Estimating elements susceptible to urban flooding using multisource data and machine learning
Year: 2024
Type: article
Abstract: Inadequate representation of flood conditioning factors could affect the accuracy of flood susceptibility prediction (FSP) and the estimation of the associated element within the susceptible areas. The present study examined 28 Earth Observation datasets representing the physical environment and extreme rainfall features of a highly urbanized Akaki catchment, which surrounds the Ethiopian capital city, Addis Ababa. The most relevant features were selected by comparing seven feature selection approaches using a flood inventory dataset as a reference. To predict susceptibility to flooding, the selected features were sequentially ingested into a Random Forest based machine learning model. Findings show that: (i) detailed representation of rainfall and anthropogenic features improved the accuracy of FSP, (ii) Gini importance was found to be the most effective feature selection method, (iii) geomorphological and physiographic features showed the highest and the lowest predictive power, respectively, and (iv) the quantile method outperformed other approaches in classifying the flood susceptibility map. Findings indicate that an area of 217 km2, 43000 buildings, 163 km of paved roads and 0.54 million inhabitants are situated in highly susceptible parts of the catchment. To improve the FSP and effectively contribute to disaster risk reduction, the present study urges that a multidimensional representation of rainfall attributes be used within the models, and that the arbitrary selection of relevant features should follow a systematic approach.
Language: English
Subfield: Global and Planetary Change
Field: Environmental Science
Domain: Physical Sciences
Open Access status: hybrid
Estimated APC paid: $2,620
Award: G8086538336
OpenAlex ID: W4405910475