What can be the minimal number of triggering factors needed for acceptable landslide susceptibility modelling on regional (or even national) scale with frequency ratio or AHP approach? In our latest research of first time LS modelling in national extent, we used only 6 factors for which we have enough reliable data: slope angle, lithology, land cover, convexity, distance to roads and distance to streams. With these factors and 302 landslides in the inventory for the entire country (25713 km sq.) we got acceptable AUC of 0.78. Still, some researches are not convinced in the results with such number of factors. We try to include precipitations as a factor, but in our country they increase generally in vertical direction - with altitude, and the landslide inventory in the higher zones is not very reliable. Thus, the FR or LSI for this factor is not very reliable too. So please help us with your opinion or experience. Thank you in advance.
I think the number of factor depends on your study area, it is necesarry to consider all the main causes. In general, together with your factors, you can consider "active faults" and "drainage system".
Lanslide historical data is one of most important for lanslide suspectabily mapping.
Dear Milevski
According to the present literature there is no universal criteria about number of landslide conditioning factors. The selection of such factors are still subjective and depends on expert opinion. The most common landslide conditioning factors using in literature are viz., Lithology, Slope angle, Slope curvature (either total curvature or profile and plan curvature separately), TWI, NDVI/Land use & land cover, Distance from streams as well as lineaments or faults, Soil texture or slope material, distance to roads and altitude. If the study area have enough number of landslides initiated by rainfall events then we can add precipitation also. Further it is important to identify the collinearity among the factors to avoid noise in the finale model. You can identify the most important factors which contribute to the occurrence of landslides by statistically through Principal component analysis, information gain ratio (IGR), LCZ, and so on (please go thorough some recent literature). If you find the important factors then create the final model using the same. Try to convince the researchers through statistical testing of variable importance. If the factors which you already used (minimal number of factors ) are statistically significant contributor then its enough else you may select 10 to 15 factors do variable importance analysis and find most significant among them. Finally you test the sensitivity of AUC values on each factors by an omission and inclusion.
Can you send or show the coordinate and the exact location or name of the area to explore and help ?
I think near ten factors depends on the terrain features in the frequency ratio model.
In addition to the factors listed above, the slope in mountainous areas and the land cover on it should be considered for acceptable level of satisfaction in landslide susceptibility modeling. I agree with above answers that the factors will depend on your study area.
In general, you need at least a geometric metric (e.g. slope angle), a hydrological metric (e.g. upstream area), a material metric (e.g. lithology, soil thickness) and, for earthquake-induced landslides, also a seismic metric (e.g. expected pga).
In my opinion, based on available data, field work analysis and data collection from previous project you can decide how many landslide causative factors are select. As we all know, landslide ussually occur in mountainous area, mountain foot, along rivers and springs and this phenomena also stronly effect by human activities such as cutting slope for contruction of roads , remove land cover on the hills and mountains... However, the most important factors is always slope degree, weathered crust, other factor that considered as triggering factors such as, pricipitation, tectonic activities that relate to faults density, lineament density, earthquake. But the critical role is quality of data and the knowledge of the study area and finally, selecting the most suitable method for application. Good luck for you!
Dear Ivica Milevski,
Landslide is controlled by several conditioning factors and so far there in no agreement about the number of the conditioning factors that should be used. However, it not necessary increasing the number of the conditioning factors can increase the obtained accuracy. I suggest you to use all the available factors that listed in the literature then assess the multicollinearity among these factors so you can remove the correlated factors. After that you should determine the best subset of the remaining factors by applying a technique such as Ant Colony Optimization (ACO).
This is depend on the nature and location of the study area. Anyway, you have an expert in land slide mapping
you can consult Prof.Beswagit. He successfully published tremounds numbers of research articles on land slides in tropical region
- Badges
- Science topic