What type of study is this?

August 21, 2025

Machine learning-integrated landslide susceptibility modelling addressing spatial correlation in grid-based environmental factors

YLYuanyao LiTianjin Institute of Metrological Supervision Testing FLFan LiChina University of Geosciences YZYing‐Jie ZhuZhejiang Chinese Medical University

Key Points

Models incorporating spatial correlation improved landslide susceptibility prediction, enhancing predictive accuracy significantly.
The machine learning model with the highest accuracy was the spatial correlation-random forest model, achieving an AUC of 94.35%.
By evaluating the receiver operating characteristic and accuracy indices, significant performance differences were observed.
Results indicate that integrating spatial correlations leads to better predictive outcomes in landslide susceptibility models.

Abstract

The grid units are assumed to independent in grid-based landslide susceptibility prediction (LSP). However, interactions exist due to the continuity and correlation of landslide environmental factors. Hence, the influences of spatial correlation among these factors are considered in the study. The domain analysis in ARCGIS 10.2 software is adopted to calculate the mean, standard deviation and range of fifteen environmental factors. Then, it incorporates spatial correlation into the dataset. Three machine learning models, such as support vector machine (SVM), C5.0 Decision Tree (C5.0 DT) and random forest (RF), are utilised to establish spatial relationship models. Subsequently, prediction performance is evaluated with the receiver operating characteristic (ROC) curve, frequency ratio accuracy and statistical indices of landslide susceptibility indexes (LSIs). Results show: Models considering spatial correlation outperform those that do not, with an increase in area of ROC curve (AUC) values by 2–6%, frequency ratio accuracy by 0.7–1.3 and smaller LSI mean and larger standard deviation. Moreover, the spatial correlation-RF model had the highest accuracy with AUC of 94.35%. The LSP accuracy can be enhanced by considering spatial correlation of environment factors.

Ask AI

Helpful

Bookmark

View Full Paper