Efficacy of Machine Learning in Simulating Precipitation and Its Extremes Over the Capital Cities in North Indian States

Efficacy of Machine Learning in Simulating Precipitation and Its Extremes Over the Capital Cities in North Indian States | 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 Efficacy of Machine Learning in Simulating Precipitation and Its Extremes Over the Capital Cities in North Indian States Aayushi Tandon, Amit Awasthi, Kanhu Charan Pattnayak This is a preprint; it has not been peer reviewed by a journal. https://doi.org/ 10.21203/rs.3.rs-4339400/v1 This work is licensed under a CC BY 4.0 License Status: Published Journal Publication published 25 Mar, 2025 Read the published version in Scientific Reports → Version 1 posted 10 You are reading this latest preprint version Abstract Climate change-induced precipitation extremes have become a pressing global concern. This study investigate the predictability of precipitation patterns and its extremes using MERRA2 datasets across North Indian states for the period 1984 to 2022 utilizing machine learning (ML) models. A strong positive correlations of precipitation 0.4 was found with dew point temperature and relative humidity significant at 0.05. In simulating precipitation, Random Forest Classifier (RFC) achieved the highest accuracy (~ 83%) for Rajasthan and Uttar Pradesh, while Support Vector Classifier (SVC) performed best (79–83% accuracy) for other states. However, the ML models exhibited about 5% lower skill in higher elevated stations as compared to the lower elevated stations, its due to the different atmospheric mechanisms control differently over the lower and higher topography. For extreme precipitation events (10th and 95th percentiles of intensity), RFC consistently outperformed SVC across all states. It demonstrated superior ability to distinguish extreme from non-extreme events (Area under curve ~ 0.90) and better model calibration (Brier Scores ~ 0.01). The developed ML models successfully simulated precipitation and extreme patterns, with RFC excelling at predicting extreme precipitation events. These findings can contribute to disaster preparedness and water resource management efforts in the region with varied topography and complex terrain. Climate Change Machine Learning North Indian States Precipitation Patterns Random Forest Support Vector Machine Figures Figure 1 Figure 2 Figure 3 Figure 4 Figure 5 Figure 6 Figure 7 1. Introduction Extreme precipitation events, intensified by global warming, have become significant challenges for both society and the environment. These events, characterized by intense precipitation, can wreak havoc on agriculture, ecosystems, and human communities 1 – 5 . As atmospheric moisture content rises due to climate change, these events are projected to become more frequent and intense. Observational data since the 1950s indicate a noticeable increase in the occurrence of heavy precipitation events in many regions, a trend expected to continue, according to the Intergovernmental Panel on Climate Change 6 . The sixth assessment report (AR6) also underscores that local communities, particularly those with limited adaptive capacity, are disproportionately vulnerable to these impacts 7 . The mechanisms associated with the extreme precipitation events during the monsoon season, mostly triggered by the interactions between westward-moving monsoon systems, eastward-moving mid-tropospheric westerly troughs, and the rugged Himalayan topography, have had devastating consequences in various parts of Uttarakhand, Himachal Pradesh, and Jammu & Kashmir. Notable incidents include the Kedarnath tragedy (2013) 8 , 9 and Uttarakhand cloudburst (2022, 2017, 2012) 10 – 12 , Lahaul-Spiti(July 2021), Mandi (July 2015) 13 , 14 , Leh cloudburst (2015, 2010) 15 , 16 , and the Jammu & Kashmir floods (2015, Sonmarg, Pahalgam, Ganderbal and Baltal) 16 and many more reported in detail in a study conducted by Dimri et al., (2017). These events, often caused by the interaction of multiple atmospheric dynamics, lead to excessive precipitation, casualties, and significant infrastructural damage. In North India, extreme precipitation events result from complex dynamics involving both large-scale atmospheric influences and localized factors. The unique topography of North India, particularly the Himalayas, plays a crucial role in shaping the Indian monsoon 17 – 21 . The interplay between the mountainous terrain and atmospheric disturbances, coupled with cold air intrusion from northern latitudes, creates conditions conducive to extreme precipitation events, as seen in Jammu and Kashmir during January 2017 22 . Western disturbances, embedded in the eastward-moving upper tropospheric Rossby wave train, contribute significantly to heavy precipitation in the Western Himalayas, especially during winter 23 – 25 . The high temperatures over the mountains and neighbouring areas contribute to the formation of low-pressure systems, which extend southward across the plains of South Asia, facilitating the northward progression of the monsoon 26 – 28 . The 2013 Uttarakhand disaster, which involved rapid monsoon progression and heavy precipitation, resulted from a combination of these large-scale circulations, orographic lifting, and intense convective activity 29 . These extreme precipitation events are known to exacerbate due to climate change with increased frequency and intensity 30 , 16 . Traditional modeling approaches struggle to capture the complex interplay of factors shaping precipitation patterns, including climate change, topography, and atmospheric dynamics 31 – 33 . In response, integrating machine learning (ML) techniques has emerged as a promising avenue for understanding precipitation 34 – 37 and extreme precipitation 38 – 41 variability and enhancing forecasting accuracy leveraging large meteorological datasets and computational power to improve predictive capabilities amid climate change uncertainties 34 – 36 . For India, a few literatures are available for extreme precipitation analysis 30 , 42 – 44 , however relatively fewer for extreme precipitation modelling using machine learning 37 , 45 , 46 . The limited availability of high-quality and comprehensive meteorological data in India and complexity of the regional topography and climate are also prominent contributing factor for limited number of research. A study by Ray et al. (2022) suggest that machine learning techniques, can effectively predict precipitation by correlating meteorological parameters with precipitation events 47 . In light of aforesaid reasons, our study aims to explore key atmospheric variables affecting precipitation patterns and extreme events in North India and studies link between these variables and precipitation intensity across regions. It further examines the performance of machine learning models in predicting precipitation and extreme events in each state and analyzed how different models simulate precipitation and capture extreme thresholds. Lastly, the study investigates the ability of machine learning models to distinguish between extreme and non-extreme precipitation events and their calibration and discrimination for extreme event prediction. Ultimately, the goal is to harness ML to improve the accuracy and reliability of precipitation forecasts, helping policymakers and communities prepare for and mitigate the impact of extreme weather events in the context of climate change. 2. Study Area Encompassing a vast region in North India, this study area covers seven states: Himachal Pradesh (HP), Jammu and Kashmir (JK), Punjab and Haryana (PH) (considered together), Rajasthan (RJ), Uttarakhand (UK), and Uttar Pradesh (UP) (Fig. 1 ). The dramatic elevation range stretches from a low point of 60 meters in Uttar Pradesh to a staggering 8,611 meters at K2 Peak in Jammu and Kashmir. This translates to significant temperature variations across the region. The plains experience scorching summers with highs reaching 45°C, particularly in the Rajasthan Thar Desert, while winters can be mild with lows around 0°C. In contrast, the hilly regions offer cooler summers with highs of 25°C, but winters can be harsh with temperatures dipping as low as -30°C in Jammu and Kashmir. Each state within this diverse landscape faces distinct environmental challenges 48 . The Himalayan states like Uttarakhand, Himachal Pradesh, and Jammu and Kashmir grapple with issues like forest fires, biodiversity loss, glacial retreat leading to water scarcity, and soil erosion. Meanwhile, the plains states of Uttar Pradesh, Punjab & Haryana, and Rajasthan battle water scarcity, soil degradation, and air pollution. Additionally, all states face challenges related to the impact of climate change on agriculture 49 , 50 . Understanding this interplay between natural processes and human activities across this vast region with its varying elevations and temperatures is crucial for developing effective management strategies and sustainable development practices 30 , 49 . 3. Results 3.1 State Specific Inter-Variable Analysis Understanding the intricate relationships between atmospheric variables is crucial for accurate precipitation prediction 39 . In this section these intricacies are explored by examining state-specific inter-variable interactions. For each state capital included in the study, correlation matrices were generated (Fig. 2 .). These matrices provide a valuable tool to explore the strength and direction of linear associations between the chosen atmospheric variables (temperature, humidity, pressure, etc.). By analyzing these correlations, we aim to identify recurring patterns, potential dependencies between variables within each state, and any instances of multicollinearity. Multicollinearity occurs when variables exhibit high correlations, potentially leading to issues during machine learning model training. Examining these state-specific relationships allows for a more nuanced understanding of how atmospheric variables interact and influence precipitation patterns across diverse geographical regions within North India 68 . Across all states, a consistent theme emerges: Dew Point Temperature, Surface Pressure and Relative humidity tend to be the strongest allies of precipitation. Strong Positive correlations of Dew Point Temperature, Relative humidity (~ 0.4) and Strong negative correlations (~ -0.3) for the aforesaid variables highlights their potential role as contributors to increased precipitation. The influence of temperature revealed a more precise observation. While states like HP and UK exhibit positive correlations (0.013, 0.04) while PH, JK, and UP show weaker negative (-0.0028 to -0.04) associations. In almost all states, solar irradiance and surface pressure acts as a counterpoint to precipitation, exhibiting negative correlations ranging from − 0.18 in PH (Solar irradiance) to a more pronounced − 0.3 in HP, RJ, UK and UP (Surface pressure). The impact of wind speed on precipitation varies across states. While RJ, UP and JK show modest positive correlations (0.04, 0.11 and 0.24), other states like HP, PH and UK exhibit weakly negative associations (-0.072, -0.054 and 0.037) respectively with wind (Fig. 2 .). Overall precipitation patterns across the study region are influenced by a complex interplay of various atmospheric factors, both directly and indirectly. In the direct category, elevated levels of dew point temperature, relative humidity and surface pressure emerge as significant contributors to increased precipitation in the studies locations. These states consistently exhibit positive correlations between dew, pressure, humidity, and precipitation, underscoring the direct impact of moisture content on precipitation. Additionally, surface pressure and solar irradiance displays a notable indirect relationship, with lower level of the events associated with higher precipitation events across the study location, however correlation with surface pressure was only statistically significant at significance level of 0.05. Conversely, temperature and wind demonstrates a varied impact, with both positive and negative correlations observed in different states, indicating that warmer and windier conditions may contribute to precipitation in some regions while having the opposite effect in others (Fig. 3 .). 3.2 Model Performance for Precipitation Prediction Four machine learning models – Support Vector Classifier (SVC), Random Forest Classifier (RFC), XGBoost, and K Nearest Neighbors (KNN) – were developed to analyze precipitation and its extreme patterns using the Modern-Era Retrospective Analysis for Research and Applications, version 2 (MERRA-2) reanalysis data. The models were trained over the data to achieve the highest possible accuracies. The developed model was trained to predict not only simple precipitation events, but extreme precipitation events at different threshold levels also. The developed model was trained and tested at different data split ratios and were evaluated based on their accuracy in predicting precipitation events. Figure 3 . illustrates the accuracy achieved by various machine learning algorithms across different states for different train test split ratios. State wise observations are presented. 3.2.1 Himachal Pradesh The analysis revealed that SVC consistently achieved the highest accuracy scores across all train-test split ratios in Himachal Pradesh, ranging from 83.35–83.50%. This suggests the effectiveness of SVC in capturing the complex interplay between various atmospheric variables and precipitation patterns in this state. The mountainous terrain and diverse climate of Himachal Pradesh necessitate a robust model capable of handling intricate relationships. RFC followed closely with accuracy scores ranging from 82.84–82.94%, demonstrating strong and reliable performance. This indicates the suitability of both SVC and RFC for precipitation prediction in Himachal Pradesh. XGBoost maintained competitive accuracy (82.28% − 82.75%), while KNN also achieved a reliable scores (81.62% − 81.74%) (Fig. 3 .) Here, the consistent high accuracy of SVC and RFC suggests their ability to effectively model precipitation patterns in Himachal Pradesh's diverse topography and climatic zones. 3.2.2 Jammu and Kashmir Similar to Himachal Pradesh, SVC emerged as the most effective algorithm for precipitation prediction in Jammu and Kashmir (Fig. 3 .). The average accuracy across all split ratios for SVC was 79.79%, closely followed by RFC with an average accuracy of 79.61%. This indicates the suitability of both algorithms for capturing precipitation patterns in this state characterized by the Himalayas and Kashmir Valley with its unique weather systems. XGBoost and KNN achieved average accuracies of 78.69% and 77.64%, respectively. The effectiveness of SVC and RFC in Jammu and Kashmir underscores their ability to handle the complex interactions between atmospheric variables in a region with significant topographic variations. 3.2.3 Punjab and Haryana The analysis revealed SVC as the most effective algorithm across all split ratios for predicting precipitation patterns in Punjab and Haryana (Fig. 3 .). The average accuracy for SVC was 82.06%, followed closely by XGBoost with an average accuracy of 81.77%. RFC and KNN achieved slightly lower average accuracies of 80.65% and 81.03%, respectively. Interestingly, SVC exhibited superior performance, particularly at higher split ratios, showcasing its robustness in handling varying data distributions. This highlights its suitability for capturing precipitation patterns influenced by diverse factors such as the proximity to the Himalayas and the Indus River basin. The strong performance of SVC in Punjab and Haryana suggests its effectiveness in modeling precipitation patterns in the region's predominantly flat plains. 3.2.4 Rajasthan In contrast to other states, RFC emerged as the top performer for precipitation prediction in Rajasthan. The consistently high accuracy scores achieved by RFC, ranging from 83.55–83.71%, indicate its effectiveness in capturing precipitation patterns in this arid state. SVC displayed competitive performance with accuracy scores ranging from 82.34–82.79%. XGBoost also showcased consistent performance with accuracy scores between 82.78% and 83.22%. While KNN achieved lower scores (82.00% − 82.06%), it still offers reasonable predictive capabilities. 3.2.5 Uttarakhand Similar to Himachal Pradesh, SVC emerged as the leader for precipitation prediction in Uttarakhand. Its accuracy scores ranged from a high of 84.47% to a low of 84.16% across different split ratios. This consistency suggests SVC's effectiveness in capturing the intricate relationships between atmospheric variables and precipitation patterns in this state characterized by the Himalayan foothills and diverse microclimates. RFC followed closely with accuracy scores ranging from 83.84–83.96%, demonstrating strong performance. XGBoost displayed competitive accuracy (83.33% − 83.76%), while KNN achieved respectabe scores (82.93% − 83.25%). 