Climate-Aware Predictive Forecasting for Maharashtra Crops: Integrating Statistical and Machine Learning Approaches

Abstract

Maharashtra is a key contributor to India\'s agricultural sector, with its productivity heavily influenced by climatic and environmental factors. This study examines the relationship between crop yield, production, and critical variables such as temperature, rainfall, irrigation, and nutrient consumption using data from 1966 to 2023. Correlations between yield, production, and factors like weather, fertilizers, and soil nutrients are analyzed. SHAP (SHapley Additive exPlanations) identifies the most influential factors, while the Apriori algorithm uncovers associations between agricultural attributes. For forecasting, machine learning models—RFR (Random Forest Regressor), SVR (Support Vector Regressor), and GBR (Gradient Boosting Regressor)—are compared, with GBR emerging as the best. STL (Seasonal and Trend decomposition using Loess) is applied to GBR\'s time series data to reveal trends and seasonal patterns. This comprehensive approach provides actionable insights for enhancing agricultural productivity and sustainability in Maharashtra.

Introduction

Agriculture is vital to India’s economy, providing livelihoods and food security, with Maharashtra being a key agricultural state featuring diverse crops like sugarcane, cotton, pulses, and cereals. However, the sector faces severe climate vulnerabilities, including erratic rainfall, rising temperatures, and frequent droughts, with regions such as Marathwada and Vidarbha experiencing extreme heat affecting soil and water resources. Climate change also drives innovation, prompting adoption of resilient crops, precision farming, and advanced irrigation. About 77% of Maharashtra’s cropped area is climate-vulnerable, according to a study analyzing 1966–2023 data using correlation, STL decomposition, and the Apriori algorithm, alongside machine learning models like Random Forest (RFR), Support Vector Regressor (SVR), and Gradient Boosting Regressor (GBR) for crop yield forecasting.

Related Work:
Prior studies focused on rainfall prediction, crop forecasting, and climate modeling using machine learning (ML) techniques such as MLP, SVM, Random Forest, LSTM, and Bayesian inference. While these studies demonstrated high predictive accuracy, they often suffered from limitations like outdated datasets, small sample sizes, lack of real-time weather integration, or omission of deep learning comparisons. Unlike previous research, this study provides a district-level, crop-specific, and climate-sensitive analysis for Maharashtra, combining advanced ML, pattern mining (Apriori), and STL decomposition for both retrospective and predictive insights.

Methodology:
Data from ICRISAT on agricultural outputs, climate, irrigation, fertilizers (NPK), and land use (1966–2023) was collected. Exploratory Data Analysis (EDA) and correlation analysis identified relationships between yield/production and climatic or nutrient factors. SHAP values explained feature importance, Apriori revealed associations between climate/fertilizers and yield, and machine learning models (RFR, SVR, GBR) predicted future yield trends. STL decomposition separated trends, seasonality, and residuals to validate forecasts.

Implementation:

• Data: Crop and aggregate datasets included major crops, pulses, oilseeds, cash crops, climate variables, fertilizer usage, and land-use metrics. Missing values were forward-filled.

• Statistical & Pattern Analysis: Correlation, SHAP, and Apriori identified key relationships and influential factors.

• Regression Models: RFR, SVR, and GBR were tuned and trained on district-level data, with performance evaluated using R² and RMSE.

• Forecasting & STL Decomposition: Models generated projections up to 2040, with STL validating seasonal and trend components.

Results:

• Correlation: Nitrogen and phosphate positively influenced crop yields, while extreme temperatures and unbalanced fertilizers negatively impacted production.

• SHAP Analysis: Key drivers of production included Nitrogen consumption and irrigated area, with precipitation being critical for rice production.

• Apriori Algorithm: Association rules highlighted how combinations of climatic parameters and fertilizers influence crop yields, providing insights into temperature and irrigation effects.

