Comparative Assessment of Advanced Machine Learning Algorithms for Predicting Canopy Cover, Biomass and Soil Water Content Under Stagewise Deficit Irrigation in Tomato Cultivation

Abstract

ABSTRACT This research focuses on the performance of six advanced machine learning models, namely, artificial neural network (ANN), convolutional neural network (CNN), deep neural network (DNN), recurrent neural network (RNN), random forest (RF) and long short‐term memory (LSTM), for estimating canopy cover (CC), biomass (BM) and soil water content (SWC) in tomato under different irrigation regimes. Model performance was evaluated using the coefficient of determination ( R 2 ), Nash–Sutcliffe efficiency (NSE), root mean square error (RMSE) and normalised RMSE (nRMSE) across training and validation phases. The RF model achieved the highest accuracy for canopy cover prediction. The CNN outperformed all models for biomass estimation with respect to spatial variability, while the DNN achieved superior performance for soil water content prediction, demonstrating its strength in modelling nonlinear, high‐dimensional data. The LSTM model consistently underperformed across all parameters. Among irrigation regimes, moderate deficit irrigation yielded the lowest prediction errors (nRMSE = 4.00%) and strongest correlation ( R 2 = 0.99) for CC, indicating optimal water use efficiency. In contrast, severe stress conditions during flowering and ripening produced the highest errors. Overall, RF, CNN and DNN emerged as the most reliable models for predicting tomato growth parameters under variable irrigation conditions.