An Explainable Deep Learning Framework for Evapotranspiration Prediction Using Hydro-Meteorological Data
Year: 2026
Type: preprint
Abstract: Accurate estimation of evapotranspiration (ET) and potential evapotranspiration (PET) is fundamental for hydrological modelling, flood risk assessment, irrigation management, and climate adaptation. Conventional empirical approaches often require extensive meteorological observations and calibration, limiting their applicability in data-constrained environments. This study proposes an explainable deep learning framework that integrates Artificial Neural Networks (ANN), Convolutional Neural Networks (CNN), Gated Recurrent Units (GRU), Long Short-Term Memory (LSTM), Geographic Information Systems (GIS), and SHapley Additive exPlanations (SHAP) to predict ET and PET using secondary hydro-meteorological and spatial data. The framework combines advanced deep learning techniques with GIS-based spatial analysis to capture complex nonlinear relationships among rainfall, temperature, humidity, wind speed, solar radiation, and geographical variables, while SHAP enhances model transparency by quantifying the contribution of individual predictors. Comparative qualitative evaluation indicates that recurrent architectures are better suited for modelling temporal hydrological patterns, whereas CNN improves spatial feature representation. By integrating predictive accuracy, spatial visualization, and explainable artificial intelligence within a unified framework, the proposed approach addresses key limitations of conventional black-box models and provides a robust decision-support tool for sustainable water resource management, flood preparedness, environmental planning, and climate-resilient policy development.
Source: Research Square
Authors: Kazi Abdul Mannan, Md. Swopon Ali
Language: English
Subfield: Environmental Engineering
Field: Environmental Science
Domain: Physical Sciences
SDG: Climate action
Open Access status: green
OpenAlex ID: W7204496254