تخمين تغييرات تراز آب زيرزميني بااستفاده از شبكه عصبي پيچشي و بكارگيري داده¬هاي ماهواره محور

Groundwater level changes , Convolutional Neural Network , satellite-based data , spatial downscaling , Gravity Recovery an‎d Climate Experiment (GRACE) data , Semirom

Estimation of Groundwater Level Changes Using Convolutional Neural Networks an‎d Satellite-Based Data

مهندسي عمران

Groundwater, due to lower contamination an‎d easier access compared to surface water sources, holds very high value. About 33% of the world population uses groundwater for daily drinking water. Global climate changes, population growth, improper management, an‎d changes an‎d improvements in human lifestyle an‎d hygiene have contributed to the depletion of groundwater reserves. The reduction in groundwater levels can lead to serious environmental an‎d ecological problems. Therefore, identifying groundwater storage locations an‎d optimizing its allocation is critically important. This study aimed to analyze an‎d estimate groundwater level changes in the Samirim region of Isfahan Province, leveraging the integration of ground-based observational data an‎d satellite data. The datasets used included comprehensive information on precipitation, evapotranspiration, surface runoff, groundwater level, an‎d storage changes, collected from reputable sources. In the first step, a correlation analysis among variables was conducted to identify the influential factors. To enhance prediction accuracy, satellite GRACE data were downscaled using two machine learning models, namely Convolutional Neural Networks (CNN) an‎d Support Vector Machines (SVM). The eva‎luation results indicated that the CNN model, with a coefficient of determination (R² = 0.71), performed substantially better than the SVM model (R² = 0.60). Moreover, to estimate missing GRACE data, three models—CNN, ANN (Artificial Neural Network), an‎d SVM—were employed. In this section as well, CNN demonstrated superior performance (R² = 0.70) due to its ability to account for neighborhood patterns in the data (matrix inputs). One of the key applications of GRACE data is the water balance calculation in hydro basins. In this study, downscaled GRACE data were used to estimate groundwater level changes. Finally, using the Q-Q plot, the estimated groundwater level change values derived from satellite data were compared an‎d validated against the ground-based observational data. The results of this study demonstrate the high potential of machine learning models, especially CNN, to improve the accuracy of hydrological analyses an‎d the management of groundwater resources.

6

نرجس سلماني