3.2.6 Uttar Pradesh In Uttar Pradesh, the analysis revealed RFC as the dominant algorithm for precipitation prediction. It achieved the highest accuracy scores across all splits, ranging from 84.70–85.05%. XGBoost followed closely with accuracy scores ranging from 83.93–84.50%. SVC maintained competitive scores (83.78% − 84.16%). KNN achieved respectable accuracy (83.57% − 83.73%). The dominance of RFC in Uttar Pradesh, the most populous state in India, can be attributed to its ability to effectively model precipitation patterns influenced by diverse factors such as the Gangetic Plain's topography and proximity to the Himalayas. The strong performance of both RFC and XGBoost suggests promising avenues for further exploration and potential implementation in operational forecasting systems. Plain's topography and proximity to the Himalayas. The strong performance of both RFC and XGBoost suggests promising avenues for further exploration and potential implementation in operational forecasting systems. Overall, SVM and RF emerged as strong contenders in our analysis of mapping precipitation patterns across various states in North India (Fig. 4 ). Notably, SVM consistently outperformed RF in the majority of states, showcasing its effectiveness in accurately predicting precipitation patterns. However, interestingly, RF exhibited superior performance in Rajasthan and Uttarakhand, where it achieved the highest accuracy scores among all algorithms tested. This observation underscores the nuanced nature of regional climatic patterns and highlights RF's capability to excel in certain geographical contexts. Upon aggregating the results from multiple iterations and train-test split ratios, we constructed line plot (Fig. 4 .) for each state individually, providing a comprehensive overview of algorithm performance. These plots affirmed the dominance of SVM across most states, but both SVM and RF demonstrated comparable performance, indicating a close competition between the two algorithms. Additionally, the study revealed an inverse correlation between elevation and the performance of machine learning algorithms in simulating precipitation and its extremes over the capital cities in North Indian states. Specifically, in regions with lower elevations, the models exhibited higher skill (accuracy) in simulating precipitation, while in areas with higher elevations, the model skill was relatively lower, with a skill difference of approximately 5%. This observation highlights the influence of topographic features and associated meteorological complexities on the models' ability to accurately capture precipitation patterns. 3.3 Evaluation of Model Performance for Extreme Precipitation Building on our previous model evaluation, which demonstrated competitive performance between RFC and SVC models for general precipitation patterns across most states, we sought to specifically evaluate their effectiveness in predicting extreme precipitation events. For this we implemented a two-pronged evaluation for this purpose. First, both models were evaluated using all initial variables (Fig. 5 .). and then only for statistically significant variables (Fig. 6 .). across different states. Receiver Operating Characteristic (ROC) curves were generated for each scenario to visualize model discrimination between extreme events below 10th and above 95th percentiles (Figs. 5 and 6 ). While AUC of an ROC curve provides a good overview of model performance, it's valuable to consider additional metrics for a more comprehensive evaluation and robustness check of the build models. Here, we considered Brier Score in order to assesses the overall calibration of the model's predicted probabilities with actual outcomes (Fig. 7 .) By comparing ROC values and Brier Scores across models and variable selection approaches, we aimed to identify the most effective model for extreme precipitation prediction. Overall Performance: Both models (RFC and SVC) exhibit good skill in predicting extreme precipitation events, with AUC values generally exceeding 0.85 across states and scenarios (Figs. 5 & 6 .). This indicates a strong ability to discriminate between extreme and non-extreme events. Model Comparison: While there are some variations, the performance of RFC and SVM models is often comparable, with neither model consistently outperforming the other across all states and scenarios. Impact of Variable Selection: Focusing on statistically significant variables sometimes results in slightly improved AUC values for the SVM model (e.g., state JK at the 95th percentile threshold) (Fig. 6 ). However, the differences are generally small, suggesting that the additional variables included in the "all variables" scenario might not significantly impact model performance for extreme event prediction. State-Specific Variations: There is some variation in model performance across states. States like RJ and UP show consistently high AUC values (> 0.90) for both models in both scenarios, indicating exceptional skill in predicting extreme events. Conversely, states like HP and PH show slightly lower but still good performance (AUC around 0.87–0.90). This might be due to regional differences in extreme precipitation patterns or require further investigation into specific factors influencing those states. Threshold Dependence: As expected, AUC values are generally higher for the 95th percentile threshold compared to the 10th percentile threshold. This is because the 95th percentile represents a more extreme precipitation event, which might be easier for the models to predict accurately. Overall Brier Scores: The Brier Scores range from approximately 0.07 to 0.12 across states and models. These are generally considered good scores, indicating the models are providing reasonably accurate probability estimates for extreme precipitation events. Model Comparison: Similar to the AUC results, there are no significant differences between RF and SVM models in terms of Brier Scores for most states. Both models achieve comparable performance Impact of Variable Selection: Focusing on statistically significant variables (often including humidity and surface pressure) sometimes leads to slightly higher Brier Scores (worse performance) compared to using all variables. This suggests that while some variables might not be statistically significant, they could still contribute to the model's ability to calibrate its predictions. State-Specific Variations: Brier Scores also show some variation across states. States like RJ and UP consistently have lower Brier Scores (better performance) compared to states like HP and JK. This aligns with the observations from AUC values, suggesting these states might have more predictable extreme precipitation patterns or benefit from the additional information captured by all initial variables. 4. Discussion Our findings align with the growing popularity of machine learning approaches in India. The study over Western Ghats region in India by 69 showcases the effectiveness of XGBoost and RF models in multi-model ensembles for capturing the behaviour of climate change on precipitation patterns. This study emphasizes the importance of comprehensive testing and validation, especially for regional investigations with diverse precipitation mechanisms 69 . Machine learning offers a distinct advantage in simulating extreme precipitation events, which are notoriously difficult to predict using traditional methods. The success of the improved K-Nearest Neighbor model in simulating such events in New Delhi 70 highlights its potential for vulnerability assessments in flood-prone regions. The study suggested expanding on this concept, future research could explore the application of other ML algorithms, like LSTMs, which excel at capturing temporal dependencies, to improve extreme event prediction. ML methods have been successfully employed for various precipitation-related applications across India. Studies in Nellore Station, Andhra Pradesh 71 demonstrated the feasibility of ML for short-term forecasting. 72 showcased the potential of Extreme Learning Machines for probabilistic predictions of monsoon precipitation, crucial for agricultural planning. Additionally, the study by 73 underlines the effectiveness of RF and LSTM in creating multi-model ensembles for precipitation prediction in river basins. This highlights the versatility of ML in tackling diverse precipitation-related challenges. Unsupervised learning approaches, like the one employed by 74 to analyze changes in extreme precipitation due to storm dynamics, offer valuable insights into the complex interplay of factors influencing these events. The study suggested further research could explore the integration of unsupervised and supervised learning approaches to gain a more holistic understanding of precipitation dynamics. The success of ML in North India aligns relatively well with global trends. Studies across the United States by 39 employed RF, XGBoost, and ANNs to explore the key variables influencing extreme precipitation. These models demonstrated improved performance in regions with distinct seasonal variations, suggesting their potential applicability in other parts of the world with similar climatic patterns. Similar to our findings with RF, international research highlights its effectiveness in extreme event forecasting. 75 in China employed RF models to forecast days with a high probability of extreme precipitation events. Additionally, studies by 76 and 77 showcase the success of RF models in predicting extreme precipitation events in other regions. This global consistency suggests inherent advantages of RF for this specific task. In regions with lower elevations, such as Uttar Pradesh with an elevation of 123mt., the models exhibited higher accuracy in precipitation simulations (84%). Conversely, areas with higher elevations like Jammu and Kashmir, with an average of 1585mt., posed greater challenges, and the models displayed relatively lower skill (79%). The approximate 5% difference in skill between low and high elevation regions highlights the influence of topographic features and associated meteorological complexities. The intricate terrain and orographic effects in mountainous regions like JK can contribute to more complex precipitation patterns, localized atmospheric circulations, and rapid weather changes, which machine learning algorithms may struggle to capture accurately. Additionally, the availability and quality of observational data used for training the models could be limited in remote, high-elevation areas, further impacting their performance. In contrast, the relatively simpler topography and more predictable meteorological conditions in lower elevation regions, coupled with potentially better data availability, may have facilitated improved model skill in simulating precipitation patterns accurately. Our study identified RF as the most effective model for simulating precipitation extremes in North India. This finding aligns with international research, where RF consistently demonstrates high accuracy in extreme event prediction. While SVM has proven successful in various applications, studies specifically focused on extreme precipitation simulation do not explicitly mention its effectiveness. This suggests that RF might be a superior choice for this particular task due to model’s ability in handling high dimensionality, non-linearity at a high pace 39 , 75 , 78 , 79 . 5. Conclusion In this work, a comprehensive study was conducted to investigate the intricate interplay between atmospheric variables and extreme precipitation events across the seven states in North India. The study leverages MERRA-2 reanalysis data spanning from 1984 to 2022 and employed four machine learning models (Support Vector Classifier (SVC), Random Forest Classifier (RFC), XGBoost, and K Nearest Neighbors (KNN)). The analysis revealed insightful correlations between key predictor variables and extreme precipitation intensity for each region. Dew Point Temperature and Relative Humidity exhibited strong positive correlations (~ 0.4) with precipitation across all states, while Temperature exhibited regional variations with positive correlations in Himachal Pradesh, Punjab and Haryana, and Uttarakhand (~ 0.2), and weaker negative associations in Jammu and Kashmir, Rajasthan, and Uttar Pradesh (-0.1 to -0.2). Solar irradiance and Surface Pressure often acted as counterpoints to precipitation, with negative correlations ranging from − 0.18 to -0.3. The significance of these variables was taken into account while performing predictive modelling in case of extreme precipitation pattern detection. The predictive modeling aspect, employed machine learning algorithms across all states which depicted competitive performance among algorithms. SVC and RFC emerged as powerful tools for precipitation prediction, with SVC dominating in Himachal Pradesh, Jammu and Kashmir, Uttarakhand, and Punjab and Haryana, while RFC excelled in Rajasthan and Uttar Pradesh. The models exhibited higher skill in simulating precipitation over lower elevation regions compared to higher elevation areas, with a skill difference of around 5%, potentially due to the influence of topographic complexity on meteorological phenomena. Furthermore, the analysis of extreme precipitation events revealed that Random Forest models consistently outperformed Support Vector Machines, achieving higher Area Under the Curve (AUC) values (~ 0.90), and lower Brier Scores (~ 0.01), across all states and precipitation thresholds. Despite the promising results, the study acknowledges limitations such as reliance on reanalysis data, limited atmospheric variables, and coarse spatial and temporal resolutions. The study's success with Random Forest models for simulating precipitation extremes paves the way for further advancements. Future research should explore more advance algorithms like LSTMs and ensemble learning approaches, as well data augmentation strategies. Integrating high-resolution data, climate change projections, and atmospheric processes into the models can enhance their accuracy and robustness. Ultimately, these advancements should be translated into practical tools for real-time flood and drought forecasting, optimizing agricultural practices, and informing adaptation strategies. By considering both atmospheric processes and anthropogenic influences, we can develop comprehensive and holistic models for understanding and predicting precipitation extremes, leading to a more sustainable future in the face of climate change. 6. Data For the study, data was sourced from NASA’s Prediction of Worldwide Energy Resources (POWER) project spanning from 1984 to 2022 51 . The data collection process was centred around obtaining meteorological observations from the capital cities of each states in North India. These capital cities were strategically chosen as representative locations to assess and characterize the atmospheric conditions prevalent across their respective states. This approach aimed to capture the regional variations in climate and weather patterns that influence precipitation dynamics in the region. The study encompassed a comprehensive set of seven atmospheric variables, which were meticulously recorded and analyzed. These variables included maximum temperature, relative humidity, surface pressure, wind speed, dew point temperature, precipitation from Modern-Era Retrospective Analysis for Research and Applications, version 2 (MERRA-2), and solar irradiance from Clouds and the Earth’s Radiant Energy System Project (CERES). The data was a combination of MERRA-2 52–56 and CERES SYN 57 – 60 and has been widely used by researchers for various climate and atmospheric studies. The comprehensive coverage of atmospheric and climate variables, alongside its high spatial and temporal resolution, is underscored by its assimilation of a wide range of satellite observations and other data sources, making it a valuable asset for climate and atmospheric research. MERRA-2 served a grid resolution of 0.5*0.625 degrees and demonstrates advancements in addressing known deficiencies, such as reducing spurious trends and jumps associated with changes in the observing system, as well as mitigating biases and imbalances in aspects of the water cycle 51 . CERES-SYN served a grid resolution of 1*1 degrees. This product provides accurate satellite-retrieved estimates of Earth's radiation budget components, including downwelling longwave radiation at the surface and other atmospheric levels from 2000 to the present. It utilizes radiation transfer model along with improved cloud property retrievals and consistent temperature/humidity data to estimate radiative fluxes more accurately. As a representative global satellite product covering various timescales, CERES-SYN serves as a valuable data source for evaluating reanalysis radiation estimates over regions lacking sufficient ground observations. 