A deep learning model for predicting climate-induced disasters

The increased severity and frequency of Climate-Induced Disasters (CID) including those attributed to hydrological, meteorological, and climatological effects have been testing the resilience of cities worldwide. The World Economic Forum highlighted—in its 2020 Global Risk Report—that from 2018 to 2020, three of the top five risks with respect to likelihood and impact are climate related with extreme weather events being the highest ranked risk in terms of likelihood. To alleviate the adverse impacts of CID on cities, this paper aims at predicting the occurrence of CID by linking different climate change indices to historical disaster records. In this respect, a deep learning model was developed for spatial–temporal disaster occurrence prediction. To demonstrate its application, flood disaster data from the Canadian Disaster Database was linked to climate change indices data in Ontario in order to train, test and validate the developed model. The results of the demonstration application showed that the model was able to predict flood disasters with an accuracy of around 96%. In addition to its association with precipitation indices, the study results affirm that flood disasters are closely linked to temperature-related features including the daily temperature gradient, and the number of days with minimum temperature below zero. This work introduces a new perspective in CID prediction, based on historical disaster data, global climate models, and climate change metrics, in an attempt to enhance urban resilience and mitigate CID risks on cities worldwide.