A hybrid model using data mining and multi-criteria decision-making methods for landslide risk mapping at Golestan Province, Iran

The accurate modelling of landslide risk is essential pre-requisite for the development of reliable landslide control and mitigation strategies. However, landslide risk depends on the poorly known environmental and socio-economic factors for regional patterns of landslide occurrence probability and vulnerability, which constitute still a matter of research. Here, a hybrid model is described that couples data mining and multi-criteria decision-making methods for hazard and vulnerability mapping and presents its application to landslide risk assessment in Golestan Province, Northeastern Iran. To this end, landslide probability is mapped using three state-of-the-art machine learning (ML) algorithms – Maximum Entropy, Support Vector Machine and Genetic Algorithm for Rule Set Production – and combine the results with Fuzzy Analytical Hierarchy Process computations of vulnerability to obtain the landslide risk map. Based on obtained results, a discussion is presented on landslide probability as a function of the main relevant human-environmental conditioning factors in Golestan Province. In particular, from the response curves of the machine learning algorithms, it can be found that the probability 𝑝 of landslide occurrence decreases nearly exponentially with the distance 𝑥 to the next road, fault or river. Specifically, the results indicated that 𝑝≈exp(−𝜆𝑥), where the length-scale 𝜆 is about 0.0797 km −1 for road, 0.108 km −1 for fault and 0.734 km −1 for river. Furthermore, according to the results, 𝑝 follows, approximately, a lognormal function of elevation, while the equation 𝑝=𝑝0−𝐾∙(𝜃−𝜃 0 ) 2 fits well the dependence of landslide modeling on the slope-angle 𝜃, with 𝑝 0 ≈0.64, 𝜃 0 ≈25.6° and |𝐾|≈6.6×10 −4 . However, the highest predicted landslide risk levels in Golestan Province are located in the south and southwest areas surrounding Gorgan City, owing to the combined effect of dense local human occupation and strongly landslide-prone environmental conditions. Obtained results provide insights for quantitative modelling of landslide risk, as well as for priority planning in landslide risk management.