2050 UTC
The RAP Convective Weather Group conducted operational tests of automated thunderstorm forecasting systems at three different locations during the summer of 1997. The NCAR Automated Thunderstorm Nowcasting System (Auto-nowcaster) was installed at the White Sands Missile Range in New Mexico and the National Weather Service Weather Forecast Office in Sterling, Virginia. Output from a modified version of NCAR's radar echo extrapolation software package called TITAN was displayed at the Aviation Weather Center (AWC) in Kansas City MO. The AWC product was a 60 minute forecast on a national scale that defined regions of thunderstorms based on the combined use of lightning and radar data. The Auto-nowcaster produced 30 minute forecasts of the location of convective radar reflectivity echoes in excess of 35 dBZ (which typically corresponds to the presence of thunderstorms). This was for a region roughly within 125 km of the radar at each location. These forecasts included the initiation, growth, extrapolation and dissipation of radar echoes. The primary data input for all three field efforts was the WSR-88D radar data. The unique feature of the Auto-nowcaster is the ability to forecast thunderstorm initiation. This is possible by detecting and extrapolating boundary layer convergence lines and detecting and monitoring cumulus clouds prior to their development into thunderstorms. Results from the field tests emphasize the need to further improve the detection of convergence lines and small cumulus. In addition, it was apparent that forecasts beyond 30 minutes will require refinements to the echo extrapolation techniques to account for different life time expectancies due to the spatial scales of the phenomena. A brief description of the field operations are given in the following sections.
S. Henry, C. Mueller, D. Megenhardt, A. Crook, T. Betancourt and M. Dixon fielded the Autonowcast system at the White Sands Missile Range to assist forecasters with daily weather forecasts, as well as special forecasts for meteorological conditions expected during to missile. The White Sands Missile Range Auto-nowcaster used radar, sounding and surface mesonet data, along with output from TITAN (the thunderstorm detection and extrapolation program), COLIDE (the convergence line detection and extrapolation algorithm) and TREC (a 3D boundary layer wind field derived from single Doppler data) algorithms, to generate place-specific, 30 minute forecasts of thunderstorms. The system also included radar and integrated data displays that offered easily accessable real-time and archived data for the forecasters to view.
As expected, forecasting for the White Sands Missile Range proved to be quite challenging due to the complex terrain. The San Andres Mountains occupied the western half of the forecast region (orange polygon in Figure 1) and the Sacramento Mountains were located just east of the forecast region. The mountainous areas are apparent by the sudden lack of radar coverage, leaving a narrow region (typically < 50 km wide) of clear air available for tracking features in the valley. This was the first time the Auto-nowcaster attempted to forecast directly over the mountains. Figure 2 shows an example of a WSMR thunderstorm forecast (red polygons), forecast of wind shifts from COLIDE (white line) and TREC winds (yellow vectors) from 14 August 1997 at 2240 UTC and the verification 30 minutes later. Quite often storms developed quickly over the mountains and were very short-lived. Consequently, TITAN had difficulty extrapolating existing storms for 30 minutes. Forecasting thunderstorm initiation over the mountains was also challenging because of the difficulty of detecting convergence lines over the mountains. Future enhancements to the Auto-nowcaster include adding a terrain interest map that will be used to anticipate the likelihood of thunderstorm activity over the mountains.
The National Weather Service modernization goals for the next 10 years are to improve the accuracy of short-term forecasts and warnings and to extend the range of predictability for all weather elements. Given these motivations, a collaborative effort was formulated between the National Weather Service Techniques Development Laboratory, Operations Support Facility, the National Severe Storms Laboratory, and NCAR scientists R. Roberts, J. Wilson and T. Betancourt under the heading of System for Convection Analysis and Nowcasting (SCAN). SCAN goals are 1) to detect, analyze and monitor convection and generate short-term probabilistic forecasts and warning guidance within AWIPS (NWS's Advanced Weather Interactive Processing System), and 2) to combine previous research and development efforts (NSSL's Warning Decision Support System, NCAR's Auto-nowcaster, NWS Thunderstorm Product and hydrological algorithms) into one integrated approach to forecasting convection. The first phase of SCAN was to run the above algorithms at the NWS WFO in Sterling, Virginia (this office forecasts for N. Virginia, Washington D.C. and western Maryland) from August through September 1997. The NWS Thunderstorm Product and the thunderstorm forecasts and related interest fields produced by NCAR's Auto-nowcaster algorithm were displayed on the NSSL display as a first step towards integrating these algorithms into AWIPS.
Output from TITAN, COLIDE, and TREC algorithms and sounding information were used by the Auto-nowcaster to produce 0-30 min, site-specific thunderstorm forecasts. A spectrum of weather events occurred on 12 days during the field test. Most of the intense convective activity was associated with large scale synoptic situations; i.e., lines of storms moving from a northwesterly direction associated with frontal activity or a squall line ahead of a front. Examples of real-time forecasts (white polygons) produced by the auto-nowcaster algorithm at Sterling can be seen in the
2050 UTC
and
2104 UTC
radar reflectivity images from 17 August as an E-W boundary sagged south through the WFO region ahead of an approaching squall line from the northwest. In these images, the COLIDE detections (magenta contours) and 30 min extrapolated boundary positions (cyan contours) are also overlaid. The E-W boundary can be seen south of the radar. Forecasts shown in the above plots are for both initiation of new storms and for persistence of existing storms over the next 30 min period. The Auto-nowcaster performance was surprisingly good, in light of the complexity of the event.
Some of the challenges in running the Auto-nowcaster at Sterling were related to the local terrain (the Appalachian Mountains and the Chesapeake Bay) and the relatively poor clear air radar return (< 40-50 km in range) in that area of the country. While this caused some initial concern about how COLIDE would perform, results of the field test showed that strong outflow boundaries were well detected. However, the reflectivity features associated with the bay breeze were often very subtle. While COLIDE had difficultly detecting these boundaries, the convergence of TREC winds due to the bay breeze was very apparent, as can be seen in this time series of reflectivity plots illustrating the westward progression of the bay breeze with time and its subsequent collision with an eastward-moving gust front. TREC winds and human-inserted boundary locations are overlaid. The TREC and surface mesonet winds will be combined with the convergence and buoyancy fields from the Adjoint numerical model (NCAR/MMM) to provide an improved boundary detection product for the Auto-nowcaster next summer at Sterling. During this second SCAN field test, the Auto-nowcaster will also use information from satellite and an area echo tracker to produce 1-2 hour forecasts of thunderstorms.
RAP and the National Weather Service - Aviation Weather Center (AWC) are collaborating under FAA funding to develop an automated, graphical Convective Sigmet Forecast product. The AWC issues warnings, forecasts and analyses of hazardous weather to the aviation community. Convective SIGMETs are issued for any convective situation which the forecaster feels is hazardous to all categories of aircraft. Currently Convective SIGMETS are updated hourly and are only available as text describing broad polygonal regions via aviation VORTAC locations.
As a first step in the development of the automated Convective SIGMET, C. Mueller, N. Rehak and M. Dixon ran a prototype algorithm in real time this summer that used radar and lightning data to generate a forecast product that was sent to the AWC. The product was a 60 minute extrapolation forecast of thunderstorm activity on a national scale (shown as the grey shaded region along with the verification interest field). This automated product aided the on-line forecasters in producing Convective Sigmets by providing information about the movement and intensity of convective systems. During the next few years, we hope to combine automation and forecaster interpretation into an accurate and timely graphical forecast of significant convection.
