Dear all,
I want to replicate landslide susceptibility map using weight value determined in a small area (60 km2) to a large area. What is the suitable method for this? and what are the parameters required for this?
Thank you.
Dear Pradhan,
It depends from the modeling approach and the GIS program you use.
For your susceptibility map, what is the format of your settings parameters (predsposing factors)?
Dear Mastere,
I want to use any bivariate or multivariate techniques.
The variables used in the study are Dist to drain, aspect, curvature, plane curvature, profile curvature, soil depth, soil type, geology, forest type, forest age, forest diensity, internal relief,soil drain characteristics, slope, sti, spi, twi and dem
Dear Pradhan,
First step is to find out which of the many used parameters in your study have really influence. For this purpose you should apply the AUC approach, introduced by Chung & Fabri 2003. In our experience a few parameters of the theoretical huge list of inherent factors play a decisive role. The importance of factors depends on the investigation area naturally. But, suppose you found 3 to 5 factors (parameters) responsible and you can validate a good model for instance (AUC > 80) than you can assign your calculated weights on the respective combinations to the sourroundings of your investigation area. Typically the spatial validity ends with the change in lithological properties, soil cover and vegetation. You have to think about the extend specifically.
For application of AUC applied to factor of safety see my paper first link and applied on weight of evidence (unfortunately in German) see in the second link.
Best regards Michael
http://www.sciencedirect.com/science/article/pii/S0169555X14003717
Conference Paper Torizin. J, Fuchs, M., Balzer, D., Kuhn, D., Arifianti, Y., ...
Dear Pradhan,
Replication of LSM (landslide susceptibility map) from small area to higher area suppose that the litho-structural properties and geomorphic features are supposed scalary invariant in your study area. Is it the case ? Another precaution is to be sure that the factors you put in your modelling process are conditionnaly independent (see Bonham-Carter ant its book on modelling process in geology); You can test this independancy with khi2 test or Cramer's V test. If it is not, your model will be wrong with some information redundancy. You can validate your model by build your model with the landslide in the small area and validate it with landslides locating in the higher area with the AUC, as Michael Fuchs suggest.
Best regards
Alexandre Poiraud
Dear Pradhan
meanwhile, you can use of some data mining methods such as Entropy or random forest.
in the mentioned models, you will get effect of each factors and then will select some conditioning factors that have the highest effect on landslide occurrence. finally, you can validate your built model based on ROC curve.
currently, we are working on some data mining models and comparison between them.
- Badges
- Science topic
- Similar topics
- Geoscience
- Geoinformatics