| Landslide poses an incremental threat to the life of individuals,livestock,and properties.It is solemnly true in Enugu state,located in the southeastern part of Nigeria,which has suffered numerous landslides.In recent times,the complex nature of landslides caused by natural instability by both climatic,influencing factors,and human activities in Enugu State has increased landslide processes and occurrence,which imposes some modifications in the evolution of landscape architecture.Hence,with the numerous landslide experienced in the region and the hazardous nature of the area,their controlled and triggered mechanisms have remained less studied.Therefore there is a need for a comprehensive investigation of landslide susceptibility assessment which is crucial for sustainable environmental development and management.A landslide susceptibility assessment study was carried out,aiming to 1)monitor the landslide situations of Enugu state and its environs with emphasis on the significant research advances of landslide recorded and landslide distribution in the region so far,and1)Comprehensively determine possible factors responsible for the landslide occurrence.Its primary innovations are established on landslide inventory map,the quality of factors used,factors selected and screening methods and the successful application of remote sensing,field observation,geographic information system tools and techniques,and the implementation of machine learning algorithms to understand the non-linear relationship between the landslide occurrences and the influencing factors.Uncertainty analysis was carried out to compare the models and landslide susceptibility map of Enugu state was produced,and localities prone to high risk of future landslides were determined and highlighted from the various models.To buttress the achievable contents used in this paper lies on five basic stages;1)After research advances on landslide susceptibility in Enugu state from past literature and comprehensive investigation,suitable landslide data was extracted from numerous sources to prepare the Landslide inventory map.296 locations were verified.Thus,the landslide inventory datasets were randomly assigned to a group of 70% for training the models,and 30% was allocated for testing purposes.2)For suitable landslide susceptibility assessment,seven topographical factors(elevation,slope,aspect,plan curvature,profile curvature,stream power index,and topographic wetness index),two hydrological factors(rainfall,distance to drainage),two lithological factors(geology,distance to fault),four environmental factors(soil,land use,distance to road and NDVI),which is a collective of Fifteen influencing factors were collected for landslide susceptibility assessment First,the author test the predictive ability of each of the influencing factors obtained from various sources using Information gain ratio to effectively select factors that contribute to landslide occurrence in the study area.And the result using the Information gain ratio techniques revealed that rainfall has the highest contribution to landslide occurrence.In contrast,geology and soil have the least contributing ability and were excluded for further analysis.Secondly,For the Multicollinearity Check,Pearson Correlation coefficient was used to judge the correlation of the influencing factors,and the result revealed that elevation and distance to drainage have the highest correlation value(moderately correlated)3)Four machine learning models,including extreme gradient boosting(XGBOOST),random forest(RF),na(?)ve Bayes(NB),and k-nearest neighbor(KNN),supported with R statistical program environment,were used to design the models,and the variable importance using each of the models was plotted.The studied landslide susceptibility map was divided into five classes(very low,low,moderate,high,and very high susceptible class)using natural break in R and completed in the Arc GIS platform.4)The different models were quantitatively evaluated and compared using the receiver operating characteristics curve(ROC)and area under the curve(AUC);statistical analysis of the model results of AUROC values of RF,XGBOOST,NB,and KNN was0.935,0.922,0.894 and 0.868 for the prediction rate curve.In comparison,it is demonstrated that the RF model has the highest accuracy for landslide analysis in the study area,followed by XGBOOST,NB,and KNN.The four machine learning models shown to be suitable for landslide susceptibility assessment in Enugu state.5)In Conclusion,this work can serve as a point of reference for susceptibility risk assessment and management in Enugu state.And it is also recommended that to prevent future landslide risks,more advanced techniques should be studied and implemented for steady,continuous monitoring to aid in planning adaptation,sustainable environmental development,and adequate policy making. |
PDF Full Text Request