Prediction of SAR and TDS parameters using LSTM- RNN model: A case study on Aran station, Iran

Surface water quality is of particular importance due to its drinking, industrial, and agricultural water sources. Changes in rainfall, temperature and river discharge can affect surface water quality. In this study, the effect of CANESM۲, FIO, GFDL, MIROC climate models and weight composition model of CMIP۵ (Coupled Model Intercomparison Project) under representative concentration pathways (RCP) of ۴.۵, ۶, ۸.۵ scenarios on rainfall and temperature were investigated and then monthly discharge of the Aran river in Iran during ۲۰۲۰-۲۰۵۲ and ۲۰۵۳-۲۰۸۵ is predicted using the IHACRES runoff model. Next, the LSTM (Long Short-Term Memory network)-RNN (Recurrent Neural Networks) model were used to predict the total dissolved solids (TDS), sodium adsorption ratio (SAR) for the period ۲۰۲۰-۲۰۳۰. The results showed that the long-term monthly rainfall under the RCP۸.۵ scenario reported a further decrease compared to the RCP۴.۵ and RCP۶, and the rainfall fluctuations were higher than the other two scenarios. Temperature changes in the second period are higher than the first period, so that in the first period under the scenarios of RCP۴.۵, RCP۶ and RCP۸.۵ showed respectively ۱, ۱.۵ and ۲ degrees Celsius increase, while in the second period, ۲, ۳ and ۴ degrees Celsius is predicted. The average discharge shows by ۱۵.۸ % and ۲۰.۹۷ % respectively decrease under the RCP۴.۵ scenario in the first and second periods, while by ۸.۵۱ % and ۲۷.۵۵ % under the RCP۶ scenario and ۶.۳۸ % and ۳۹.۸۹ % under the RCP۸.۵ scenario compared to the observed discharge. The mean, maximum, and minimum TDS parameters under RCP۴.۵ scenario are, respectively, ۲۹۵, ۴۱۰, and ۲۶۳, and ۳۰۲, ۴۱۰, and ۲۵۷ under RCP۶ scenario while ۲۹۲, ۴۱۰, and ۲۵۷ mg, for RCP۸.۵ scenario. These changes are, respectively, ۰.۴۲, ۰.۹۳ and ۰.۱۴ for the SAR parameter in RCP۴.۵ scenario, and equal to ۰.۴۴, ۰.۹۴ and ۰.۱ in scenario ۶, while ۰.۴۲, ۰.۹۳ and ۰.۱۵, respectively, for RCP۸.۵ scenario in Khorramrood river.

Keywords:

CMIP۵ Scenarios , SAR , TDS , Rainfall-runoff , Prediction , LSTM-RNN