National-scale assessment of pan evaporation models across different climatic zones of China

Abstract

Evaporation is an essential component in hydrological processes, and accurate estimation of evaporation is of importance for sustainable management of water resources. This study conducted a national-scale assessment of different models for pan evaporation (Epan) estimation at 178 meteorological stations across different climatic zones of China, including the temperate continental zone (TCZ), temperate monsoon zone (TMZ), mountain plateau zone (MPZ) and subtropical monsoon zone (SMZ). Firstly, three data-driven models, including extreme learning machine (ELM), artificial neural networks optimized by particle swarm optimization (PSO-ANN) and genetic algorithm (GA-ANN), were trained with nine input combinations of climatic variables. The performance of the 27 proposed models along with the empirical Stephens and Stewart (SS) and physically-based PenPan models were investigated and compared using relative root mean square error (RRMSE), Nash-Sutcliffe coefficient (NS) and mean absolute error (MAE). The three statistical indicators were further normalized to global performance indicator (GPI), by which all the evaluated models can be easily ranked. The results showed that the data-driven models with complete inputs generally obtained more accurate Epan estimation, where the ELM model with full input data provided the best accuracy, with average RRMSE of 12.5%–15.2%, NS of 0.909–0.936 and MAE of 11.7–19.9 mm/m. Air temperature was found to be the most influential parameter to data-driven models, followed by sunshine duration, wind speed and relative humidity. The SS model provided slightly better results in MPZ and TCZ, and slightly less accurate results in SMZ and TMZ, compared with the data-driven models under the same input conditions. Overall, ELM was recommended as the best model for Epan estimation when all the selected climatic data are available, while temperature-based PSO-ANN is recommended in MPZ and SMZ and temperature-based GA-ANN is recommended in TCZ and TMZ when the other climatic data are missing.