Mostly, landslides are mapped using aerial photographs, satellite images and other higher remote sensing techniques. Field based mapping is next approach in mapping the landslides. While we generate landslide susceptibility or hazard model using the inventory, we have to validate the model. This can be done by randomly partitioning data in training and testing data-sets (for success and prediction rate). ROC plots/AUC can be used to validate the model.
Is it MUST/NECESSARY to verify each and every landslides (in the inventory) in field? How to deal with this shortcomings in the quality of inventory maps?
It is not necessary to validate all the landslides if the inventory is large you can do some random sampling and check your results. One best approach is classification as you may have classified the landslide zone in different categories you can validate one or two from each class. Mostly importantly validate those zones which can directly effect the human component in your area or the most crucial areas like close to habitation/Roads etc.
I am agree with Dr Rafiq. It is not mandatory but you should check as much as you can and mostly the ones that can have effect on the population or human infrastructures.
Furthermore, you would need more testing point or less depends on the quality of your DEM/ images that you are using.
I hope this will help you
Depending on the size of your inventory, it might be impossible to check them all. But it is important to get a feeling for the quality of your data. Are the landslides mapped as polygons or as points? If polygons, then you should check for instance if the entire landslide was mapped or just the part that has moved (without the crown). If points, it is important to check where the points are located. For susceptibility modelling is may seem rather useless if they are right in the middle of the landslide - you should prefer points at the headscarp.
And please do not only pay attention to the ROC value for estimating the quality of your statistical output. It should also make sense from a geomorphological point of view and should delineate high susceptibility where this is really the case.
You have to see the model you used already if it has already been validated by other researchers.The more you choose points on the map to validate them on the field plus the model is robust
The point for such modelling approaches usually is that you can analyse a larger area/dataset than what is possible to validate in the field (otherwise you could choose the reliable field approach from the beginning). However, it is important to validate the model to have controll over the quality of the model. For this, I agree with the others, it is important to choose a subset, which is representative for all classes and check those in the field. Of course, the more landslides you check, the more robust the validation. Another important part are the statistics that you choose to validate your model. Make sure you don't tweek the results to a good fit, but deal with whatever results you get. If they don't fit well, try to find out why.
Dear Badal,
Field survey is essential, but certainly not for everyone. Mohammad correctly identied the use of random sampling. However, could I suggest that any sampling takes into consideration factors such as Rock type present, slope angle, weather and climatic factors, latitudinal location, the presence or absence of glacial or volcanic phenomena. There may be other factors to consider as well.
For example, survey possibly 3 landslides from a group located on one rock type, eg granite or schist or sandstone, or clay. Choose one smaller example, one larger and one of apparently average size. Apply the same set of rules related to the measurements you are carrying out. In this way you should be able make cross comparisons from different areas.
kind regards
George
I forgot to add that you must be able to justify any sampling and why you included a,b and c but not x, y and z. Good sampling is paramount or you could bias your findings
George
I think it is very important to verify in the field a statistically significant portion of the inventory in order to identify possible systematic errors.
Thank you so much all for your valuable comments. It helped me.
Regards,
Badal
- Badges
- Science topic
- Similar topics
- Political Science
- International History and Politics
- Globalization