Multi hazard risk assessment through geospatial analysis and machine learning

Climate change, increased frequency of natural hazards and growing urban populations all have created a greater necessity for conducting comprehensive multi-hazard assessments using Geographical Information Systems (GIS). Therefore, this research paper proposes developing a machine learning and GIS based methodology to identify and map zones of multi-hazard risk through use of integrated geospatial and environmental variables. The framework incorporates multiple types of geospatial data sets that can be drawn from various sources which include; hydrology, topography, environment, climate and anthropological conditions that have been identified as being critical variables in assessing potential landslides. Some of the key data sets included in the analysis are; proximity to water bodies, TWI, Elevation, NDVI, Impervious Surface Ratio, Rainfall, Wind Speed, Road Networks and Historical Landslide Data. All of these data sets were analyzed within a GIS environment to standardize them prior to training a random forest classifier for predicting multi-hazard risk at a grid level. Results indicated that the model was able to accurately identify regional hazard susceptibilities and categorize the study area into four categories of risk. It is also important to note that the experimental results show good predictions of high risk areas with a total classification accuracy of 82.6% and an ROC-AUC score of 0.84. Additionally, the final high resolution maps indicate that areas of low elevation, high rainfall, large amounts of impervious surfaces and those proximal to water bodies represent the highest risk areas. The integration of GIS with machine learning has demonstrated that it represents a viable data driven approach for conducting multi-hazard risk assessments and therefore will provide significant value to emergency planners and managers responsible for mitigating disaster impacts through improved preparedness and resilience planning, urban development and climate resiliency planning.