Work

Reanalysis-assisted AI framework for regional pan evaporation estimation in Taiwan without ground-based meteorological observations

Year: 2025

Type: article

Abstract: Taiwan, a subtropical island in East Asia with complex topography and diverse climatic conditions. Motivated by the sparse distribution of Class A evaporation pans in Taiwan, this study develops an artificial intelligence-based framework to estimate daily pan evaporation across Taiwan without relying on ground-based meteorological station data. By integrating high-resolution reanalysis inputs with station metadata, the framework enables spatially continuous estimation of evaporation patterns. The performance and interpretability of the proposed framework were systematically evaluated to ensure its robustness for ungauged estimation. XGBoost achieved the best overall performance, with MAE = 0.92 mm/day, CC = 0.72, CE = 0.50, KGE = 0.58, and PBIAS = 1.30 %, highlighting its strong potential for estimating pan evaporation in ungauged regions. The framework not only captured dominant meteorological drivers such as radiation, temperature, humidity, and wind speed but also revealed hidden influences like precipitation and seasonal cycles through SHapley Additive exPlanations (SHAP) analysis. Regional case studies confirmed that the model can reproduce both spatial and seasonal variability of evaporation, with higher rates in coastal lowlands and lower rates in mountainous areas. By bridging sparse ground observations with gridded reanalysis inputs, this study provides spatially continuous and interpretable evaporation estimates, offering practical benefits for agricultural water management, reservoir operation, and climate adaptation in Taiwan’s complex hydrological settings. • Used high-resolution reanalysis data to estimate pan evaporation in Taiwan. • Compared five AI models for ungauged pan evaporation estimation. • Leave-one-station-out validation assessed model generalization ability. • SHAP analysis revealed nonlinear effects of rainfall, and time. • Framework overcomes sparse station limitations for water resources management.

Source: Journal of Hydrology Regional Studies

DOI: 10.1016/j.ejrh.2025.102863

Authors: Hsuan-Yu Lin, Shu-Yun Lai, Yi-Ju Lin

Institution: National Chiayi University

Language: English

Topic: Hydrological Forecasting Using AI

Subfield: Environmental Engineering

Field: Environmental Science

Domain: Physical Sciences

SDG: Climate action

Open Access status: gold

Estimated APC paid: $2,600