Estimation of Daily and Weekly Evapotranspiration Using Hybrid Approaches of Soft Computing

Since a considerable part of the available water is wasted due to evapotranspiration, a precise estimation of it in short-term and long-term periods is of great importance. In this paper, the capability of the M5 model tree and random forest (RF), as the artificial intelligence approaches, and in combination with the wavelet preprocessing, investigated to estimate the potential daily and weekly evapotranspiration in the synoptic station of Babolsar. Given the time series structure of the input data, the two functions of Coiflet mother wavelet and Daubechies 6 wavelet in the decomposition levels of 3 to 8 were chosen. The four indices of correlation coefficient (R), index of agreement (Ia), Nash-Sutcliffe model efficiency coefficient (NSE) and root mean square error (RMSE) used to evaluate the presented models. The obtained results indicated that although all individual models have desirable efficiency in modeling the evapotranspiration, the use of wavelet preprocessing enhances the performance of individual models in all cases while allows the simpler input scenarios to provide more desirable results. For instance, in the third input scenario (wind speed, maximum temperature, relative humidity, and dew point), the use of Daubechies 6 wavelet in the decomposition level of 5 increased the correlation coefficient of the daily model from 0.908 to 0.928 while reduced the RMSE from 0.833 mm/day to 0.722 mm/day. Similarly, the use of Coiflet-4 mother wavelet in the decomposition level of 5 raised the correlation coefficient of the weekly model from 0.948 to 0.961 while lowered the RMSE from 4.55 mm/week to 4 mm/week. Therefore, in the present study, the efficiency of both individual and hybrid approaches in estimating the evapotranspiration of daily and weekly periods is satisfactory. However, if the hybrid approaches employed, even the use of simpler and more accessible meteorological parameters will provide satisfactory results.