+N2G8swGZNrIbhtuawiDKG27dvT1OEyCdS7ESuee6555IbN24EZag/oAyy8dpxa3Nzc/L2228XFcTFixf3UaxieEaeu8GZM2eShQsXprFe2Fzffffd4lEeG+mhQ4dCOItbt26F+4wZM8J9qJg4cWK4e4Vi8+bNRaUoC+Zlzpw5aWzwYaxZDxcuXEhT+vLBBx8k//rXv9JYEvKR3xQgGI79jhmOa27+/PlByRYiz0ixE7mGzWbr1q3JunXrKvpFeWKLySeffBKsJdOnT++XVaMaYl8ku2L/QOPSpUvJlClT0lgvWEOampqCAlHJF4tnb7zxRtLS0lJUJLyvIX5TlcDHD6uhKb4DBeWgq6srzA8yVPLBoj0sqijeQF/M15DyJpdZW/GpsvkizefhmYUpz526DcKk4RsWjyf1XL16NY39Cmvso48+SpYsWZKmJCEf+T219puwzZf1qRL0v9SaGgi1rjlkRnb6QHkuxtv6Qhhsjuiz5YFSY+DXBHk8kyZNCvf+fBcIMexIj2TvOKMVYrjj17T5++Cr5P2AgHzmy2QQj9PxE8KvCF8n6jN/KSBv7GMHVk89KdUW4D9U2GiDjN6XiT5Tzq7Vq1enT3qhPmSlT/StGqxMPcBPC5li2Wz87Ir7hY+a+anZvBI3PyrClo7fmvlb+jVgYfOfM8yfjXvcT/L5vAZjEo9tqbww0H6D1UlfS62HmFiOuB1/VTu3A11zjCv5WXM21rb2GBfCJoOVJe77mjUG3KmXq1R/q+2bEMMB1rRHFjsxItixY0fwk/vuu+/SlPLcvn073O3IDLD+ccT6/fffh2OxN998M33SOOBHhEWPo9qVK1emqb0UNjk+/cFStGvXrpLWNvppYPEwq5Vh1pB9+/alKX0xKwpXtdZNLFj4rhU24yBbbMEqbMRJQQEI1iGsXAbHemapsWNA0r799tsQ9nz11VcVj8fHjh2bhnr7DrNmzUruueeeEK4ER5P4YlbLQPsdM2bMmHD31iqri3s8hwbWM9YEF+ujoBQV4+WOhT0DXXOMK76GrDcba8L04eTJk8nzzz8fPq/EYwt1FjYGfEbxWeTy8ynESEGKnRgRsEkV/upPXn/99TSlPG+99VY4NmLjZTPiCMhg8+HoB1+3euAVIX/159gMeTxsmqV8idiI2XCXL1+epvQyb968cNS8YcOGNCUJjuYosjjJ23Hgww8/HDZrlI0sTDHgKueHCCh+/riMzRiy/NcYd1PQrQzzShmUFlPwSGNuOG7jBRbmsNLRG+ONYvLOO++kKb2gVOAbZnKVwxSpahSiWvsNxJH7s88+K76E8fHHHyePPvpoUAbx9YOXX345zOF7770X4vWi1jW3YsWKoICzzhlnYG2h7DHm+BMyp998803y4IMPhucxWWNgcNT+0ksvpTEhRg5S7EQusY3Cg08QL1J44nwoAChxWAz8RuitDeTBQsEGClltGWb5K6dYeEXIX6Uc0bMUKt7SNOsY8pw9ezZpa2sLcYhlxDLjywDWIfwRDx48mKYkYaOeMGFCsBadOHEi+eGHH4qbLBt5PUCRtPGxMUYJABs/A6WJfjFH1ic2dMYqtmgxh7/5zW+C0kR48uTJ6ZM7YbxR5FEiDOrfuHFjeKkhCxRGD38MvPrqq2nsV+J8Rq39Ji8KHPNo9aAQ/f3vfw9zhq8foBhyVaOc9oda1xwyMbakM87A2mKd0Q9TyFF2SynLWWMAKMAo9aX67C3xQuSOwhdawAWFGPbgs8Oa5u7xPleAbxD57MJPiDL45xiFjaOnvb29mJc8xA1ry/sYmX+Pr5e4r3egUE/sr0S/8Emy9pAJue2ZpeOLZNAH0qgLvyTzO6KfVpY6LQzURV7SaKNWXyXGg/YZH2ShbS+jTzcKm3hI5/rnP/8Z6kAen4e+2TzznH74sfdrwMLUSz4DuagXv7y4n7TFOBgmkx8rg3xeNqi135Q3uSln65H54fJQB3LRB9/vWql1zdn4+r4Q92NBnX6c6YOPZ40B9VkeSzPIR/tC5Il4TUuxE2KYwWYVKwq1Qn1sosDmZ4qQbdCmHNjGy3PkYHMeSpDVlCC/4aOMedn8eNEvytA3HwbKoCyQTp12ecUOBYs0exEDyo0F+chPuXphCgpygs2HzQ/PrE82hzZWVmaoQV7kQiY/lvTFxoo5RW6DMSY/lBoD1jHpXLEiS3nyCZEnWOueUfxTSAx+CmlQCNHAcORU2LzDjy/zO30caYnBgePFp59+Ohz9nTp1Kk3t/WmOY8eOlTz6wx+NixclSh3NirsLx934kRaUyLv2G5NCDAWx/iYfOyGGGSgG3d3d4Xf18KUy/yxxd8E3jJdLwPs/4s+Fol3Oh4254g1N8nm/RjE4sPE9++yz4e1cKXUi78hiJ4QQQggxTJHFTgghhBAip0ixE0IIIYTICVLshBBCCCFyghQ7IYQQQoicIMVOCCGEECInSLETQgghhMgJUuyEEEIIIXKCFDshhBBCiJwgxU4IIYQQIidIsRNCCCGEyAlS7IQQQgghcoIUOyGEEEKInCDFTgghhBAiJ0ixE0IIIYTICVLshBBCCCFyghQ7IYQQQoicIMVONATLli1LRo0alSxatChN6WXatGkhfffu3WlKkly/fj1Zs2ZNMm7cuPCMPMQNnq9fv77Pc8qfP38+xLni/HEbe/fuLealnK8LfF128Wzbtm3heQz5Z8+encbqB/UyZrR//Pjxkn0H7l5eLmQ6cOBAeN7oXLlypbhOTHb6DD7dLtIYn6GA9v3YlwL5bE0ZrD36Rh0wnPpdC+X62Whrl/armV8hhoSeFBcUYtC5du1aT1NTU1iH+/fvT1N7erq7u3taWlrSWC/EuS5fvhzinZ2dfdZvR0dHn+erV6/uaW1tDWGexWu9q6srpJHPQxmeATKRB3mM9vb2PnWZHNw9lu7z1gP6x5jRJ6Nc34FnFmfMeY5c586dC2mNDHIjL3Jz0RfrO3H6YXH6z3PGZ6iwdWVrKAv64+cva50Nt34PlHL9BOKNtHarmV8hBgPWoacYix8IMdiwBlGWmpubwxc3sMHZlznYl6kpLgZf+gb5/YZAXVYH5SjvN042CC7aNShDPi9HXI42SPMgR1tbWxrrzcNmbeXriY2Vp1zfgbCPWz9jZbQRicefcfV95bmPm5I0lEor68qvTY+NvV/L5KWMZzj2G2K5K1Gpn424dsvNrxCDBZ8Dj45iRUPx3HPPJTdu3Ejef//9NKUvn332WVL4ck+mTp2apvRy6tSpNJSEZ++++2442oHx48cnhw4dCmGeFZSh5MSJEyEOR44cSVatWpUUNthimW+++SYpfGmHsv3h5s2byaxZs9JYkqxduzZZvHhxGqsvZ86cSRYuXJjGeinX9yxu3boV7jNmzAj3RoZ5e/vtt8NxMzCujG8pbt++He4zZ84M96Fg/vz5yenTp9NYXz7++OOk8EdAn7VMXsp4hmO/B0J/+9kIa7fc/AoxVEixEw0FisjWrVuTdevWFZUTz88//3yHUhezadOmpKmpKWwU+NLZRmGgDB09ejSE8UUiziZImf/+978h/T//+U/ypz/9KYSrBX8blNIlS5akKf0j9iGyK/Y7NC5dupRMmTIljfVSqe8enr3xxhtJS0tLMnfu3GKa+ezhp9dIvlqffPJJUMKnT59e0pfRYO1s3749aW9vT1N6fdfoE30jjI+U+bfRTxtny0ea+XICYT8nfmwpk+VvNWnSpHDPWsuvv/568uSTT6axX/NYGaPWfjfynHr608+stRvPr82d9dfmkThhnwfqPb9CDBVS7ETDgfWML2y+aKvBvsxtk0Y5xILX1dVV3Cj8ZobCdvjw4RDmbhYSFLyvvvoqhClXraXNNoP33nsv+f777ysqnqXo6QmuEXdcpSxuWBhjKvUd6DPyTpgwIcS9tXPjxo3Jww8/HNp98cUXk19++SV9MvSgfDO+yPXaa68F53m/+QJ/ENA3FNvnn38+2bx5c/okCQoP5bu7u5MffvghKBJm8bx48WKxrgULFgRrLe098cQToT0g3NHREcaGOaYM4MBPfWZByuKnn35KQ73YSwF+jcV5jFr7PZhzikJmnweYN29en3g5qulnubUbz+/OnTvDeFAvihdhxp04L2lw3a35FWIokWInGpIdO3aEL/HvvvsuTfmV+K/jzz//PNz37NkT7gZf3Fi12LxXrlyZpibJAw88EO58ye/bty9s5PDQQw8FZYh02/CrgY3g3LlzQdEqdYQ82JTqO3CUjcwof7t27SoqGYBlwo6/uGMNQTFkM7VNG3jzljB3IA+KNUo282NlLF+sXIJXAriqAcUVudjAOfZ+88030ye92MaMYsYfBl4xYGP/8ccfQ5+ogw3eFAM2bcYKOTmGf+SRR0J6Oe65555wRzl78MEHQ7haPvzww+Spp55KY5Wppd/VzCn5zapnb5pSjvlEwQIsVuQhblauGOpHDi5AyfLxSlTqZ7m1G88vUB99O3nyZFB4+T4hHlu6s6hlfoUYSqTYiYaEL2c2KY6rPPfff39Q+PzGFWOWOwOfN+8Hw5c9FsGDBw8W4/D73/8+5CO9v8ewKAn85V/qCLkaTMGJLzbTaqnUdw/KHxvl8uXL05ReH8EY+kY+LsYNpQAllg0WPz/g+BwLGEo2vmNW5rHHHks6OzvDnMV4JaDSxs8G7i24zBl9tfZjOJIGrFXGli1bwvyYMgo29/C73/0uWGmw2rL+sjDLGP3jGgisXZSSao7s69HvauaUOXv00UeTa9euJR988EHIYxawpUuXBjmwpPPHAhZklGCvVNVKf/uZtXaz5hdLJXOK4j5nzpyi4l5KUavH/Aox1EixEw1BlqLGJoXPmmfFihXBh4yNy8r4FyEApcN8dMhz9uzZ4KTuYWNiI2DTMvgip2423fgY1hzSPfHRDLJhNfAblGHlyyl9XsnxV6mjWNqKqdT3eJyx9PgyHqwyvi+MDxYujvI++uijsAGisLEp4/t43333haM3818EytTLud1baBhHlAyUEYj7hWKAsunLcLzGWE6ePDlYo4CNn/CYMWOS3/72t8m3334bwqVgvLDOcuzeHyZOnJiGkmDV5Y+W+Mje5/HU2m9PqTllLf/9738PR5zMLfAHAXGUHfuMoWzRhn9BqF6U6ydUWrtZ84sCh+z0j+fUe+HChZKKez3mV4ihRoqdaAiefvrpcOcvccM2KQ9fzseOHQubEZuOHQnx17sdnWAh4ogVxcN8cd55551wN+yojXIeFD4vg/G3v/0t3M1CwF/+KIZgGyXyvvXWW0HZMcsZ+bC4/fGPfwxxrCPESa8V2rh69Woa66Vc3zlKY7NGPtv42NhxtGfzph8ogeYkjvKB31EM40w+UzzZJMeOHRs2xHKKaC2gdCEn40/fGEdkwIoEtn7wzbKxNV9NlE366+cVnyn4n//5n2TDhg3BmsNYYCEqp9gB+TiOrObHcTkaBK/E8dZy1rhaHisD9eh3NXNKn1k7NqdAHVjwiNvRZn+wtVENlfpZzdrNml/qZbxR9oB1Wola51eIIafw4Qu4oBBiGFDY1O74Hbta4bfPqLOpqSnUz++Kkcb3A78haBQ23ZDGHchHGa64DHlIH2oKykKQif5dTn87jt9C82NYUPT7/K5cQREqyu7D5KEc/ePyY+ahDO0a/Dabby+GvJSpJ9XMqY0Dafa7cNxtvHhOnDooy93mvlHIml8g3eIdHR191nG951eIoYB17xnFP4XE8FdSGhRCDAM4VipsOMFqwe//YTEUjQPHiljNCspc8WgfqxIv6ZglKgZr1OOPPx5eDsiyHIvGIWt+hRgKYv1Nip0Qwxg2l2effTYcU/EGYrVHX+LuYseJvIlpCpop4hxvllPCOXbkpaHL6QsqovHIml8hhgopdkIIIYQQOSHW3/TyhBBCCCFETpBiJ4QQQgiRE6TYCSGEEELkBCl2QgghhBA5QYqdEEIIIUROkGInhBBCCJETpNgJIYQQQuQEKXZCCCGEEDlBip0QQgghRE6QYieEEEIIkROk2AkhhBBC5AQpdkIIIYQQOUGKnRBCCCFETpBiJ4QQQgiRE6TYCSGEEELkBCl2oqE5f/58MmrUqHAdP348TU2SvXv3hrRx48aFuM/nr0WLFoVnBnXMnj27+HzZsmXJlStXwp049fn869evL7ZhTJs2rVh29+7dxbqQCawuf5Hm6wXymyy0QVv1hPqQlTGAUn0H2rd0k2fNmjXJ9evXw/NGYbj0Ydu2bXfIVwry2NoB5CUN+akH8jB3Bp8Z+wyZrEYj9aU/cyhEQ9GT4oJCNBTt7e09TU1NPS0tLWlKL62trT379+9PY735SDMuX77c09zcXCx37dq1UE9nZ2eInzt3LsS7u7uLzzo6OsIzg/J8NqjLID/pBvX7dqmLMlYXZclD/YbVYfV2dXWFMtzrAWPh6y/Xd2Acad/i3Hne1tYW4o3AcOsDY8+68GslhjzISN8M1goyW1oe5s4w2UxWW/fGcJxDIYYa/xkCWexEwzNmzJjwV3PhSzb8te8ZPXp0GurN55k6dWpS2BCSmzdvhvjFixdDeNWqVSE+c+bMZPXq1SE8fvz40MbRo0dDHPgrvbDJJoWNJTl58mSamiQHDx4M9RqU9cRx5HjllVdC22a1mzt3bnLp0qXwDGgbfvjhh3CvBawbW7ZsSV599dVi/eX6Dn4cAfl4fuTIkTRl6BlufWDsd+7cGdbtgQMH0tS+vP/++2Hubc2Q7/Tp02H+LK2R+41VyyzC1XDixIlkzpw5QUag73zGjOE4h0I0GlLsxLBgypQpyYsvvpi89NJLVR/LcHy1a9euoODAxIkTw92Ot2Dz5s3FTWb+/PnJ4cOHQxi++OKL5KGHHkoWLlyYfPnll2lqEjaZFStWpLHquH37drizKWdhfZo8eXK41wKKALCBGpX6ngXKhK9jqBmOfUAxaG5uTi5cuJCm9IX1aQobkI/8ppBDHubOYH3zGeOzaZw6dSoNZdPocyhEoyHFTgwbUKbwudm4cWOacidsGuafM2/evKAMmjWML+iurq5k3bp1wccn/gvcNg/bdI4dO5YsWLAgKHe2+Zivjd94K0GZ7du3J+3t7WnKnWC5wXJhsnrY0K1P8eU3SOPHH38Mdy9jpb7HUC9+X88880yIY2lk7K1d8wekLq9w3E1q7QPKM9Yl5Ke89cl8vLAGW19I83l45seAu/eLYzxI9/5iBvVcvXo1jf0K5VEYvLJPPvJ7au03YZs3Lzd1DdbcGaxvLHB8NmP/1ywafQ6FaEjSI9k7zmiFaBTwVTN/Ne+Dg++L+eIAebyvm/nn4G/mwW+msLmEerh78G+ytsyvBp8m8uLrhJ+T+ToZtOnbBfL7C1+iUvCMdr2fVS0gP21mUarvjJWXl3Gj3x7rI3IiL/i5GSwG2gc/d9YX4jY3hC2ddcYFfm4tjAwWpl3KMi62Zjzk83UYpMXrolReqMfceblrmTvfnr+qrQ+ZkYMyfgyG2xwK0QjwWfHIYieGFYsXL04KX7DheHXs2LFpajYcVWGxw1/JgwUEv5nChhGOwrwFhGNX/OxIMz86LCqFDSb55ptvkk8//TRY8aqhsMnxaQsWCqwAWUfIWBCw5h06dOgO37y7Qbm+Q2GDSwqbW7AkYSEqBc89ZgkB+mp+V1hESOeNzizMksRVrfVooH2YMWNGkBMYbyDt22+/DWHPV199FdZaOWz9sc44TmX+vL9YObDi4rdZ7VqCeszdL7/8ktx///1prBfmyObLW/HKzR3rmos1zufR4mvXrk1zlIcxYw4KylTy+OOP3/HZGA5zKESjIsVODDvY3Dhyxcm8P7Bh+aMXOwLzvjN/+MMfQt1sCoQNFD7S2IDYYPvDpk2bwj0+QmaD+utf/1pRqUNurwD5K+soNotq+m4gy44dO8I4+DJAmxMmTEiefPLJNKUX6rOj7CeeeCLcAaX1+++/T5YuXZopqykE1SgFtfaBzZsyKCqmHJB25syZoGjde++9YW7tuL0UjAHr4Z133klTeqF/f/nLX9JYeeKXJspRj7kjjty4FcT+of74/6mnnkpDleduIKBA+vFFKQPzC/U0+hwK0ahIsRMNz61bt9JQL3xxYym4fPlymtJLnI8v6ddeey1YFIwNGzYUv/Rts/LPH3jggXDHyuf/2mdDJG3lypVpyq/E1oY4zga1devWYGXxGyR1ffLJJ8XNnY0q67fsUHi8AuQvNrWYUi9glOu7vdxhUC8Wy9jSSJtYUt566600pTwo3yiCWF14I7JWau0DVhkU6diihfXsN7/5TVAaCJd7iYUxwKqDBddg7uhflo8k+D8GkIW18MILL6Qpv1Lqj4Za+01e5GY+yvl6euo9dwbtm1zUizXc1vFwmUMhGprCAg+4oBANA/41rE0uwkZBuQi+MPjGgM9nF/45+CKR1/IQJ53nlI99nKCwCd7xu1lWP345HvxyrD2ri7LWvpcZvzTLx2Xl/DVQnycPY0JdXtZKfffpBuVJt34wLsB40hdAXpOZ57RNvZaXfDb+tVJrH/75z3+GOmztGPhgmrw8R2buhj0DC1OvjQH1IRfE/pxAW35e/fjEkM/LBrX2m/Jebgv7uSOf9cPS6jl3Bj5v9vngise6Ul8aZQ6FaCT4zHik2AmRM9ik+Dz7zb9W2CRt0+WyzY8NzzZYU3Jpl7yUsTTy1FtJ6C8mF/L4TRplwzZ18AqD9Zt++DBQBiXBK/dcHpQH0syJH1AmSs0N+chPuXpBWyYb8lvbfu5Qtqxv8Xw2wtwZjTKHQjQS8ZodxT+FxOBzkAaFEMMcjq74vb09e/YUj7nE4MLR4dNPPx1cBuzncjju41j/xo0bIZ4FLzBw4Uuq47+hJWsOhWg0Yv1NPnZC5BBe2MA36c9//nPxjUcxePDCw/Tp00PY3t4EfktttftfI7JAIedtzebm5lCPGBpKzaEQjY4sdkIIIYQQwxRZ7IQQQgghcooUOyGEEEKInCDFTgghhBAiJ0ixE0IIIYTICVLshBBCCCFyghQ7IYQQQoicIMVOCCGEECInSLETQgghhMgJUuyEEEIIIXKCFDshhBBCiJwgxU4IIYQQIidIsRNCCCGEyAlS7IQQQgghcoIUOyGEEEKInCDFTgghhBAiJ0ixE0IIIYTICVLshBBCCCFywqieAiEwalRIEEIIIYQQw4dUlQv0Uez8AyGEEEII0djE+puOYoUQQgghcoIUOyGEEEKInCDFTgghhBAiJ0ixE0IIIYTICVLshBBCCCFyghQ7IYQQQoicIMVOCCGEECInSLETQgghhMgJUuyEEEIIIXKCFDshhBBCiJwgxU4UOX/+fPivSUpdA+H48ePJsmXLknHjxhXrWbRoUWirv+zevbuPPP6inaGAdmm/0WB8GWcbH4N0m4uRwLZt28I4wN69e0PfLV4vGnUNlIL+My6VmD17duhXqc8W48nzeo9nuTb7y92S8W6wZs2aIGs1c9OfvGLkIcVOFJk5c2bS0dERwq2treH/nuPq6uoKaQPhxIkTyZEjR5Jjx44l165dS9ra2pLDhw8nf/3rX9Mc1bNq1aqkpaUlyGiyUSdpg8Fw2sAZ30cffTSMT3Nzc5raO8f/+7//m8aGJ15Z6w/8gfHiiy+msYEx3JS4Wjh16lQayobxbG9vT2ONQbw2GlHGUuzcuTMNVYa8g/W9J4YfUuxERfhy9P/BcH9Yu3ZtcuPGjaBQjB8/Ppk1a1ZIJ20gUIeHOBvQ3Llz05S7x48//piGfoV2Bzo2d5PLly8nM2bMCONz6dKlNLWX0aNHp6HhydWrV9NQZVh/hw4dSmO1M5zWQCkYD8alHowZMyYN1Q/GcqCf56y1cTdkbATi70IhDCl2oiwcf8bWEdKmTZsWLBccbdmxQDXcunUr3FevXh3utXD9+vXQvoc0k4dr/fr1QVaIj2WyjiWvXLnSp7wddVCWdLBne/bsyTzWpE1L5zjrwIEDId23T5qNIekGMtkRmJXPwuezOaDvYGXmzZsXnpfDjwFyW5rVjYwmv59z8GVZEzE8p6885zK5kJM/FiydMHmh0hgh465du4LV18r7MoRNJp8e4+v2/bayhuXBUkd9ldaAr8P338oZfo3YFcvp54C6uOLx4G5p3Fm/YGn002RgLdva9kd49M3yIJPNt3H79u1iOZOlFH5uqcu342XlsnH32HM7ivXyclEnV5YMWWvDQ/tW3sYPysns8WNq+Rk3ytM28Xh8CJf6nIIfk3j+IX5ebuyFKFL46yjggmIE09HREdaCv1pbW9OnPT379+8Pac3NzT3Xrl3rOXfuXDFfJcjf1NTU09LSkqbcibVfCmSx9rLa5Tn10xZXW1tbnzw89/2J20O+7u7uEI6fZclG3jgPdVy+fDnECwpseG5xe97Z2Rni7e3tIW4gG2nA2Ppnho1jnI+2DNq0fsTEMtNmV1dXCFvdyAnISdzGk3ImO5CPMY6xekwmWzdgc2B1Ml+sJ6OaMeLyWBn6wVhbW6T7vHHdNhbWf577sbHnNpbxc4jH0+LITf+s78wT0Db9RU6eI5+fOw/54vHmAtIZe+qgLvLaM6BNG2fqt2ekWRiZvCzU58fL6vDP47ny+QmbTPbdYGNHmLEAxtuX8/gyQD7aJI16rY0seBbXi4y2NiDug9WXJXOMHw/LS5ww5anXPpfEabfU59TWicllcZsbi5ssyOj7RtjyipEN68Qji53IpPClwUpJCptHmtLLBx98EO5PPfVUOArgiLVann766WTOnDk1H40VvsyCbIUvzqTwRZmm9lrb+Gt9x44dQTYuO/qtFo6I7RjowQcfDPf+8O677wY/rqlTp4a4+c188cUX4Q6MAf6C8MgjjyQ3b94MYRg7dmxy5syZ0BfGNuvImroos3nz5hAnH216K0Q1mKVi+fLl4Q4mpx3VISdtXbx4MYxnYVMK/pLG0aNHky1btqSxXzEZrf+LFy8Oc2ZztGHDhuIcMV+FjbKPNaLcGJWCMvSDsaetUhQ232LdzDVr/csvvwzxesL80D/6Dr/88ku4Y7XGCoOcPH/44YeLlraYwmaevPfee2msd32tWLEihOkDc04d1EWdMTbOzEPW8Strh6N6kyXr8/LMM88Un//jH/8Ic5Ulr83tSy+9FPJSN2OLny3wWf3qq6+K664/3wN83zBX1Mt4/fzzz+mT6rC1AU8++WToA1SSOQvGlPGwvMhDmPLMAZ9fqPQ5/fDDD0N5kys+fuY582/pjD2yClEJKXaiLGwe/gu4v1+oBkcQHEVQF1+ApWDzKbcpe6jHKz4//fRTGho4HAHxRcsXNEeZ/cU2jBg7gq7EO++8EzYNlI9SRy8//PBDGupLNcqPZ8KECclHH32UTJo0KU3prZt6OPqxC2yje+KJJ0IZNmc7LkPemFIylpsjU3zuNlnrr5RidTfA5+vkyZOhTcYR5RiFPguUuNOnT4e8jPfChQuL8lOWY0M7Rh3opo+iQR18PtetW5em/kqWT2bWPFrafffdV1w7yET/gD8IUHpYd/GR5GDi+1NJ5lqo9DmttOZ4zmfN5LLvI/vcCVEKKXaiInwB88UyUNg4eMFh06ZNIU59KE71+oLCv6XaukptoMAXKV+eWC0++eSTpLu7O33SSzVO2FglspS4yZMnp6HysGljXcEaicKU9fawyRFvjCiD/cEssn/+85+LdVE39fDMX2btwXrA848//jh5++23g/Uii1Jjdc8994Q7flsxEydOTEPlKTeHA4G+33///SFcaY6rWQOVWLJkSRhDLpQc+oNCnwVrgLcfsf4w3i+88EL6JEk2btyY7Nu3L1jUWC9Yf/oLn02ULKy2KJBYw8th8/bb3/423D02t/xx49eO/WGIxYowz/k+QP5609+1UUnmWqj1c0pfsJB7ubhiy54QMVLsREX8FzDHDsCREF9YsaM1ChZKoDkCoyyxcbBpsInxjDtfpFmbAxaIckpk/CVJHIdp4AsPxYoNEGg7/ssbpY2jJ8rx3L9FZ3+9YyXBimVWKmuTt0wBKxobInfb6CwP1j7kMUsb/UGmBQsWhDhKn+9DXB6rCXKh4LERZh3FWl02L+RnPjiuAqsrS3kCn45FFiuQzRd1Mzf2MgR1MX+0Ybz66qvJ66+/Hqy3pTYZk9Ec0SnP2NAnNraXX3451M3FfKGUmOWv0hj5OTQH/LiMkZWORcb+EGD9sjY57gU/x5T77rvvQtyoZg3E8TgdpXjp0qXFjZp6sqyIxsqVK8N4sza8dZQxZe446uWonHbsj4q4TY/Ph1WJ+bDjwLNnzxbLGswPaVxvvfVWmCuT1+oBm9s33nijWAfriP4x3qwx0ukD+bLWtpXzcpPm28maU6PU2vD4usvJHFNKtlJU+pzyc0SsRfuuoE1kMf7yl78Uxw7sM2SUa1uMcApfLAEXFCMUcwYudQEOwTjxEi8oLMF51z83h9/Cl3+I4yhsz+MrC6svC5zF4zrsMgdj7oUvx5CG03JcX0Fp6fPc5EVO+obcxLkTJ0x+8M/NIZoxIM6YAHl8n3kxgHEFP77mEB+Xt/q5aNccq2NwQrd+UIfJA35+rG3DHLh5jpxg40p9PPd1czeHd4M+UoeNeSmQ3eqhX34cfD+RlzrtmaWXGiPyUC/ptOHLWJ8gqy7WA2Nl7VNPPMb2jPtA1kAc9/MB9jKFv6iftZmFyRCPN3JTp42DrXXCvk3GwfCfIdJtLInHnwfgGfX6Ptlc0Y7VZeNLfTY+lKGspfNZsPz+c+GJxypLXotbmx6eI7ONiZeRMPi+AGWyZI4pJRtx6shqq9znFOy7gjzUwbhYfeDHHhkt3X/HWJoYubAOPKP4p5AYrCRpUIh+YRa2Rlw/WIzwG9Larh9YDrDi1eO4aiSC1QVrjLe+8JMYWPHq9ftyQoiRQ6y/6ShW1A02fJFf7Citvb29pG+dqAwO8d6xnqM4HOqr9cMUQohySLETNYHlxmDDbyTYMF977bUQ9nKKgTN9+vTkoYcekgN3DXR1dYWXHvgrm4sXZPC78hY8IYQYKDqKFUIIIYQYpugoVgghhBAip0ixE0IIIYTICVLshBBCCCFyghQ7IYQQQoicIMVOCCGEECInSLETQgghhMgJUuyEEEIIIXKCFDshhBBCiJwgxU4IIYQQIidIsRNCCCGEyAlS7IQQQgghcoIUOyGEEEKInCDFTgghhBAiJ0ixE0IIIYTICVLshBBCCCFyghQ7IYQQQoicIMVOCCGEECInSLETQgghhMgJUuyEEEIIIXLCqJ4CITBqVEgQQgghhBDDh1SVCxQVOyGEEEIIMbzRUawQQgghRE6QYieEEEIIkQuS5P8DoxLoXxKd1UgAAAAASUVORK5CYII=\"\u003e\u003c/p\u003e\n \u003cp\u003eThis positive correlation also confirms the presence of surface water land uses, such as aquaculture, is likely to cause a positive bias in salinity results if not masked. Previous studies, such as Nguyen et al. \u003csup\u003e\u003cspan class=\"CitationRef\"\u003e33\u003c/span\u003e\u003c/sup\u003e in coastal Vietnam found NIR reflectance and the Vegetation Soil Salinity Index (VSSI) \u003csup\u003e\u003cspan class=\"CitationRef\"\u003e13\u003c/span\u003e\u003c/sup\u003e respectively explaining 89 percent and 77 percent of the variation in electrical conductivity. Both of these predictors are highly sensitive to water, such that as water increases, NIR decreases and VSSI increases. After accounting for surface water land uses, we also find a positive relationship between electrical conductivity and both NIR and VSSI, however, these relationships are significantly weaker as compared to that found by Nguyen et al. \u003csup\u003e\u003cspan class=\"CitationRef\"\u003e33\u003c/span\u003e\u003c/sup\u003e. NIR and VSSI only explained 20 percent and 8 percent of electrical conductivity variation, respectively, when tested on pure soil data (Supplementary 2).\u003c/p\u003e\n \u003cp\u003eThese findings collectively attest to the necessity of developing alternate measures of salinity in coastal areas than those used in dryer contexts, and the need to address the presence of water-based land uses in these approaches.\u003c/p\u003e\n \u003c/div\u003e\n\u003c/div\u003e\n\u003ch3\u003eStacked ensemble method is an effective approach for modeling coastal salinity\u003c/h3\u003e\n\u003cp\u003eLinear, log-linear, and polynomial regression models with single predictors for electrical conductivity found the best fit with NDWI, NDSI2, NIR, NRSWIR2, and NBNIR (Supplementary 2). The explanatory powers of all these predictors individually, however, was low (R-squared less than 0.40). These explanatory powers were also lower compared to what previous studies have presented, a bias potentially owing to unaccounted presence of surface water \u003csup\u003e\u003cspan class=\"CitationRef\"\u003e31\u003c/span\u003e,\u003cspan class=\"CitationRef\"\u003e33\u003c/span\u003e\u003c/sup\u003e.