An agricultural water management strategy must include reference evapotranspiration (ET0), which is frequently employed in the meaningful and efficient scheduling of irrigation events. This ensures that the limited water resources are used appropriately. The most effective design of irrigation scheduling and efficient planning and management of agricultural water resources both depend on the reliable and precise forecast of ET0. This paper suggests a new approach for estimating ET0 that combines a genetic algorithm (GA) and a neural network (NN) of group method of data handling (GMDH) type. The GA optimises the GMDH network structure’s efficacy and efficiency by allowing each neuron to look for the connections that make up its ideal set from the layer before. Based on simulation findings, the dependability of computational models was evaluated utilising three statistical metrics: mean absolute error, mean square error, coefficient of determination, and Nash–Sutcliffe efficiency. During testing, the GMDH model (MSE = 20.1149, R2 = 0.9264, and NSC = 0.921) fared worse than the GMDH-GA model (MSE = 7.3328, R2 = 0.9758, and NSC = 0.9702). The findings supported the hypothesis that GA optimisation techniques might enhance the traditional GMDH model's monthly ET0 prediction performance in Srinagar's various climates.