Comparative Evaluation of Whole-Season and Stage-Specific ANN Models for Predicting CWSI in Wheat Crop

Water scarcity is an escalating threat to global agriculture, especially in semi-arid regions. Precision irrigation, guided by crop water stress indicators, is critical for sustainable water resource management. This study evaluates the potential of artificial neural networks (ANNs) for predicting the crop water stress index (CWSI) in wheat using canopy temperature and meteorological data. Field experiments were conducted in Uttarakhand, India, across multiple irrigation treatments. Two ANN models were developed: Model 1 was trained separately for key phenological stages [crop development (Stage 2) and reproductive (Stage 3)], while Model 2 (designated the whole-season model) was trained on combined data from the entire physiologically stable crop period (Stages 2 and 3: crop development and reproductive phases). Results show that Model 1 outperformed Model 2, with higher accuracy in Stage 3 (R=0.935) and Stage 2 (R=0.903), compared to Model 2 (R=0.805). Experimental treatments [T1 (well-watered) through T6 (maximum stress)] were maintained consistently across all analyses. These findings emphasize the importance of growth-stage-specific modeling in capturing crop physiological responses for more precise irrigation scheduling. This study supports the integration of machine learning into crop water management strategies for improved agricultural sustainability.