\u003c/p\u003e\n\u003cp\u003eAdding all variables in a single polynomial regression model increased the overall adjusted R-squared as compared to that of individual variables, but the explanatory power of the model remains moderate at 0.57 percent. In this model, most bands remained statistically insignificant (except NRSWIR2 and NBNIR at 95% confidence). Results did not significantly improve with a principal component analysis (PCA) either, although it reduced overfitting with reduced dimensionality. The training and testing accuracies remained low since a PCA still treated the relationships between EC and surface reflectance variables as linear (Fig.\u0026nbsp;5 shows many non-linear relationships including NDWI).\u003c/p\u003e\n\u003cp\u003eWe applied a Lasso regression to remove multicollinear variables. The predictors selected based on the coefficients of the Lasso model included Green, SWIR2, NDWI, NDSI1, VSSI, NBNIR, NBSWIR2, NRSWIR1, NGSWIR1, and NNIRSWIR1 (Supplementary 3). We used these predictors to train 9 machine-learning models.\u003c/p\u003e\n\u003cp\u003eApplying individual machine learning models to this data improved the overall accuracy. Mean accuracy of the models was 0.8496, and the models found to be above this threshold were nnet, svmRadial, knn, glmnet (Fig. \u003cspan class=\"InternalRef\"\u003e6\u003c/span\u003e). This accuracy was similar to what was achieved in previous studies, such as Sarkar et al. \u003csup\u003e\u003cspan class=\"CitationRef\"\u003e43\u003c/span\u003e\u003c/sup\u003e, but could be improved further. We therefore used a more advanced machine learning algorithm, ensemble stacking, to arrive at the final model for CoSal-SA.\u003c/p\u003e\n\u003cp\u003eApart from the two surface reflectance bands (SWIR2 and Green), other variables that contribute the most to the CoSal-SA model in predicting salinity include Normalized Difference Water Index (NDWI), Normalized Blue and NIR (NBNIR), Normalized Red and SWIR1 (NRSWIR1), Vegetation Soil Salinity Index (VSSI) \u003csup\u003e\u003cspan class=\"CitationRef\"\u003e13\u003c/span\u003e\u003c/sup\u003e, Normalized NIR and SWIR1 (NNIRSWIR1), Normalized Difference Salinity Index (NDSI1) \u003csup\u003e\u003cspan class=\"CitationRef\"\u003e68\u003c/span\u003e\u003c/sup\u003e, Normalized Green and SWIR1 (NGSWIR1), and Normalized Blue and SWIR2 (NBSWIR2) (Fig. \u003cspan class=\"InternalRef\"\u003e7\u003c/span\u003e and Eq. 2).\u003c/p\u003e\n\u003cp\u003eOur correlation analysis (Supplementary 8) shows varying direction of relationship between predictor variables with EC and moisture. NDWI, NBNIR, Green, and VSSI have a positive relationship with EC; NNIRSWIR1, and NDSI1 have a negative relationship with EC; while NRSWIR1, NGSWIR1, and NBSWIR2 have a non-linear relationship with EC. Simultaneously, NRSWIR1, VSSI, NNIRSWIR1, NDSI1, NGSWIR1, and NBSWIR2 have a positive relationship with moisture, SWIR2 and Green have a negative relationship with moisture, while NDWI and NBNIR have a non-linear relationship with moisture.\u003c/p\u003e\n\u003cp\u003eOverall, a combination of these different indices in the CoSal-SA model appears to help identify saline soil in the presence of water. The model is validated by the quality metrics. The training accuracy of the CoSal-SA model is 91.8%, with high precision of 100% and a reasonable recall value of 71%. The high F1-score of 83% suggests it is a balanced model.\u003c/p\u003e\n\u003cp\u003eWe get a test accuracy of 85.7%, precision or proportion of accurately classified positive predictions of 80%, and recall or true positive rate of 67%. The model has an F1 score of 72.7% suggesting a balanced performance between true positives and false negatives on unseen data.\u003c/p\u003e\n\u003cp\u003eWith the Bangladesh sample, we found a testing accuracy of 69.3%, precision of 71.2%, recall of 82.2% and F1 Score of 76.3%. The accuracy drop on external data is common and expected since there are variations in data collection methods and other differences in the geography. The model maintains high precision and recall, suggesting a reasonable reliability in positive predictions of salinity and the maintained F1-score suggests our model is robust.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eLong term salinity trends suggest overall reduction across the region except in the coastal belt alongside aquaculture ponds\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eTo explore the applicability of CoSal-SA, we applied it to Jagatsinghpur district at two scales - one at the district level, and another about 200 sq.km. study area in the Ersama block in the district\u0026rsquo;s coastal belt. The study area was identified such that the (lower) half of it was around aquaculture ponds, while the (upper) half was without much aquaculture conversions (Fig. \u003cspan class=\"InternalRef\"\u003e2\u003c/span\u003e). Both areas, however, were equally flooded by the 1999 storm surge, and are equally distant from the coast with similar elevations from the sea (less than 5m). This selection was done to account for potential coastal influence on the observed salinity, with the key differentiator primarily being aquaculture.\u003c/p\u003e\n\u003cp\u003eApplying the CoSal-SA model to Jagatsinghpur district\u0026rsquo;s historical satellite imagery reveals insightful patterns of salinity change (Fig. \u003cspan class=\"InternalRef\"\u003e8\u003c/span\u003e top panel). In 1995, about 138.2 sq. km of the district\u0026rsquo;s total area (1668 sq. km) was too saline for rice cultivation. In the years after the 1999 storm surge, salinity increased towards the interiors and became more concentrated over time. It increased from 160.26 sq. km in 2001, to 230.05 sq. km in 2005. This salinity trend, however, declined in the years after 2005, especially in the interior regions, potentially owing to natural leaching processes. By 2010, saline areas had reduced to 199.05 sq. km. and remained somewhat constant up until 2014, however salinity had started to become more concentrated in the coastal belt again. It was also in this period when the district experienced over 70 percent increase in aquaculture activity in the coastal belt, expanding from 66.7 sq.km in 2014 to over 112 sq. km in 2024.\u003c/p\u003e\n\u003cp\u003eBy 2024, about 10 percent of the Jagatsinghpur district had salinity levels that were not conducive for rice cultivation, and over 6 percent of area under active aquaculture. Compared to 1995, when about 6 percent of the district had high salinity areas spread across the district, salinity seems to have almost doubled in the district and become more concentrated in the coastal belt (~\u0026thinsp;10km from the coastline) over these three decades.\u003c/p\u003e\n\u003cp\u003eSalinity changes in the study area reflect similar trends as were for the district, however, with more closer insights into the salinity dynamics with aquaculture land change in the area. Prior to the 1999 storm surge, salinity was concentrated around the southern part of the study area, and alongside the perennial rivers and deltas containing brackish water from the sea. Salinity became more widespread across both the north and south parts after the surge and became more intense at least until 2005. Over the next few years, the northern parts experienced a reduction in salinity, from 2010 to about 2014, before briefly seeing another period of increased salinity around 2017. Throughout this period, however, the southern part of the study area has sustained high salinity. It was also the same southern part that witnessed a change in land use to aquaculture in the more recent times. Aquaculture land use increased from 32 sq km. in 2010 to about 50 sq. km in 2024, or about half of the lower study region.\u003c/p\u003e\n\u003cp\u003eGiven that the two parts of the study area are similar in all ways except the one attribute of aquaculture and prevalence of salinity, there could be empirical exercises undertaken to test the causal pathways for salinity and aquaculture in coastal areas.