A Forecast-Enhanced Hybrid Model for Adaptive Irrigation Management in Arecanut Cultivation

Abstract

Arecanut cultivation in the coastal regions of India, especially in areas such as coastal Karnataka and parts of Kerala, faces increasing challenges due to uncertainty in the rainfall, increasing temperatures, and scarcity of groundwater. Although sensor-based irrigation systems have improved responsiveness to soil moisture conditions, Irrigation scheduling driven purely by real-time sensor readings usually leads to unnecessary watering before rainfall or excess moisture during continuous wet periods, increasing the risk of fungal and root-related diseases. To address these challenges, this study proposes a forecast-driven hybrid irrigation framework that integrates predictive weather insights with existing optimization-based irrigation mechanisms to enhance efficiency of usage of water and reliability of irrigation control. The framework extends a conventional optimization model by incorporating short-term rainfall and temperature forecasts generated through forecast modeling techniques. These forecasts, along with inputs such as soil moisture (TDR), reservoir water availability, and climatic variables, are processed using a hybrid stacked classification model combining Extra Trees, HistGradientBoosting, and Logistic Regression. Forecast information serves as a conditional reasoning layer to dynamically delay, advance, or suppress irrigation actions. Scenario-based simulations covering normal conditions, scarcity of water, rainfall expectation period, continuous rainfall, and post-rain recovery phases showed that the forecast-enhanced model improved decision accuracy from 96.51% to 97.59%, reduced false irrigation triggers by over 50%, and achieving approximately 8–10% additional water savings. Overall, the model demonstrated better adaptability during rainfall transitions and enhanced disease-prevention control under extended wet conditions.