Conclusion

This study analyzed the impact of climatic and agricultural factors on crop yield and production using two datasets: an aggregate dataset and a crop-specific dataset. SHAP is applied to find the most influential features per attribute, highlighting importance. The Apriori algorithm identified key associations between attributes. For Yield, there is a positive correlation with Nitrogen Consumption (tons) and a negative correlation with Minimum Temperature (Celsius). For Production, it is positively correlated with Total Consumption (tons) and negatively correlated with Minimum Temperature (Celsius). Regression models, including RFR, SVR, and GBR, were employed to predict future values, with GBR emerging as the most effective across both datasets, and GBR\'s predictions align with the original data\'s trend, ensuring the model\'s accuracy and reliability. For the Aggregate dataset, the R² scores were 0.927 for RFR, 0.989 for SVR, and 0.963 for GBR, while the RMSE values were 378 for RFR, 573.33 for SVR, and 17.16 for GBR. For the Crop dataset, the R² scores were 0.967 for RFR, 0.976 for SVR, and 0.974 for GBR, with RMSE values of 21.23 for RFR, 72.23 for SVR, and 17.5 for GBR. GBR is the best model overall because it consistently achieves the lowest RMSE while maintaining high R² scores across both datasets. By applying STL (Seasonal and Trend decomposition using Loess) to GBR\'s predictions, the data is decomposed into trend, seasonal, and residual components. The alignment with the trend shows accuracy and reliability, ensuring that GBR\'s forecasts are consistent with historical patterns and capable of providing meaningful insights for future agricultural planning. The study presents a comprehensive evaluation of agricultural trends across 20 districts in Maharashtra, revealing mounting climate-related risks and production stress. The analysis identifies declining crop yields, climate variability, soil fertility concerns, and irrigation challenges that have begun to reshape the agricultural landscape of the region. These insights offer critical guidance for stakeholders seeking to ensure long-term agricultural sustainability. One of the most prominent trends observed is the consistent decline in rice and wheat production across several districts, including Ahmednagar, Aurangabad, Amravati, Jalgaon, and Solapur. This reduction appears to stem from rising temperatures, unpredictable rainfall, and a steady decrease in irrigated land. These crops, especially wheat, are highly sensitive to heat stress and require stable water availability, making them particularly vulnerable to the current climate conditions. In contrast, maize shows a more complex pattern. While some districts report a decline, others show increased production. This inconsistency may be due to maize’s greater resilience to varying rainfall patterns, suggesting that it may be a more adaptable cereal crop in the context of climate change. Sugarcane and cotton, in comparison, demonstrate relatively greater resilience. Cotton\'s inherent drought resistance makes it less dependent on consistent irrigation, allowing it to thrive even in water-scarce regions. Sugarcane, though water-intensive, has maintained a more stable output, though it remains at risk if irrigation continues to decline. Climate trends across the districts show a troubling rise in both maximum and minimum temperatures. This temperature increase exacerbates heat stress and directly affects yields, especially for temperature-sensitive crops like wheat. In parallel, rainfall patterns have become increasingly erratic. Districts such as Jalgaon and Satara face sharp rainfall declines, while others like Akola, Pune, and Kolhapur are experiencing irregular cycles of drought and flooding. This climate unpredictability disrupts planting and harvesting schedules and threatens to destabilize annual crop output. The situation is further aggravated by the ongoing reduction in irrigation coverage, especially in districts such as Ahmednagar, Solapur, Aurangabad, and Amravati. As more land becomes reliant on rainfall, water-intensive crops face heightened production risks. Simultaneously, the use of fertilizers—particularly nitrogen, phosphate, and potash—has been decreasing in several districts. This decline may be attributed to shifts in farming practices, rising costs, or underlying soil health problems, which could contribute to long-term reductions in yield due to nutrient deficiencies. Based on the combined effect of these variables, districts have been classified into different risk categories. High-risk districts include Ahmednagar, Aurangabad, Jalgaon, Solapur, Amravati, Nanded, Osmanabad, Chandrapur, Beed, and Gadchiroli. These areas suffer from a convergence of declining rice and wheat yields, increasing temperatures, reduced irrigation, and rainfall inconsistency. Medium-risk districts such as Akola, Nashik, Pune, Satara, Wardha, Latur, Buldhana, and Yavatmal experience moderate but growing impacts from climate variability, though some of them maintain relatively better irrigation or soil health. Kolhapur and Nagpur emerge as low-risk districts, with more stable climate trends, irrigation availability, and crop production. This study reinforces that climate change is no longer a distant threat—it is already affecting agricultural systems in tangible ways. Without timely interventions, the situation will worsen, particularly for smallholder farmers who lack access to adaptive technologies or data-driven support systems. The analysis highlights the urgent need for better water management, widespread adoption of climate-resilient crop varieties, and investment in precision agriculture. Monitoring soil health and managing fertilizer use effectively are equally critical to preserving long-term productivity. The findings of this study are valuable for a broad range of stakeholders. Farmers and agricultural cooperatives can use these insights to adjust their cropping strategies, adopt efficient irrigation practices, and anticipate yield risks based on district-specific climate patterns. In many high-risk regions, farmers may benefit from transitioning to less water-intensive crops or introducing drought-tolerant varieties. For policymakers and government agencies, the results offer a roadmap for designing proactive agricultural policies. These could include targeted subsidies, improved irrigation infrastructure, climate-smart extension services, and early warning systems for extreme weather. Most importantly, this research underscores the need to shift from reactive to preventive agricultural planning to secure food and economic security for rural communities.