\u003c/p\u003e"},{"header":"Discussion and conclusion","content":"\u003cp\u003eSoil salinization is an urgent global problem in the face of changing climate and increasing food insecurity \u003csup\u003e\u003cspan additionalcitationids=\"CR2 CR3 CR4\" citationid=\"CR1\" class=\"CitationRef\"\u003e1\u003c/span\u003e\u0026ndash;\u003cspan citationid=\"CR5\" class=\"CitationRef\"\u003e5\u003c/span\u003e\u003c/sup\u003e. Coastal soil salinity is dynamic with multiple interconnected natural and anthropogenic drivers, including sea level rise, coastal erosion, rainfall patterns, land use changes, and agricultural practices \u003csup\u003e\u003cspan citationid=\"CR8\" class=\"CitationRef\"\u003e8\u003c/span\u003e,\u003cspan citationid=\"CR11\" class=\"CitationRef\"\u003e11\u003c/span\u003e\u003c/sup\u003e. Estimating coastal soil salinity with remote sensing, however, has remained challenging owing to the confounding presence of water\u003csup\u003e\u003cspan citationid=\"CR22\" class=\"CitationRef\"\u003e22\u003c/span\u003e,\u003cspan citationid=\"CR29\" class=\"CitationRef\"\u003e29\u003c/span\u003e,\u003cspan citationid=\"CR31\" class=\"CitationRef\"\u003e31\u003c/span\u003e,\u003cspan citationid=\"CR69\" class=\"CitationRef\"\u003e69\u003c/span\u003e,\u003cspan citationid=\"CR70\" class=\"CitationRef\"\u003e70\u003c/span\u003e\u003c/sup\u003e.\u003c/p\u003e \u003cp\u003eInsight into the specific channels of salinity increase at a local level, however, requires information on the historical salinity patterns, which has been limited. Important attempts at consolidating national-level data on soil quality and salinity, have been made but are dependent on a country\u0026rsquo;s participation and data collection costs (e.g. GSASMap)\u003csup\u003e\u003cspan citationid=\"CR71\" class=\"CitationRef\"\u003e71\u003c/span\u003e\u003c/sup\u003e. These attempts, however, have not addressed tracing historical salinity changes at an actionable sub-national or local scale.\u003c/p\u003e \u003cp\u003eThis study offers a framework to estimate long-term coastal soil salinity in the context of soil moisture and confounding water-based land uses. This approach offers a means to study salinity changes over time at varying geographical scales, to enable future causal inferential research.\u003c/p\u003e \u003cp\u003eWe find that soil salinity in coastal India presents itself differently in surface reflectance as compared to how it is observed in arid and semi-arid contexts \u003csup\u003e\u003cspan citationid=\"CR25\" class=\"CitationRef\"\u003e25\u003c/span\u003e,\u003cspan citationid=\"CR29\" class=\"CitationRef\"\u003e29\u003c/span\u003e\u003c/sup\u003e. This justifies our undertaking to develop a soil salinity model more suitable for coastal areas with high moisture than the models developed for arid and semi-arid regions.\u003c/p\u003e \u003cp\u003eWe also find that univariate models based on indices sensitive to the presence of water cannot be used to estimate soil salinity, due to the growing presence of water-based land uses. Our findings hold up against evidence presented by Silvestri et al.\u003csup\u003e\u003cspan citationid=\"CR31\" class=\"CitationRef\"\u003e31\u003c/span\u003e\u003c/sup\u003e that the presence of water in aquaculture ponds causes a bias in the explanatory power of certain bands and indices (e.g. NIR and VSSI). We argue that water-based land uses must be accounted for before modeling salinity in coastal areas, and over reliance on a single band or parameter must be avoided.\u003c/p\u003e \u003cp\u003eDespite their coarser resolution, we demonstrate the utility of Landsat imagery in conducting soil salinity analysis, by accounting for mixed pixels with partial proportion of water before correlating with soil salinity values \u003csup\u003e\u003cspan citationid=\"CR31\" class=\"CitationRef\"\u003e31\u003c/span\u003e\u003c/sup\u003e. The additional step of performing a supervised land use land cover classification in the framework in order to mask water-based land uses such as aquaculture, is key for reducing bias while unlocking the advantage of long-term satellite data.\u003c/p\u003e \u003cp\u003eWhile machine learning based models help estimate non-linear relationships between multiple variables and salinity, their \u0026ldquo;black box\u0026rdquo; tendencies keep the direction and magnitude of influence of different variables opaque \u003csup\u003e\u003cspan citationid=\"CR65\" class=\"CitationRef\"\u003e65\u003c/span\u003e\u003c/sup\u003e. We use Shapley feature values to assess the specific contributions of all features behind the model, enabling us to make a better judgement of the model. For instance, we find a strong contribution of NDWI in the CoSal-SA model, suggesting its ability to detect salinity in the context of high soil moisture. This holds up against the results from previous soil science studies conducted in coastal and wetland contexts with varying levels of moisture \u003csup\u003e\u003cspan citationid=\"CR29\" class=\"CitationRef\"\u003e29\u003c/span\u003e,\u003cspan citationid=\"CR69\" class=\"CitationRef\"\u003e69\u003c/span\u003e\u003c/sup\u003e. Most other indices in the model employ Red, SWIR and NIR bands, which are widely used for soil mineral and moisture detection \u003csup\u003e\u003cspan citationid=\"CR72\" class=\"CitationRef\"\u003e72\u003c/span\u003e\u003c/sup\u003e.\u003c/p\u003e \u003cp\u003eMeanwhile, the ultimate challenge related to soil salinity facing science and policy is to identify and address its causes and consequences. This requires better understanding of the various causal pathways in a context where multiple drivers, such as coastal erosion, agricultural practices, and sea-water intrusion, are at play. While we do not claim any causal inferences in this study, our findings give some insights in that direction. Previous studies have found that soil moisture loss increases top-soil salinity due to capillary action that moves salts to the surface\u003csup\u003e\u003cspan citationid=\"CR73\" class=\"CitationRef\"\u003e73\u003c/span\u003e\u003c/sup\u003e. The positive correlation between the presence of moisture and higher levels of salinity in this coastal context is perhaps due to the mediating presence of saline or brackish water that is both associated with high salinity and high moisture.\u003c/p\u003e \u003cp\u003eFurther to this end, the spatial-temporal findings from Jagatsinghpur also highlight the increasing salinity in the interiors after the storm surge, and in the coastal belt alongside growing aquaculture, both related to saline water intrusion. One could argue that salinity in the coastal belt is due to the proximity to the sea, however, our comparative study area results show that parts of the coast with no aquaculture activity have lower salinity compared to those with aquaculture transitions, despite being equally distant, with equal elevation from the coast and having similar history of storm surge impact. A farm-level temporal analysis of salinity and land change can give specific insights into the causal relationships between aquaculture and soil salinity.\u003c/p\u003e \u003cp\u003eThe CoSal-SA model is, however, not without its limitations. It currently does not factor soil type and color, which may influence the visible bands of the spectrum. Other factors, such as pH were not included because of the low correlation between EC and pH in the field samples. In other contexts where acidity and EC are more closely related, pH could be included in the assessment, similar to moisture. More soil samples could perhaps improve the model further, however, our numbers are similar to other comparable studies \u003csup\u003e\u003cspan citationid=\"CR25\" class=\"CitationRef\"\u003e25\u003c/span\u003e,\u003cspan citationid=\"CR33\" class=\"CitationRef\"\u003e33\u003c/span\u003e,\u003cspan citationid=\"CR74\" class=\"CitationRef\"\u003e74\u003c/span\u003e\u003c/sup\u003e. We substantiate the results by adding Bangladesh soil data, to gain better insights into the external generalizability of the model, which to the best of our knowledge is done for the first time in a salinity modelling setting.