Detection of geographic faults using deep learning model from DEM and remote sensing data in Djibouti

抄録

Underground water flow delineation is critical for understanding the groundwater cycling systems and their utilization at arid land areas, such as in Eastern African Djibouti. However, such regions lack essential data, such as borehole data, and this becomes a challenge. One solution is through fault detection to evaluate the possibility of fault-driven groundwater flow into the water cycling systems. This study focuses on geographic faults which exist in Djibouti where the plate tectonic activities are remarkable. Our study aimed to utilize the fault lines delineated on existing geology maps since fault lines distribution has potentiality for high correlation with groundwater flow volumes. It is essential to evaluate if the fault systems can contribute to the simulation of groundwater volume modules.

This work develops on our previous research of fault inspection using relief image in fault distribution derived from adaptive gradient filter applied on PALSAR-1/2 image data. In this study, deep learning techniques are used in fault detection analysis based on Digital Elevation Model (DEM) and remote sensing data in training a multi-input deep convolutional neural network (Deep CNN) model. We used ortho-rectified PALSAR-1 RTI and PALSAR-2 Global Mosaic, DEM data as well as curvature and slope images derived from the DEM. We used fault labels digitized and rectified from the existing geology map, specifically on the Ali Faren catchment as the target region for this study.

The result showed that using multi-input data in our derived deep CNN model, fault lines were detected. Further, we determined and derived the extent of primary faults in a higher accuracy. Results also showed promising level of groundwater flow evaluation with deep CNN detected geographically fault lines. Further, our proposed deep CNN model could be applied to other watersheds in Djibouti to help in groundwater flow model simulations and eventually help locate the potential area for groundwater resources in entire Djibouti.