An evaluation of the impact of exponential downscale input parameters with artificial intelligence method for estimation of hydrological parameters, case study: Ardabil Synoptic Station

In recent years, factors such as the growth of industrial activities and environmental destruction have led to an increase in greenhouse gases, resulting in disruption of the climate balance known as climate change. The negative impact of this phenomenon on various systems, such as water resources, agriculture and industry, has raised concerns in human society. Consequently, addressing the issue of climate change regarding water resources has become one of the primary causes of concern today. Climate change and its effects pose significant challenges to water and energy resource management, necessitating thorough investigation and developing plans to mitigate its impact on water resources. This study aims to identify the region's most suitable climate change model and assess the effectiveness of artificial intelligence methods in studying the climate change phenomenon.

One of the most reliable approaches for studying the parameters influencing hydrological phenomena under climate change is atmospheric general circulation models. To employ these models on a regional scale, downscaling operations are necessary. Given the large number of parameters derived from Earth's General Circulation Models (GCMs), selecting the most influential parameters is essential before proceeding with the exponential downscaling process. In this study, the meteorological and hydrological parameters of the Ardabil synoptic station were determined using 25 models from the fifth series of the IPCC report. The linear correlation coefficient between monthly precipitation and observed temperature with the output of GCM was used to identify the most appropriate model among the reviewed models. Artificial Neural Network (ANN) was also utilized to downscale the GCMs output. Before employing the neural network, the linear correlation coefficient, the standard information function, and the M5 decision tree were used to identify the most suitable input parameters from the parameters of the best GCMs in the region, to obtain an ideal and optimal network.

This research investigated 25 models from the fifth series of the IPCC report to explore the uncertainty of GCMs. The results indicated that three models-MRI-CGCM3, CMCC-CMS, and MPI-ESMMR-demonstrated the most suitable correlation coefficients at the Ardabil synoptic station. The findings related to determining the most appropriate input parameters for exponential downscaling, using three methods linear correlation coefficient, standard information function, and M5 decision tree, revealed that the decision tree algorithm provided the most suitable parameters. Moreover, the results obtained from the downscale analysis using the neural network with the variables selected by the decision tree method exhibited the excellent performance of this approach in selecting the effective input parameters of the neural network. Specifically, using the selected parameters of the MRI-CGCM3 model as input for the neural network as a downscaling method yielded better outcomes. The results obtained using the selected parameters of the MRI-CGCM3 model indicated that for the precipitation parameter, the values of the Determination Coefficient (DC), Root Mean Square Error (RMSE), and Correlation Coefficient (CC) for the test data were 0.39, 0.04, and 0.63, respectively. For the temperature parameter, the values of DC, RMSE, and CC for the test data of the superior model were 0.9, 0.03, and 0.95, respectively.

The performance of exponential downscaling networks is determined by the climatic conditions of the region. The superiority of a particular model in one study cannot be regarded as a valid argument for selecting that model for all regions. It is advisable to utilize different models of the general earth circulation within the region to identify an optimal model. Conducting such studies can assist researchers in investigating various hydrological phenomena that may occur in the future, which may have irreparable consequences.