Question

For Exercise 1, I evaluated the spatial pattern of economic losses resulting from a joint earthquake and tsunami event. I then deaggregated the losses by hazard (earthquake only, tsunami only, and combined) as well the intensity of the event.

For Exercise 2, I evaluated how the spatial pattern of economic losses resulting from a tsunami relates to the spatial pattern of tsunami momentum flux (a measure of velocity and inundation depth) by performing a geographically weighted regression (GWR). For this analysis, I only considered the tsunami because there is significant spatial variation in the hazard, whereas the spatial variation for the earthquake is minimal.

Tool and approach

I performed the GWR using the python library PySAL (Python Spatial Analysis Library). The independent variable was defined as the momentum flux, and the dependent variable defined as the percent loss of economic value.

Description of steps

The average losses at each building resulting from an earthquake/tsunami loss model were first converted to percent loss (loss divided by real market value), and added as an attribute to a GIS shapefile. The percent loss was used as opposed to the economic losses because each building has a different initial value. Consequently, the percent loss serves to normalize the economic losses across all buildings within Seaside. For this analysis, the results from the “1000-year” tsunami event were analyzed.

The GWR was then performed using PySAL with the momentum flux defined as the independent variable and the percent loss defined as the dependent variable. The GWR resulted in a slope and intercept at each tax lot, as well as a global r² value. Two separate maps were generated wherein each tax lot was color coded based on values of the slope and intercept.

Description of results

The results from the GWR and a global regression are shown in Figures 1 and 2 respectively. A global r-squared value of 0.575 was obtained, indicating that the data is moderately correlated. In Figure 1, it can be seen that the intercept is larger near to the ocean, and decreases as the distance to the shore increases. This can be explained by the fact that the momentum flux is the largest near to the coast, and decreases as the tsunami propagates over the land.

Similar trends would be expected for the slope coefficients; however, it can be seen that along the coast the results are negative indicating that the economic losses decrease as the momentum flux increases. This can likely be explained by inconsistent building types within Seaside. For example, concrete buildings are able to better withstand the impact of a tsunami compared to their wood counterparts. Similarly, buildings of different heights (number of floors) have different damage properties. Consequently, because the building types are not consistent within Seaside, significant variations in the percent of loss within a small spatial region can occur (e.g. a wood building is located next to a concrete building). This would lead to a decrease in percent loss for a larger momentum flux.

Figure 1: Spatial variation of slope and intercept resulting from the GWR

Figure 2: Global regression and line of best fit

Critique

While the GWR does provide a means to evaluate correlation between two variables that are within the same geographical region, there are limitations for this particular application. The results showed negative slopes in some locations, which is likely caused by the large variation in the percent loss. To alleviate this, alternative statistical models could be developed using GWR that only consider similar building types. An example of a non-spatial regression for wood buildings with 2 and 3 floors can be seen in Figure 3. The improvement in r-squared values can be observed, and would likely translate to the GWR.

Figure 3: Example of global regression considering specific building types