A Review of Predictive Models for Agro-Meteorological Data Using Machine Learning

Abstract: Karnataka's agricultural sector is very sensitive to climatic variability with regard to timing and distribution of monsoon rainfall. In response to the urgent need for reliable Agro-Meteorological forecasting, this study investigated the potential use of two supervised machine learning models Decision Tree and Random Forest for precipitation forecasting in Karnataka. The models were trained using historical meteorological datasets that were developed from the results of a number of climate parameters related to, for example, atmospheric circulation patterns, sea surface temperature, and specific monsoon-related indices. For both models, the combination of recursive partitioning and ensemble learning incorporated numerous distributed climate factors and their associated nonlinear relationships. Both models were then critically compared, with Random Forest being determined as most superior and a better overall estimation in terms of quality and reliability, trend association and being more robust to overfitting. In summary, these results demonstrate that data-driven approaches can dramatically improve the geographic and temporal quality of forecasts in this region. The framework proposed is a scalable, interpretable, and practical tool to aid climate resilient agricultural planning in Karnataka, and more generally Agro-climatic zones in India.

Keywords: Karnataka agriculture, climatic variability, monsoon rainfall forecasting, agro-meteorology, machine learning, Decision Tree, Random Forest, precipitation prediction, climate parameters, ensemble learning, nonlinear relationships, overfitting robustness, data-driven forecasting, climate-resilient agriculture, agro-climatic zones of India