Close Window

North Carolina State Univ.: "Nowcasting convection at Raleigh, North Carolina"

Final Report



The Partners Project concentrated on the development of experimental forecasts for summer rainfall in North Carolina. Two types of forecast products were developed: (1) a six-hour precipitation outlook and (2) a 60-to-90-minute nowcast. Both were developed in real time at the Raleigh NWSFO and used only guidance that was routinely available at the forecast office. The project objectives were to develop a methodology for forecasting summer rainfall, to determine if the experimental forecasts thus developed showed measurable skill, and to determine how skill was affected by forecast resolution.

The participants in the project included Dr. Allen J. Riordan, NCSU Principle Investigator and Michael Black, NCSU Research Assistant, and Kermit Keeter, NWS SOO and co-principal investigator, and Michael Moneypenny, NWS forecaster.

The responsibility of the NCSU participants involved developing, documenting and verifying the forecasts. The responsibility of the NWS partners, Kermit Keeter and Michael Moneypenny, were to ensure the smooth operation, integration, and introduction of the NCSU forecasters to the NWS staff and the facilities available at the NWS in Raleigh. The NWS partners also provided periodic advice and guidance from the operational perspective.


Outlooks were developed to forecast the percent area coverage for eight equal-area sectors within the Raleigh WSR-88D reflectivity range (230 km). These sectors were aligned with the natural climatological divisions within the region. The anticipated areal coverage of precipitation as derived by the WSR-88D for 1500-2100 UTC was forecast by 1500 UTC each weekday.

The experiment began on 20 May and ended on 12 September 1996. In this time period, 62 outlooks were developed, 50 of which experienced precipitation. The most useful support products were found to be 12Z soundings as evaluated by SHARP, hand analysis of routine surface hourly data, and Eta model output fields of surface features and upper-level vertical velocity.

Brier skill scores were computed using the average precipitation coverage in each sector as "climatology". Eliminating days when rain was not anticipated and did not occur, an overall score of 0.25 resulted. When the areal resolution was decreased to half (using four sectors instead of eight), the score increased to about 0.30. Further decrease in areal resolution to two sectors improved the score to roughly 0.40. Forecasts of percent coverage were originally given to the nearest 10%. When skill scores were re-evaluated based on a 20% resolution, the results were only slightly improved.

Finally, an interesting result was that skill scores were a strong function of the day of the week. It was no surprise that Mondays were worst, with an average score of about -0.18. Skill peaked on Wednesday and Thursday at near 0.38. This trend suggests the importance of continuity in developing outlooks.

Nowcasts were developed on an as-needed basis to pinpoint where new thunderstorm activity would occur. Development of new echoes of 40 dBZ or greater intensity within 16 x 16 km boxes within the County Warning Area were forecast for 60- to 90-minute periods. Products were scored using the Critical Success Index (CSI), Probability of Detection (POD), and False Alarm Rate (FAR), common to the NWS.

A total of 26 experimental nowcasts were produced. Most useful in the development of these short-term forecasts were the WSR-88D especially at low reflectivity levels of less than 20 dBZ, visible satellite imagery obtained via RAMSDIS, and hourly surface analysis of mesoscale features.

Results were encouraging. A POD of 0.45 was lower than obtained in Colorado by Wilson and Mueller, but the latter included persistence and extrapolation in their scheme and we did not. Our FAR of 0.58 was much lower and our CSI was comparable.

Our results may have been affected by the nature of the forcing during the experiment. The summer of 1996 was unusually cool. Synoptic forcing played much more important role than mesoscale forcing from weak pressure troughs, outflow boundaries, sea-breeze fronts, etc. all of which are usually much more common in summer in North Carolina.


Michael Black (NCSU) gave a presentation on 14 July 1997 at the National Weather Service in Raleigh. Information and results on the cooperative project were presented to the NWS staff. A question and answer session followed the presentation.

Seven conferences were held for the NCSU/NWS team throughout the year to discuss the project. These meetings typically included the NWS partners as well as members of Mike Black's research advisory committee.


Black, Michael A., Allen Riordan, Kermit Keeter, and Michael Moneypenny, 1996: Nowcasting Thunderstorms in North Carolina: A Summer of Convection. Poster Presentation: National Weather Association Meeting, Cocoa Beach, FL, 2-5 December 1996.

Black, Michael A., 1997: Experimental Forecasts of Summer Precipitation in North Carolina. M. S. Thesis. Marine, Earth and Atmospheric Sciences Department, Box 8208, North Carolina State University, Raleigh, NC 27695-8208.



The university investigators, Al Riordan and Michael Black, experienced the daily operations within a typical NWS setting. The experience was unique because it allowed research and operational forecasters to work hand-in-hand on a project. A team effort was forged not only with the designated partners in the project, but with the entire NWS staff. Information, both in research and operational fields, was exchanged as the study progressed. As a result of the NWS interactions, an improved forecast was able to be produced.