7. Methodology The dataset utilized in this study, has been extensively employed in various research investigations. Notably, it has been employed in studies for evaluation of hydrological performance of precipitation products different locations and over Basins 61 , 62 , for analysis of the diurnal cycle of summer precipitation and associated land-atmosphere interactions 63 , drought estimations 64 and other hydrometeorological application 65 – 67 . These studies demonstrate the widespread application and reliability of the MERRA-2 dataset across different geographical regions and climatic conditions. Our primary objective was to find the intricate relationships between various atmospheric variables. For this generation and analysis of correlation plots was performed. To ensure the comparability of our findings, we did feature engineering over the dataset. Leveraging the power of supervised machine learning, we applied a diverse set of algorithms including Random Forest Classifier (RFC), Support Vector Classifier (SVC), XGBoost Classifier (XGB), and K-Nearest Neighbors (KNN) to the pre-processed data. Our aim was to select the most effective model for accurately mapping precipitation patterns across the expansive Indian region. For this, we adhered to a conventional 80:20 split for training and testing the algorithm. However, to comprehensively assess the performance of each model, we systematically experimented with different split ratios including 70 − 30, 60 − 40, and 50–50. Each model underwent a rigorous iterative process, ranging from 25 to 30 iterations, to fine-tune its efficiency and effectiveness. For evaluating the performance of our models, we employed a range of metrics including accuracy, precision, recall, and F1-score. However, to know the skill of the model. we selected ‘accuracy’ for forming comparison. Furthermore, in our quest to gain deeper insights into the efficacy of each algorithm, especially in accurately representing extreme precipitation patterns over the North Indian States, we visualized model performance for different thresholds level (10th and 95th percentiles) using receiver operating characteristic (ROC) curves. These curves not only allowed us to assess sensitivity and specificity but also provided a comprehensive understanding of the Area Under the Curve (AUC), which serves as a valuable metric for quantifying the overall performance of each algorithm in distinguishing between positive and negative precipitation events. The AUC offers a nuanced perspective on the discriminatory power of the models, aiding in the identification of the most adept algorithm for capturing extreme precipitation occurrences in the North Indian region. In addition to AUC, we also analysed Brier’s Score. Brier Score measures the mean squared difference between the predicted probability of an extreme event and the actual outcome (0 or 1). A lower Brier Scores indicate better model performance, signifying predictions closer to actual occurrences. Declarations Acknowledgement All the authors acknowledge NASA’s POWER Data Access Viewer for providing the climate data. KCP acknowledge the support from the NERC (UK Natural Environment Research Council) AMAZONICA and Amazon Hydrological Cycle grants (NE/F005806/1 and NE/K01353X/1). AT and AA thankful to the University of Petroleum and Energy Studies, Dehradun for providing research facilities. Author contribution AT and KCP conceived the study. AT performed the analyses and wrote the initial draft of the manuscript. All authors contributed to the interpretation of the results, discussion of the associated mechanisms, and refinement of the paper. Conflict of Interest The authors declare no competing interest. Data Availability Statement All the data utilized in this study are openly accessible and can be obtained by contacting the first author, Aayushi Tandon, via email at [email protected] , or directly from the source: NASA’s POWER Data Access Viewer at https://power.larc.nasa.gov. References Bhattacharyya, S., Sreekesh, S. & King, A. Characteristics of extreme rainfall in different gridded datasets over India during 1983–2015. Atmospheric Research 267 , 105930 (2022). Fowler, H. J. et al. Anthropogenic intensification of short-duration rainfall extremes. Nat Rev Earth Environ 2 , 107–122 (2021). Mishra, V., Aadhar, S. & Mahto, S. S. Anthropogenic warming and intraseasonal summer monsoon variability amplify the risk of future flash droughts in India. npj Clim Atmos Sci 4 , 1–10 (2021). Suman, M. & Maity, R. Southward shift of precipitation extremes over south Asia: Evidences from CORDEX data. Sci Rep 10 , 6452 (2020). Mukherjee, S., Aadhar, S., Stone, D. & Mishra, V. Increase in extreme precipitation events under anthropogenic warming in India. Weather and Climate Extremes 20 , 45–53 (2018). Annex IX: Contributors to the IPCC Working Group I Sixth Assessment Report. in Climate Change 2021: The Physical Science Basis. Contribution of Working Group I to the Sixth Assessment Report of the Intergovernmental Panel on Climate Change (eds. Masson-Delmotte, V. et al.) 2267–2286 (Cambridge University Press, Cambridge, United Kingdom and New York, NY, USA, 2021). doi:10.1017/9781009157896.001. Pörtner, H. O. et al. IPCC, 2022: summary for policymakers. in (eds. Pörtner, H. O. et al.) 3–33 (Cambridge University Pres, Cambridge, UK and New York, NY, US, 2022). Sikka, D. R., Ray, K., Chakravarthy, K., Bhan, S. C. & Tyagi, A. Heavy rainfall in the Kedarnath valley of Uttarakhand during the advancing monsoon phase in June 2013. Current Science 109 , 353–361 (2015). Anoop Kumar Mishra & J Srinivasan. Did a cloud burst occur in Kedarnath during 16 and 17 June 2013. (2013). Rawat, K. S., Sahu, S. R., Singh, S. K. & Mishra, A. K. Cloudburst analysis in the Nainital district, Himalayan Region, 2021. Discov Water 2 , 12 (2022). Singh, B., Thapliyal, R. & IMD, M. Cloudburst events observed over Uttarakhand during monsoon season 2017 and their analysis. MAUSAM 73 , 91–104 (2022). Chevuturi, A. & Dimri, A. P. Investigation of Uttarakhand (India) disaster-2013 using weather research and forecasting model. Nat Hazards 82 , 1703–1726 (2016). DNA. Cloudburst in Himachal Pradesh’s Mandi district kills 5. https://www.dnaindia.com/india/report-cloudburst-in-himachal-pradesh-s-mandi-district-kills-5-2112258 (2015). TOI. Himachal Pradesh cloudburst: 178 stranded people rescued from Lahaul-Spiti - India Today. https://www.indiatoday.in/india/story/himachal-pradesh-cloudburst-178-stranded-people-rescued-from-lahaul-spiti-1835222-2021-07-31 (2021). Bhan, S., Devrani, A. & Sinha, V. An analysis of monthly rainfall and the meteorological conditions associated with cloudburst over the dry region of Leh (Ladakh), India. Mausam 66578540 , 107–122 (2015). Dimri, A. P. et al. Cloudbursts in Indian Himalayas: A review. Earth-Science Reviews 168 , 1–23 (2017). Yadav, M., Dimri, A. P., Mal, S. & Maharana, P. Elevation-dependent precipitation in the Indian Himalayan Region. Theor Appl Climatol 155 , 815–828 (2024). Adhikari, P. & Mejia, J. F. Influence of aerosols on clouds, precipitation and freezing level height over the foothills of the Himalayas during the Indian summer monsoon. Clim Dyn 57 , 395–413 (2021). Mal, S. et al. Determining the quasi monsoon front in the Indian Himalayas. Quaternary International 599–600 , 4–14 (2021). Rangwala, I., Palazzi, E. & Miller, J. R. Projected Climate Change in the Himalayas during the Twenty-First Century. in Himalayan Weather and Climate and their Impact on the Environment (eds. Dimri, A. P., Bookhagen, B., Stoffel, M. & Yasunari, T.) 51–71 (Springer International Publishing, Cham, 2020). doi:10.1007/978-3-030-29684-1_4. Acosta, R. P. & Huber, M. Competing Topographic Mechanisms for the Summer Indo-Asian Monsoon. Geophysical Research Letters 47 , e2019GL085112 (2020). J. A. Para et al. Large-scale dynamics of western disturbances caused extreme precipitation on 24–27 January 2017 over Jammu and Kashmir, India. Modeling Earth Systems and Environment 6 , 99–107 (2020). Aggarwal, D., Chakraborty, R. & Attada, R. Investigating bi-decadal precipitation changes over the Northwest Himalayas during the pre-monsoon: role of Pacific decadal oscillations. Clim Dyn 62 , 1203–1218 (2024). Hunt, K. M. R., Turner, A. G. & Schiemann, R. K. H. How Interactions between Tropical Depressions and Western Disturbances Affect Heavy Precipitation in South Asia. Monthly Weather Review 149 , 1801–1825 (2021). Kumar, P. V., Naidu, C. V. & Prasanna, K. Recent unprecedented weakening of Indian summer monsoon in warming environment. Theor Appl Climatol 140 , 467–486 (2020). Singh, R., Jaiswal, N. & Kishtawal, C. M. Rising surface pressure over Tibetan Plateau strengthens indian summer monsoon rainfall over northwestern India. Sci Rep 12 , 8621 (2022). Gupta, A. K., Prakasam, M., Dutt, S., Clift, P. D. & Yadav, R. R. Evolution and Development of the Indian Monsoon. in Geodynamics of the Indian Plate: Evolutionary Perspectives (eds. Gupta, N. & Tandon, S. K.) 499–535 (Springer International Publishing, Cham, 2020). doi:10.1007/978-3-030-15989-4_14. Hahn, D. G. & Manabe, S. The Role of Mountains in the South Asian Monsoon Circulation. Journal of the Atmospheric Sciences 32 , 1515–1541 (1975). Anu Xavier, M. G. Manoj, & K. Mohankumar. On the dynamics of an extreme rainfall event in northern India in 2013. Journal of Earth System Science 127 , 30 (2018). Pattnayak, K. C., Awasthi, A., Sharma, K. & Pattnayak, B. B. Fate of Rainfall Over the North Indian States in the 1.5 and 2°C Warming Scenarios. Earth and Space Science 10 , e2022EA002671 (2023). Alizadeh, O. Advances and challenges in climate modeling. Climatic Change 170 , 18 (2022). Morrison, H. et al. Confronting the Challenge of Modeling Cloud and Precipitation Microphysics. Journal of Advances in Modeling Earth Systems 12 , e2019MS001689 (2020). Wang, H. et al. Predicting climate anomalies: A real challenge. Atmospheric and Oceanic Science Letters 15 , 100115 (2022). Yakubu, A. T., Abayomi, A. & Chetty, N. Machine Learning-Based Precipitation Prediction Using Cloud Properties. in Hybrid Intelligent Systems (eds. Abraham, A. et al.) 243–252 (Springer International Publishing, Cham, 2022). doi:10.1007/978-3-030-96305-7_23. Gibson, P. B. et al. Training machine learning models on climate model output yields skillful interpretable seasonal precipitation forecasts. Commun Earth Environ 2 , 1–13 (2021). Mansfield, L. A. et al. Predicting global patterns of long-term climate change from short-term simulations using machine learning. npj Clim Atmos Sci 3 , 1–9 (2020). Tandon, A., Awasthi, A. & Pattnayak, K. C. Comparison of different Machine Learning methods on Precipitation dataset for Uttarakhand. in 2023 2nd International Conference on Ambient Intelligence in Health Care (ICAIHC) 1–6 (2023). doi:10.1109/ICAIHC59020.2023.10431402. Wang, B., Wei, W., Yin, Z. & Xu, L. Using machine learning to analyze the changes in extreme precipitation in southern China. Atmospheric Research 302 , 107307 (2024). Lin, X., Fan, J., Hou, Z. J. & Wang, J. Machine Learning of Key Variables Impacting Extreme Precipitation in Various Regions of the Contiguous United States. Journal of Advances in Modeling Earth Systems 15 , e2022MS003334 (2023). Vitanza, E., Dimitri, G. M. & Mocenni, C. A multi-modal machine learning approach to detect extreme rainfall events in Sicily. Sci Rep 13 , 1–15 (2023). Hu, H. & Ayyub, B. M. Machine Learning for Projecting Extreme Precipitation Intensity for Short Durations in a Changing Climate. Geosciences 9 , 209 (2019). Chauhan, A. S., Singh, S., Maurya, R. K. S., Rani, A. & Danodia, A. Spatio-temporal trend analysis and future projections of precipitation at regional scale: a case study of Haryana, India. Journal of Water and Climate Change 13 , 2143–2170 (2022). Malik, A. & Kumar, A. Spatio-temporal trend analysis of rainfall using parametric and non-parametric tests: case study in Uttarakhand, India. Theor Appl Climatol 140 , 183–207 (2020). Jaswal, A., Bhan, S., Karandikar, A. & Gujar, M. Seasonal and annual rainfall trends in Himachal Pradesh during 1951-2005. Mausam 66 , 247–264 (2015). Agarwal, S., Mukherjee, D. & Debbarma, N. Analysis of extreme annual rainfall in North-Eastern India using machine learning techniques. AQUA - Water Infrastructure, Ecosystems and Society 72 , 2201–2215 (2023). Praveen, B. et al. Analyzing trend and forecasting of rainfall changes in India using non-parametrical and machine learning approaches. Sci Rep 10 , 10342 (2020). Ray, R. & Chakraborty, S. A Discussion on Machine Learning Approach of Rainfall Prediction. in 2022 URSI Regional Conference on Radio Science (USRI-RCRS) 1–4 (2022). doi:10.23919/URSI-RCRS56822.2022.10118437. Lighthill, J. & Pearce, R. P. Monsoon Dynamics . (Cambridge University Press, 1981). Awasthi, A., Pattnayak, K. C., Tandon, A., Sarkar, A. & Chakraborty, M. Implications of climate change on surface temperature in North Indian states: evidence from CMIP6 model ensembles. Front. Environ. Sci. 11 , (2023). Climate-Know India, I. Profile - Climate - Know India: National Portal of India. https://knowindia.india.gov.in/profile/climate.php (2024). Gelaro, R. et al. The Modern-Era Retrospective Analysis for Research and Applications, Version 2 (MERRA-2). Journal of Climate 30 , 5419–5454 (2017). Agarwal, P., Stevenson, D. S. & Heal, M. R. Evaluation of WRF-Chem-simulated meteorology and aerosols over northern India during the severe pollution episode of 2016. Atmospheric Chemistry and Physics 24 , 2239–2266 (2024). Dai, Z. et al. Long-Term MERRA-2 Reanalysis Data Indicate Atmospheric Environmental Changes for Three Major Concentrating-Solar-Power-Plant Project Areas in Xinjiang, China. Atmosphere 14 , 1700 (2023). Wen, A. et al. Evaluation of MERRA-2 land surface temperature dataset and its application in permafrost mapping over China. Atmospheric Research 279 , 106373 (2022). Rajeev, A., Singh, C., Singh, S. K. & Chauhan, P. Assessment of WRF-CHEM Simulated Dust Using Reanalysis, Satellite Data and Ground-Based Observations. J Indian Soc Remote Sens 49 , 1545–1559 (2021). Gupta, P. et al. Validation of Surface Temperature Derived From MERRA-2 Reanalysis Against IMD Gridded Data Set Over India. Earth and Space Science 7 , e2019EA000910 (2020). Feng, C. et al. Comprehensive assessment of global atmospheric downward longwave radiation in the state-of-the-art reanalysis using satellite and flux tower observations. Clim Dyn 60 , 1495–1521 (2023). Schmeisser, L., Hinkelman, L. M. & Ackerman, T. P. Evaluation of Radiation and Clouds From Five Reanalysis Products in the Northeast Pacific Ocean. Journal of Geophysical Research: Atmospheres 123 , 7238–7253 (2018). Doelling, D. R. et al. Advances in Geostationary-Derived Longwave Fluxes for the CERES Synoptic (SYN1deg) Product. Journal of Atmospheric and Oceanic Technology 33 , 503–521 (2016). Kato, S. et al. Surface Irradiances Consistent with CERES-Derived Top-of-Atmosphere Shortwave and Longwave Irradiances. Journal of Climate 26 , 2719–2740 (2013). Hamal, K. et al. Evaluation of MERRA-2 Precipitation Products Using Gauge Observation in Nepal. Hydrology 7 , 40 (2020). Xu, X., Frey, S. K. & Ma, D. Hydrological performance of ERA5 and MERRA-2 precipitation products over the Great Lakes Basin. Journal of Hydrology: Regional Studies 39 , 100982 (2022). Song, Y. & Wei, J. Diurnal cycle of summer precipitation over the North China Plain and associated land–atmosphere interactions: Evaluation of ERA5 and MERRA-2. International Journal of Climatology 41 , 6031–6046 (2021). Le, M.-H. et al. Assessment of drought conditions over Vietnam using standardized precipitation evapotranspiration index, MERRA-2 re-analysis, and dynamic land cover. Journal of Hydrology: Regional Studies 32 , 100767 (2020). Igbawua, T., Zhang, J., Yao, F. & Zhang, D. Assessment of moisture budget over West Africa using MERRA-2’s aerological model and satellite data. Clim Dyn 52 , 83–106 (2019). Cuo, L. & Zhang, Y. Spatial patterns of wet season precipitation vertical gradients on the Tibetan Plateau and the surroundings. Sci Rep 7 , 5057 (2017). Wei, J., Dirmeyer, P. A., Wisser, D., Bosilovich, M. G. & Mocko, D. M. Where Does the Irrigation Water Go? An Estimate of the Contribution of Irrigation to Precipitation Using MERRA. Journal of Hydrometeorology 14 , 275–289 (2013). Senthilnathan, S. Usefulness of Correlation Analysis. SSRN Electronic Journal (2019) doi:10.2139/ssrn.3416918. Shetty, S., Umesh, P. & Shetty, A. The effectiveness of machine learning-based multi-model ensemble predictions of CMIP6 in Western Ghats of India. International Journal of Climatology 43 , 5029–5054 (2023). Sharif, M. Simulation of Extreme Precipitation Events Using an Improved K-Nearest Neighbor Model . 389 (2023). doi:10.1061/9780784484852.036. K, K., N C, V., P, P. & Selvapandian, D. An Efficient Rainfall Forecasting System using Machine Learning Methods. in 2023 Second International Conference on Electrical, Electronics, Information and Communication Technologies (ICEEICT) 01–06 (2023). doi:10.1109/ICEEICT56924.2023.10157395. Acharya, N. & Hall, K. A Machine Learning Approach for Probabilistic Multi-Model Ensemble Predictions of Indian Summer Monsoon Rainfall. MAUSAM 74 , 421–428 (2023). Jose, D. M., Vincent, A. M. & Dwarakish, G. S. Improving multiple model ensemble predictions of daily precipitation and temperature through machine learning techniques. Sci Rep 12 , 4678 (2022). Mooers, G., Beucler, T., Pritchard, M. & Mandt, S. Understanding Extreme Precipitation Changes through Unsupervised Machine Learning. Preprint at https://doi.org/10.48550/arXiv.2211.01613 (2023). Wei, W. et al. Seasonal prediction of summer extreme precipitation over the Yangtze River based on random forest. Weather and Climate Extremes 37 , 100477 (2022). Hill, A. J. & Schumacher, R. S. Forecasting Excessive Rainfall with Random Forests and a Deterministic Convection-Allowing Model. Weather and Forecasting 36 , 1693–1711 (2021). Pham, Q. B., Yang, T.