Accurately predicting reference crop evapotranspiration (ETo) is important for ensuring the sustainable use of water resources, especially in dry regions where increasing water scarcity, aridification, and sparse observations pose formidable challenges in estimating ET, like the Amu Darya Irrigation Basin (ADB). This study innovatively hybridized four supervised learning algorithms: Decision Trees (DT), Generalized Linear Models (GLM), KNearest Neighbours (KNN), and Support Vector Regression (SVR), to achieve an enhanced standalone stacking ensemble (stkENS) model that utilizes fewer inputs yet can be implemented across different prediction scenarios. Results showed that stkENS mean ET estimates are consistent with the FAO56-PM ETo for cotton (3 mm day(-1)) and Sorghum (3 mm day(-1)), and a slight variance of 0.01 mm day(-1) for rice. stkENS outperformed other models in daily ETo prediction, with an optimal R-2 > 0.96, the lowest bias of 0.01 mm d(-1), RMSE: 0.65 mm d(-1) and MAE: 0.42 mm d(-1). This indicates superior performance relative to DT (R-2: 0.73, RMSE: 1.12 mm d(-1), and MAE: 0.86 mm d(-1)), GLM (R2: 0.85, RMSE: 0.83 mm d(-1), and MAE: 0.60 mm d(-1)), KNN (R-2: 0.89, RMSE: 0.72 mm d(-1), and MAE: 0.46 mm d(-1)), and SVR (R-2: 0.85, RMSE: 0.84 mm d(-1) and MAE: 0.59 mm d(-1)). The relative uncertainty of stkENS in ADB was 10.44%, whereas KNN, SVR, GLM, and DT were 14.77%, 16.95%, 23.29%, and 34.49%, respectively. Moreover, stkENS has robust generalizability and can be operationalized to enable seamless predictions for unhindered access to daily ETo estimates that are essential for improving irrigation scheduling and water use efficiency in cotton, rice, and sorghum production in ungauged arid croplands. The study highlights the importance of expanding the stacked generalization method to increase deeper understanding of its applicability across diverse dry ecological regions in the context of changing environmental conditions.