\u003c/p\u003e \u003cp\u003eThis research identifies the distinctive property of coastal soils that exhibit both high salinity and moisture levels. Through the innovative CoSal framework, the study offers a thorough methodology for precisely tracking historical patterns of coastal salinity across various geographical settings, including areas where new coastal land uses like aquaculture are emerging. To the best of our knowledge, this is the first framework that comprehensively addresses the presence of water-based land uses in coastal areas into a soil salinity assessment. Additional coastal soil data, with similar salinity and moisture measurement approaches, could help adapt the framework to other geographical and historical contexts in the future. The ability to map salinity at a detailed level opens the door to critically needed research on the specific local mechanisms driving coastal salinity, which can guide focused interventions to address significant agricultural productivity losses, food insecurity, and the resulting poverty in affected regions.\u003c/p\u003e"},{"header":"Declarations","content":"\u003cp\u003e \u003ch2\u003eCompeting Interest Statement\u003c/h2\u003e \u003cp\u003eThe author(s) declare no competing interests.\u003c/p\u003e \u003c/p\u003e\u003ch2\u003eAuthor Contribution\u003c/h2\u003e\u003cp\u003eG.J., W.Y., A.E.F., and D.S.C. conceptualized the study; G.J., W.Y., and S.K.S. performed the methodology; G.J., W.Y., and S.K.S. supported data collection and visuals; all authors reviewed and edited the manuscript.\u003c/p\u003e\u003ch2\u003eAcknowledgement\u003c/h2\u003e\u003cp\u003eThe authors would like to thank Germ\u0026aacute;n Silva, Krishnachandran Balakrishnan, Teja Malladi, Pratyush Tripathy, Jiwon Jang, Nicole Corcoran, and B L Turner II for their inputs at various stages of the research. We would also like to thank Greenforce Labs and Deepthi Nagappa for their support with the field data collection. This study is supported by the US National Science Foundation (NSF Award Number 2409616), Intergovernmental Panel for Climate Change (IPCC), Cuomo Foundation, and Horowitz Foundation for Social Policy. This study would not have been possible without the support of Arizona State University and its resources.\u003c/p\u003e\u003ch2\u003eData Availability\u003c/h2\u003e\u003cp\u003eAll data, codes, and supplementary material from this study are available on the GitHub repository: https://github.com/garimajain2002/coastal_salinity_index.git. Further information may be available from the corresponding author based on request.\u003c/p\u003e"},{"header":"References","content":"\u003col\u003e\u003cli\u003e\u003cspan\u003eD\u0026iacute;az, S. \u003cem\u003eet al. Summary for Policymakers of the IPBES Global Assessment Report on Biodiversity and Ecosystem Services\u003c/em\u003e. 56 (2019). \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://ipbes.net/system/tdf/ipbes_global_assessment_report_summary_for_policymakers.pdf?file=1\u0026amp;type=node\u0026amp;id=35329\u003c/span\u003e\u003cspan address=\"https://ipbes.net/system/tdf/ipbes_global_assessment_report_summary_for_policymakers.pdf?file=1\u0026amp;type=node\u0026amp;id=35329\" targettype=\"URL\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eIPCC. \u003cem\u003eSummary for Policymakers. In: Climate Change and Land: An IPCC Special Report on Climate Change, Desertification, Land Degradation, Sustainable Land Management, Food Security, and Greenhouse Gas Fluxes in Terrestrial Ecosystems\u003c/em\u003e (Cambridge University Press, 2019).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eOlsson, L. et al. Land degradation. in Climate Change and Land: an IPCC special report on climate change, desertification, land degradation, sustainable land management, food security, and greenhouse gas fluxes in terrestrial ecosystems 345\u0026ndash;436 (Cambridge University Press, (2022).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eQadir, M. et al. Economics of salt-induced land degradation and restoration. \u003cem\u003eNat. Resour. Forum\u003c/em\u003e. \u003cb\u003e38\u003c/b\u003e, 282\u0026ndash;295 (2014).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eWicke, B. et al. The global technical and economic potential of bioenergy from salt-affected soils. \u003cem\u003eEnergy Environ. Sci.\u003c/em\u003e \u003cb\u003e4\u003c/b\u003e, 2669\u0026ndash;2681 (2011).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eFAO. Global Map of salt-affected soils. \u003cem\u003eGSASMap v1.0\u003c/em\u003e (2021). \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://openknowledge.fao.org/server/api/core/bitstreams/31be1fac-a057-4b6b-80ea-a4554910368c/content\u003c/span\u003e\u003cspan address=\"https://openknowledge.fao.org/server/api/core/bitstreams/31be1fac-a057-4b6b-80ea-a4554910368c/content\" targettype=\"URL\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eAmoako Johnson, F., Hutton, C. W., Hornby, D., L\u0026aacute;z\u0026aacute;r, A. N. \u0026amp; Mukhopadhyay, A. Is shrimp farming a successful adaptation to salinity intrusion? A geospatial associative analysis of poverty in the populous Ganges-Brahmaputra-Meghna Delta of Bangladesh. \u003cem\u003eSustain. Sci.\u003c/em\u003e \u003cb\u003e11\u003c/b\u003e, 423\u0026ndash;439 (2016).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eHossain, M. S., Uddin, M. J. \u0026amp; Fakhruddin, A. N. M. Impacts of shrimp farming on the coastal environment of Bangladesh and approach for management. \u003cem\u003eRev. Environ. Sci. Biotechnol.\u003c/em\u003e \u003cb\u003e12\u003c/b\u003e, 313\u0026ndash;332 (2013).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003evan Schie, D. et al. Addressing non-economic loss and damage: learning from autonomous responses in Bangladesh. \u003cem\u003eClim. Change\u003c/em\u003e. \u003cb\u003e177\u003c/b\u003e, 1\u0026ndash;22 (2024).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eChen, J. \u0026amp; Mueller, V. Salt of the earth: Migration, adaptation, and soil salinity in coastal Bangladesh. \u003cem\u003eNat. Clim. Change\u003c/em\u003e (2018).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eEswar, D., Karuppusamy, R. \u0026amp; Chellamuthu, S. Drivers of soil salinity and their correlation with climate change. \u003cem\u003eCurr. Opin. Environ. Sustain.\u003c/em\u003e \u003cb\u003e50\u003c/b\u003e, 310\u0026ndash;318 (2021).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eSen, R. Salt in the wound: embodied everyday adaptations to salinity intrusion in the Sundarbans. \u003cem\u003eEcol. Soc.\u003c/em\u003e \u003cb\u003e28\u003c/b\u003e, (2023).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eDehni, A. \u0026amp; Lounis, M. Remote sensing techniques for salt affected soil mapping: Application to the Oran region of Algeria. \u003cem\u003eProcedia Eng.\u003c/em\u003e \u003cb\u003e33\u003c/b\u003e, 188\u0026ndash;198 (2012).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eS\u0026aacute;, T. D., de Sousa, R. R., Rocha, \u0026Iacute;. R. C., de Lima, G. C. \u0026amp; Costa, F. H. F. Brackish shrimp farming in Northeastern Brazil: The environmental and Socio-economic impacts and sustainability. \u003cem\u003eNatural Resources\u003c/em\u003e 538\u0026ndash;550 (2013). (2013).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eAli, A. M. S. Rice to shrimp: Land use/land cover changes and soil degradation in Southwestern Bangladesh. \u003cem\u003eLand. Use Policy\u003c/em\u003e. \u003cb\u003e23\u003c/b\u003e, 421\u0026ndash;435 (2006).