-C., Kuo, C.-M., Tseng, H.-W. & Yu, P.-S. Combing Random Forest and Least Square Support Vector Regression for Improving Extreme Rainfall Downscaling. Water 11 , 451 (2019). Grazzini, F., Grams, C. & Craig, G. Classification and prediction of days with extreme rainfall using random forest approach. EGU22-11954 (2022) doi:10.5194/egusphere-egu22-11954. Li, K., Huang, G., Wang, S., Razavi, S. & Zhang, X. Development of a Joint Probabilistic Rainfall-Runoff Model for High-to-Extreme Flow Projections Under Changing Climatic Conditions. Water Resources Research 58 , e2021WR031557 (2022). Additional Declarations No competing interests reported. Cite Share Download PDF Status: Published Journal Publication published 25 Mar, 2025 Read the published version in Scientific Reports → Version 1 posted Editorial decision: Revision requested 12 Jun, 2024 Reviews received at journal 07 Jun, 2024 Reviews received at journal 26 May, 2024 Reviewers agreed at journal 13 May, 2024 Reviewers agreed at journal 09 May, 2024 Reviewers invited by journal 07 May, 2024 Editor assigned by journal 07 May, 2024 Editor invited by journal 02 May, 2024 Submission checks completed at journal 02 May, 2024 First submitted to journal 28 Apr, 2024 You are reading this latest preprint version Research Square lets you share your work early, gain feedback from the community, and start making changes to your manuscript prior to peer review in a journal. As a division of Research Square Company, we’re committed to making research communication faster, fairer, and more useful. We do this by developing innovative software and high quality services for the global research community. Our growing team is made up of researchers and industry professionals working together to solve the most critical problems facing scientific publishing. Also discoverable on Platform About Our Team In Review Editorial Policies Advisory Board Help Center Resources Author Services Accessibility API Access RSS feed Manage Cookie Preferences © Research Square 2026 | ISSN 2693-5015 (online) Privacy Policy Terms of Service Do Not Sell My Personal Information {"props":{"pageProps":{"initialData":{"identity":"rs-4339400","acceptedTermsAndConditions":true,"allowDirectSubmit":false,"archivedVersions":[],"articleType":"Article","associatedPublications":[],"authors":[{"id":299873521,"identity":"32189a05-0f55-45ee-a6a7-41a786c966b1","order_by":0,"name":"Aayushi Tandon","email":"","orcid":"","institution":"University of Petroleum \u0026 Energy Studies","correspondingAuthor":false,"prefix":"","firstName":"Aayushi","middleName":"","lastName":"Tandon","suffix":""},{"id":299873523,"identity":"6fa3fd0a-e8c8-4975-9e57-427135726406","order_by":1,"name":"Amit Awasthi","email":"","orcid":"","institution":"University of Petroleum \u0026 Energy Studies","correspondingAuthor":false,"prefix":"","firstName":"Amit","middleName":"","lastName":"Awasthi","suffix":""},{"id":299873525,"identity":"cc195f94-a185-4685-a662-2c09ff062a3d","order_by":2,"name":"Kanhu Charan Pattnayak","email":"data:image/png;base64,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","orcid":"","institution":"University of Leeds","correspondingAuthor":true,"prefix":"","firstName":"Kanhu","middleName":"Charan","lastName":"Pattnayak","suffix":""}],"badges":[],"createdAt":"2024-04-28 22:09:14","currentVersionCode":1,"declarations":"","doi":"10.21203/rs.3.rs-4339400/v1","doiUrl":"https://doi.org/10.21203/rs.3.rs-4339400/v1","draftVersion":[],"editorialEvents":[{"content":"https://doi.org/10.1038/s41598-024-84360-w","type":"published","date":"2025-03-25T15:56:51+00:00"}],"editorialNote":"","failedWorkflow":false,"files":[{"id":56090899,"identity":"d6ee99cd-e6fb-45e1-92d2-6ca4d53d48df","added_by":"auto","created_at":"2024-05-08 12:19:31","extension":"png","order_by":1,"title":"Figure 1","display":"","copyAsset":false,"role":"figure","size":188200,"visible":true,"origin":"","legend":"\u003cp\u003eStudy area map highlighting the geographical region under investigation\u003c/p\u003e","description":"","filename":"Fig.1.png","url":"https://assets-eu.researchsquare.com/files/rs-4339400/v1/10fe9c5d809e869faad48d3a.png"},{"id":56091424,"identity":"50dda55a-f295-4996-b0d5-6b01d59aefea","added_by":"auto","created_at":"2024-05-08 12:27:30","extension":"png","order_by":2,"title":"Figure 2","display":"","copyAsset":false,"role":"figure","size":150164,"visible":true,"origin":"","legend":"\u003cp\u003eMatrix\u003cstrong\u003e illustrating relationships between precipitation and other atmospheric variable for each of the states\u003c/strong\u003e\u003c/p\u003e","description":"","filename":"Fig.2.png","url":"https://assets-eu.researchsquare.com/files/rs-4339400/v1/05b9c1db6d5f8d594cd8f596.png"},{"id":56090893,"identity":"b21344dc-10c1-431d-850d-7dd450228ceb","added_by":"auto","created_at":"2024-05-08 12:19:30","extension":"png","order_by":3,"title":"Figure 3","display":"","copyAsset":false,"role":"figure","size":2545481,"visible":true,"origin":"","legend":"\u003cp\u003eMachine Learning Accuracy Across Various Train-Test Splits Over North Indian States\u003c/p\u003e","description":"","filename":"Fig.3.png","url":"https://assets-eu.researchsquare.com/files/rs-4339400/v1/f48203f66178de719be19428.png"},{"id":56091797,"identity":"79961f70-1d7d-44d4-b83a-69e123dae8c6","added_by":"auto","created_at":"2024-05-08 12:35:30","extension":"png","order_by":4,"title":"Figure 4","display":"","copyAsset":false,"role":"figure","size":211865,"visible":true,"origin":"","legend":"\u003cp\u003eOverall Comparison of machine learning performance across the states in North India.\u003c/p\u003e","description":"","filename":"Fig.4.png","url":"https://assets-eu.researchsquare.com/files/rs-4339400/v1/29d9e83e3ef1c2ef17737eb3.png"},{"id":56090896,"identity":"f18f03b2-2320-45a8-9ae0-24dba99523fe","added_by":"auto","created_at":"2024-05-08 12:19:30","extension":"png","order_by":5,"title":"Figure 5","display":"","copyAsset":false,"role":"figure","size":1508821,"visible":true,"origin":"","legend":"\u003cp\u003eIllustration of ROC curves when all variables are taken into consideration. The illustration showcasing AUC achieved for the 10th percentiles (Class 1) and 95th percentiles (Class 2) in each state. The figures are labelled according to the state initials followed by the model’s name. For instance, HP-RF denotes the Random Forest Model applied to Himachal Pradesh\u003c/p\u003e","description":"","filename":"Fig.5.png","url":"https://assets-eu.researchsquare.com/files/rs-4339400/v1/cf6a1cc3ac45491416388ce6.png"},{"id":56090897,"identity":"09e36bda-92c6-47e3-a355-23948e26b705","added_by":"auto","created_at":"2024-05-08 12:19:30","extension":"png","order_by":6,"title":"Figure 6","display":"","copyAsset":false,"role":"figure","size":1529359,"visible":true,"origin":"","legend":"\u003cp\u003eIllustration of ROC curves when statistically significant variables are taken into consideration. The illustration showcasing AUC achieved for the 10th percentiles (Class 1) and 95th percentiles (Class 2) in each state. The figures are labelled according to the state initials followed by the model’s name. For instance, HP-RF denotes the Random Forest Model applied to Himachal Pradesh.\u003c/p\u003e","description":"","filename":"Fig.6.png","url":"https://assets-eu.researchsquare.com/files/rs-4339400/v1/76b81597cff4e911e9fb1ba6.png"},{"id":56090898,"identity":"5e21daf0-7ff6-4337-9a66-5e1a2009c30f","added_by":"auto","created_at":"2024-05-08 12:19:30","extension":"png","order_by":7,"title":"Figure 7","display":"","copyAsset":false,"role":"figure","size":268689,"visible":true,"origin":"","legend":"\u003cp\u003eModel Skill Evaluation using Brier’s Score (a) considering all variables, (b) considering statistically significant variables only\u003c/p\u003e","description":"","filename":"Fig.7.png","url":"https://assets-eu.researchsquare.com/files/rs-4339400/v1/1f1c3099ad358be0eb429c9b.png"},{"id":79604707,"identity":"4f03e59a-bd7d-493d-81a4-7b5521ac38ef","added_by":"auto","created_at":"2025-03-31 15:59:53","extension":"pdf","order_by":0,"title":"","display":"","copyAsset":false,"role":"manuscript-pdf","size":7293539,"visible":true,"origin":"","legend":"","description":"","filename":"manuscript.pdf","url":"https://assets-eu.researchsquare.com/files/rs-4339400/v1/fb297919-29c1-4dcb-945d-55e7ea1c6731.pdf"}],"financialInterests":"No competing interests reported.","formattedTitle":"Efficacy of Machine Learning in Simulating Precipitation and Its Extremes Over the Capital Cities in North Indian States","fulltext":[{"header":"1. Introduction","content":"\u003cp\u003eExtreme precipitation events, intensified by global warming, have become significant challenges for both society and the environment. These events, characterized by intense precipitation, can wreak havoc on agriculture, ecosystems, and human communities \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. As atmospheric moisture content rises due to climate change, these events are projected to become more frequent and intense. Observational data since the 1950s indicate a noticeable increase in the occurrence of heavy precipitation events in many regions, a trend expected to continue, according to the Intergovernmental Panel on Climate Change \u003csup\u003e\u003cspan citationid=\"CR6\" class=\"CitationRef\"\u003e6\u003c/span\u003e\u003c/sup\u003e. The sixth assessment report (AR6) also underscores that local communities, particularly those with limited adaptive capacity, are disproportionately vulnerable to these impacts \u003csup\u003e\u003cspan citationid=\"CR7\" class=\"CitationRef\"\u003e7\u003c/span\u003e\u003c/sup\u003e.\u003c/p\u003e \u003cp\u003eThe mechanisms associated with the extreme precipitation events during the monsoon season, mostly triggered by the interactions between westward-moving monsoon systems, eastward-moving mid-tropospheric westerly troughs, and the rugged Himalayan topography, have had devastating consequences in various parts of Uttarakhand, Himachal Pradesh, and Jammu \u0026amp; Kashmir. Notable incidents include the Kedarnath tragedy (2013) \u003csup\u003e\u003cspan citationid=\"CR8\" class=\"CitationRef\"\u003e8\u003c/span\u003e,\u003cspan citationid=\"CR9\" class=\"CitationRef\"\u003e9\u003c/span\u003e\u003c/sup\u003e and Uttarakhand cloudburst (2022, 2017, 2012) \u003csup\u003e\u003cspan additionalcitationids=\"CR11\" citationid=\"CR10\" class=\"CitationRef\"\u003e10\u003c/span\u003e\u0026ndash;\u003cspan citationid=\"CR12\" class=\"CitationRef\"\u003e12\u003c/span\u003e\u003c/sup\u003e, Lahaul-Spiti(July 2021), Mandi (July 2015) \u003csup\u003e\u003cspan citationid=\"CR13\" class=\"CitationRef\"\u003e13\u003c/span\u003e,\u003cspan citationid=\"CR14\" class=\"CitationRef\"\u003e14\u003c/span\u003e\u003c/sup\u003e, Leh cloudburst (2015, 2010) \u003csup\u003e\u003cspan citationid=\"CR15\" class=\"CitationRef\"\u003e15\u003c/span\u003e,\u003cspan citationid=\"CR16\" class=\"CitationRef\"\u003e16\u003c/span\u003e\u003c/sup\u003e, and the Jammu \u0026amp; Kashmir floods (2015, Sonmarg, Pahalgam, Ganderbal and Baltal) \u003csup\u003e\u003cspan citationid=\"CR16\" class=\"CitationRef\"\u003e16\u003c/span\u003e\u003c/sup\u003e and many more reported in detail in a study conducted by Dimri et al., (2017). These events, often caused by the interaction of multiple atmospheric dynamics, lead to excessive precipitation, casualties, and significant infrastructural damage.\u003c/p\u003e \u003cp\u003eIn North India, extreme precipitation events result from complex dynamics involving both large-scale atmospheric influences and localized factors. The unique topography of North India, particularly the Himalayas, plays a crucial role in shaping the Indian monsoon \u003csup\u003e\u003cspan additionalcitationids=\"CR18 CR19 CR20\" citationid=\"CR17\" class=\"CitationRef\"\u003e17\u003c/span\u003e\u0026ndash;\u003cspan citationid=\"CR21\" class=\"CitationRef\"\u003e21\u003c/span\u003e\u003c/sup\u003e. The interplay between the mountainous terrain and atmospheric disturbances, coupled with cold air intrusion from northern latitudes, creates conditions conducive to extreme precipitation events, as seen in Jammu and Kashmir during January 2017 \u003csup\u003e22\u003c/sup\u003e. Western disturbances, embedded in the eastward-moving upper tropospheric Rossby wave train, contribute significantly to heavy precipitation in the Western Himalayas, especially during winter \u003csup\u003e\u003cspan additionalcitationids=\"CR24\" citationid=\"CR23\" class=\"CitationRef\"\u003e23\u003c/span\u003e\u0026ndash;\u003cspan citationid=\"CR25\" class=\"CitationRef\"\u003e25\u003c/span\u003e\u003c/sup\u003e. The high temperatures over the mountains and neighbouring areas contribute to the formation of low-pressure systems, which extend southward across the plains of South Asia, facilitating the northward progression of the monsoon \u003csup\u003e\u003cspan additionalcitationids=\"CR27\" citationid=\"CR26\" class=\"CitationRef\"\u003e26\u003c/span\u003e\u0026ndash;\u003cspan citationid=\"CR28\" class=\"CitationRef\"\u003e28\u003c/span\u003e\u003c/sup\u003e. The 2013 Uttarakhand disaster, which involved rapid monsoon progression and heavy precipitation, resulted from a combination of these large-scale circulations, orographic lifting, and intense convective activity \u003csup\u003e\u003cspan citationid=\"CR29\" class=\"CitationRef\"\u003e29\u003c/span\u003e\u003c/sup\u003e. These extreme precipitation events are known to exacerbate due to climate change with increased frequency and intensity \u003csup\u003e\u003cspan citationid=\"CR30\" class=\"CitationRef\"\u003e30\u003c/span\u003e,\u003cspan citationid=\"CR16\" class=\"CitationRef\"\u003e16\u003c/span\u003e\u003c/sup\u003e.\u003c/p\u003e \u003cp\u003eTraditional modeling approaches struggle to capture the complex interplay of factors shaping precipitation patterns, including climate change, topography, and atmospheric dynamics \u003csup\u003e\u003cspan additionalcitationids=\"CR32\" citationid=\"CR31\" class=\"CitationRef\"\u003e31\u003c/span\u003e\u0026ndash;\u003cspan citationid=\"CR33\" class=\"CitationRef\"\u003e33\u003c/span\u003e\u003c/sup\u003e. In response, integrating machine learning (ML) techniques has emerged as a promising avenue for understanding precipitation \u003csup\u003e\u003cspan additionalcitationids=\"CR35 CR36\" citationid=\"CR34\" class=\"CitationRef\"\u003e34\u003c/span\u003e\u0026ndash;\u003cspan citationid=\"CR37\" class=\"CitationRef\"\u003e37\u003c/span\u003e\u003c/sup\u003e and extreme precipitation\u003csup\u003e\u003cspan additionalcitationids=\"CR39 CR40\" citationid=\"CR38\" class=\"CitationRef\"\u003e38\u003c/span\u003e\u0026ndash;\u003cspan citationid=\"CR41\" class=\"CitationRef\"\u003e41\u003c/span\u003e\u003c/sup\u003e variability and enhancing forecasting accuracy leveraging large meteorological datasets and computational power to improve predictive capabilities amid climate change uncertainties \u003csup\u003e\u003cspan additionalcitationids=\"CR35\" citationid=\"CR34\" class=\"CitationRef\"\u003e34\u003c/span\u003e\u0026ndash;\u003cspan citationid=\"CR36\" class=\"CitationRef\"\u003e36\u003c/span\u003e\u003c/sup\u003e.\u003c/p\u003e \u003cp\u003eFor India, a few literatures are available for extreme precipitation analysis \u003csup\u003e\u003cspan citationid=\"CR30\" class=\"CitationRef\"\u003e30\u003c/span\u003e,\u003cspan additionalcitationids=\"CR43\" citationid=\"CR42\" class=\"CitationRef\"\u003e42\u003c/span\u003e\u0026ndash;\u003cspan citationid=\"CR44\" class=\"CitationRef\"\u003e44\u003c/span\u003e\u003c/sup\u003e, however relatively fewer for extreme precipitation modelling using machine learning \u003csup\u003e\u003cspan citationid=\"CR37\" class=\"CitationRef\"\u003e37\u003c/span\u003e,\u003cspan citationid=\"CR45\" class=\"CitationRef\"\u003e45\u003c/span\u003e,\u003cspan citationid=\"CR46\" class=\"CitationRef\"\u003e46\u003c/span\u003e\u003c/sup\u003e. The limited availability of high-quality and comprehensive meteorological data in India and complexity of the regional topography and climate are also prominent contributing factor for limited number of research. A study by Ray et al. (2022) suggest that machine learning techniques, can effectively predict precipitation by correlating meteorological parameters with precipitation events \u003csup\u003e\u003cspan citationid=\"CR47\" class=\"CitationRef\"\u003e47\u003c/span\u003e\u003c/sup\u003e.