Assessment of Wildfire Landslide Risk using Spatial Analytics and Deep Learning Techniques for Rudraprayag Forest Division, Uttarakhand

Abstract

Rudraprayag Forest Division of Uttarakhand State in India has a rich biodiversity; however, it experiences high frequency of wildfire incidents every year. According to Forest Survey of India, a total of 2609 Fires were recorded during the last one decade. Also, 1893 Landslides were detected by ISRO, and over 327 Landslides were recorded by Geological Survey of India during 2013-15. Wildfires are known to harm not only trees and plants, but they also decrease water retention capacity of topsoil leading to higher rainfall runoff resulting in significant erosion and debris flow. This attributes to potential causes for land sliding activity and flash floods. In this paper, we attempt to assess relation between wildfires and landslide hazards using GIS based spatial data sciences and subsequently deep learning techniques to identify sub-basin-wise landslide prone areas post wildfire incidents in the Mandakini valley, Uttarakhand. Freely available Landsat 8/Sentinel 2 Imagery, SRTM/ASTER DEM, Landuse/Landcover data were used along with near live feeds from MODIS and NASA. Analysis was carried out using Esri ArcGIS Pro and Python Scripts in Jupyter Notebook. In the study area, about 63-69% landslides were observed near forest fires suggestive of role of wildfires in triggering landslides. Many landslides were observed near roads and drainage establishing that landslides in the region are not only correlated to fire occurrences but are also influenced by other factors. The current study protocol may be used to detect wildfire generated landslide hazards in other areas for developing site specific disaster management plan in the identified susceptible watersheds.