Investigating the effect of precipitation series decomposition on the simulation of electrical conductivity of river flow (case study: Eskandari sub-basin)

In this study, the effect of time series decomposition of precipitation in the simulation of electrical conductivity values of the surface water in Eskandari sub-basin in the northwest of the of Zayandeh Rood Dam Basin in the period of 1990-2020 was investigated by two algorithms based on random tree and random forest. Decomposition of observation precipitation series was done using wavelet theory, Daubechies 4 and level 2. At first, the simulation of electrical conductivity values in the studied sub-basin was done using two random tree and random forest algorithms in two phases of training and testing according to the rainfall values corresponding to the electrical conductivity values on a daily scale. The results showed that the efficiency of the model was 0.67 and 0.73 in the training phase, respectively, for random forest and random tree algorithms, and the efficiency was 0.59 and 0.55 in the testing phase for the mentioned algorithms by the Nash-Sutcliffe statistic. By decomposing the rainfall series into an approximate series and two detail series and increasing the simulation dimension to 4 dimensions, the results showed that the combination of wavelet theory with random forest and random tree algorithms was able to reduce the simulation error (RMSE) of the conductivity values in the training phase compared to the random forest and random tree algorithms by about 77.5 and 54%, respectively. These numbers in the testing phase are about 10 and 22 percent, respectively.