Problems encountered during the project were not related to the COMET program or with any problems with the NWS staff, all of who showed enthusiasm and support for our project. Rather, the challenges we faced were mainly due to equipment limitations or conflicts of mission. For example, one technical problem related to the access to clear-air mode of the WSR-88D. Distant echoes or anomalous propagation nearly always prevented all but very brief access to this high-sensitivity mode even during the pre-convective morning hours. As a result, we were usually unable to use a product that has been demonstrated to be very useful in identifying pre-convective signatures.

A second difficulty dealt with the archival of WSR-88D data. At times, indications that the archival process had been successfully completed were found to be in error, and, for example, only one scan was archived from a 3-hour period. We did not discover that most of the data was missing until we later tried to review the products. Fortunately, the problem was discovered early in the experiment, and through efforts by the NWS staff, was corrected.

The challenge of conflicting objectives was encountered in the nowcasting portion of the project. It was quickly found that soon after convective activity began, some storms often approached severe limits. As soon as these storms began to affect the County Warning Area for the Raleigh NWS, the use of the PUP was re-assigned to NWS staff. This necessary re-assignment was anticipated, since warning responsibilities must certainly take priority over nowcasting. However, it was also soon learned that PUP products and surveillance techniques for severe-weather detection were not compatible with nowcasting. For example, forecasters examining cells for severe storm potential normally filtered low reflectivity values necessary for nowcasting. Further, extended periods were often spent examining or obtaining cross sections of specific strong cells and little time was spent viewing areas of less intense activity.

Since developing nowcasts is a new objective of the NWS mission, a possible solution for the conflict between nowcasting and warning would be to allow separate access to products on a second independent workstation. This would alleviate the strain on both the storm warning forecaster and nowcaster, both of whom rely heavily on timely WSR-88D imagery.

We found that both outlook development and nowcasting benefited greatly from the RAMSDIS display of satellite imagery. The high spatial resolution of visible images was particularly helpful in identifying arc lines and early convection for nowcasts. It is noteworthy that this imagery was helpful even though it was often over 60 minutes old when obtained. We understand that measures have since been taken to allow earlier access to current products. This more timely display should greatly benefit future nowcasting efforts.


For the past few years, the modernized NWS has mandated its field offices to issue nowcasts and encouraged the issuance of short-term convective outlooks. During this time, little was documented about the skill of field offices to produce these products or about the procedures used to do so. To date, the findings from a few pertinent research studies have been typically carried out in settings with resources beyond the scope of the typical NWS field office.

This project has done much to provide some answers and direction to important NWS operational considerations regarding the issuance of nowcasts and convective outlooks. The project results were generally encouraging. The results demonstrated that both the nowcasts and convective outlooks were produced with skill using less than ideal resources.

The biggest resource need is for an integrated workstation (AWIPS) that incorporates multiple data sources into one platform and provides more timely satellite imagery; however, the addition of AWIPS will not likely resolve the conflict that was documented to exist between using the 88-D data for nowcasting versus issuing warnings.

Both nowcasting and severe weather surveillance require close and nearly constant monitoring of 88-D data. Warning assessment requires viewing a number of different radar products. Filtering the lower reflectivity returns is useful for quick identification of possible severe cells especially when there are multiple returns to consider. On the other hand, nowcasting new or intensifying convection requires a need to constantly view unfiltered base reflectivity returns. This conflict for simultaneously using the PUP display for different operational missions needs to be resolved before field forecast offices can maximize their nowcasting skill. Is there a cost effective means where field offices can constantly display unfiltered base reflectivity returns without conflicting with the severe warning evaluation process?

The experiences reported by the project's participants were enlightening and provided positive feedback for some of the current procedures used by the NWS Raleigh office to produce nowcasts and convective outlooks. Specifically, this forecast office will continue to recommend hourly surface mesoscale analysis as a means to detect meteorological features that act to focus convection. Using model data to modify the morning sounding data will also continue to be extensively employed.

Other experiences hint at possible procedural improvements. Specifically, this office needs to evaluate better the utility of NCEP mesoscale model output in the convective outlook process. Also, the poorer forecast skill associated with the first day back is not just peculiar to this project but is widely acknowledged in a subjective sense by NWS forecasters. This office should consider more stringent briefing producers for the forecasters first day back on the operational desk as means to compensate for the skill deficit.

Finally, the project's findings and the experiences of forecasters at Raleigh point to a need for further investigation of storm initiation and intensification as a result of boundary interaction. Previous COMET funded research at this forecast office has shown North Carolina to be a haven for a wide variety of low-level convergence boundaries. Contrary to the results generally reported in the literature, boundary interactions do not always lead to increased convection. The results from this project support the subjective impressions of the forecasters at Raleigh that boundary interaction may not lead to new convection or infrequently diminish the initiation of new convection or even weaken or suppress convection. Moreover, the effects of boundary interaction may vary according to a number of factors including boundary type, boundary depth, time of day, and the presence of a thermal cap. Understanding these considerations could do much to improve the nowcasting and short term forecasting of convection.