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eBelton, B. Shrimp, prawn and the political economy of social wellbeing in rural Bangladesh. \u003cem\u003eJ. Rural Stud.\u003c/em\u003e \u003cb\u003e45\u003c/b\u003e, 230\u0026ndash;242 (2016).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eZhang, W. et al. Aquaculture will continue to depend more on land than sea. \u003cem\u003eNature\u003c/em\u003e \u003cb\u003e603\u003c/b\u003e, E2\u0026ndash;E4 (2022).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eOttinger, M., Clauss, K., Kuenzer, C. \u0026amp; Aquaculture Relevance, distribution, impacts and spatial assessments \u0026ndash; A review. \u003cem\u003eOcean. Coast Manag\u003c/em\u003e. \u003cb\u003e119\u003c/b\u003e, 244\u0026ndash;266 (2016).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eAzad, A. K., Jensen, K. R. \u0026amp; Lin, C. K. Coastal aquaculture development in Bangladesh: unsustainable and sustainable experiences. \u003cem\u003eEnviron. Manage.\u003c/em\u003e \u003cb\u003e44\u003c/b\u003e, 800\u0026ndash;809 (2009).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003ePrimavera, J. H. Overcoming the impacts of aquaculture on the coastal zone. \u003cem\u003eOcean. Coast Manag\u003c/em\u003e. \u003cb\u003e49\u003c/b\u003e, 531\u0026ndash;545 (2006).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eGorji, T., Yildirim, A., Sertel, E. \u0026amp; Tanik, A. Remote sensing approaches and mapping methods for monitoring soil salinity under different climate regimes. \u003cem\u003eInt. J. Environ. Geoinformatics\u003c/em\u003e. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003e10.30897/IJEGEO.500452\u003c/span\u003e\u003cspan address=\"10.30897/IJEGEO.500452\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e (2019).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eMetternicht, G. I. \u0026amp; Zinck, J. A. Remote sensing of soil salinity: potentials and constraints. \u003cem\u003eRemote Sens. Environ.\u003c/em\u003e \u003cb\u003e85\u003c/b\u003e, 1\u0026ndash;20 (2003).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eMorshed, M. M., Islam, M. T. \u0026amp; Jamil, R. Soil salinity detection from satellite image analysis: an integrated approach of salinity indices and field data. \u003cem\u003eEnviron. Monit. Assess.\u003c/em\u003e \u003cb\u003e188\u003c/b\u003e, 119 (2016).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eFarifteh, J., Van der Meer, F., Atzberger, C. \u0026amp; Carranza, E. J. M. Quantitative analysis of salt-affected soil reflectance spectra: A comparison of two adaptive methods (PLSR and ANN). \u003cem\u003eRemote Sens. Environ.\u003c/em\u003e \u003cb\u003e110\u003c/b\u003e, 59\u0026ndash;78 (2007).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eAbuelgasim, A. \u0026amp; Ammad, R. Mapping soil salinity in arid and semi-arid regions using Landsat 8 OLI satellite data. \u003cem\u003eRemote Sens. Applications: Soc. Environ.\u003c/em\u003e \u003cb\u003e13\u003c/b\u003e, 415\u0026ndash;425 (2019).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eAllbed, A. \u0026amp; Kumar, L. Soil Salinity Mapping and Monitoring in Arid and Semi-Arid Regions Using Remote Sensing Technology: A Review. \u003cem\u003eAdvances in Remote Sensing\u003c/em\u003e (2013). (2013).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eBannari, A. \u0026amp; Al-Ali, Z. M. Assessing climate change impact on soil salinity dynamics between 1987\u0026ndash;2017 in arid landscape using Landsat TM, ETM\u0026thinsp;+\u0026thinsp;and OLI data. \u003cem\u003eRemote Sens. (Basel)\u003c/em\u003e. \u003cb\u003e12\u003c/b\u003e, 2794 (2020).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eElshewy, M. A., Mohamed, M. H. A. \u0026amp; Refaat, M. Developing a soil salinity model from Landsat 8 satellite bands based on advanced machine learning algorithms. \u003cem\u003eJ. Ind. Soc. Remote Sens.\u003c/em\u003e \u003cb\u003e52\u003c/b\u003e, 617\u0026ndash;632 (2024).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eYang, X. \u0026amp; Yu, Y. Estimating soil salinity under various moisture conditions: An experimental study. \u003cem\u003eIEEE Trans. Geosci. Remote Sens.\u003c/em\u003e \u003cb\u003e55\u003c/b\u003e, 2525\u0026ndash;2533 (2017).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eAllbed, A., Kumar, L. \u0026amp; Aldakheel, Y. Y. Assessing soil salinity using soil salinity and vegetation indices derived from IKONOS high-spatial resolution imageries: Applications in a date palm dominated region. \u003cem\u003eGeoderma\u003c/em\u003e \u003cb\u003e230\u0026ndash;231\u003c/b\u003e, 1\u0026ndash;8 (2014).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eSilvestri, S., Nguyen, D. N. \u0026amp; Chiapponi, E. Comment on \u0026lsquo;Soil salinity assessment by using near-infrared channel and Vegetation Soil Salinity Index derived from Landsat 8 OLI data: a case study in the Tra Vinh Province, Mekong Delta, Vietnam\u0026rsquo; by Kim-Anh Nguyen, Yuei-An Liou, Ha-Phuong Tran, Phi-Phung Hoang and Thanh-Hung Nguyen. \u003cem\u003eProgress Earth Planet. Sci.\u003c/em\u003e \u003cb\u003e9\u003c/b\u003e, 1\u0026ndash;8 (2022).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eLee, S. \u0026amp; Lathrop, R. G. Sub-pixel estimation of urban land cover components with linear mixture model analysis and Landsat Thematic Mapper imagery. \u003cem\u003eInt. J. Remote Sens.\u003c/em\u003e \u003cb\u003e26\u003c/b\u003e, 4885\u0026ndash;4905 (2005).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eNguyen, K. A., Liou, Y. A., Tran, H. P., Hoang, P. P. \u0026amp; Nguyen, T. H. Soil salinity assessment by using near-infrared channel and Vegetation Soil Salinity Index derived from Landsat 8 OLI data: a case study in the Tra Vinh Province, Mekong Delta, Vietnam. \u003cem\u003eProgress Earth Planet. Sci.\u003c/em\u003e \u003cb\u003e7\u003c/b\u003e, 1\u0026ndash;16 (2020).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eMorshed, M. M., Sarkar, S. K., Zzaman, M. R. U. \u0026amp; Islam, M. M. Application of remote sensing for salinity based coastal land use zoning in Bangladesh. \u003cem\u003eSpat. Inf. Res.\u003c/em\u003e \u003cb\u003e29\u003c/b\u003e, 353\u0026ndash;364 (2021).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eSarkar, S. K., Rudra, R. R., Nur, M. S. \u0026amp; Das, P. C. Partial least-squares regression for soil salinity mapping in Bangladesh. \u003cem\u003eEcol. Indic.\u003c/em\u003e \u003cb\u003e154\u003c/b\u003e, 110825 (2023).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eMahajan, G. R. et al. Monitoring properties of the salt-affected soils by multivariate analysis of the visible and near-infrared hyperspectral data. \u003cem\u003eCatena\u003c/em\u003e \u003cb\u003e198\u003c/b\u003e, 105041 (2021).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eAn, D. et al. Hyperspectral field estimation and remote-sensing inversion of salt content in coastal saline soils of the Yellow River Delta. \u003cem\u003eInt. J. Remote Sens.\u003c/em\u003e \u003cb\u003e37\u003c/b\u003e, 455\u0026ndash;470 (2016).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eDavis, E., Wang, C. \u0026amp; Dow, K. Comparing Sentinel-2 MSI and Landsat 8 OLI in soil salinity detection: a case study of agricultural lands in coastal North Carolina. \u003cem\u003eInt. J. Remote Sens.\u003c/em\u003e \u003cb\u003e40\u003c/b\u003e, 6134\u0026ndash;6153 (2019).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eEldeiry, A. A. \u0026amp; Garcia, L. A. Detecting soil salinity in alfalfa fields using spatial modeling and remote sensing. \u003cem\u003eSoil. Sci. Soc. Am. J.\u003c/em\u003e \u003cb\u003e72\u003c/b\u003e, 201\u0026ndash;211 (2008).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003ePang, G., Wang, T., Liao, J. \u0026amp; Li, S. Quantitative model based on field-derived spectral characteristics to estimate soil salinity in minqin county, China. \u003cem\u003eSoil. Sci. Soc. Am. J.