A Hybrid Variable Weight Theory Approach of Hierarchical Analysis and Multi-Layer Perceptron for Landslide Susceptibility Evaluation: A Case Study in Luanchuan County, China

Abstract

Landslides, which can cause significant losses of lives or property damages, result from several different environmental factors whose influences are very complex. Thus, the statistical multi-layer perceptron (MLP) and heuristic analytical hierarchy process (AHP) are employed in the evaluation of landslide susceptibility. However, the landslide susceptibility maps drawn by these two methods are always affected by subjectivity and randomness. In the present study, we introduce variable weight theory (VW) to improve the MLP and AHP methods, and two novel hybrid models, AHP-VW and MLP-VW, are respectively proposed. VW theory is used to redistribute the weights of the factors in the two constant weight evaluations. This is so that the weights of the factors change with different evaluation units, thereby eliminating the subjectivity and randomness problems. The landslide susceptibility maps of the study area were categorized into very low, low, moderate, high, and very high susceptibility grades. The landslide susceptibility maps of the four models are validated by the receiver operating characteristic (ROC) curve. The area under the curve (AUC) is 0.825 for the AHP model, 0.879 for the MLP model, 0.873 for the AHP-VW model, and 0.915 for the MLP-VW model. The results show that the landslide susceptibility map drawn by statistical MLP is better than that drawn by heuristic AHP, which is consistent with many other current research results. Furthermore, VW can significantly improve the performance of constant-weight single models. Landslide susceptibility maps drawn by the statistical MLP model hybrid VW can be used for regional land use planning and landslide hazard mitigation purposes.