Introduction:
Land use around rivers is one of the key factors contributing to their pollution. Burying waste and garbage in the soil, agricultural activities, and the use of various fertilizers and pesticides ultimately lead to river pollution. Spatial and temporal changes in the chemical quality of river water depend on various factors. Due to the importance of quantitative parameters in determining the expected response of the river, it is essential to measure these parameters using models. Measuring these parameters in large volumes is time-consuming, costly, and requires high accuracy; so, the need for some indirect methods to estimate these parameters is already apparent. Principal component analysis (PCA) is one of the methods used to overcome this problem. So that the most accurate answer is achieved with less cost and with the least information. Artificial neural networks (ANNs) are suitable for water quality assessment due to their ability to solve nonlinear problems compared to conventional techniques. The aim of this research is to investigate the efficiency of parameter reduction using PCA and the effect of land use and precipitation on some water quality factors of the Zayandeh-Rood River using ANN.
Materials and Methods:
To evaluate the effect of land use and precipitation on some water quality factors, data were collected from the Water and Environment Organization, the General Directorate of Natural Resources and Watershed Management, and the Provincial Research Center. These data were collected daily from 12 hydrological stations in the two provinces of Chaharmahal and Bakhtiari and Isfahan for the Zayandeh-Rood River over a 10-year period. According to the hydrological stations monitoring the water quality of the river from upstream to downstream, the Zayandeh-Rood River basin was divided into 12 sub-basins. The area of each sub-basin was calculated using partitioning software. PCA was applied to some water quality parameters, including total dissolved solids (TDS), electrical conductivity (EC), water reaction (pH), carbonate (CO3), bicarbonate (HCO3), chlorine (Cl), sulfate (SO4), calcium

(Ca), magnesium (Mg), sodium (Na), potassium (K), and sodium absorption ratio (SAR), as well as discharge. This was done to reduce the main data arrays affecting the water quality of the Zayandeh-Rood River and to identify pollution sources using XLSTAT 2018 software. To check the suitability of the dataset for PCA, the Kaiser-Meyer-Olkin test and the Bartlett sphericity test were used. After extracting land use and rainfall data for the Zayandeh-Rood basin, modeling was performed using MATLAB software, with 60% of the data randomly selected for network training and the remaining 40% for validation and entered into the network. The effect of these two factors on some river water properties was analyzedusing the backpropagation algorithm, and by changing the number of neurons and layers, the mean square error, coefficient of determination, and Mean Absolute Error (MAE) were calculated for each factor.
Result and Discussion:
According to the results, PCA can clarify the order of the data in detail by eliminating unimportant data as the Kaiser Meyer Olkin (KMO) sample adequacy index was 0.618 and Bartlett's Sphericity test was 0.001, indicating that the number of samples for the test was sufficient. The results of PCA analysis showed that the first six components accounted for 94.8% of the total variance. This indicates that PCA is effective in reducing the slope of change among variables. Because the sum of the variances of the first six components is close to one. This implies a linear correlation among variables, because PCA uses a linear transformation to transform variables into principal components. The ANN results showed that the total coefficient of determination was 0.79 and the root mean square error was 0.20, and the coefficient of determination for SAR, TDS, EC, and Na were 0.88, 0.84, 0.83, and 0.83, respectively, indicating the high accuracy of the network predictions. The lowest coefficient of determination for the three factors pH, CO3, and discharge (0.16, 0.10, and 0.43, respectively), were all below0.50, and for other water quality factors, the coefficient of determination was above 0.50. The results indicated that the coefficient of determination for land use and rainfall on the three factors of discharge, pH and CO3 is very weak and the model provides very little explanation of the variance of the dependent variable. The results of the Nash-Sutcliffe coefficient showed that it is appropriate for all measured factors (except pH, CO₃ and discharge) and is above 0.6 and this value indicates that the model has been able to explain the observed changes well and provide relatively accurate predictions. The outcomes also revealed that the reduced values in PCA are the same factors predicted with the highest coefficient of determination in the ANN model.