Application Of Machine Learning To Map Flood Susceptibility In Chadereka Ward, Muzarabani District, Zimbabwe
- Author
- Tawonezvi, Irene
- Title
- Application Of Machine Learning To Map Flood Susceptibility In Chadereka Ward, Muzarabani District, Zimbabwe
- Abstract
- Floods are a leading disaster that often result in loss of thousands of human lives, affect millions of people, damage property and infrastructure worth several billions of US dollars globally. The increase in the intensity, duration and frequency hence number of flooding events due to a combination of climate change and anthropogenic factors motivated the need to explore accurate flood mapping using a robust model. The aim of the study was to model flood susceptibility using robust machine learning in the flood-prone Chadereka Ward in Muzarabani District, Zimbabwe. Fieldwork was conducted using transects across floodplain for ground truthing. Landsat 8 image was secondary data downloaded from online repository. The study calculated the normalised difference water index (NDWI) and implemented Random Forest algorithm, a decision tree machine learning using Semi-automatic Plugin (SCP) in QGIS to model flood-prone areas validated by field observations. The study used boundary data sets from Diva GIS, digitised physical infrastructure in Google Earth image in order to visualise the spatial distribution of phenomena. The spatial overlay of the physical infrastructure on a Random Forest produced flood extent map defined vulnerability and identified the infrastructure at risk of flooding. The study achieved its first objective by identifying key infrastructures such as the clinic, business centre, Chadereka primary school, and nearby homesteads predominantly located between the Hoya and Nzoumvunda rivers, indicating significant flood risk due to their proximity to these water bodies. For the second objective, the study utilized the Random Forest algorithm to model flood extent in Chadereka Ward, revealing increased flood risk along river areas with an impressive overall model accuracy of 97.75%, thus providing reliable predictions of flood-prone zones. The study addressed the third objective by overlaying the flood susceptibility map with the distribution of physical infrastructure map, revealing that infrastructures along the river banks are particularly vulnerable to flooding, effectively determining flood vulnerability in Chadereka Ward. The study demonstrated that Random Forest machine learning flood susceptibility modelling has great potential for proactive flood risk management in similar flood-prone areas. Therefore, the study recommends the testing of transfer learning techniques to model flood susceptibility in similar areas.
- Date
- MAY 2024
- Publisher
- BUSE
- Keywords
- Machine Learning, Flood Susceptibility,Muzarabani District.
- Supervisor
- Mr Pedzisai
- Item sets
- Department of Geosciences
