Date of Award
Master of Science
Geological and Atmospheric Sciences
William A. Gallus
The goal of this study was to determine if radar data assimilation might be able to aid in the prediction of flash flood events before the corresponding heavy rain events occur. Twelve heavy rainfall events that occurred over the state of Iowa during the warm season between 2001 and 2011 were simulated using the Weather Research and Forecasting (WRF) model. Two 24-hour high resolution (4 km) WRF simulations were run for each heavy rainfall event, one where radar data was not assimilated (cold start) and one where radar data was assimilated at the time of model initialization (hot start). In order to get a measure of the forecast accuracy, equitable threat scores (ETS) and bias were calculated for the hot start and the cold start rainfall forecasts for each case. The ETS and bias were calculated for several different rainfall thresholds in order to see if forecasting skill might differ between the different rainfall thresholds. Other statistics such as rainfall volume, rate, and areal coverage were analyzed along with plots of the rainfall amounts and post processed reflectivity. The degree of accuracy present in the cold start and hot start model quantitative precipitation forecasts (QPFs) was discussed along with some problems that were consistently present in the model simulation for all of the heavy rain events. Then Plots of a Flood Severity Index (showing areas where flash flooding would be predicted) were created for the heavy rainfall cases using a hydrology model forced with Stage IV rainfall (high resolution rainfall analysis), hot start run QPF, and cold start run QPF for comparison and analysis. Additionally, hot start and cold start WRF simulations (these were only 12-hours simulations) were run for two more heavy rain cases that occurred in the warm season of 2010. A probability matched mean rainfall forecast was also produced from a mixed physics model and compared with QPFs from hot start and cold start model runs for these two additional heavy rain cases in order to see if one method provided greater improvement in forecast skill.
Overall, the radar data assimilation resulted in increased accuracy in model QPF for the initial 12 heavy rainfall cases simulated. As a result the rainfall forecasts from the hot start runs yielded higher ETSs than for the cold start runs, although the heaviest rainfall areas generally saw less of an increase in ETS from cold start to hot start and had fairly low ETSs for most cases. A couple significant problems were noted in the hot start runs, one being that the runs were far too wet in the first hour or two for each case and another being that thunderstorms/rain areas created from radar-assimilation and not present in the cold start simulation tend to dissipate too soon. Analyzing the hydrology model-predicted flash flooding for all 12 cases revealed that as one might expect the rainfall from the hot start runs yielded fairly accurate flash flood predictions (in terms of general placement and intensity) for a few of the better forecasted (in terms of rainfall) cases, while placement and intensity discrepancies existed for many of the cases when compared with the flash flooding produced via the Stage IV rainfall. One encouraging finding regarding the hydrology model runs forced with the hot start rainfall was that they predicted flash flooding at some point during the 24-hour simulations in all 12 cases, and this was also true when the Stage IV rainfall was used. Implying that one might be able to say with fairly good certainty that a rain event will produce flash flooding (even if location and intensity isn't known) based on the hot start runs, whereas the same can't be said for the cold start.
Comparison of hot start, cold start and PM rainfall forecasts for the additional two heavy rain cases revealed that the hot start and PM rainfall forecasts both had a greater level of skill and were more accurate than the cold start rainfall forecasts for the two events. The hot start rainfall forecasts did a better job of placement and areal coverage (and had higher ETSs) for the entire precipitating region (lower rainfall thresholds), while the PM forecasts generally was more skillful (higher ETSs) in placement of the heaviest rainfall areas (highest rainfall thresholds). This is an important finding since it implies that the PM approach may be better than radar data assimilation alone for forecasting the heavy rainfall areas that are usually the most important when it comes to the potential for flash flooding.
Moser, Ben, "The impact of radar data assimilation on warm season rainfall forecasts for use in hydrologic models: examples from extreme rain events in Iowa" (2014). Graduate Theses and Dissertations. 13673.