\u003c/p\u003e \u003cp\u003eIn light of aforesaid reasons, our study aims to explore key atmospheric variables affecting precipitation patterns and extreme events in North India and studies link between these variables and precipitation intensity across regions. It further examines the performance of machine learning models in predicting precipitation and extreme events in each state and analyzed how different models simulate precipitation and capture extreme thresholds. Lastly, the study investigates the ability of machine learning models to distinguish between extreme and non-extreme precipitation events and their calibration and discrimination for extreme event prediction. Ultimately, the goal is to harness ML to improve the accuracy and reliability of precipitation forecasts, helping policymakers and communities prepare for and mitigate the impact of extreme weather events in the context of climate change.\u003c/p\u003e"},{"header":"2. Study Area","content":"\u003cp\u003eEncompassing a vast region in North India, this study area covers seven states: Himachal Pradesh (HP), Jammu and Kashmir (JK), Punjab and Haryana (PH) (considered together), Rajasthan (RJ), Uttarakhand (UK), and Uttar Pradesh (UP) (Fig.\u0026nbsp;\u003cspan refid=\"Fig1\" class=\"InternalRef\"\u003e1\u003c/span\u003e). The dramatic elevation range stretches from a low point of 60 meters in Uttar Pradesh to a staggering 8,611 meters at K2 Peak in Jammu and Kashmir. This translates to significant temperature variations across the region. The plains experience scorching summers with highs reaching 45\u0026deg;C, particularly in the Rajasthan Thar Desert, while winters can be mild with lows around 0\u0026deg;C. In contrast, the hilly regions offer cooler summers with highs of 25\u0026deg;C, but winters can be harsh with temperatures dipping as low as -30\u0026deg;C in Jammu and Kashmir.\u003c/p\u003e \u003cp\u003eEach state within this diverse landscape faces distinct environmental challenges \u003csup\u003e\u003cspan citationid=\"CR48\" class=\"CitationRef\"\u003e48\u003c/span\u003e\u003c/sup\u003e. The Himalayan states like Uttarakhand, Himachal Pradesh, and Jammu and Kashmir grapple with issues like forest fires, biodiversity loss, glacial retreat leading to water scarcity, and soil erosion. Meanwhile, the plains states of Uttar Pradesh, Punjab \u0026amp; Haryana, and Rajasthan battle water scarcity, soil degradation, and air pollution. Additionally, all states face challenges related to the impact of climate change on agriculture \u003csup\u003e\u003cspan citationid=\"CR49\" class=\"CitationRef\"\u003e49\u003c/span\u003e,\u003cspan citationid=\"CR50\" class=\"CitationRef\"\u003e50\u003c/span\u003e\u003c/sup\u003e. Understanding this interplay between natural processes and human activities across this vast region with its varying elevations and temperatures is crucial for developing effective management strategies and sustainable development practices \u003csup\u003e\u003cspan citationid=\"CR30\" class=\"CitationRef\"\u003e30\u003c/span\u003e,\u003cspan citationid=\"CR49\" class=\"CitationRef\"\u003e49\u003c/span\u003e\u003c/sup\u003e.\u003c/p\u003e \u003cp\u003e \u003c/p\u003e"},{"header":"3. Results","content":"\u003cdiv id=\"Sec4\" class=\"Section2\"\u003e \u003ch2\u003e3.1 State Specific Inter-Variable Analysis\u003c/h2\u003e \u003cp\u003eUnderstanding the intricate relationships between atmospheric variables is crucial for accurate precipitation prediction \u003csup\u003e\u003cspan citationid=\"CR39\" class=\"CitationRef\"\u003e39\u003c/span\u003e\u003c/sup\u003e. In this section these intricacies are explored by examining state-specific inter-variable interactions. For each state capital included in the study, correlation matrices were generated (Fig.\u0026nbsp;\u003cspan refid=\"Fig2\" class=\"InternalRef\"\u003e2\u003c/span\u003e.). These matrices provide a valuable tool to explore the strength and direction of linear associations between the chosen atmospheric variables (temperature, humidity, pressure, etc.). By analyzing these correlations, we aim to identify recurring patterns, potential dependencies between variables within each state, and any instances of multicollinearity. Multicollinearity occurs when variables exhibit high correlations, potentially leading to issues during machine learning model training. Examining these state-specific relationships allows for a more nuanced understanding of how atmospheric variables interact and influence precipitation patterns across diverse geographical regions within North India \u003csup\u003e\u003cspan citationid=\"CR68\" class=\"CitationRef\"\u003e68\u003c/span\u003e\u003c/sup\u003e.\u003c/p\u003e \u003cp\u003eAcross all states, a consistent theme emerges: Dew Point Temperature, Surface Pressure and Relative humidity tend to be the strongest allies of precipitation. Strong Positive correlations of Dew Point Temperature, Relative humidity (~\u0026thinsp;0.4) and Strong negative correlations (~ -0.3) for the aforesaid variables highlights their potential role as contributors to increased precipitation. The influence of temperature revealed a more precise observation. While states like HP and UK exhibit positive correlations (0.013, 0.04) while PH, JK, and UP show weaker negative (-0.0028 to -0.04) associations.\u003c/p\u003e \u003cp\u003eIn almost all states, solar irradiance and surface pressure acts as a counterpoint to precipitation, exhibiting negative correlations ranging from \u0026minus;\u0026thinsp;0.18 in PH (Solar irradiance) to a more pronounced \u0026minus;\u0026thinsp;0.3 in HP, RJ, UK and UP (Surface pressure). The impact of wind speed on precipitation varies across states. While RJ, UP and JK show modest positive correlations (0.04, 0.11 and 0.24), other states like HP, PH and UK exhibit weakly negative associations (-0.072, -0.054 and 0.037) respectively with wind (Fig.\u0026nbsp;\u003cspan refid=\"Fig2\" class=\"InternalRef\"\u003e2\u003c/span\u003e.).\u003c/p\u003e \u003cp\u003eOverall precipitation patterns across the study region are influenced by a complex interplay of various atmospheric factors, both directly and indirectly. In the direct category, elevated levels of dew point temperature, relative humidity and surface pressure emerge as significant contributors to increased precipitation in the studies locations. These states consistently exhibit positive correlations between dew, pressure, humidity, and precipitation, underscoring the direct impact of moisture content on precipitation. Additionally, surface pressure and solar irradiance displays a notable indirect relationship, with lower level of the events associated with higher precipitation events across the study location, however correlation with surface\u003c/p\u003e \u003cp\u003e \u003c/p\u003e \u003cp\u003epressure was only statistically significant at significance level of 0.05. Conversely, temperature and wind demonstrates a varied impact, with both positive and negative correlations observed in different states, indicating that warmer and windier conditions may contribute to precipitation in some regions while having the opposite effect in others (Fig.\u0026nbsp;\u003cspan refid=\"Fig3\" class=\"InternalRef\"\u003e3\u003c/span\u003e.).\u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec5\" class=\"Section2\"\u003e \u003ch2\u003e3.2 Model Performance for Precipitation Prediction\u003c/h2\u003e \u003cp\u003eFour machine learning models \u0026ndash; Support Vector Classifier (SVC), Random Forest Classifier (RFC), XGBoost, and K Nearest Neighbors (KNN) \u0026ndash; were developed to analyze precipitation and its extreme patterns using the Modern-Era Retrospective Analysis for Research and Applications, version 2 (MERRA-2) reanalysis data. The models were trained over the data to achieve the highest possible accuracies. The developed model was trained to predict not only simple precipitation events, but extreme precipitation events at different threshold levels also. The developed model was trained and tested at different data split ratios and were evaluated based on their accuracy in predicting precipitation events. Figure\u0026nbsp;\u003cspan refid=\"Fig3\" class=\"InternalRef\"\u003e3\u003c/span\u003e. illustrates the accuracy achieved by various machine learning algorithms across different states for different train test split ratios. State wise observations are presented.\u003c/p\u003e \u003cdiv id=\"Sec6\" class=\"Section3\"\u003e \u003ch2\u003e3.2.1 Himachal Pradesh\u003c/h2\u003e \u003cp\u003eThe analysis revealed that SVC consistently achieved the highest accuracy scores across all train-test split ratios in Himachal Pradesh, ranging from 83.35\u0026ndash;83.50%. This suggests the effectiveness of SVC in capturing the complex interplay between various atmospheric variables and precipitation patterns in this state. The mountainous terrain and diverse climate of Himachal Pradesh necessitate a robust model capable of handling intricate relationships. RFC followed closely with accuracy scores ranging from 82.84\u0026ndash;82.94%, demonstrating strong and reliable performance. This indicates the suitability of both SVC and RFC for precipitation prediction in Himachal Pradesh. XGBoost maintained competitive accuracy (82.28% \u0026minus;\u0026thinsp;82.75%), while KNN also achieved a reliable scores (81.62% \u0026minus;\u0026thinsp;81.74%) (Fig.\u0026nbsp;\u003cspan refid=\"Fig3\" class=\"InternalRef\"\u003e3\u003c/span\u003e.) Here, the consistent high accuracy of SVC and RFC suggests their ability to effectively model precipitation patterns in Himachal Pradesh's diverse topography and climatic zones.\u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec7\" class=\"Section3\"\u003e \u003ch2\u003e3.2.2 Jammu and Kashmir\u003c/h2\u003e \u003cp\u003eSimilar to Himachal Pradesh, SVC emerged as the most effective algorithm for precipitation prediction in Jammu and Kashmir (Fig.\u0026nbsp;\u003cspan refid=\"Fig3\" class=\"InternalRef\"\u003e3\u003c/span\u003e.). The average accuracy across all split ratios for SVC was 79.79%, closely followed by RFC with an average accuracy of 79.61%. This indicates the suitability of both algorithms for capturing precipitation patterns in this state characterized by the Himalayas and Kashmir Valley with its unique weather systems. XGBoost and KNN achieved average accuracies of 78.69% and 77.64%, respectively. The effectiveness of SVC and RFC in Jammu and Kashmir underscores their ability to handle the complex interactions between atmospheric variables in a region with significant topographic variations.\u003c/p\u003e \u003cp\u003e \u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec8\" class=\"Section3\"\u003e \u003ch2\u003e3.2.3 Punjab and Haryana\u003c/h2\u003e \u003cp\u003eThe analysis revealed SVC as the most effective algorithm across all split ratios for predicting precipitation patterns in Punjab and Haryana (Fig.\u0026nbsp;\u003cspan refid=\"Fig3\" class=\"InternalRef\"\u003e3\u003c/span\u003e.). The average accuracy for SVC was 82.06%, followed closely by XGBoost with an average accuracy of 81.77%. RFC and KNN achieved slightly lower average accuracies of 80.65% and 81.03%, respectively. Interestingly, SVC exhibited superior performance, particularly at higher split ratios, showcasing its robustness in handling varying data distributions. This highlights its suitability for capturing precipitation patterns influenced by diverse factors such as the proximity to the Himalayas and the Indus River basin. The strong performance of SVC in Punjab and Haryana suggests its effectiveness in modeling precipitation patterns in the region's predominantly flat plains.\u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec9\" class=\"Section3\"\u003e \u003ch2\u003e3.2.4 Rajasthan\u003c/h2\u003e \u003cp\u003eIn contrast to other states, RFC emerged as the top performer for precipitation prediction in Rajasthan. The consistently high accuracy scores achieved by RFC, ranging from 83.55\u0026ndash;83.71%, indicate its effectiveness in capturing precipitation patterns in this arid state. SVC displayed competitive performance with accuracy scores ranging from 82.34\u0026ndash;82.79%. XGBoost also showcased consistent performance with accuracy scores between 82.78% and 83.22%. While KNN achieved lower scores (82.00% \u0026minus;\u0026thinsp;82.06%), it still offers reasonable predictive capabilities.\u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec10\" class=\"Section3\"\u003e \u003ch2\u003e3.2.5 Uttarakhand\u003c/h2\u003e \u003cp\u003eSimilar to Himachal Pradesh, SVC emerged as the leader for precipitation prediction in Uttarakhand. Its accuracy scores ranged from a high of 84.47% to a low of 84.16% across different split ratios. This consistency suggests SVC's effectiveness in capturing the intricate relationships between atmospheric variables and precipitation patterns in this state characterized by the Himalayan foothills and diverse microclimates. RFC followed closely with accuracy scores ranging from 83.84\u0026ndash;83.96%, demonstrating strong performance. XGBoost displayed competitive accuracy (83.33% \u0026minus;\u0026thinsp;83.76%), while KNN achieved respectabe scores (82.93% \u0026minus;\u0026thinsp;83.25%).\u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec11\" class=\"Section3\"\u003e \u003ch2\u003e3.2.6 Uttar Pradesh\u003c/h2\u003e \u003cp\u003eIn Uttar Pradesh, the analysis revealed RFC as the dominant algorithm for precipitation prediction. It achieved the highest accuracy scores across all splits, ranging from 84.70\u0026ndash;85.05%. XGBoost followed closely with accuracy scores ranging from 83.93\u0026ndash;84.50%. SVC maintained competitive scores (83.78% \u0026minus;\u0026thinsp;84.16%). KNN achieved respectable accuracy (83.57% \u0026minus;\u0026thinsp;83.73%). The dominance of RFC in Uttar Pradesh, the most populous state in India, can be attributed to its ability to effectively model precipitation patterns influenced by diverse factors such as the Gangetic Plain's topography and proximity to the Himalayas. The strong performance of both RFC and XGBoost suggests promising avenues for further exploration and potential implementation in operational forecasting systems. Plain's topography and proximity to the Himalayas. The strong performance of both RFC and XGBoost suggests promising avenues for further exploration and potential implementation in operational forecasting systems.\u003c/p\u003e \u003cp\u003e \u003c/p\u003e \u003cp\u003eOverall, SVM and RF emerged as strong contenders in our analysis of mapping precipitation patterns across various states in North India (Fig.\u0026nbsp;\u003cspan refid=\"Fig4\" class=\"InternalRef\"\u003e4\u003c/span\u003e). Notably, SVM consistently outperformed RF in the majority of states, showcasing its effectiveness in accurately predicting precipitation patterns. However, interestingly, RF exhibited superior performance in Rajasthan and Uttarakhand, where it achieved the highest accuracy scores among all algorithms tested. This observation underscores the nuanced nature of regional climatic patterns and highlights RF's capability to excel in certain geographical contexts. Upon aggregating the results from multiple iterations and train-test split ratios, we constructed line plot (Fig.\u0026nbsp;\u003cspan refid=\"Fig4\" class=\"InternalRef\"\u003e4\u003c/span\u003e.) for each state individually, providing a comprehensive overview of algorithm performance. These plots affirmed the dominance of SVM across most states, but both SVM and RF demonstrated comparable\u003c/p\u003e \u003cp\u003e \u003c/p\u003e \u003cp\u003eperformance, indicating a close competition between the two algorithms. Additionally, the study revealed an inverse correlation between elevation and the performance of machine learning algorithms in simulating precipitation and its extremes over the capital cities in North Indian states. Specifically, in regions with lower elevations, the models exhibited higher skill (accuracy) in simulating precipitation, while in areas with higher elevations, the model skill was relatively lower, with a skill difference of approximately 5%. This observation highlights the influence of topographic features and associated meteorological complexities on the models' ability to accurately capture precipitation patterns.\u003c/p\u003e \u003c/div\u003e \u003c/div\u003e \u003cdiv id=\"Sec12\" class=\"Section2\"\u003e \u003ch2\u003e3.3 Evaluation of Model Performance for Extreme Precipitation\u003c/h2\u003e \u003cp\u003eBuilding on our previous model evaluation, which demonstrated competitive performance between RFC and SVC models for general precipitation patterns across most states, we sought to specifically evaluate their effectiveness in predicting extreme precipitation events. For this we implemented a two-pronged evaluation for this purpose. First, both models were evaluated using all initial variables (Fig.\u0026nbsp;\u003cspan refid=\"Fig5\" class=\"InternalRef\"\u003e5\u003c/span\u003e.). and then only for statistically significant variables (Fig.\u0026nbsp;\u003cspan refid=\"Fig6\" class=\"InternalRef\"\u003e6\u003c/span\u003e.).\u003c/p\u003e \u003cp\u003eacross different states. Receiver Operating Characteristic (ROC) curves were generated for each scenario to visualize model discrimination between extreme events below 10th and above 95th percentiles (Figs.