Evaporation and precipitation prediction for future time frames via combined machine learning-climate change models (case study: Quri Gol Wetland, Iran)

Abstract

Evaporation plays a fundamental role in water resources management. However, due to the interplay of meteorological variables in evaporation calculations, several nonlinear relationships have been proposed. Their effectiveness can be discussed based on the climatic conditions of each region. Therefore, in the present study, the efficiency of machine learning methods, including random forest and gradient boosted tree, generalized linear model, support vector machine, Gaussian process regression, and deep learning under different scenarios resulting from the combination of meteorological variables in forecasting and modeling evaporation in Quri Gol wetland located in East Azarbaijan province of Iran was investigated for a period of thirty years (1991-2020). In the second part, by applying climate change models including LARS-WG and SDSM in three scenarios, the values of meteorological variables were calculated for future periods (2021-2050, 2051-2080 and 2081-2100) and then, in order to estimate evaporation, these variables were used as input in the machine learning model. The output of the machine learning models showed that among the studied meteorological variables, temperature has the most influence in modeling. Also, the accuracy of the performance of all studied models was evaluated as suitable, and among these models the performance of RF and DL models brought the best results. The results of the climate change models also showed that in most scenarios for all three time periods, the cumulative values of evaporation and precipitation have increased, with the difference that the ratio of evaporation increase was higher than precipitation.