\u003c/em\u003e \u003cb\u003e78\u003c/b\u003e, 546\u0026ndash;555 (2014).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eFadl, M. E. et al. Soil salinity assessing and mapping using several statistical and distribution techniques in arid and semi-arid ecosystems, Egypt. \u003cem\u003eAgron. (Basel)\u003c/em\u003e. \u003cb\u003e13\u003c/b\u003e, 583 (2023).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eG\u0026oacute;mez, A. M. R. et al. Digital mapping of the soil available water capacity: tool for the resilience of agricultural systems to climate change. \u003cem\u003eSci. Total Environ.\u003c/em\u003e \u003cb\u003e882\u003c/b\u003e, (2023).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eSarkar, S. K. et al. Coupling of machine learning and remote sensing for soil salinity mapping in coastal area of Bangladesh. \u003cem\u003eSci. Rep.\u003c/em\u003e \u003cb\u003e13\u003c/b\u003e, 17056 (2023).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eGrattan, S. R., Zeng, L., Shannon, M. C. \u0026amp; Roberts, S. R. Rice is more sensitive to salinity than previously thought. \u003cem\u003eCalif. Agric. (Berkeley)\u003c/em\u003e. \u003cb\u003e56\u003c/b\u003e, 189\u0026ndash;195 (2002).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eFAO. The state of food security and nutrition in the world 2024. Preprint at (2024). \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.4060/cd1254en\u003c/span\u003e\u003cspan address=\"10.4060/cd1254en\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eSharma, D. K. \u0026amp; Singh, A. Salinity research in India - achievements, challenges and future prospects. \u003cem\u003eWater Energy Int.\u003c/em\u003e 35\u0026ndash;45 (2015).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eKumar, P. \u0026amp; Sharma, P. K. Soil Salinity and Food Security in India. \u003cem\u003eFront. Sustainable Food Syst.\u003c/em\u003e \u003cb\u003e4\u003c/b\u003e, (2020).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eNRSC-ISRO. \u003cem\u003eStatus of Land Degradation in India: 2015-16\u003c/em\u003e. (2019). \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://bhuvan-app1.nrsc.gov.in/2dresources/thematic/ld0506/ATLASLD.pdf\u003c/span\u003e\u003cspan address=\"https://bhuvan-app1.nrsc.gov.in/2dresources/thematic/ld0506/ATLASLD.pdf\" targettype=\"URL\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eCensus of India. District Statistical Handbook. Preprint at (2011). \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://jagatsinghpur.odisha.gov.in/sites/default/files/2023-05/2018051173.pdf\u003c/span\u003e\u003cspan address=\"https://jagatsinghpur.odisha.gov.in/sites/default/files/2023-05/2018051173.pdf\" targettype=\"URL\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e (2011).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eGovernment of Odisha. Odisha District Gazetteers - Jagatsinghpur. Preprint at (2018). \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://jagatsinghpur.odisha.gov.in/sites/default/files/2023-05/2018051050.pdf\u003c/span\u003e\u003cspan address=\"https://jagatsinghpur.odisha.gov.in/sites/default/files/2023-05/2018051050.pdf\" targettype=\"URL\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eMahata, K. R., Singh, D. P., Saha, S., Ismail, A. M. \u0026amp; Haefele, S. M. Improving rice productivity in the coastal saline soils of the Mahanadi Delta of India through integrated nutrient management. in Tropical deltas and coastal zones: Food production, communities and environment at the land and water interface 239\u0026ndash;248 (CABI, UK, (2010).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eRafi, S., Mourya, N. K. \u0026amp; Balasani, R. Evaluation of shoreline alteration along theJagatsinghpur district coast, India (1990\u0026ndash;2020) using DSAS. \u003cem\u003eOcean. Coast Manag\u003c/em\u003e. \u003cb\u003e253\u003c/b\u003e, 107132 (2024).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eMohapatra, M., Mandal, G. S., Bandyopadhyay, B. K., Tyagi, A. \u0026amp; Mohanty, U. C. Classification of cyclone hazard prone districts of India. \u003cem\u003eNat. Hazards\u003c/em\u003e. \u003cb\u003e63\u003c/b\u003e, 1601\u0026ndash;1620 (2012).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eMandal, U. K. et al. Delineation of saline soils in coastal India using satellite remote sensing. \u003cem\u003eCurr. Sci.\u003c/em\u003e \u003cb\u003e125\u003c/b\u003e, 1339\u0026ndash;1353 (2023).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eDDMA. District Disaster Management Plan: Jagatsinghpur. Preprint at (2019). \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://www.osdma.org/districtplan/jagatsinghpur/#gsc.tab=0\u003c/span\u003e\u003cspan address=\"https://www.osdma.org/districtplan/jagatsinghpur/#gsc.tab=0\" targettype=\"URL\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eChhotray, V. A supercyclone, landscapes of \u0026lsquo;emptiness\u0026rsquo; and shrimp aquaculture: The lesser-known trajectories of disaster recovery in coastal Odisha, India. \u003cem\u003eWorld Dev.\u003c/em\u003e \u003cb\u003e153\u003c/b\u003e, 105823 (2022).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003ePekel, J. F., Cottam, A., Gorelick, N. \u0026amp; Belward, A. S. High-resolution mapping of global surface water and its long-term changes. \u003cem\u003eNature\u003c/em\u003e \u003cb\u003e540\u003c/b\u003e, 418\u0026ndash;422 (2016).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eUSGS. How do I use a scale factor with Landsat Level-2 science products? \u003cem\u003eUSGS\u003c/em\u003e (2023). \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://www.usgs.gov/faqs/how-do-i-use-a-scale-factor-landsat-level-2-science-products\u003c/span\u003e\u003cspan address=\"https://www.usgs.gov/faqs/how-do-i-use-a-scale-factor-landsat-level-2-science-products\" targettype=\"URL\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eUSGS. Landsat Project Documents. \u003cem\u003eUSGS\u003c/em\u003e \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://www.usgs.gov/landsat-missions/landsat-project-documents\u003c/span\u003e\u003cspan address=\"https://www.usgs.gov/landsat-missions/landsat-project-documents\" targettype=\"URL\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eUSGS. Why are negative values observed over water in some Landsat Surface Reflectance products? \u003cem\u003eUSGS\u003c/em\u003e (2022). \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://www.usgs.gov/faqs/why-are-negative-values-observed-over-water-some-landsat-surface-reflectance-products\u003c/span\u003e\u003cspan address=\"https://www.usgs.gov/faqs/why-are-negative-values-observed-over-water-some-landsat-surface-reflectance-products\" targettype=\"URL\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eBreiman, L. Bagging predictors. \u003cem\u003eMach. Learn.\u003c/em\u003e \u003cb\u003e24\u003c/b\u003e, 123\u0026ndash;140 (1996).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eDeane-Mayer, Z. A., Knowles, J. E. \u0026amp; L\u0026oacute;pez, A. \u003cem\u003ecaretEnsemble: Ensembles of Caret Models\u003c/em\u003e. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://cran.r-project.org/web/packages/caretEnsemble/caretEnsemble.pdf\u003c/span\u003e\u003cspan address=\"https://cran.r-project.org/web/packages/caretEnsemble/caretEnsemble.pdf\" targettype=\"URL\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e (2024). \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003e10.32614/CRAN.package.caretEnsemble\u003c/span\u003e\u003cspan address=\"10.32614/CRAN.package.caretEnsemble\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eXu, Y. et al. Classification using ensemble learning under weighted misclassification loss: Ensemble Learning under Weighted Misclassification Loss. \u003cem\u003eStat. Med.