\u0026nbsp;\u003cspan refid=\"Fig5\" class=\"InternalRef\"\u003e5\u003c/span\u003e and \u003cspan refid=\"Fig6\" class=\"InternalRef\"\u003e6\u003c/span\u003e). While AUC of an ROC curve provides a good overview of model performance, it's valuable to consider additional metrics for a more comprehensive evaluation and robustness check of the build models. Here, we considered Brier Score in order to assesses the overall calibration of the model's predicted probabilities with actual outcomes (Fig.\u0026nbsp;\u003cspan refid=\"Fig7\" class=\"InternalRef\"\u003e7\u003c/span\u003e.)\u003c/p\u003e \u003cp\u003eBy comparing ROC values and Brier Scores across models and variable selection approaches, we aimed to identify the most effective model for extreme precipitation prediction.\u003c/p\u003e \u003cp\u003e \u003cul\u003e \u003cli\u003e \u003cp\u003eOverall Performance: Both models (RFC and SVC) exhibit good skill in predicting extreme precipitation events, with AUC values generally exceeding 0.85 across states and scenarios (Figs.\u0026nbsp;\u003cspan refid=\"Fig5\" class=\"InternalRef\"\u003e5\u003c/span\u003e \u0026amp; \u003cspan refid=\"Fig6\" class=\"InternalRef\"\u003e6\u003c/span\u003e.). This indicates a strong ability to discriminate between extreme and non-extreme events.\u003c/p\u003e \u003c/li\u003e \u003cli\u003e \u003cp\u003eModel Comparison: While there are some variations, the performance of RFC and SVM models is often comparable, with neither model consistently outperforming the other across all states and scenarios.\u003c/p\u003e \u003c/li\u003e \u003cli\u003e \u003cp\u003eImpact of Variable Selection: Focusing on statistically significant variables sometimes results in slightly improved AUC values for the SVM model (e.g., state JK at the 95th percentile threshold) (Fig.\u0026nbsp;\u003cspan refid=\"Fig6\" class=\"InternalRef\"\u003e6\u003c/span\u003e). However, the differences are generally small,\u003c/p\u003e \u003c/li\u003e \u003c/ul\u003e \u003c/p\u003e \u003cp\u003e \u003c/p\u003e \u003cp\u003e \u003cdiv class=\"BlockQuote\"\u003e \u003cp\u003esuggesting that the additional variables included in the \"all variables\" scenario might not significantly impact model performance for extreme event prediction.\u003c/p\u003e \u003c/div\u003e \u003c/p\u003e \u003cp\u003e \u003cul\u003e \u003cli\u003e \u003cp\u003eState-Specific Variations: There is some variation in model performance across states. States like RJ and UP show consistently high AUC values (\u0026gt;\u0026thinsp;0.90) for both models in both scenarios, indicating exceptional skill in predicting extreme events. Conversely, states like HP and PH show slightly lower but still good performance (AUC around 0.87\u0026ndash;0.90). This might be due to regional differences in extreme precipitation patterns or require further investigation into specific factors influencing those states.\u003c/p\u003e \u003c/li\u003e \u003cli\u003e \u003cp\u003eThreshold Dependence: As expected, AUC values are generally higher for the 95th percentile threshold compared to the 10th percentile threshold. This is because the 95th percentile represents a more extreme precipitation event, which might be easier for the models to predict accurately.\u003c/p\u003e \u003c/li\u003e \u003c/ul\u003e \u003c/p\u003e \u003cp\u003e \u003c/p\u003e \u003cp\u003e \u003cul\u003e \u003cli\u003e \u003cp\u003eOverall Brier Scores: The Brier Scores range from approximately 0.07 to 0.12 across states and models. These are generally considered good scores, indicating the models are providing reasonably accurate probability estimates for extreme precipitation events.\u003c/p\u003e \u003c/li\u003e \u003cli\u003e \u003cp\u003eModel Comparison: Similar to the AUC results, there are no significant differences between RF and SVM models in terms of Brier Scores for most states. Both models achieve comparable performance\u003c/p\u003e \u003c/li\u003e \u003cli\u003e \u003cp\u003eImpact of Variable Selection: Focusing on statistically significant variables (often including humidity and surface pressure) sometimes leads to slightly higher Brier Scores (worse performance) compared to using all variables. This suggests that while some variables might not be statistically significant, they could still contribute to the model's ability to calibrate its predictions.\u003c/p\u003e \u003c/li\u003e \u003cli\u003e \u003cp\u003eState-Specific Variations: Brier Scores also show some variation across states. States like RJ and UP consistently have lower Brier Scores (better performance) compared to states like HP and JK. This aligns with the observations from AUC values, suggesting these states might have more predictable extreme precipitation patterns or benefit from the additional information captured by all initial variables.\u003c/p\u003e \u003c/li\u003e \u003c/ul\u003e \u003c/p\u003e \u003c/div\u003e"},{"header":"4. Discussion","content":"\u003cp\u003eOur findings align with the growing popularity of machine learning approaches in India. The study over Western Ghats region in India by \u003csup\u003e\u003cspan citationid=\"CR69\" class=\"CitationRef\"\u003e69\u003c/span\u003e\u003c/sup\u003e showcases the effectiveness of XGBoost and RF models in multi-model ensembles for capturing the behaviour of climate change on precipitation patterns. This study emphasizes the importance of comprehensive testing and validation, especially for regional investigations with diverse precipitation mechanisms \u003csup\u003e\u003cspan citationid=\"CR69\" class=\"CitationRef\"\u003e69\u003c/span\u003e\u003c/sup\u003e.\u003c/p\u003e \u003cp\u003eMachine learning offers a distinct advantage in simulating extreme precipitation events, which are notoriously difficult to predict using traditional methods. The success of the improved K-Nearest Neighbor model in simulating such events in New Delhi \u003csup\u003e\u003cspan citationid=\"CR70\" class=\"CitationRef\"\u003e70\u003c/span\u003e\u003c/sup\u003e highlights its potential for vulnerability assessments in flood-prone regions. The study suggested expanding on this concept, future research could explore the application of other ML algorithms, like LSTMs, which excel at capturing temporal dependencies, to improve extreme event prediction.\u003c/p\u003e \u003cp\u003eML methods have been successfully employed for various precipitation-related applications across India. Studies in Nellore Station, Andhra Pradesh \u003csup\u003e\u003cspan citationid=\"CR71\" class=\"CitationRef\"\u003e71\u003c/span\u003e\u003c/sup\u003e demonstrated the feasibility of ML for short-term forecasting. \u003csup\u003e\u003cspan citationid=\"CR72\" class=\"CitationRef\"\u003e72\u003c/span\u003e\u003c/sup\u003e showcased the potential of Extreme Learning Machines for probabilistic predictions of monsoon precipitation, crucial for agricultural planning. Additionally, the study by \u003csup\u003e\u003cspan citationid=\"CR73\" class=\"CitationRef\"\u003e73\u003c/span\u003e\u003c/sup\u003e underlines the effectiveness of RF and LSTM in creating multi-model ensembles for precipitation prediction in river basins. This highlights the versatility of ML in tackling diverse precipitation-related challenges.\u003c/p\u003e \u003cp\u003eUnsupervised learning approaches, like the one employed by \u003csup\u003e\u003cspan citationid=\"CR74\" class=\"CitationRef\"\u003e74\u003c/span\u003e\u003c/sup\u003e to analyze changes in extreme precipitation due to storm dynamics, offer valuable insights into the complex interplay of factors influencing these events. The study suggested further research could explore the integration of unsupervised and supervised learning approaches to gain a more holistic understanding of precipitation dynamics.\u003c/p\u003e \u003cp\u003eThe success of ML in North India aligns relatively well with global trends. Studies across the United States by \u003csup\u003e\u003cspan citationid=\"CR39\" class=\"CitationRef\"\u003e39\u003c/span\u003e\u003c/sup\u003e employed RF, XGBoost, and ANNs to explore the key variables influencing extreme precipitation. These models demonstrated improved performance in regions with distinct seasonal variations, suggesting their potential applicability in other parts of the world with similar climatic patterns.\u003c/p\u003e \u003cp\u003eSimilar to our findings with RF, international research highlights its effectiveness in extreme event forecasting. \u003csup\u003e\u003cspan citationid=\"CR75\" class=\"CitationRef\"\u003e75\u003c/span\u003e\u003c/sup\u003e in China employed RF models to forecast days with a high probability of extreme precipitation events. Additionally, studies by \u003csup\u003e76\u003c/sup\u003e and \u003csup\u003e77\u003c/sup\u003e showcase the success of RF models in predicting extreme precipitation events in other regions. This global consistency suggests inherent advantages of RF for this specific task.\u003c/p\u003e \u003cp\u003eIn regions with lower elevations, such as Uttar Pradesh with an elevation of 123mt., the models exhibited higher accuracy in precipitation simulations (84%). Conversely, areas with higher elevations like Jammu and Kashmir, with an average of 1585mt., posed greater challenges, and the models displayed relatively lower skill (79%). The approximate 5% difference in skill between low and high elevation regions highlights the influence of topographic features and associated meteorological complexities. The intricate terrain and orographic effects in mountainous regions like JK can contribute to more complex precipitation patterns, localized atmospheric circulations, and rapid weather changes, which machine learning algorithms may struggle to capture accurately. Additionally, the availability and quality of observational data used for training the models could be limited in remote, high-elevation areas, further impacting their performance. In contrast, the relatively simpler topography and more predictable meteorological conditions in lower elevation regions, coupled with potentially better data availability, may have facilitated improved model skill in simulating precipitation patterns accurately.\u003c/p\u003e \u003cp\u003eOur study identified RF as the most effective model for simulating precipitation extremes in North India. This finding aligns with international research, where RF consistently demonstrates high accuracy in extreme event prediction. While SVM has proven successful in various applications, studies specifically focused on extreme precipitation simulation do not explicitly mention its effectiveness. This suggests that RF might be a superior choice for this particular task due to model\u0026rsquo;s ability in handling high dimensionality, non-linearity at a high pace \u003csup\u003e\u003cspan citationid=\"CR39\" class=\"CitationRef\"\u003e39\u003c/span\u003e,\u003cspan citationid=\"CR75\" class=\"CitationRef\"\u003e75\u003c/span\u003e,\u003cspan citationid=\"CR78\" class=\"CitationRef\"\u003e78\u003c/span\u003e,\u003cspan citationid=\"CR79\" class=\"CitationRef\"\u003e79\u003c/span\u003e\u003c/sup\u003e.\u003c/p\u003e"},{"header":"5. Conclusion","content":"\u003cp\u003eIn this work, a comprehensive study was conducted to investigate the intricate interplay between atmospheric variables and extreme precipitation events across the seven states in North India. The study leverages MERRA-2 reanalysis data spanning from 1984 to 2022 and employed four machine learning models (Support Vector Classifier (SVC), Random Forest Classifier (RFC), XGBoost, and K Nearest Neighbors (KNN)).\u003c/p\u003e \u003cp\u003eThe analysis revealed insightful correlations between key predictor variables and extreme precipitation intensity for each region. Dew Point Temperature and Relative Humidity exhibited strong positive correlations (~\u0026thinsp;0.4) with precipitation across all states, while Temperature exhibited regional variations with positive correlations in Himachal Pradesh, Punjab and Haryana, and Uttarakhand (~\u0026thinsp;0.2), and weaker negative associations in Jammu and Kashmir, Rajasthan, and Uttar Pradesh (-0.1 to -0.2). Solar irradiance and Surface Pressure often acted as counterpoints to precipitation, with negative correlations ranging from \u0026minus;\u0026thinsp;0.18 to -0.3. The significance of these variables was taken into account while performing predictive modelling in case of extreme precipitation pattern detection. The predictive modeling aspect, employed machine learning algorithms across all states which depicted competitive performance among algorithms.\u003c/p\u003e \u003cp\u003eSVC and RFC emerged as powerful tools for precipitation prediction, with SVC dominating in Himachal Pradesh, Jammu and Kashmir, Uttarakhand, and Punjab and Haryana, while RFC excelled in Rajasthan and Uttar Pradesh. The models exhibited higher skill in simulating precipitation over lower elevation regions compared to higher elevation areas, with a skill difference of around 5%, potentially due to the influence of topographic complexity on meteorological phenomena. Furthermore, the analysis of extreme precipitation events revealed that Random Forest models consistently outperformed Support Vector Machines, achieving higher Area Under the Curve (AUC) values (~\u0026thinsp;0.90), and lower Brier Scores (~\u0026thinsp;0.01), across all states and precipitation thresholds. Despite the promising results, the study acknowledges limitations such as reliance on reanalysis data, limited atmospheric variables, and coarse spatial and temporal resolutions.\u003c/p\u003e \u003cp\u003eThe study's success with Random Forest models for simulating precipitation extremes paves the way for further advancements. Future research should explore more advance algorithms like LSTMs and ensemble learning approaches, as well data augmentation strategies. Integrating high-resolution data, climate change projections, and atmospheric processes into the models can enhance their accuracy and robustness. Ultimately, these advancements should be translated into practical tools for real-time flood and drought forecasting, optimizing agricultural practices, and informing adaptation strategies. By considering both atmospheric processes and anthropogenic influences, we can develop comprehensive and holistic models for understanding and predicting precipitation extremes, leading to a more sustainable future in the face of climate change.\u003c/p\u003e"},{"header":"6. Data","content":"\u003cp\u003eFor the study, data was sourced from NASA\u0026rsquo;s Prediction of Worldwide Energy Resources (POWER) project spanning from 1984 to 2022 \u003csup\u003e51\u003c/sup\u003e. The data collection process was centred around obtaining meteorological observations from the capital cities of each states in North India. These capital cities were strategically chosen as representative locations to assess and characterize the atmospheric conditions prevalent across their respective states. This approach aimed to capture the regional variations in climate and weather patterns that influence precipitation dynamics in the region. The study encompassed a comprehensive set of seven atmospheric variables, which were meticulously recorded and analyzed. These variables included maximum temperature, relative humidity, surface pressure, wind speed, dew point temperature, precipitation from Modern-Era Retrospective Analysis for Research and Applications, version 2 (MERRA-2), and solar irradiance from Clouds and the Earth\u0026rsquo;s Radiant Energy System Project (CERES).\u003c/p\u003e \u003cp\u003eThe data was a combination of MERRA-2 \u003csup\u003e52\u0026ndash;56\u003c/sup\u003e and CERES SYN \u003csup\u003e\u003cspan additionalcitationids=\"CR58 CR59\" citationid=\"CR57\" class=\"CitationRef\"\u003e57\u003c/span\u003e\u0026ndash;\u003cspan citationid=\"CR60\" class=\"CitationRef\"\u003e60\u003c/span\u003e\u003c/sup\u003e and has been widely used by researchers for various climate and atmospheric studies. The comprehensive coverage of atmospheric and climate variables, alongside its high spatial and temporal resolution, is underscored by its assimilation of a wide range of satellite observations and other data sources, making it a valuable asset for climate and atmospheric research. MERRA-2 served a grid resolution of 0.5*0.625 degrees and demonstrates advancements in addressing known deficiencies, such as reducing spurious trends and jumps associated with changes in the observing system, as well as mitigating biases and imbalances in aspects of the water cycle \u003csup\u003e\u003cspan citationid=\"CR51\" class=\"CitationRef\"\u003e51\u003c/span\u003e\u003c/sup\u003e. CERES-SYN served a grid resolution of 1*1 degrees. This product provides accurate satellite-retrieved estimates of Earth's radiation budget components, including downwelling longwave radiation at the surface and other atmospheric levels from 2000 to the present. It utilizes radiation transfer model along with improved cloud property retrievals and consistent temperature/humidity data to estimate radiative fluxes more accurately. As a representative global satellite product covering various timescales, CERES-SYN serves as a valuable data source for evaluating reanalysis radiation estimates over regions lacking sufficient ground observations.