A novel hybrid approach for landslide susceptibility mapping integrating analytical hierarchy process and normalized frequency ratio methods with the cloud model

Abstract

Landslides, which could cause huge losses of lives or property damages, result from several different environmental factors whose influences on landslides are very complex. Therefore, it is essential to understand the relationships between these environmental factors and landslides. Thus, the integration of the analytical hierarchy process (AHP) with the normalized frequency ratio (NFR) is evaluated for landslide susceptibility analyses. However, in addition to these complex relationships, the randomness and fuzziness always affect landslide susceptibility mapping. This study introduces the cloud model (CM) to improve the integrated AHP-NFR method, and proposes a novel hybrid AHP-NFR-CM method for landslide susceptibility analyses, which can better address issues of the randomness and fuzziness. Firstly, ten environmental parameters are selected as landslide impact factors, and their values for all the landslides identified in the study area are obtained through the remote sensing (RS) and geographical information system (GIS) technologies. The AHP method is used to obtain the weight of each landslide impact factor, and the NFR method is used to obtain the weight of each subclass in each landslide impact factor, which can reflect the relationship between the landslide impact factor and landslide occurrence. After applying an appropriate compositional operation between the weights of the landslide impact factors and the weights of the subclasses of the impact factors, a landslide susceptibility index (LSI) for each grid divided via the attribution-based spatial information multi-grid method (ASIMG) can be computed. To solve the inevitable issues of randomness and fuzziness in landslide susceptibility analyses, a cloud model that uses three numerical features (expectation, entropy and hyper-entropy) to represent the intension of the concept, is adopted to improve the methods of AHP and NFR. The relative importance of two landslide impact factors is scaled with the cloud model rather than the Saaty criteria. Pair-wise comparison matrixes of landslide impact factors given by each expert are described by the normal cloud model, and the floating cloud model is used to aggregate all experts' judgments. The weight of each landslide impact factor is also expressed with the cloud model rather than a certain value. In improving the NFR, the weight of each subclass of each landslide impact factor is expressed with the cloud model rather than a certain value. In the improvement of the landslide susceptibility results, the domain of landslide risk assessment results is also displayed with the cloud model instead of a series of definite intervals. As the study area examined is large, several grids would need to be divided, meaning that it would take a considerable amount time to subject the entire study area to landslide susceptibility mapping. Thus, we propose a new attribute-based spatial information multi-grid (ASIMG) division method and introduce grid-computing technology to improve the calculation efficiency during the process. Finally, the proposed hybrid AHP-NFR-CM-ASIMG approach is validated and applied in the study area. It's concluded that the new integration of AHP and NFR methods with the cloud model can consider both randomness and fuzziness and therefore can increase the robustness of landslide susceptibility analyses, while the ASIMG technology can enhance the calculation efficiency in regional landslide susceptibility mapping.