\u003c/em\u003e \u003cb\u003e38\u003c/b\u003e, 2002\u0026ndash;2012 (2019).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eAznar-Gimeno, R., Esteban, L. \u0026amp; Sanz, G. del-Hoyo-Alonso, R. Comparing the min-max-Median/IQR approach with the min-max approach, logistic regression and XGBoost, maximising the Youden index. \u003cem\u003eSymmetry (Basel)\u003c/em\u003e. \u003cb\u003e15\u003c/b\u003e, 756 (2023).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eRodr\u0026iacute;guez-P\u0026eacute;rez, R. \u0026amp; Bajorath, J. Interpretation of machine learning models using shapley values: application to compound potency and multi-target activity predictions. \u003cem\u003eJ. Comput. Aided Mol. Des.\u003c/em\u003e \u003cb\u003e34\u003c/b\u003e, 1013\u0026ndash;1026 (2020).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eShapley, L. S. In Contributions to the Theory of Games. Annals of Mathematical Studies. (1953).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eRichards, J. A. \u0026amp; Xiuping, J. \u003cem\u003eRemote Sensing Digital Image Analysis: An Introduction\u003c/em\u003e (Springer, 2005).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eHenrich, V., Krauss, G., G\u0026ouml;tze, C. \u0026amp; Sandow, C. IDB - Index DataBase. \u003cem\u003eIndex. DataBase\u003c/em\u003e (2012). \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://www.indexdatabase.de/\u003c/span\u003e\u003cspan address=\"https://www.indexdatabase.de/\" targettype=\"URL\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eShahmoradi, S., Malmiri, G., Sharifi Pichoon, M. \u0026amp; H. R. \u0026amp; Modeling and mapping of soil salinity and moisture using spectral and radar remote sensing. \u003cem\u003eAppl. Soil. Res.\u003c/em\u003e \u003cb\u003e10\u003c/b\u003e, 43\u0026ndash;65 (2022).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eGuo, Y. et al. Integrating remote sensing and proximal sensors for the detection of soil moisture and salinity variability in coastal areas. \u003cem\u003eJ. Integr. Agric.\u003c/em\u003e \u003cb\u003e12\u003c/b\u003e, 723\u0026ndash;731 (2013).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eFAO. GSASmap. (2021). \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://www.fao.org/global-soil-partnership/gsasmap/en\u003c/span\u003e\u003cspan address=\"https://www.fao.org/global-soil-partnership/gsasmap/en\" targettype=\"URL\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eUSGS. What are the best Landsat spectral bands for use in my research? \u003cem\u003eUSGS\u003c/em\u003e (2022). \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://www.usgs.gov/faqs/what-are-best-landsat-spectral-bands-use-my-research\u003c/span\u003e\u003cspan address=\"https://www.usgs.gov/faqs/what-are-best-landsat-spectral-bands-use-my-research\" targettype=\"URL\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eFu, Z. et al. Composition, seasonal variation, and salinization characteristics of soil salinity in the Chenier Island of the Yellow River Delta. \u003cem\u003eGlob Ecol. Conserv.\u003c/em\u003e \u003cb\u003e24\u003c/b\u003e, e01318 (2020).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eAbbas, A., Khan, S., Hussain, N., Hanjra, M. A. \u0026amp; Akbar, S. Characterizing soil salinity in irrigated agriculture using a remote sensing approach. \u003cem\u003ePhys. Chem. Earth (2002)\u003c/em\u003e 55\u0026ndash;57, 43\u0026ndash;52 (2013).\u003c/span\u003e\u003c/li\u003e\u003c/ol\u003e"}],"fulltextSource":"","fullText":"","funders":[],"hasAdminPriorityOnWorkflow":false,"hasManuscriptDocX":true,"hasOptedInToPreprint":true,"hasPassedJournalQc":"","hasAnyPriority":false,"hideJournal":true,"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":"researchsquare","isNatureJournal":false,"hasQc":true,"allowDirectSubmit":true,"externalIdentity":"","sideBox":"","snPcode":"","submissionUrl":"/submission","title":"Research Square","twitterHandle":"researchsquare","acdcEnabled":true,"dfaEnabled":false,"editorialSystem":"","reportingPortfolio":"","inReviewEnabled":false,"inReviewRevisionsEnabled":true},"keywords":"Satellite data, ensemble model, salt-affected soil, climate adaptation, shrimp farming, Odisha","lastPublishedDoi":"10.21203/rs.3.rs-6173117/v1","lastPublishedDoiUrl":"https://doi.org/10.21203/rs.3.rs-6173117/v1","license":{"name":"CC BY 4.0","url":"https://creativecommons.org/licenses/by/4.0/"},"manuscriptAbstract":"\u003cp\u003eCoastal salinity represents a critical global environmental crisis that threatens agricultural productivity and food security. Traditional remote sensing methods to measure soil salinity in coastal areas are confounded by the presence of soil moisture and ubiquitous water-based land uses. This study introduces CoSal, a remote sensing and machine learning framework for mapping long-term coastal soil salinity trends while accounting for soil moisture and aquaculture, a fast-growing land-based practice of fish farming. We apply CoSal in South Asia (CoSal-SA), where salinity and aquaculture acutely impact agriculture, where we integrate Landsat imagery with soil data from coastal India and Bangladesh. Using 28 metrics and a stacked ensemble of nine machine learning models, CoSal-SA identifies saline soils in waterlogged coastal areas with over 91% accuracy. Applying CoSal-SA to a coastal district in India reveals that 10 percent of the area in 2024 had salinity levels unsuitable for rice cultivation. While interior regions showed decreasing salinity between 1995\u0026ndash;2024, the coastal belt experienced intensifying salinity alongside increased aquaculture adoption. CoSal can be adopted for diverse coastal contexts and time periods with additional soil data. CoSal enables crucial research on salinity dynamics at different geographical scales that can guide targeted interventions to ultimately address agricultural productivity losses, food insecurity, and poverty in vulnerable coastal regions.\u003c/p\u003e","manuscriptTitle":"CoSal: A remote sensing and machine learning framework for mapping coastal soil salinity trends around aquaculture in South Asia","msid":"","msnumber":"","nonDraftVersions":[{"code":1,"date":"2025-03-25 10:02:29","doi":"10.21203/rs.3.rs-6173117/v1","editorialEvents":[{"type":"communityComments","content":0}],"status":"published","journal":{"display":true,"email":"[email protected]","identity":"researchsquare","isNatureJournal":false,"hasQc":true,"allowDirectSubmit":true,"externalIdentity":"","sideBox":"","snPcode":"","submissionUrl":"/submission","title":"Research Square","twitterHandle":"researchsquare","acdcEnabled":true,"dfaEnabled":false,"editorialSystem":"","reportingPortfolio":"","inReviewEnabled":false,"inReviewRevisionsEnabled":true}}],"origin":"","ownerIdentity":"2752eb9c-9b94-426b-acfd-de630358329e","owner":[],"postedDate":"March 25th, 2025","published":true,"recentEditorialEvents":[],"rejectedJournal":[],"revision":"","amendment":"","status":"posted","subjectAreas":[{"id":45918007,"name":"Earth and environmental sciences/Environmental sciences/Environmental chemistry"},{"id":45918008,"name":"Earth and environmental sciences/Environmental sciences/Environmental impact"},{"id":45918009,"name":"Earth and environmental sciences/Environmental social sciences/Climate change impacts"}],"tags":[],"updatedAt":"2025-09-04T11:54:05+00:00","versionOfRecord":[],"versionCreatedAt":"2025-03-25 10:02:29","video":"","vorDoi":"","vorDoiUrl":"","workflowStages":[]},"version":"v1","identity":"rs-6173117","journalConfig":"researchsquare"},"__N_SSP":true},"page":"/article/[identity]/[[...version]]","query":{"redirect":"/article/rs-6173117","identity":"rs-6173117","version":["v1"]},"buildId":"XKTyCvWXoU3ODBz1xrDgd","isFallback":false,"isExperimentalCompile":false,"dynamicIds":[84888],"gssp":true,"scriptLoader":[]}