\u003c/p\u003e"},{"header":"7. Methodology","content":"\u003cp\u003eThe dataset utilized in this study, has been extensively employed in various research investigations. Notably, it has been employed in studies for evaluation of hydrological performance of precipitation products different locations and over Basins \u003csup\u003e\u003cspan citationid=\"CR61\" class=\"CitationRef\"\u003e61\u003c/span\u003e,\u003cspan citationid=\"CR62\" class=\"CitationRef\"\u003e62\u003c/span\u003e\u003c/sup\u003e, for analysis of the diurnal cycle of summer precipitation and associated land-atmosphere interactions \u003csup\u003e\u003cspan citationid=\"CR63\" class=\"CitationRef\"\u003e63\u003c/span\u003e\u003c/sup\u003e, drought estimations \u003csup\u003e\u003cspan citationid=\"CR64\" class=\"CitationRef\"\u003e64\u003c/span\u003e\u003c/sup\u003e and other hydrometeorological application \u003csup\u003e\u003cspan additionalcitationids=\"CR66\" citationid=\"CR65\" class=\"CitationRef\"\u003e65\u003c/span\u003e\u0026ndash;\u003cspan citationid=\"CR67\" class=\"CitationRef\"\u003e67\u003c/span\u003e\u003c/sup\u003e. These studies demonstrate the widespread application and reliability of the MERRA-2 dataset across different geographical regions and climatic conditions.\u003c/p\u003e \u003cp\u003eOur primary objective was to find the intricate relationships between various atmospheric variables. For this generation and analysis of correlation plots was performed. To ensure the comparability of our findings, we did feature engineering over the dataset. Leveraging the power of supervised machine learning, we applied a diverse set of algorithms including Random Forest Classifier (RFC), Support Vector Classifier (SVC), XGBoost Classifier (XGB), and K-Nearest Neighbors (KNN) to the pre-processed data. Our aim was to select the most effective model for accurately mapping precipitation patterns across the expansive Indian region. For this, we adhered to a conventional 80:20 split for training and testing the algorithm. However, to comprehensively assess the performance of each model, we systematically experimented with different split ratios including 70\u0026thinsp;\u0026minus;\u0026thinsp;30, 60\u0026thinsp;\u0026minus;\u0026thinsp;40, and 50\u0026ndash;50. Each model underwent a rigorous iterative process, ranging from 25 to 30 iterations, to fine-tune its efficiency and effectiveness.\u003c/p\u003e \u003cp\u003eFor evaluating the performance of our models, we employed a range of metrics including accuracy, precision, recall, and F1-score. However, to know the skill of the model. we selected \u0026lsquo;accuracy\u0026rsquo; for forming comparison. Furthermore, in our quest to gain deeper insights into the efficacy of each algorithm, especially in accurately representing extreme precipitation patterns over the North Indian States, we visualized model performance for different thresholds level (10th and 95th percentiles) using receiver operating characteristic (ROC) curves. These curves not only allowed us to assess sensitivity and specificity but also provided a comprehensive understanding of the Area Under the Curve (AUC), which serves as a valuable metric for quantifying the overall performance of each algorithm in distinguishing between positive and negative precipitation events. The AUC offers a nuanced perspective on the discriminatory power of the models, aiding in the identification of the most adept algorithm for capturing extreme precipitation occurrences in the North Indian region. In addition to AUC, we also analysed Brier\u0026rsquo;s Score. Brier Score measures the mean squared difference between the predicted probability of an extreme event and the actual outcome (0 or 1). A lower Brier Scores indicate better model performance, signifying predictions closer to actual occurrences.\u003c/p\u003e"},{"header":"Declarations","content":"\u003cp\u003e\u003cstrong\u003eAcknowledgement\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eAll the authors acknowledge NASA\u0026rsquo;s POWER Data Access Viewer for providing the climate data. KCP acknowledge the support from the NERC (UK Natural Environment Research Council) AMAZONICA and Amazon Hydrological Cycle grants (NE/F005806/1 and NE/K01353X/1). AT and AA thankful to the University of Petroleum and Energy Studies, Dehradun for providing research facilities. \u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eAuthor contribution \u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eAT and KCP conceived the study. AT performed the analyses and wrote the initial draft of the manuscript. All authors contributed to the interpretation of the results, discussion of the associated mechanisms, and refinement of the paper. \u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eConflict of Interest \u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eThe authors declare no competing interest.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eData Availability Statement\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eAll the data utilized in this study are openly accessible and can be obtained by contacting the first author, Aayushi Tandon, via email at [email protected], or directly from the source: NASA\u0026rsquo;s POWER Data Access Viewer at https://power.larc.nasa.gov.\u003c/p\u003e"},{"header":"References","content":"\u003col\u003e\n\u003cli\u003eBhattacharyya, S., Sreekesh, S. \u0026amp; King, A. Characteristics of extreme rainfall in different gridded datasets over India during 1983\u0026ndash;2015. \u003cem\u003eAtmospheric Research\u003c/em\u003e \u003cstrong\u003e267\u003c/strong\u003e, 105930 (2022).\u003c/li\u003e\n\u003cli\u003eFowler, H. J. \u003cem\u003eet al.\u003c/em\u003e Anthropogenic intensification of short-duration rainfall extremes. \u003cem\u003eNat Rev Earth Environ\u003c/em\u003e \u003cstrong\u003e2\u003c/strong\u003e, 107\u0026ndash;122 (2021).\u003c/li\u003e\n\u003cli\u003eMishra, V., Aadhar, S. \u0026amp; Mahto, S. S. Anthropogenic warming and intraseasonal summer monsoon variability amplify the risk of future flash droughts in India. \u003cem\u003enpj Clim Atmos Sci\u003c/em\u003e \u003cstrong\u003e4\u003c/strong\u003e, 1\u0026ndash;10 (2021).\u003c/li\u003e\n\u003cli\u003eSuman, M. \u0026amp; Maity, R. Southward shift of precipitation extremes over south Asia: Evidences from CORDEX data. \u003cem\u003eSci Rep\u003c/em\u003e \u003cstrong\u003e10\u003c/strong\u003e, 6452 (2020).\u003c/li\u003e\n\u003cli\u003eMukherjee, S., Aadhar, S., Stone, D. \u0026amp; Mishra, V. Increase in extreme precipitation events under anthropogenic warming in India. \u003cem\u003eWeather and Climate Extremes\u003c/em\u003e \u003cstrong\u003e20\u003c/strong\u003e, 45\u0026ndash;53 (2018).\u003c/li\u003e\n\u003cli\u003eAnnex IX: Contributors to the IPCC Working Group I Sixth Assessment Report. in \u003cem\u003eClimate Change 2021: The Physical Science Basis. Contribution of Working Group I to the Sixth Assessment Report of the Intergovernmental Panel on Climate Change\u003c/em\u003e (eds. Masson-Delmotte, V. et al.) 2267\u0026ndash;2286 (Cambridge University Press, Cambridge, United Kingdom and New York, NY, USA, 2021). doi:10.1017/9781009157896.001.\u003c/li\u003e\n\u003cli\u003eP\u0026ouml;rtner, H. O. \u003cem\u003eet al.\u003c/em\u003e IPCC, 2022: summary for policymakers. in (eds. P\u0026ouml;rtner, H. O. et al.) 3\u0026ndash;33 (Cambridge University Pres, Cambridge, UK and New York, NY, US, 2022).\u003c/li\u003e\n\u003cli\u003eSikka, D. R., Ray, K., Chakravarthy, K., Bhan, S. C. \u0026amp; Tyagi, A. Heavy rainfall in the Kedarnath valley of Uttarakhand during the advancing monsoon phase in June 2013. \u003cem\u003eCurrent Science\u003c/em\u003e \u003cstrong\u003e109\u003c/strong\u003e, 353\u0026ndash;361 (2015).\u003c/li\u003e\n\u003cli\u003eAnoop Kumar Mishra \u0026amp; J Srinivasan. Did a cloud burst occur in Kedarnath during 16 and 17 June 2013. (2013).\u003c/li\u003e\n\u003cli\u003eRawat, K. S., Sahu, S. R., Singh, S. K. \u0026amp; Mishra, A. K. Cloudburst analysis in the Nainital district, Himalayan Region, 2021. \u003cem\u003eDiscov Water\u003c/em\u003e \u003cstrong\u003e2\u003c/strong\u003e, 12 (2022).\u003c/li\u003e\n\u003cli\u003eSingh, B., Thapliyal, R. \u0026amp; IMD, M. Cloudburst events observed over Uttarakhand during monsoon season 2017 and their analysis. \u003cem\u003eMAUSAM\u003c/em\u003e \u003cstrong\u003e73\u003c/strong\u003e, 91\u0026ndash;104 (2022).\u003c/li\u003e\n\u003cli\u003eChevuturi, A. \u0026amp; Dimri, A. P. Investigation of Uttarakhand (India) disaster-2013 using weather research and forecasting model. \u003cem\u003eNat Hazards\u003c/em\u003e \u003cstrong\u003e82\u003c/strong\u003e, 1703\u0026ndash;1726 (2016).\u003c/li\u003e\n\u003cli\u003eDNA. Cloudburst in Himachal Pradesh\u0026rsquo;s Mandi district kills 5. https://www.dnaindia.com/india/report-cloudburst-in-himachal-pradesh-s-mandi-district-kills-5-2112258 (2015).\u003c/li\u003e\n\u003cli\u003eTOI. Himachal Pradesh cloudburst: 178 stranded people rescued from Lahaul-Spiti - India Today. https://www.indiatoday.in/india/story/himachal-pradesh-cloudburst-178-stranded-people-rescued-from-lahaul-spiti-1835222-2021-07-31 (2021).\u003c/li\u003e\n\u003cli\u003eBhan, S., Devrani, A. \u0026amp; Sinha, V. An analysis of monthly rainfall and the meteorological conditions associated with cloudburst over the dry region of Leh (Ladakh), India. \u003cem\u003eMausam\u003c/em\u003e \u003cstrong\u003e66578540\u003c/strong\u003e, 107\u0026ndash;122 (2015).\u003c/li\u003e\n\u003cli\u003eDimri, A. P. \u003cem\u003eet al.\u003c/em\u003e Cloudbursts in Indian Himalayas: A review. \u003cem\u003eEarth-Science Reviews\u003c/em\u003e \u003cstrong\u003e168\u003c/strong\u003e, 1\u0026ndash;23 (2017).\u003c/li\u003e\n\u003cli\u003eYadav, M., Dimri, A. P., Mal, S. \u0026amp; Maharana, P. Elevation-dependent precipitation in the Indian Himalayan Region. \u003cem\u003eTheor Appl Climatol\u003c/em\u003e \u003cstrong\u003e155\u003c/strong\u003e, 815\u0026ndash;828 (2024).\u003c/li\u003e\n\u003cli\u003eAdhikari, P. \u0026amp; Mejia, J. F. Influence of aerosols on clouds, precipitation and freezing level height over the foothills of the Himalayas during the Indian summer monsoon. \u003cem\u003eClim Dyn\u003c/em\u003e \u003cstrong\u003e57\u003c/strong\u003e, 395\u0026ndash;413 (2021).\u003c/li\u003e\n\u003cli\u003eMal, S. \u003cem\u003eet al.\u003c/em\u003e Determining the quasi monsoon front in the Indian Himalayas. \u003cem\u003eQuaternary International\u003c/em\u003e \u003cstrong\u003e599\u0026ndash;600\u003c/strong\u003e, 4\u0026ndash;14 (2021).\u003c/li\u003e\n\u003cli\u003eRangwala, I., Palazzi, E. \u0026amp; Miller, J. R. Projected Climate Change in the Himalayas during the Twenty-First Century. in \u003cem\u003eHimalayan Weather and Climate and their Impact on the Environment\u003c/em\u003e (eds. Dimri, A. P., Bookhagen, B., Stoffel, M. \u0026amp; Yasunari, T.) 51\u0026ndash;71 (Springer International Publishing, Cham, 2020). doi:10.1007/978-3-030-29684-1_4.\u003c/li\u003e\n\u003cli\u003eAcosta, R. P. \u0026amp; Huber, M. Competing Topographic Mechanisms for the Summer Indo-Asian Monsoon. \u003cem\u003eGeophysical Research Letters\u003c/em\u003e \u003cstrong\u003e47\u003c/strong\u003e, e2019GL085112 (2020).\u003c/li\u003e\n\u003cli\u003eJ. A. Para \u003cem\u003eet al.\u003c/em\u003e Large-scale dynamics of western disturbances caused extreme precipitation on 24\u0026ndash;27 January 2017 over Jammu and Kashmir, India. \u003cem\u003eModeling Earth Systems and Environment\u003c/em\u003e \u003cstrong\u003e6\u003c/strong\u003e, 99\u0026ndash;107 (2020).\u003c/li\u003e\n\u003cli\u003eAggarwal, D., Chakraborty, R. \u0026amp; Attada, R. Investigating bi-decadal precipitation changes over the Northwest Himalayas during the pre-monsoon: role of Pacific decadal oscillations. \u003cem\u003eClim Dyn\u003c/em\u003e \u003cstrong\u003e62\u003c/strong\u003e, 1203\u0026ndash;1218 (2024).\u003c/li\u003e\n\u003cli\u003eHunt, K. M. R., Turner, A. G. \u0026amp; Schiemann, R. K. H. How Interactions between Tropical Depressions and Western Disturbances Affect Heavy Precipitation in South Asia. \u003cem\u003eMonthly Weather Review\u003c/em\u003e \u003cstrong\u003e149\u003c/strong\u003e, 1801\u0026ndash;1825 (2021).\u003c/li\u003e\n\u003cli\u003eKumar, P. V., Naidu, C. V. \u0026amp; Prasanna, K. Recent unprecedented weakening of Indian summer monsoon in warming environment. \u003cem\u003eTheor Appl Climatol\u003c/em\u003e \u003cstrong\u003e140\u003c/strong\u003e, 467\u0026ndash;486 (2020).\u003c/li\u003e\n\u003cli\u003eSingh, R., Jaiswal, N. \u0026amp; Kishtawal, C. M. Rising surface pressure over Tibetan Plateau strengthens indian summer monsoon rainfall over northwestern India. \u003cem\u003eSci Rep\u003c/em\u003e \u003cstrong\u003e12\u003c/strong\u003e, 8621 (2022).\u003c/li\u003e\n\u003cli\u003eGupta, A. K., Prakasam, M., Dutt, S., Clift, P. D. \u0026amp; Yadav, R. R. Evolution and Development of the Indian Monsoon. in \u003cem\u003eGeodynamics of the Indian Plate: Evolutionary Perspectives\u003c/em\u003e (eds. Gupta, N. \u0026amp; Tandon, S. K.) 499\u0026ndash;535 (Springer International Publishing, Cham, 2020). doi:10.1007/978-3-030-15989-4_14.\u003c/li\u003e\n\u003cli\u003eHahn, D. G. \u0026amp; Manabe, S. The Role of Mountains in the South Asian Monsoon Circulation. \u003cem\u003eJournal of the Atmospheric Sciences\u003c/em\u003e \u003cstrong\u003e32\u003c/strong\u003e, 1515\u0026ndash;1541 (1975).\u003c/li\u003e\n\u003cli\u003eAnu Xavier, M. G. Manoj, \u0026amp; K. Mohankumar. On the dynamics of an extreme rainfall event in northern India in 2013. \u003cem\u003eJournal of Earth System Science\u003c/em\u003e \u003cstrong\u003e127\u003c/strong\u003e, 30 (2018).\u003c/li\u003e\n\u003cli\u003ePattnayak, K. C., Awasthi, A., Sharma, K. \u0026amp; Pattnayak, B. B. Fate of Rainfall Over the North Indian States in the 1.5 and 2\u0026deg;C Warming Scenarios. \u003cem\u003eEarth and Space Science\u003c/em\u003e \u003cstrong\u003e10\u003c/strong\u003e, e2022EA002671 (2023).\u003c/li\u003e\n\u003cli\u003eAlizadeh, O. Advances and challenges in climate modeling. \u003cem\u003eClimatic Change\u003c/em\u003e \u003cstrong\u003e170\u003c/strong\u003e, 18 (2022).\u003c/li\u003e\n\u003cli\u003eMorrison, H. \u003cem\u003eet al.\u003c/em\u003e Confronting the Challenge of Modeling Cloud and Precipitation Microphysics. \u003cem\u003eJournal of Advances in Modeling Earth Systems\u003c/em\u003e \u003cstrong\u003e12\u003c/strong\u003e, e2019MS001689 (2020).\u003c/li\u003e\n\u003cli\u003eWang, H. \u003cem\u003eet al.\u003c/em\u003e Predicting climate anomalies: A real challenge. \u003cem\u003eAtmospheric and Oceanic Science Letters\u003c/em\u003e \u003cstrong\u003e15\u003c/strong\u003e, 100115 (2022).\u003c/li\u003e\n\u003cli\u003eYakubu, A. T., Abayomi, A. \u0026amp; Chetty, N. Machine Learning-Based Precipitation Prediction Using Cloud Properties. in \u003cem\u003eHybrid Intelligent Systems\u003c/em\u003e (eds. Abraham, A. et al.) 243\u0026ndash;252 (Springer International Publishing, Cham, 2022). doi:10.1007/978-3-030-96305-7_23.\u003c/li\u003e\n\u003cli\u003eGibson, P. B. \u003cem\u003eet al.\u003c/em\u003e Training machine learning models on climate model output yields skillful interpretable seasonal precipitation forecasts. \u003cem\u003eCommun Earth Environ\u003c/em\u003e \u003cstrong\u003e2\u003c/strong\u003e, 1\u0026ndash;13 (2021).\u003c/li\u003e\n\u003cli\u003eMansfield, L. A. \u003cem\u003eet al.\u003c/em\u003e Predicting global patterns of long-term climate change from short-term simulations using machine learning. \u003cem\u003enpj Clim Atmos Sci\u003c/em\u003e \u003cstrong\u003e3\u003c/strong\u003e, 1\u0026ndash;9 (2020).\u003c/li\u003e\n\u003cli\u003eTandon, A., Awasthi, A. \u0026amp; Pattnayak, K. C. Comparison of different Machine Learning methods on Precipitation dataset for Uttarakhand. in \u003cem\u003e2023 2nd International Conference on Ambient Intelligence in Health Care (ICAIHC)\u003c/em\u003e 1\u0026ndash;6 (2023). doi:10.1109/ICAIHC59020.2023.10431402.\u003c/li\u003e\n\u003cli\u003eWang, B., Wei, W., Yin, Z. \u0026amp; Xu, L. Using machine learning to analyze the changes in extreme precipitation in southern China. \u003cem\u003eAtmospheric Research\u003c/em\u003e \u003cstrong\u003e302\u003c/strong\u003e, 107307 (2024).\u003c/li\u003e\n\u003cli\u003eLin, X., Fan, J., Hou, Z. J. \u0026amp; Wang, J. Machine Learning of Key Variables Impacting Extreme Precipitation in Various Regions of the Contiguous United States. \u003cem\u003eJournal of Advances in Modeling Earth Systems\u003c/em\u003e \u003cstrong\u003e15\u003c/strong\u003e, e2022MS003334 (2023).\u003c/li\u003e\n\u003cli\u003eVitanza, E., Dimitri, G. M. \u0026amp; Mocenni, C. A multi-modal machine learning approach to detect extreme rainfall events in Sicily. \u003cem\u003eSci Rep\u003c/em\u003e \u003cstrong\u003e13\u003c/strong\u003e, 1\u0026ndash;15 (2023).