Hotspot analysis of critical infrastructure risk to debris flow hazard scenarios using Geo AI approach

Abstract

The Indian Himalayas are facing rapid construction of critical infrastructure (CI) which is significant for the functioning of mountainous society. At the same time, these infrastructures are disrupted frequently during monsoon season particularly due to the onset of debris flows. It causes soaring economic losses and cutting off essential services like food, water, and health during disasters. In the background, mapping detailed CI and assessing their risk to debris flows is essential. However, one of the challenges of estimating debris flow is that it not only damages infrastructure at its source area but far beyond wherever it travels as run-out. The literature shows the risk of CI under such a run-out scenario is limited. Therefore, the study proposes a spatial framework for assessing CI risk to debris flow hazard scenarios. The frame is constituted in two parts, the first one focuses on developing a debris flow hazard model by integrating source and their runout areas. The second part concentrates on a systematic mapping of exposed CI and its risk-hotspot zonation. In the debris flow hazard modelling, an inventory database of sources and runouts are generated covering the year 2005 to 2022. The conditioning factors cover topographic, hydrological, geological, and environmental variables that are generated from multiple data sources such as DEM (ALOS PALSAR), Google Earth images, and high-resolution satellite images (Planet Scope). Next, the susceptibility of the debris flow source area is estimated by using Stacking Random Forest Model. Further, the runout area is simulated using the Flow R model in which susceptible debris flow source areas are considered as input. We generate two debris flow scenarios; one is considered normal rainfall-induced debris flows and another is a worst-case scenario that is developed considering extreme rainfall-induced debris flows. The models are validated using a confusion matrix and further, applied to CI risk analysis. In the second part of the paper, the detail of twelve category of CI is identified and mapped using GIS. These CI is treated as hard assets such as transportation, electricity, water lines, telecommunication, hospitals, schools, waste management, dam, recreation areas, hotels, helipads, and evacuation shelters. The spatial data of critical infrastructure are collected from multiple sources data such as Open Street Map, My Maps, Google Earth Images, Toposheet and various published reports. Then, the density of each CI at each village is generated and it is overlayed with the debris flow hazard scenarios for estimating risk. Finally, the hotspot of CI risk is analyzed using Global Moran's I method. The modelling framework is applied in the Sikkim Himalayas which is the one of sensitive debris flow regions of the world. We find a positive linearity with remoteness and debris flow hazard. However, a non-linearity exists with remoteness and CI risk. The findings and output map of the study can be used for financing and policy-making towards disaster-resilient infrastructure development. Keywords: Critical infrastructure; Debris flow; Runout; Geo AI; Stacking Random Forest; Global Moran's I