Actual evapotranspiration (ETa) is governed by both atmospheric and surface conditions. However, most previous studies have focused on atmospheric drivers, while surface controls and the seasonal dynamics of key drivers of ETa remain largely unexplored, particularly in monsoon-driven climates where significant seasonal changes prevail. Explainable Machine Learning approach was employed to identify the key parameters influencing ETa in a wheat–paddy cropping system, using eddy covariance (EC) observations. Five Machine-Learning (ML) models were evaluated, among which the Random Forest (RF) model outperformed the other approaches. RF predicted LE values were compared using physics-based models (PenmanMonteith (PM) and Priestley Taylor (PT)) and EC observations for one crop season. The relative error associated with the RF model was 18%, compared to 27.9% for PT and 33.3% for PM. Explainable ML techniques (RF based feature importance, permutation feature importance, recursive feature elimination, Partial Differential Plots, SHapley Additive exPlanations (SHAP) and Global Sensitivity Analysis (GSA)) were used to identify the most important ETa drivers and their controlling behavior. ETa and its controlling variables exhibit pronounced non-linear behavior. Results indicates that, across all seasons, incoming solar radiation is the main driver of ETa, followed by Air Temperature and Soil Moisture. The relative importance of the secondary ETa drivers differs in the pre-monsoon season compared to the monsoon, winter, and post-monsoon seasons. Overall, these findings enhance process-level understanding of ETa dynamics and demonstrate the value of Explainable ML approaches for interpreting complex land–atmosphere interactions.