Generation of erosion hotspot zones along the streams by incorporating significant factors and deep neural network

Abstract An erosion-induced quick clay landslide (QCL) is a geological hazard that poses a severe risk to human life and property. The recent disaster in Norway at Gjerdrum in 2020 caused loss of 11 lives and destroyed many infrastructures overnight. One of the reasons for the Gjerdrum landslide was continuous erosion that led to deterioration of the slope’s stability. However, not many researchers have worked on identifying and mapping this erosion along the streams. Hence, the present study attempts to identify precise erosion hotspots zones of a study area close to Gjerdrum incorporating significant erosion causing factors and deep learning model. To achieve this, first, erosion and non-erosion data were extracted using 1 m resolution lidar data and incorporating change detection technique. Second, twenty erosion-causing factors were correlated with 70% of the data using a Deep Neural Network (DNN) model, and the accuracy of the model was evaluated using the remaining 30%. Lastly, the trained model was used to generate erosion susceptibility map highlighting the hotspots erosion zones along the streams. The results show that about 21% of the study area comes under very high susceptibility erosion hotspot zones. Also, these hotspots can extend up to 300 m from one side of the streams while on the other side they may not exist. The DNN model accomplished an accuracy of 0.88 utilizing fourteen significant factors. The study concludes that the erosion hotspot zones require more in-depth site-specific investigation along with numerical slope stability analysis considering erosion under cutting the slope.