Machine-learning surrogates of FAO-56 Penman–Monteith reference evapotranspiration for the 21 districts of North Bihar, India

This deposit contains the data, code, results, and figures behind the study.Machine-learning surrogates ofFAO-56 Penman-Monteith reference evapotranspiration for the 21 districts ofNorth Bihar, India" Daily NASA POWER data for 2001-2024 were obtained for a representative point.in each of the 21 districts of North Bihar, India. FAO-56 Penman-MonteithReference evapotranspiration (ET0) was computed at the daily step andaveraged to calendar months, giving 6,048 district months. Fivemachine-learning models (multiple linear regression, random forest, gradientboosting, support vector machine, and artificial neural network) were trainedas surrogates of ET0 under eleven input combinations built from maximum andminimum air temperature, relative humidity, wind speed, and solar radiation.Hyperparameters were tuned by expanding-window cross-validation within2001-2019, and the models were tested once on the withheld period 2020-2024.In total 1,155 models were fitted and evaluated. The deposit includes the download and analysis scripts (Python); the monthlyand daily input data with computed ET0; the complete results for everydistrict, input combination, and algorithm, including the selectedhyperparameters and the test-period predictions; the summary tables of thepaper; and the figures at publication resolution. Meteorological data are from the NASA Prediction Of Worldwide EnergyResources (POWER) project (https://power.larc.nasa.gov). District boundariesare Census of India 2011 maps distributed by DataMeet under CC BY 2.5 IN.