Development of integrated meta-heuristic support vector and analytical models in predicting evaporation from dam reservoir (case study: Dez Dam)

Evaporation from lakes and reservoirs of dams as well as soil is one of the most important processes in hydrological engineering. One of the procedures required for proper and effective management of reservoirs, water resources, basin stability, and agricultural operations is the accurate prediction of evaporation. Weather phenomena that are non-stationary, unpredictable, and non-linear generally have an impact on evaporation. It is evident that these issues hinder the development of precise prediction models. Evaporation is a natural process based on energy supply and air exchange, during which molecules and atoms find the necessary energy to leave the liquid phase and enter the gas phase. On the other hand, climate changes can affect the evaporation parameter, so the prediction of evaporation from the reservoirs of dams is vital in the discussion of water resources management

This investigation uses the SVR-ABC, Wavelet-SVR, and SVM models with six wavelet functions to predict the amount of evaporation from the Dez dam reservoir in Khuzestan province, Iran. The data series from 1350–1396 for 46 years arises from the Dez dam meteorological station in Iran. The study also uses precipitation, maximum temperature, minimum temperature, average temperature, absolute maximum temperature, and absolute minimum temperature and Using the findings of Shannon's entropy, 5 parameters Tmax, Tmin, Tave, Tamax, Tamin are divided into 5 groups with 1 parameter, 2 parameters, 3 parameters, 4 parameters and 5 parameters respectively as the most effective parameters for scenario planning, which in total 28 scenarios are studied.

According to modeling results based on RMSE, MAE and WI evaluation indices show that the performance of SVR-ABC meta-heuristic model with RMSE=82.219, MAE=53.977 and WI=0.815 is better than Wavelet-SVR model with RMSE=93.637, MAE= 69.360 and WI=0.762. Additionally, based on the violin diagram, the single SVR model with periodic and non-periodic inputs estimated the monthly evaporation lower than the observed values, and as a result, the average data has decreased compared to the observed values.

Furthermore, according to the results of the six wavelet functions that were used in the study, the Wavelet-SVR axis decomposition model with the Haar wavelet function at decomposition level 1 and values of 93.637 mm/month, 69.360 mm/month, and 0.762, respectively, shows the most appropriate result among the six wavelet functions based on the RMSE, MAE, and WI evaluation indices.