SpaceMika

Although I’ve found a surprising number of landslide-bloggers (my favourite is Dave’s), google searches on the DAN-W and DAN3D software packages seem to drop people here. I’m fairly regularly getting comments asking how to go about modeling particular landslides, or acquire the software, or related queries. To speed up response time, I’ve developed an F.A.Q.

1. How can I get DAN-W or DAN3D?
DAN-W is owned by Oldrich Hungr. Please see his software website and contact him directly with any inquires about acquiring the software. DAN3D was developed by Scott McDougall as part of his phD thesis. To the best of my knowledge, it is for research purposes only and not currently available commercially, but again, Oldrich Hungr knows for sure.

2. What information do I need about a landslide to model its runout?
For DAN-W, you need a profile of the travel path (including entrainment zones) and the source area, and the width of the path. This can be either a list of coordinates or a to-scale sketch which you can enter directly into the software. For DAN3D, you need digital terrain models (topography) of the area before and after the landslide. You will need to format this as ASCII grid files of the path, the source area, and any entrainment.

3. What rheology should I use?
If you’re doing a back-analysis, you use whatever rheology and parameters provide the appropriate runout distance, debris distribution, and velocity profile. If you’re doing a forward prediction, you can follow the suggestions in my thesis (currently TBA, sorry!), or back-analyze cases similar to your target and use that range of parameters in your prediction.

4. Tell me more about a particular landslide.
If I’ve personally modeled a landslide, it should be floating around this site somewhere. Most are linked off the Thesis page, although the latest versions haven’t been translated from thesis-formatting to website-formatting and hopped online yet.

5. What about modeling this specific landslide not on your website?
If you’re working on modeling a landslide I haven’t seen before, I’m curious. Tell me about it!

• analysis of landslide motion
• delineation of impact area
• estimation of hazard intensity parameters within impact area
• characterization of secondary effects

In runout analysis, landslide motion is analyzed, followed by outlining the runout area — all the area that impacted by a landslide. Then, the annalist estimates the intensity of hazard parameters within the impact area such as how deep is the flow and how fast did it travel at a given point. Finally, one characterizes any secondary effects* resulting from the landslide, such as a landslide dam giving way to an outburst flood.

* I don’t actually know what other secondary effects would be — I need to look this up!

1. compile data on path geometry and character
2. debris distribution & velocities
3. select similar cases to the target slide (forward prediction only)
4. run program to obtain required runout
5. compare debris thickness, velocity distribution
6. select “best fit” rheology and parameters

from the Geohazards IV paper:

For all case histories, the runout path is based on detailed topographic maps. The material rheologies are determined through empirical calibration. Systematic trial and error back-analyses of similar previous events are used to constrain the rheological relationship and the specific resistance parameter values. Validity of the model is assessed through comparison of calculated and observed landslide characteristics, such as deposit distribution and extent, velocity profiles, and volume. In the December 2007 benchmarking exercises at the International Landslide Risk Management Forum in Hong Kong, the DAN-W and DAN3D models produced results consistent with those of other models from several countries around the world

I compile data from papers describing case histories. In these papers, sometimes there is a profile of the event that may be manually input into DAN-W for a quasi-3D analysis. If this information is not available, I search for digital data sources. My primary source is HeyWhatsThat, which extracts path profiles using a google interface. If I want to continue preliminary analysis from DAN-W into a true-3D analysis in DAN-3D, I will use any public-domain digital elevation model. Most of these topography sets are provided by government geologic surveys.

Other data I extract from the papers is debris distributions, velocity profiles, runout distances and geometries, and any additional information on the runout characteristics. When I am modelling the landslide, the debris distribution provides the first clue for the appropriate rheology (frictional flows have a tapered distribution thicker towards the source, Voellmy has an even distribution, and Bingham has a leading bulge*) while the runout distance, time, and velocities determine the exact parameters. I do several runs, comparing model results to the reported case study parameters as I do, and settle on a model with parameters resulting in a best fit of landslide characteristics.

Questioning the accuracy of case history reports is outside the bounds of my thesis. What is reported is all that I’m working with. Without field investigations of my own, I have no basis to judge for myself. If multiple papers contradict each others’ findings, I use the most-accepted interpretation. This is usually the most recent paper.

* confirm that the rheologies have the appropriate debris distributions!

I use custom in-house software for runout analysis, DAN-W and DAN-3D. These are the assumptions behind why they (or most dynamic analysis software) work

• landslides are complex, but predictions are possible
• must concentrate on the external aspects of behaviour
• must be open-minded about rheologicial character
• analysis must consider the character of path material
• material entrainment should be considered
• model verification and calibration are essential

Landslides are complex, both in material and in motion. By concentrating on external behaviour, bypassing internal aspects that are challenging to gather direct evidence for, a landslide expert may still make predictions about landslide behaviour. Instead of attempting to mimic the heterogeneous muddle of real landslide rheology including mud, tree trunks, boulders, and anything else that is entrained in the flow, the DAN models use simple homogenous fluids that have external behaviours matching real landslide flow. A landslide is affected not just by the material that fails, but by the material of the path it flows over and the material it entrains. As direct laboratory tests to determine the rheological properties of materials are not practical (how could you gather a sample, and what would you test?), verification and calibration of the models against back-analysis of real case histories is necessary. All the landslide case studies in this thesis could all be well-modelled using frictional, Voellmey, or Bingham fluids.

From the Geohazards IV paper

DAN-W and DAN3D are dynamic analysis models designed to predict the velocity and extent of motion of rapid landslides such as debris flows and avalanches, flow slides and rock avalanches. DAN-W works on a two-dimensional slope profile and requires input of any changes in flow path width (Hungr 1995), while DAN3D is a three-dimensional extension of the model which requires input of a digital terrain model. (McDougall and Hungr 2004 and 2005) Both models can simulate landslides which entrain path material and have variable rheology. The models typically produce similar results when consistent input parameters are used. The models diverge somewhat when there is significant deflection of the flow. This is likely because the DAN-W model requires the input of a pre-defined path profile and may not accurately simulate the energy losses associated with sharp changes in direction.

DAN-W in action

This is how I spend my nights: clicking little dots onto other little dots, adjusting with more dots to make the spline even around tight turns, and hoping all the external data I’m using is cross-consistent. I’ve been trying to get the 1987 Val Pola, Italy profile correct, but so far no matter what I use for input (figures from papers, slices taken from DAN-3D terrain, or profiles extracted from googlemaps) my volume is ending up at 60-70 million cubic meters instead of the 40 million cubic meters I should get. Sigh.