Comparative evaluation of deep learning and machine learning in modelling pan evaporation using limited inputs

Estimation of pan evaporation (Epan) is an important issue for planning and management of available water resources. In the present study, the accuracy of a new deep learning method, long short-term memory (LSTM) with grey wolf optimization (GWO), in modelling Epan using limited climatic variables as input is investigated. The outcomes of the LSTM-GWO are compared with the single LSTM and advanced machine learning (ML) methods. Minimum and maximum temperatures and extra-terrestrial radiation are used as inputs to the models. Three data splitting scenarios are considered and the outcomes of the abovementioned methods are also compared with the Stephen-Stewart (SS) and calibrated Hargreaves-Samani (CHS) empirical methods. The results reveal that the LSTM-GWO method has a better ability in estimating Epan using limited inputs compared to other ML and empirical methods. They also indicate that an increase in the amount of training data used improves the accuracy of the models.