\u003c/li\u003e\n\u003cli\u003eHu, H. \u0026amp; Ayyub, B. M. Machine Learning for Projecting Extreme Precipitation Intensity for Short Durations in a Changing Climate. \u003cem\u003eGeosciences\u003c/em\u003e \u003cstrong\u003e9\u003c/strong\u003e, 209 (2019).\u003c/li\u003e\n\u003cli\u003eChauhan, A. S., Singh, S., Maurya, R. K. S., Rani, A. \u0026amp; Danodia, A. Spatio-temporal trend analysis and future projections of precipitation at regional scale: a case study of Haryana, India. \u003cem\u003eJournal of Water and Climate Change\u003c/em\u003e \u003cstrong\u003e13\u003c/strong\u003e, 2143\u0026ndash;2170 (2022).\u003c/li\u003e\n\u003cli\u003eMalik, A. \u0026amp; Kumar, A. Spatio-temporal trend analysis of rainfall using parametric and non-parametric tests: case study in Uttarakhand, India. \u003cem\u003eTheor Appl Climatol\u003c/em\u003e \u003cstrong\u003e140\u003c/strong\u003e, 183\u0026ndash;207 (2020).\u003c/li\u003e\n\u003cli\u003eJaswal, A., Bhan, S., Karandikar, A. \u0026amp; Gujar, M. Seasonal and annual rainfall trends in Himachal Pradesh during 1951-2005. \u003cem\u003eMausam\u003c/em\u003e \u003cstrong\u003e66\u003c/strong\u003e, 247\u0026ndash;264 (2015).\u003c/li\u003e\n\u003cli\u003eAgarwal, S., Mukherjee, D. \u0026amp; Debbarma, N. Analysis of extreme annual rainfall in North-Eastern India using machine learning techniques. \u003cem\u003eAQUA - Water Infrastructure, Ecosystems and Society\u003c/em\u003e \u003cstrong\u003e72\u003c/strong\u003e, 2201\u0026ndash;2215 (2023).\u003c/li\u003e\n\u003cli\u003ePraveen, B. \u003cem\u003eet al.\u003c/em\u003e Analyzing trend and forecasting of rainfall changes in India using non-parametrical and machine learning approaches. \u003cem\u003eSci Rep\u003c/em\u003e \u003cstrong\u003e10\u003c/strong\u003e, 10342 (2020).\u003c/li\u003e\n\u003cli\u003eRay, R. \u0026amp; Chakraborty, S. A Discussion on Machine Learning Approach of Rainfall Prediction. in \u003cem\u003e2022 URSI Regional Conference on Radio Science (USRI-RCRS)\u003c/em\u003e 1\u0026ndash;4 (2022). doi:10.23919/URSI-RCRS56822.2022.10118437.\u003c/li\u003e\n\u003cli\u003eLighthill, J. \u0026amp; Pearce, R. P. \u003cem\u003eMonsoon Dynamics\u003c/em\u003e. (Cambridge University Press, 1981).\u003c/li\u003e\n\u003cli\u003eAwasthi, A., Pattnayak, K. C., Tandon, A., Sarkar, A. \u0026amp; Chakraborty, M. Implications of climate change on surface temperature in North Indian states: evidence from CMIP6 model ensembles. \u003cem\u003eFront. Environ. Sci.\u003c/em\u003e \u003cstrong\u003e11\u003c/strong\u003e, (2023).\u003c/li\u003e\n\u003cli\u003eClimate-Know India, I. Profile - Climate - Know India: National Portal of India. https://knowindia.india.gov.in/profile/climate.php (2024).\u003c/li\u003e\n\u003cli\u003eGelaro, R. \u003cem\u003eet al.\u003c/em\u003e The Modern-Era Retrospective Analysis for Research and Applications, Version 2 (MERRA-2). \u003cem\u003eJournal of Climate\u003c/em\u003e \u003cstrong\u003e30\u003c/strong\u003e, 5419\u0026ndash;5454 (2017).\u003c/li\u003e\n\u003cli\u003eAgarwal, P., Stevenson, D. S. \u0026amp; Heal, M. R. Evaluation of WRF-Chem-simulated meteorology and aerosols over northern India during the severe pollution episode of 2016. \u003cem\u003eAtmospheric Chemistry and Physics\u003c/em\u003e \u003cstrong\u003e24\u003c/strong\u003e, 2239\u0026ndash;2266 (2024).\u003c/li\u003e\n\u003cli\u003eDai, Z. \u003cem\u003eet al.\u003c/em\u003e Long-Term MERRA-2 Reanalysis Data Indicate Atmospheric Environmental Changes for Three Major Concentrating-Solar-Power-Plant Project Areas in Xinjiang, China. \u003cem\u003eAtmosphere\u003c/em\u003e \u003cstrong\u003e14\u003c/strong\u003e, 1700 (2023).\u003c/li\u003e\n\u003cli\u003eWen, A. \u003cem\u003eet al.\u003c/em\u003e Evaluation of MERRA-2 land surface temperature dataset and its application in permafrost mapping over China. \u003cem\u003eAtmospheric Research\u003c/em\u003e \u003cstrong\u003e279\u003c/strong\u003e, 106373 (2022).\u003c/li\u003e\n\u003cli\u003eRajeev, A., Singh, C., Singh, S. K. \u0026amp; Chauhan, P. Assessment of WRF-CHEM Simulated Dust Using Reanalysis, Satellite Data and Ground-Based Observations. \u003cem\u003eJ Indian Soc Remote Sens\u003c/em\u003e \u003cstrong\u003e49\u003c/strong\u003e, 1545\u0026ndash;1559 (2021).\u003c/li\u003e\n\u003cli\u003eGupta, P. \u003cem\u003eet al.\u003c/em\u003e Validation of Surface Temperature Derived From MERRA-2 Reanalysis Against IMD Gridded Data Set Over India. \u003cem\u003eEarth and Space Science\u003c/em\u003e \u003cstrong\u003e7\u003c/strong\u003e, e2019EA000910 (2020).\u003c/li\u003e\n\u003cli\u003eFeng, C. \u003cem\u003eet al.\u003c/em\u003e Comprehensive assessment of global atmospheric downward longwave radiation in the state-of-the-art reanalysis using satellite and flux tower observations. \u003cem\u003eClim Dyn\u003c/em\u003e \u003cstrong\u003e60\u003c/strong\u003e, 1495\u0026ndash;1521 (2023).\u003c/li\u003e\n\u003cli\u003eSchmeisser, L., Hinkelman, L. M. \u0026amp; Ackerman, T. P. Evaluation of Radiation and Clouds From Five Reanalysis Products in the Northeast Pacific Ocean. \u003cem\u003eJournal of Geophysical Research: Atmospheres\u003c/em\u003e \u003cstrong\u003e123\u003c/strong\u003e, 7238\u0026ndash;7253 (2018).\u003c/li\u003e\n\u003cli\u003eDoelling, D. R. \u003cem\u003eet al.\u003c/em\u003e Advances in Geostationary-Derived Longwave Fluxes for the CERES Synoptic (SYN1deg) Product. \u003cem\u003eJournal of Atmospheric and Oceanic Technology\u003c/em\u003e \u003cstrong\u003e33\u003c/strong\u003e, 503\u0026ndash;521 (2016).\u003c/li\u003e\n\u003cli\u003eKato, S. \u003cem\u003eet al.\u003c/em\u003e Surface Irradiances Consistent with CERES-Derived Top-of-Atmosphere Shortwave and Longwave Irradiances. \u003cem\u003eJournal of Climate\u003c/em\u003e \u003cstrong\u003e26\u003c/strong\u003e, 2719\u0026ndash;2740 (2013).\u003c/li\u003e\n\u003cli\u003eHamal, K. \u003cem\u003eet al.\u003c/em\u003e Evaluation of MERRA-2 Precipitation Products Using Gauge Observation in Nepal. \u003cem\u003eHydrology\u003c/em\u003e \u003cstrong\u003e7\u003c/strong\u003e, 40 (2020).\u003c/li\u003e\n\u003cli\u003eXu, X., Frey, S. K. \u0026amp; Ma, D. Hydrological performance of ERA5 and MERRA-2 precipitation products over the Great Lakes Basin. \u003cem\u003eJournal of Hydrology: Regional Studies\u003c/em\u003e \u003cstrong\u003e39\u003c/strong\u003e, 100982 (2022).\u003c/li\u003e\n\u003cli\u003eSong, Y. \u0026amp; Wei, J. Diurnal cycle of summer precipitation over the North China Plain and associated land\u0026ndash;atmosphere interactions: Evaluation of ERA5 and MERRA-2. \u003cem\u003eInternational Journal of Climatology\u003c/em\u003e \u003cstrong\u003e41\u003c/strong\u003e, 6031\u0026ndash;6046 (2021).\u003c/li\u003e\n\u003cli\u003eLe, M.-H. \u003cem\u003eet al.\u003c/em\u003e Assessment of drought conditions over Vietnam using standardized precipitation evapotranspiration index, MERRA-2 re-analysis, and dynamic land cover. \u003cem\u003eJournal of Hydrology: Regional Studies\u003c/em\u003e \u003cstrong\u003e32\u003c/strong\u003e, 100767 (2020).\u003c/li\u003e\n\u003cli\u003eIgbawua, T., Zhang, J., Yao, F. \u0026amp; Zhang, D. Assessment of moisture budget over West Africa using MERRA-2\u0026rsquo;s aerological model and satellite data. \u003cem\u003eClim Dyn\u003c/em\u003e \u003cstrong\u003e52\u003c/strong\u003e, 83\u0026ndash;106 (2019).\u003c/li\u003e\n\u003cli\u003eCuo, L. \u0026amp; Zhang, Y. Spatial patterns of wet season precipitation vertical gradients on the Tibetan Plateau and the surroundings. \u003cem\u003eSci Rep\u003c/em\u003e \u003cstrong\u003e7\u003c/strong\u003e, 5057 (2017).\u003c/li\u003e\n\u003cli\u003eWei, J., Dirmeyer, P. A., Wisser, D., Bosilovich, M. G. \u0026amp; Mocko, D. M. Where Does the Irrigation Water Go? An Estimate of the Contribution of Irrigation to Precipitation Using MERRA. \u003cem\u003eJournal of Hydrometeorology\u003c/em\u003e \u003cstrong\u003e14\u003c/strong\u003e, 275\u0026ndash;289 (2013).\u003c/li\u003e\n\u003cli\u003eSenthilnathan, S. Usefulness of Correlation Analysis. \u003cem\u003eSSRN Electronic Journal\u003c/em\u003e (2019) doi:10.2139/ssrn.3416918.\u003c/li\u003e\n\u003cli\u003eShetty, S., Umesh, P. \u0026amp; Shetty, A. The effectiveness of machine learning-based multi-model ensemble predictions of CMIP6 in Western Ghats of India. \u003cem\u003eInternational Journal of Climatology\u003c/em\u003e \u003cstrong\u003e43\u003c/strong\u003e, 5029\u0026ndash;5054 (2023).\u003c/li\u003e\n\u003cli\u003eSharif, M. \u003cem\u003eSimulation of Extreme Precipitation Events Using an Improved K-Nearest Neighbor Model\u003c/em\u003e. 389 (2023). doi:10.1061/9780784484852.036.\u003c/li\u003e\n\u003cli\u003eK, K., N C, V., P, P. \u0026amp; Selvapandian, D. An Efficient Rainfall Forecasting System using Machine Learning Methods. in \u003cem\u003e2023 Second International Conference on Electrical, Electronics, Information and Communication Technologies (ICEEICT)\u003c/em\u003e 01\u0026ndash;06 (2023). doi:10.1109/ICEEICT56924.2023.10157395.\u003c/li\u003e\n\u003cli\u003eAcharya, N. \u0026amp; Hall, K. A Machine Learning Approach for Probabilistic Multi-Model Ensemble Predictions of Indian Summer Monsoon Rainfall. \u003cem\u003eMAUSAM\u003c/em\u003e \u003cstrong\u003e74\u003c/strong\u003e, 421\u0026ndash;428 (2023).\u003c/li\u003e\n\u003cli\u003eJose, D. M., Vincent, A. M. \u0026amp; Dwarakish, G. S. Improving multiple model ensemble predictions of daily precipitation and temperature through machine learning techniques. \u003cem\u003eSci Rep\u003c/em\u003e \u003cstrong\u003e12\u003c/strong\u003e, 4678 (2022).\u003c/li\u003e\n\u003cli\u003eMooers, G., Beucler, T., Pritchard, M. \u0026amp; Mandt, S. Understanding Extreme Precipitation Changes through Unsupervised Machine Learning. Preprint at https://doi.org/10.48550/arXiv.2211.01613 (2023).\u003c/li\u003e\n\u003cli\u003eWei, W. \u003cem\u003eet al.\u003c/em\u003e Seasonal prediction of summer extreme precipitation over the Yangtze River based on random forest. \u003cem\u003eWeather and Climate Extremes\u003c/em\u003e \u003cstrong\u003e37\u003c/strong\u003e, 100477 (2022).\u003c/li\u003e\n\u003cli\u003eHill, A. J. \u0026amp; Schumacher, R. S. Forecasting Excessive Rainfall with Random Forests and a Deterministic Convection-Allowing Model. \u003cem\u003eWeather and Forecasting\u003c/em\u003e \u003cstrong\u003e36\u003c/strong\u003e, 1693\u0026ndash;1711 (2021).\u003c/li\u003e\n\u003cli\u003ePham, Q. B., Yang, T.-C., Kuo, C.-M., Tseng, H.-W. \u0026amp; Yu, P.-S. Combing Random Forest and Least Square Support Vector Regression for Improving Extreme Rainfall Downscaling. \u003cem\u003eWater\u003c/em\u003e \u003cstrong\u003e11\u003c/strong\u003e, 451 (2019).\u003c/li\u003e\n\u003cli\u003eGrazzini, F., Grams, C. \u0026amp; Craig, G. Classification and prediction of days with extreme rainfall using random forest approach. EGU22-11954 (2022) doi:10.5194/egusphere-egu22-11954.\u003c/li\u003e\n\u003cli\u003eLi, K., Huang, G., Wang, S., Razavi, S. \u0026amp; Zhang, X. Development of a Joint Probabilistic Rainfall-Runoff Model for High-to-Extreme Flow Projections Under Changing Climatic Conditions. \u003cem\u003eWater Resources Research\u003c/em\u003e \u003cstrong\u003e58\u003c/strong\u003e, e2021WR031557 (2022).\u003c/li\u003e\n\u003c/ol\u003e"}],"fulltextSource":"","fullText":"","funders":[],"hasAdminPriorityOnWorkflow":false,"hasManuscriptDocX":true,"hasOptedInToPreprint":true,"hasPassedJournalQc":"","hasAnyPriority":false,"hideJournal":false,"highlight":"","institution":"","isAcceptedByJournal":true,"isAuthorSuppliedPdf":false,"isDeskRejected":"","isHiddenFromSearch":false,"isInQc":false,"isInWorkflow":false,"isPdf":false,"isPdfUpToDate":true,"isWithdrawnOrRetracted":false,"journal":{"display":true,"email":"[email protected]","identity":"scientific-reports","isNatureJournal":false,"hasQc":true,"allowDirectSubmit":false,"externalIdentity":"scirep","sideBox":"Learn more about [Scientific Reports](http://www.nature.com/srep/)","snPcode":"","submissionUrl":"","title":"Scientific Reports","twitterHandle":"","acdcEnabled":true,"dfaEnabled":true,"editorialSystem":"stoa","reportingPortfolio":"Scientific Reports","inReviewEnabled":true,"inReviewRevisionsEnabled":true},"keywords":"Climate Change, Machine Learning, North Indian States, Precipitation Patterns, Random Forest, Support Vector Machine","lastPublishedDoi":"10.21203/rs.3.rs-4339400/v1","lastPublishedDoiUrl":"https://doi.org/10.21203/rs.3.rs-4339400/v1","license":{"name":"CC BY 4.0","url":"https://creativecommons.org/licenses/by/4.0/"},"manuscriptAbstract":"\u003cp\u003eClimate change-induced precipitation extremes have become a pressing global concern. This study investigate the predictability of precipitation patterns and its extremes using MERRA2 datasets across North Indian states for the period 1984 to 2022 utilizing machine learning (ML) models. A strong positive correlations of precipitation 0.4 was found with dew point temperature and relative humidity significant at 0.05. In simulating precipitation, Random Forest Classifier (RFC) achieved the highest accuracy (~\u0026thinsp;83%) for Rajasthan and Uttar Pradesh, while Support Vector Classifier (SVC) performed best (79\u0026ndash;83% accuracy) for other states. However, the ML models exhibited about 5% lower skill in higher elevated stations as compared to the lower elevated stations, its due to the different atmospheric mechanisms control differently over the lower and higher topography. For extreme precipitation events (10th and 95th percentiles of intensity), RFC consistently outperformed SVC across all states. It demonstrated superior ability to distinguish extreme from non-extreme events (Area under curve\u0026thinsp;~\u0026thinsp;0.90) and better model calibration (Brier Scores\u0026thinsp;~\u0026thinsp;0.01). The developed ML models successfully simulated precipitation and extreme patterns, with RFC excelling at predicting extreme precipitation events. These findings can contribute to disaster preparedness and water resource management efforts in the region with varied topography and complex terrain.\u003c/p\u003e","manuscriptTitle":"Efficacy of Machine Learning in Simulating Precipitation and Its Extremes Over the Capital Cities in North Indian States","msid":"","msnumber":"","nonDraftVersions":[{"code":1,"date":"2024-05-08 12:19:25","doi":"10.21203/rs.3.rs-4339400/v1","editorialEvents":[{"type":"communityComments","content":0},{"type":"decision","content":"Revision requested","date":"2024-06-12T04:04:59+00:00","index":"","fulltext":""},{"type":"editorInvitedReview","content":"","date":"2024-06-07T18:23:38+00:00","index":"hide","fulltext":""},{"type":"editorInvitedReview","content":"","date":"2024-05-27T02:01:33+00:00","index":"hide","fulltext":""},{"type":"reviewerAgreed","content":"12202137733349679259501058440226373237","date":"2024-05-13T13:32:07+00:00","index":"hide","fulltext":""},{"type":"reviewerAgreed","content":"190853698189011538209043195793937652775","date":"2024-05-10T00:40:06+00:00","index":"hide","fulltext":""},{"type":"reviewersInvited","content":"","date":"2024-05-07T11:30:06+00:00","index":"","fulltext":""},{"type":"editorAssigned","content":"","date":"2024-05-07T11:14:53+00:00","index":"","fulltext":""},{"type":"editorInvited","content":"","date":"2024-05-02T13:26:32+00:00","index":"","fulltext":""},{"type":"checksComplete","content":"","date":"2024-05-02T13:23:34+00:00","index":"","fulltext":""},{"type":"submitted","content":"Scientific Reports","date":"2024-04-28T21:58:18+00:00","index":"","fulltext":""}],"status":"published","journal":{"display":true,"email":"[email protected]","identity":"scientific-reports","isNatureJournal":false,"hasQc":true,"allowDirectSubmit":false,"externalIdentity":"scirep","sideBox":"Learn more about [Scientific Reports](http://www.nature.com/srep/)","snPcode":"","submissionUrl":"","title":"Scientific Reports","twitterHandle":"","acdcEnabled":true,"dfaEnabled":true,"editorialSystem":"stoa","reportingPortfolio":"Scientific Reports","inReviewEnabled":true,"inReviewRevisionsEnabled":true}}],"origin":"","ownerIdentity":"f2595725-cb34-428f-88fa-72ae2ecfd666","owner":[],"postedDate":"May 8th, 2024","published":true,"recentEditorialEvents":[],"rejectedJournal":[],"revision":"","amendment":"","status":"published-in-journal","subjectAreas":[],"tags":[],"updatedAt":"2025-03-31T15:58:11+00:00","versionOfRecord":{"articleIdentity":"rs-4339400","link":"https://doi.org/10.1038/s41598-024-84360-w","journal":{"identity":"scientific-reports","isVorOnly":false,"title":"Scientific Reports"},"publishedOn":"2025-03-25 15:56:51","publishedOnDateReadable":"March 25th, 2025"},"versionCreatedAt":"2024-05-08 12:19:25","video":"","vorDoi":"10.1038/s41598-024-84360-w","vorDoiUrl":"https://doi.org/10.1038/s41598-024-84360-w","workflowStages":[]},"version":"v1","identity":"rs-4339400","journalConfig":"researchsquare"},"__N_SSP":true},"page":"/article/[identity]/[[...version]]","query":{"redirect":"/article/rs-4339400","identity":"rs-4339400","version":["v1"]},"buildId":"qtupq5eGEP_6zYnWcrvyt","isFallback":false,"isExperimentalCompile":false,"dynamicIds":[84888],"gssp":true,"scriptLoader":[]}