Multi-hazard assessment in Kyrgyzstan’s Osh Region using Maximum Entropy
Date
Authors
ORCID
Journal Title
Journal ISSN
Volume Title
Publisher
Abstract
Purpose:
Climate change impacts natural processes that lead to increased warming and extreme precipitation. As global temperatures continue to rise, an increase in the frequency of climate- and weather-related disasters is expected. Over the past decade, approximately 200 million people were affected by disaster events, with 81,000 deaths per year on average. Majority of the impacts occurred in the 40 most mountainous countries, including Kyrgyzstan in Central Asia. More than 80% of the land area in Kyrgyzstan is mountainous and highly hazardous. The Osh Region in Kyrgyzstan, in particular, is a site that suffers multiple types of natural hazards, such as floods, landslides, earthquake, and drought. These hazards pose a great risk to the mountain communities. Currently, the susceptibility distribution of the multiple hazards in the Osh Region, and the populations exposed to it remain to be assessed. The goal of this study was to harmonize three natural hazards – flood, landslides, and wildfire – of the Osh Region in a generalized multi-hazard susceptibility map (MHSM) that incorporates bioclimatic and geo-environmental factors for disaster risk management and response planning.
Methods:
Inventory maps for single hazard susceptibility were prepared by processing thematic layers from remotely-sensed data, hazard catalogs, and bioclimatic data. A total of 37 covariates (19 bioclimatic variables and 18 geo-environmental factors) were selected as predictors using Maximum Entropy (MaxEnt) machine learning algorithm. Accuracy metric of the predictive model was evaluated using the "receiver operating characteristic” (ROC) curve and computing for the "area under the ROC curve” (AUC-ROC). Moreover, MaxEnt was able to estimate percent variable contributions and permutation importance for each of the predictors. The generated single-hazard susceptibility maps were harmonized into a multi-hazard susceptibility map in ArcGIS 10.8.
Results:
The results show significant predictive performance and degree of fitting of MaxEnt for flood, landslides, and wildfire, obtaining high AUC-ROC (> 0.9). The land cover covariate contributed to wildfire and landslide. Elevation covariate occurred most to wildfire and flood susceptibility. Distance to faults contributed to landslides, while precipitation of the coldest quarter contributed to flood. A MHSM was then generated after overlaying and fitting the single-hazard maps. The MHSM showed that 37% of Osh Region’s area is susceptible to the three hazards. Within this area, 33% is susceptible to landslides, 17% to flood, and 5% to wildfire. The population exposed to these hazards will be investigated in a future study.
Conclusion:
The multi-hazard susceptibility map can be a useful planning tool for government administrators in the Osh Region to identify areas susceptible to hazards at a regional scale. This information can promote risk-informed policy and investment decisions to minimize disaster-induced losses and damages, such as fatalities and infrastructure damage, in the long term.