Spatial Modelling of Landslide Susceptibility Using Logistic Regression Model in the Bamenda Escarpment Zone, NW Cameroon

Abstract

The Bamenda escarpment zone which is about 8.4 km²is one of the most hazardous zones along the North-West stretch of the Cameroon Volcanic Line. Landslide events in this area cause harm to humans and economic disruptions almost on a yearly basis. The impact of this hazard is highly felt in Bamenda I, II and III municipalities especially those communities leaving at close proximity to the escarpment. The steep slopes dominated by volcanic rocks and highly weathered residual soils are easily pulled down by gravity during the rainy season.  The objective of this study is to use a quantitative technique (logistic regression model) in assessing landslides susceptibility in the area. Landslides were identified using aerial photographs, satellite images and systematic field recognition. A total of 110 landslides mostly shallow translational landslides were registered during the landslide inventory process. This landslide dataset was randomly divided into two groups and the first group was used in building the predictive model and the second in its validation. Ten landslide conditioning factors were initially considered for the assessment and after being tested using the accountability and reliability indices, six out of the ten factors were retained for the susceptibility modeling. The slope was the most important geo-environmental factor influencing landslides and had a coefficient of 0.927. The model was validated using the success and prediction rates methods and the Area Under the Curve (AUC) was used to evaluate the performance of the model. The training model had a success rate of 81% and the validation model had a prediction rate 88%. The predictive power and accuracy of the model was evaluated using the error index of the ROC curve. The model had a True Positive rate of 0.554, a True Negative Rate of 0.879, an accuracy of 0.879 with a precision of 0.006. From this result, we concluded that landslide remains a serious threat in the area and can be predicted statistically.