References

[1] Rainfall Prediction for Enhancing Crop-Yield based on Machine Learning Techniques, S. Malathy, C. N. Vanitha, Kotteswari, S. V, S. P and M. E, 2022. [Online]. Available: https://doi.org/10.1109/ICAAIC53929.2022.9792793 [2] A Creative Use of Machine Learning for Crop Prediction and Analysis, R. C. Mahore and N. G. Gadge, 2022. [Online]. Available: https://doi.org/10.1109/ICWITE57052.2022.10176256 [3] Influence of Causal Inference For Crop Prediction, K. Srivastava, A. Sawarkar, S. Nawale and A. R. Agrawal, 2024. [Online]. Available: https://doi.org/10.1109/ICCCNT61001.2024.10725416 [4] A Case Study on the Application of Machine Learning to the Process of Crop Forecasting, S. A. Jiwani, J. Shinde, B. Bag, R. Nayak, R. K. Shial and U. Ghugar, 2024. [Online]. Available: https://doi.org/10.1109/OTCON60325.2024.10688298 [5] Climate Forecasting: Long Short Term Memory Model using Global Temperature Data, P. Akhila, R. L. S. Anjana and M. Kavitha, 2022. [Online]. Available: https://doi.org/10.1109/ICCMC53470.2022.9753779 [6] What is Exploratory Data Analysis?, GeeksforGeeks, 2024. [Online]. Available: https://www.geeksforgeeks.org/what-is-exploratory-data-analysis/ [7] ICRISAT - Science of Discovery to Science of Delivery, International Crops Research Institute for the Semi-Arid Tropics, 2024. [Online]. Available: https://www.icrisat.org/ [8] SHAP: A Comprehensive Guide to Shapley Additive Explanations, GeeksforGeeks, 2024. [Online]. Available: https://www.geeksforgeeks.org/shap-a-comprehensive-guide-to-shapley-additive-explanations/ [9] Apriori Algorithm, IBM, 2024. [Online]. Available: https://www.ibm.com/think/topics/apriori-algorithm [10] Random Forest Regression in Python, GeeksforGeeks, 2024. [Online]. Available: https://www.geeksforgeeks.org/random-forest-regression-in-python/ [11] RandomForestRegressor, scikit-learn, 2024. [Online]. Available: https://scikit-learn.org/1.5/modules/generated/sklearn.ensemble.RandomForestRegressor.html [12] Support Vector Regression (SVR) Using Linear and Non-Linear Kernels, GeeksforGeeks, 2024. [Online]. Available: https://www.geeksforgeeks.org/support-vector-regression-svr-using-linear-and-non-linear-kernels-in-scikit-learn/ [13] SVR, scikit-learn, 2024. [Online]. Available: https://scikit-learn.org/1.5/modules/generated/sklearn.svm.SVR.html [14] Gradient Boosting in Machine Learning, GeeksforGeeks, 2024. [Online]. Available: https://www.geeksforgeeks.org/ml-gradient-boosting/ [15] GradientBoostingRegressor, scikit-learn, 2024. [Online]. Available: https://scikit-learn.org/1.5/modules/generated/sklearn.ensemble.GradientBoostingRegressor.html#sklearn.ensemble.GradientBoostingRegressor [16] Regression Metrics, GeeksforGeeks, 2024. [Online]. Available: https://www.geeksforgeeks.org/regression-metrics/ [17] STL, ArcGIS Insights, 2024. [Online]. Available: https://doc.arcgis.com/en/insights/latest/analyze/stl.htm

Copyright

Copyright © 2025 Dr. Gresha Bhatia, Vishakha Singh, Manasi Sharma, Anushka Shirode. This is an open access article distributed under the Creative Commons Attribution License, which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly cited.