Modelling reference evapotranspiration using a new wavelet conjunction heuristic method: Wavelet extreme learning machine vs wavelet neural networks

Abstract

Evapotranspiration is an important parameter in linking ecosystem functioning, climate and carbon feedbacks, agricultural management, and water resources. This study investigates the applicability of wavelet extreme learning machine (WELM) model which uses discrete wavelet transform and ELM methods in estimating daily reference evapotranspiration (ET0). Various combination of climatic data of temperature, solar radiation, relative humidity and wind speed from two stations, Ankara and Kirikkale, located in central Anatolia region of Turkey were used as inputs to the WELM models. The WELM estimates were compared with wavelet artificial neural networks (WANN) and single artificial neural network (ANN), ELM and online sequential ELM (OS-ELM) models. The results indicate that the models comprising four input variables as inputs provide better accuracy than the models with less inputs. Solar radiation was found to be the most effective variable on ET0. Wavelet conjunction models (e.g. WELM and WANN) generally show better accuracy compared to the single models and WELM model is found to be the best model in estimating ET0. The root mean square error and mean relative error accuracies of the ELM, ANN and WANN models were improved by 28–25%, 32–32% and 27–26% for the Ankara Station and by 14–14%, 58–58% and 32–36% for the Kirikkale Station.