Skip to main content

Lund University Publications

LUND UNIVERSITY LIBRARIES

Flood susceptibility mapping in the Nyabarongo Catchment, Rwanda, based on data analysis and modeling

Nzabonantuma, Leonard LU ; Nduwayezu, Gilbert LU orcid ; Naghibi, Amir LU ; Nilsson, Erik LU ; Wali, Umaru Garba LU and Larson, Magnus LU (2025) In Geomatics, Natural Hazards and Risk 16(1).
Abstract

Rwanda’s Nyabarongo catchment frequently experiences floods, highlighting the need for effective flood susceptibility analysis and management. This study mapped flood susceptibility in the catchment using the random forest (RF), support vector machine (SVM), eXtreme Gradient Boosting (XGBoost), SHapley Additive exPlanations (SHAP) models, as well as various conditioning factors including elevation, curvature, aspect, distance to river (DTRiver), Distance to road (DTRoad), normalized difference vegetation index, slope, curve number, topographic wetness index (TWI) and rainfall. RF was the best performing model with an area under curve (AUC) of 0.968 and an F1-score of 0.92, demonstrating its high performance and robustness in flood... (More)

Rwanda’s Nyabarongo catchment frequently experiences floods, highlighting the need for effective flood susceptibility analysis and management. This study mapped flood susceptibility in the catchment using the random forest (RF), support vector machine (SVM), eXtreme Gradient Boosting (XGBoost), SHapley Additive exPlanations (SHAP) models, as well as various conditioning factors including elevation, curvature, aspect, distance to river (DTRiver), Distance to road (DTRoad), normalized difference vegetation index, slope, curve number, topographic wetness index (TWI) and rainfall. RF was the best performing model with an area under curve (AUC) of 0.968 and an F1-score of 0.92, demonstrating its high performance and robustness in flood susceptibility analysis. In addition, RF, combined with SHAP, provided both robust and interpretable results. The study found that DTRiver, TWI, DTRoad, and slope had the highest influence on model predictions, while curve number had the least. RF classified the area into five flood susceptibility classes: very high (6%), high (9.6%), moderate (15.1%), low (26.6%), and very low (42.7%), accurately reflecting environmental and geo-topographic conditions. Based on these findings, mitigation measures can be designed to reduce flood risk in the Nyabarongo catchment. Additionally, the models have potential for application across Rwanda to improve flood susceptibility management and could be adapted for use globally.

(Less)
Please use this url to cite or link to this publication:
author
; ; ; ; and
organization
publishing date
type
Contribution to journal
publication status
published
subject
keywords
eXtreme gradient Boosting, flood conditioning factors, flood inventory, random forest model, Shapley Additive exPlanations
in
Geomatics, Natural Hazards and Risk
volume
16
issue
1
article number
2556987
publisher
Taylor & Francis
external identifiers
  • scopus:105016996713
ISSN
1947-5705
DOI
10.1080/19475705.2025.2556987
language
English
LU publication?
yes
id
c864be96-cbbd-4072-9821-8738e1be2d28
date added to LUP
2025-12-09 08:30:56
date last changed
2025-12-10 03:44:07
@article{c864be96-cbbd-4072-9821-8738e1be2d28,
  abstract     = {{<p>Rwanda’s Nyabarongo catchment frequently experiences floods, highlighting the need for effective flood susceptibility analysis and management. This study mapped flood susceptibility in the catchment using the random forest (RF), support vector machine (SVM), eXtreme Gradient Boosting (XGBoost), SHapley Additive exPlanations (SHAP) models, as well as various conditioning factors including elevation, curvature, aspect, distance to river (DTRiver), Distance to road (DTRoad), normalized difference vegetation index, slope, curve number, topographic wetness index (TWI) and rainfall. RF was the best performing model with an area under curve (AUC) of 0.968 and an F1-score of 0.92, demonstrating its high performance and robustness in flood susceptibility analysis. In addition, RF, combined with SHAP, provided both robust and interpretable results. The study found that DTRiver, TWI, DTRoad, and slope had the highest influence on model predictions, while curve number had the least. RF classified the area into five flood susceptibility classes: very high (6%), high (9.6%), moderate (15.1%), low (26.6%), and very low (42.7%), accurately reflecting environmental and geo-topographic conditions. Based on these findings, mitigation measures can be designed to reduce flood risk in the Nyabarongo catchment. Additionally, the models have potential for application across Rwanda to improve flood susceptibility management and could be adapted for use globally.</p>}},
  author       = {{Nzabonantuma, Leonard and Nduwayezu, Gilbert and Naghibi, Amir and Nilsson, Erik and Wali, Umaru Garba and Larson, Magnus}},
  issn         = {{1947-5705}},
  keywords     = {{eXtreme gradient Boosting; flood conditioning factors; flood inventory; random forest model; Shapley Additive exPlanations}},
  language     = {{eng}},
  number       = {{1}},
  publisher    = {{Taylor & Francis}},
  series       = {{Geomatics, Natural Hazards and Risk}},
  title        = {{Flood susceptibility mapping in the Nyabarongo Catchment, Rwanda, based on data analysis and modeling}},
  url          = {{http://dx.doi.org/10.1080/19475705.2025.2556987}},
  doi          = {{10.1080/19475705.2025.2556987}},
  volume       = {{16}},
  year         = {{2025}},
}