Simulating Snow Cover Extent by Combined Principal Component Analysis and Artificial Intelligence Approaches Using Climatic Parameters

Snow cover holds significant importance in hydrology as it plays a vital role in the water cycle and water resource management. Acting as a natural reservoir, snow stores water during winter and gradually releases it as it melts. This process contributes to streamflow, groundwater recharge, and overall water availability. Main goal of this study is the modeling and prediction of the changes in snow cover extent in Baranduz River basin, in Iran. Accurate modeling of snow cover area is crucial in hydrology as it enables precise predictions and assessments of water resources. These models incorporate snow accumulation, melt rates, and distribution, allowing informed decision-making for water management, agriculture, and ecosystem preservation. Therefore, the snow cover extent of the basin was extracted from MODIS 8-day maximum snow extent production from 2000 to 2019. Forty meteorological parameters, 20 satellite based and 20 surface stationary collected data, were used as the independent variables. The PCA was performed to parameters, and the PCA6 vector was used as input to the machine learning models. ANN, SVM, CART, and RF machine learning approaches were performed in this study. The results showed, all machine learning models had satisfactory performance and efficiency in modeling and predicting the snow cover extent. The PCA-RF model showed the highest accuracy. The RMSE and R2 values for the PCA-RF model were 0.345 and 0.895, respectively, in the testing phase. Despite the fact that models have not been able to predict some of the boundary points accurately, they have still demonstrated acceptable performance.