Hurricane Georges Rainfall 1998

Hurricane Georges: A Comparison of Gage Rainfall and WSR-88D Storm Total Precipitation

Mark W. Rose
National Weather Service
Birmingham, Alabama

INTRODUCTION

Hurricane Georges proved to be a costly and destructive tropical cyclone for much of the Carribean and Florida Keys. For the central Gulf Coast States, however, the hurricane was known more for its excessive rainfall. The brunt of the heavy rainfall (up to 2 feet in some areas) occurred over the panhandle of Florida and southern Alabama (Fig. 1). The purpose of this study is to determine how well the WSR-88D's Precipitation Processing System performed during this heavy rainfall event.

One of the benefits of the WSR-88D is its ability to estimate rainfall amounts based on the relationship between reflectivity and rainfall rates, commonly referred to as a Z-R relationship. Many radar sites have reported the WSR-88D's Precipitation Processing System was underestimating rainfall amounts in tropical environments. A study in 1995 by the WSR-88D Operational Support Facility found that during tropical/warm process rain events, the Rosenfeld Z-R relationship (Z=250R^1.2) worked better than the default Z-R relationship (Z=300R^1.4). Radar sites located in or near tropical climates have the authority to change from the default to the Rosefeld Z-R relationship when experiencing warm/tropical rain events. The radars sites used in this study; Birmingham (KBMX), Maxwell AFB (KMXX), and Fort Rucker (KEOX), were utilizing the Rosenfeld Z-R relationship during the heavy rainfall event caused by Hurricane Georges.

DATA COLLECTION/ANALYSES

The first task was to collect rainfall reports across the region so that a good sampling of data was achieved from the area significantly impacted by Hurricane Georges. Cooperative observers (co-ops) provide a great spatial network of daily rainfall reports. In addition to co-ops, rainfall data was also collected from several ASOS (Automated Surface Observing System) and military sites.

The next step was to match gage sites with rainfall totals from the WSR-88D's Storm Total Precipitation (STP) products;

• Figure 2. STP KBMX or Shelby County Airport
• Figure 3. STP KMXX or Maxwell AFB Radar just east of Montgomery
• Figure 4. STP KEOX or DOD Fort Rucker radar site.

The WSR-88D's precipitation processing system inputs accumulated rainfall amounts into 1^o X .54 nmi bins. Adjacent bins along a radar beam radial are averaged to get the final STP product resolution, which is 1^o x 1.1 nmi. The rainfall bins are assigned color codes, according to pre-determined rainfall ranges. Since accumulated rainfall bins from the STP product are assigned ranges and not singular values, a system of estimating rainfall totals had to be established. If a gage site was clearly within a colored range bin, then the estimated radar total was assigned the mid-point of the range (e.g., range 2.5-2.9 inches; estimated total equals 2.7 inches). If the gage site was near the boundary of two colors, the estimated radar total was assigned the minimum threshold of the higher range (e.g., gage site between 2.0-2.4 and 2.5-2.9 inches; estimated total equals 2.5 inches).

The WSR-88D's graphical display system has a mouse-controlled cursor which can be set to display latitude and longitude (L/L) coordinates. The National Climatic Data Center (NCDC) has a web site, http://www.ncdc.noaa.gov/ which contains geographical and historical information on various types of weather observing stations. L/L coordinates for each gage site used in this study were acquired from the NCDC. The L/L coordinates for each gage site can be manually entered into the radar display system, allowing the operator to project the cursor to the exact location of the gage site.

A simple mathematical comparison (R_{radar -} R_gage) was done between estimated WSR-88D rainfall and gage rainfall. The difference (inches) for each gage site was plotted on a map of Alabama and northwest Florida (Figs. 5,6,7 see below). Negative values indicate WSR-88D totals were less than gage totals. The value (ND) means that the gage site was either outside the 125 nmi. range of the radar, or that the data was obscured due to forced clutter suppression close to the radar site. The forced clutter suppression is indicated by wedge shaped areas extending southwest and northeast from the radar site. These anomalies are along the zero isodop The isodop is a line a equal wind speed., and can be clearly seen on the KBMX and KMXX STP products (Figs. 2,3).

• Figure 5: The KBMX radar estimated totals were lower than gage totals at 79% of the stations. As expected, the deficiencies were cumulative, with the highest deficiencies at those stations with the higher rainfall amounts. This shows the highest radar/gage differences were located four counties south of the radar, near the northern edge of maximum rainfall totals. Radar/gage differences were not range dependent, as some of the smallest differences were located over southeast and southwest Alabama, near the edge of the radar range.

• Figure 6: The KMXX radar was just the opposite of KBMX, with the radar totals higher than gage totals at 70% of the stations. Unfortunately, the maximum rainfall total for the STP product was set at 15 inches. Therefore, it was not possible to compare rainfall data over much of southwest Alabama and the panhandle of Florida. However, there were several gage sites across southwest Alabama that reported below 15 inches, and so it can be assumed that the radar was also overestimating in this area. The majority of the gage sites in which the radar totals were less than gage totals were to the south and southeast of the radar site.

• Figure 7: The KEOX radar showed a slightly negative bias, with radar estimated totals lower than gage totals at 61% of the stations. Negative and positive numbers appear to be evenly dispersed throughout the radar umbrella, except for two areas. An area over southwest Alabama, just west of the maximum rainfall, shows mostly negative values. This would likely be caused by the radar beam energy being absorbed by the heavy rainfall. Another area is located north-northwest of the radar near the 125 nmi. range limit. Radar estimates of six inches or more were not supported by gage reports or other radar estimates.

Another way to compare radar/gage rainfall totals is to compare radar estimates against each other, by looking at individual gage sites (Table 1). The table lists the three radar totals after each gage total. In a head-to-head competition, no radar proved superior with respect to estimating rainfall. At those gage sites where radar totals were available from all three radars, each radar came out with about the same number of total wins.

CONCLUSIONS

There are limitations to applying mathematical equations to predict what is occurring in the atmosphere. The Z-R equation is just an estimate, based on reflectivity values averaged over time and space, assuming a given drop-size distribution. If the drop-size distribution deviates greatly from the standard, this will lead to inaccurate rainfall estimates.

Even though each radar was utilizing the Rosenfeld Z-R relationship, each radar's performance was different from the others. If the KBMX radar is underestimating rainfall and the KMXX radar is overestimating rainfall, it is not likely that the Z-R relationship is the main source of error. Since rainfall radar estimates are directly derived from radar reflectivity, it is possible reflectivity values are the main culprit. Prior to Hurricane Georges, forecasters at the National Weather Service office in Birmingham have observed that the KBMX radar was operating "cold" with respect to radar reflectivities, compared to other radars in Alabama. This would explain why the radar underestimated rainfall totals. It is possible that KMXX radar had the opposite problem, with the radar reflectivity values too high. Even though KEOX radar did not show a bias, its accuracy rate was not any better than the other two radars. There were many locations within KEOX radar umbrella where radar estimates were significantly different from gage reports. Interestingly, in those instances, the difference was primarily negative.

The data presented in this study provides a clue as to how individual radars handled a significant rainfall event. The WSR-88D can archive data, and now that a nationwide deployment of the radars has been completed, the resources are available to greatly improve the radar's rainfall prediction methods.

Back to home page

Weather Hazards | Graphical Forecast | Present Weather
Past Weather | Tropical Weather | Contact Us