Multi-Criteria Models

What can social vulnerability models can teach us about building other multi-criteria models?

My interest in multi-criteria models

Although I have not applied social vulnerability models to my field of study, I am very interested in modeling geohazards to make predictions that can aid public safety. Specifically, I am interested in modeling mass wasting events such as landslides and debris-flows in the spatial and temporal dimensions and also for size. I would assert that “false negatives” have a higher cost in hazards modeling that in social vulnerability modeling. If a geohazard is predicted to be less likely than it actually is, people will be unprepared and damages will be more costly. If a community is predicted to be more vulnerable than it actually is, the community will get more funding than it actually needs. However, the cost here lies in that this might take away funding from other more vulnerable communities.

Strategies for modeling landslides

In recent months, I have been thinking a lot about landslide modeling and mapping because I’ve attended various talks at conferences that take different approaches to landslide modeling. One method is to build an empirical model based on quantitative factors known to be correlated to landslides (slope angle, vegetation cover, soil properties, recovery from wildfire, etc). This method was popular in the American Southwest and Pacific Northwest because of the availablity of high resolution data (in particular, LiDAR DEMs and soil surveys) and the vast amount of localized research about landslide mechanics in these environments.

A second method is to take data of landslides that have happened in an area and compute the statistical probability that a landslide will occur on a given slope based on this training data. This method was used when extensive data was not available—only the occurence of a landslide was recorded. One researcher used one such inventory in Southeast Asia along with stream and road data to try to estimate landslide risk.

Learnings from social vulnerability modeling

Reading Rufat et al. (2019) made me think about these two approaches to building models from a validity perspective. The first model (modeling underlying physical properties) would likely hold up well in an empirical validation. The second model (based on statistics) would likely hold up well in a robustness validation. The way we choose to validate models is just as important as model structure, and the choice of validation method lead to a biased assessment of model results.

Tate (2013)’s framework for uncertainty analysis in a social vulnerability index can be applied to other multi-criteria models. I reflected on a project I did creating a landslide hazard map based on rainfall rates in the context of this framework. Using empirical factors like slope angle, catchment area, and various soil properties, my goal was to calculate the rainfall rate needed to trigger a landslide. My DEM had 1m resolution, so my model calculated a rainfall rate for each pixel. My results were extremely noisy, even when I resampled my pixel size to a larger resolution. 1m resolution is not reasonable for this scenario because landslides are often tens of meters across in the initiation zone, and small areas of topographic steepness won’t be as susceptible as larger areas. This emphasizes the importance of analysis scale. I initially thought of this parameter as “smaller is better”, but this is an example of that not being the case.

The need for pairing social vulnerability and hazards modeling

Cutter et al. (2003) eluded to the need to couple social vulnerability indices (specifically modeling vulnerability to climate change) with information about hazard frequency and severity. I was surprised that the study found that Yellowstone National Park County, MT (among others) had one of the lowest national social vulnerability ratings. In June 2022, severe flooding was declared a FEMA disaster, and the communities in and around the park experienced huge losses struggled to recover (aka were highly vulnerable and not very resilient). Despite indicators that scored the county a low social vulnerability score, this area seems highly socially vulnerabile due to its remoteness and dependence of the local economy on tourism in and around the park.

Related Reading

Cutter, S. L., Boruff, B. J., & Shirley, W. L. (2003). Social vulnerability to environmental hazards. Social Science Quarterly, 84(2), 242–261. https://doi.org/10.1111/1540-6237.8402002

Rufat, S., Tate, E., Emrich, C. T., & Antolini, F. (2019). How Valid Are Social Vulnerability Models? Annals of the American Association of Geographers, 109(4), 1131–1153. https://doi.org/10.1080/24694452.2018.1535887

Tate, E. 2013. Uncertainty Analysis for a Social Vulnerability Index. Annals of the Association of American Geographers 103 (3):526–543. https://doi.org/10.1080/00045608.2012.700616.