Precise estimation of reference crop evapotranspiration (RET) is critical for effective water resource management and irrigation across various climatic regions. The present investigation used CROPWAT, a software tool developed by the Land and Water Development Division of Food and Agriculture Organization (FAO), employing the Penman–Monteith (FAO56-PM) method for RET calculation. However, the FAO56-PM method and related software are not practical when dealing with a restricted number of meteorological components. Hence, it is crucial to create a new method for measuring RET that uses fewer parameters. To address this challenge, climate data of 30-years were collected from a meteorological station of Skardu located in humid region of Pakistan. Firstly, CROPWAT 8.0 was used to compute the RET with the help of the weather factors as input. Secondly, correlation analysis (Pearson, Spearman, and Kendall) was performed to identify the prime climatic input factors. Afterward, two tree-based machine learning (ML) algorithms [extreme gradient boosting (XGBoost) and random forest (RF)] were employed to develop a model that describe relationship between weather observations and RET. The comparison between predicted (ML approaches) and actual (FAO56-PM) RET observations were evaluated using scatter plots and Taylor diagrams. Upon analysis, RET estimated by ML approaches coincide well with standard FAO56-PM method using effective parameters. In conclusion, this study recommends the use of ML approaches in water resource management across diverse climatic regions (humid, semi-arid, and arid conditions) using limited meteorological data (temperature and sunshine hours only). In order to achieve water sustainability, these ML techniques should give preference to areas with higher RET values and execute suitable irrigation scheduling strategies for crops.