Reference Evapotranspiration Modelling Using Artificial Neural Networks Under Scenarios of Limited Weather Data: A Case Study in the Malwa Region of Punjab

Precise estimation of reference evapotranspiration (ET o ) is crucial for efficient management of water resources. However, the complex and nonlinear dependence of the ET o on several weather parameters makes its estimation challenging. In this study, artificial neural networks (ANNs) were used to simulate ET o at three diverse stations: Ludhiana, Bathinda, and Faridkot in the Malwa region of Punjab. Different input combinations of weather parameters, represented as (S1-S7) scenarios, were used to test the model against the standard FAO-56 Penman–Monteith (PM) method. The model was judged based on statistical parameters, such as r , R 2 , MSE, RMSE, NRMSE, and NSE. The results indicate that S1, with maximum number of parameters, performed the best with RMSE (0.154, 0.246, and 0.298) and NSE (0.9933, 0.9849 and 0.9709) values, followed by S2 (T max , T min , n, u 2 ) in terms of RMSE (0.218, 0.316, and 0.307) and NSE (0.9877, 0.9751, and 0.9699) for Ludhiana, Bathinda, and Faridkot stations, respectively. Scenario S6, consisting of only air temperature and relative humidity, performed the worst among the rest of the scenarios. Moreover, the developed ANN models significantly outperformed conventional empirical models, even when similar input variables were used, demonstrating their superior performance. The ANN models at Ludhiana station performed the best in almost all the scenarios because of the consistent dataset. The inconsistent dataset with higher missing values resulted in comparatively lower performance of the ANN models for the Bathinda and Faridkot stations. The developed ANN model will aid in effective crop planning, irrigation, and water resource management, especially in areas where data availability is constrained.