Enhancing Flow Direction in Geothermal Fields Using Sentinel-1 Data for Sustainability Water Management

Abstract

This study develops a flow direction prediction model using Sentinel-1 satellite imagery during rainy and dry seasons through the Random Forest machine learning algorithm. The pre-processing stage includes radiometric calibration, terrain flattening, speckle filtering, and Doppler terrain correction. The processed DEM data is used to extract key topographic parameters: elevation, slope, and curvature, which are then utilized in the model. The model is built with 500 trees (n.trees), using a mtry of 2 for the rainy season and 3 for the dry season, and out-of-bag (OOB) error estimates of 8.76% and 9.32%, respectively. Model evaluation, conducted through a confusion matrix, reveals high performance, with average Overall Accuracy, Kappa Accuracy, User Accuracy, Sensitivity, and Specificity all at 0.98 or above. The analysis shows that during the rainy season, flow direction predominantly shifts northeast (16.48%), while in the dry season, it shifts northwest (16.85%). Slope significantly influences flow direction, with feature importance scores of 60.76% in the rainy season and 63.53% in the dry season. Slope is crucial as it dictates the speed and direction of water flow under gravity. This model could significantly contribute to geothermal field management by accurately predicting surface water flow, enhancing monitoring, and promoting sustainable water resource management.