Note: You may not read this as I made this just so it's more accessible for me, but if you'd like, here's the link for the first part. Thank you.
I noticed this was too long, so I decided to cut it and put it in another narrative as part 2. Btw, yeah I sometimes re-read my entries here that's why. :)
Part 2:
The coverage metric employed was the fraction of vertical volume observed (FVO) between 0 and 20 kft, and it was because the WSR-88D has a perfect coverage above 20 kft. The cross-radial horizontal resolution (CHR) was also employed because it was relevant in taking the angular resolution. During the analysis period, the radar network underwent changes and they addressed it accordingly.
In order to relate the relationship between the warning performance and the quality of radar coverage, they matched each flood event to the appropriate catchment basin using the US Geological Survey (USGS) National Hydrographic Dataset which contains 19,031 stream gauges with corresponding catchment boundaries. They searched for a stream gauge located inside the event polygon and computed the mean radar coverage metric matching the catchment basin. With this step, they were able to match 24,236 flash flood events over the analysis period. This was their basis for all the analyses conducted in flash flood events. In detection probability dependence of flash flood on radar coverage, for FVO and CHR, the data for each were binned using cumulative distribution percentage intervals. Result shows that flash flood probability of detection (POD) increases with FVO and decreases with CHR, which is an important result because it shows connection between radar coverage and warning performance and allows a functional mapping between the two.
To compute for the flash flood warning flase alarm ratio dependence on radar coverage, they matched each warning to the corresponding catchment basins, just like what they did in mapping the flash flood events. They replaced the event polygon with the warning polygon and were able to match 32,438 flash flood warning to source basins over the analysis period. Because the casualty regression analysis did not yield a statistically meaningful relationship between FAR and casualty rate, they did not include this in the benefit model. In the dual polarization upgrade, the mean flash flood warning values did not yield statistically meaningful differences, which is in contrast with case studies that showed otherwise. This led to difficulties in obtaining QPE data so they used an algorithm technique that will not cause much error in the QPE. They also investigated the relationship between radar coverage and non-flash flood warning performance using the same procedures but it showed no significant relationship so it was not included in the benefit model.
For the relationship of flash flood warning and casualty rate, the factors that are thought to affect this relationship are population, time of day, building type, catchment basin size, water flow velocity and depth, rate of water-level rise, and the warning lead time. Only variables that could be geospatially characterized were included. For the casualty count, they used a negative binomial distribution model instead if a Poisson distribution because the casualty variance was larger than the mean statistics over the analysis period. The variables that they tried for the regression analysis were logarithm of the population, fraction of population in mobile housing, historical flash flood warning FAR, catchment basin size, flash flood warning presence and warning lead time. These variables were tested both individually and in combination to see any cross-relation effects. Population data was acquired from the Center for International Earth Science Information Network.
The value of statistical life was used to monetize the casualties. They used the equation as per guidance by the US Department of Transportation. In estimating the value provided by the radar network, they computed the modeled casualty costs in three scenarios: the current WSR-88D network, no radar coverage and perfect WSR-88D-like coverage. The results show that the current WSR-88D network provide over $300 million annual benefits and the perfect radar coverage yields $13 million annual benefits. For the purposes of QPE to support flash flood warning decisions, the result provided by the radar is very good.
Instrumentation
The instruments used in this study are the WSR-88D network wherein they ran the model for the three scenarios stated. Datasets were also used like the USGS Dataset in mapping the flash flood event on its corresponding basin. NWS storm database was used to compute for the mean annual flash flood occurrence rate for each CONUS grid cell. CONUS radar network was also used in running the model with no flash flood warnings and with flash flood warning for the purpose of estimating the benefit provided by flash flood warnings independent of radar coverage.
Conclusions and Recommendations
A geospatial model was constructed for computing meteorological radar network benefits for flash flood casualty reduction. It was established that casualty rate decreases by 44% when there is flash flood warning. Combining these effects, the model showed that the WSR-88D radar network provided over $300 million of casualty reduction per year, with $13 million per year coverage for improvements which is an indicative of its efficiency, and a maximum benefit pool of $69 million per year. A benefit model for flash flood property damage was a good idea but a preliminary analysis regarding property value did not give statistically meaningful relationship between warning performance and property damage so further study like analysis for geospatial data of property type and value was recommended.
