1. Background

Predicting the initiation and location of new convection 30 min to 2 h in advance is one of the challenges of convective weather forecasting, along with anticipating the merger of thunderstorm cells and determining when thunderstorms will begin to dissipate. RAP has focused recent efforts on developing thunderstorm nowcasting systems on the local-scale, the Convective Storm Nowcast System (ANC), and the national-scale, National Convective Weather Forecast (NCWF). The ANC combines feature detection and forecast algorithms with a numerical cloud model to ingest and process all available operational data sets and output features identified as relevant to thunderstorm evolution. The NCWF uses national mosaic radar information, echo top mosaics, cloud-to-ground lightning data and GOES satellite/RUC model-retrieved cloud-top heights combined together to produce a convective hazards detection field. A major scientific focus this year has been the evaluation of both algorithms in operational environments, and extention of the forecast period.

2. FAA Regional Convective Weather Forecast (RCWF) Demonstration

The ANC is a permanent installation at the Weather Forecast Offices at the White Sands Missile Range (WSMR) in New Mexico and the Redstone Technical Test Center (RTTC) in Huntsville, Alabama, as part of the Army Test and Evaluation Command (ATEC) 4DWX program. WSMR received full system upgrades in the spring of 2001. This was the sixth summer of operations at WSMR and the fourth at RTTC. In addition, a subset of the ANC system that runs uses level-3 WSR-88D data was installed at Dugway Proving Ground, Yuma, and Aberdeen ARMY Test Ranges.

Since 1999, the NCWF system runs operationally at the Aviation Weather Center (AWC) and is disseminated to the community as an "operational" product.

This summer, experimental versions of ANC ran over a multi-radar domain, and a regionalized version of NCWF (RCWF) ran in Boulder using data from the northeastern United States as part of an FAA Regional Convective Weather Forecast (RCWF) demonstration. The demonstration was a collaborative effort that included MIT Lincoln Laboratories and National Severe Storm Laboratory.

The ANC portion of the summer field project included, for the first time, a multi-radar domain covering a 755,000 km2 area. The engineering infrastructure to support this effort included 12 independent Linux nodes running 74 analysis processes, 27 data ingest processes, 36 data management processes, and 6 display processes. The ANC system was able to leverage off of the CRAFT (Collaborative Radar Acquisition Field Test) radar network for internet delivery of level II radar data. This system also ran an off-line nowcast using MIT's MIGFA algorithm for boundary detection.

The RCWF portion of the summer field project included a new thunderstorm growth algorithm and a new stability field calculation. The algorithms were run over a smaller domain than before, using higher resolution data. The engineering infrastructure to support this effort included 8 independent Linux nodes running 38 data management processes, 18 data ingest processes, 85 analysis processes, 13 data distribution processes, and 2 display processes.

Added to the base RCWF and autoNowcast systems described above, two additional workstations were engaged for relaying data between MIT and NCAR which allowed RAP to run displays at MIT.

The forecasts for both systems were supplied to the Forecast Systems Lab via an anonymous FTP site for realtime statistical verification.

3. Variational Doppler Ananlysis System (VDRAS)

Realtime retrieval of low level winds for RCWF

Our primary focus this year was setting up and running the Variational Doppler Radar Analysis System (VDRAS) in realtime for the RCWF project. This year, two VDRAS domains were set up to run on separate processors of a dual-processor Linux PC. The two domains were relocatable, working on either a northern pair of radars (KLOT/KIWX) or a southern pair (KIND/KILN). For the summer project, VDRAS ran exclusively on the northern pair.

The system this year was robust and ran reliably throughout the field program, requiring minimal human intervention. The system was capable of retrieving the low-level convergence signature associated with phenomena such as gust fronts and lake breezes off Lake Michigan. An important issue this year was ensuring a smooth transition of the wind field between the two assimilation domains. This is a subject of ongoing research and will be one of our main foci next year.

Adding an evaporative cooling term to VDRAS

A second focus this year was adding an evaporative cooling term to the current dry version of VDRAS. In the presence of convective precipitation, a dry boundary layer model is deficient because it cannot represent moist processes. In complex situations, intense convective cells generate gust fronts that propagate, interact and produce new convective elements. These features are driven primarily by the buoyancy field in the boundary layer. On the other hand, the most important moist process occurring in the boundary layer during these phenomena is the evaporation of rain in sub-saturated air. In a first step toward the use of a deep convection model with full microphysics, a parameterization of the effect of evaporative cooling is added to the dry boundary layer model to improve the temperature field.

The evaporative cooling is parameterized using the precipitation deduced from the observed reflectivity. A first guess of temperature and humidity is derived from the RUC sounding at the radar site. The 3D structure of the humidity field is retrieved during the data assimilation but is constant with time. At the end of the 10-minute assimilation window, a 1-hour forecast is issued. During the forecast, the evaporative cooling is applied assuming steady state conditions for humidity and precipitation.

The evaporative cooling scheme was applied to a moist convective case from this year's RCWF program. Figure 1(a,b,c) shows the analyzed velocity and convergence fields along with 30 and 60-min forecasts. The color contours show the low level reflectivity from the KIWX radar. The model convergence field does a very good job at forecasting the location of the gust front at 60 min.

Figure 1a

Figure 1b

Figure 1c

One hour forecast of the low level wind and convergence field using VDRAS. The forecast is initialized with a VDRAS analysis at 1836 UTC on 25 June 2002 using data from the KIWX radar, Fig. 1 (a) Shown is the analyzed wind and convergence field overlaid on the low level KIWX reflectivity field, (b) 30-min forecast, and (c) 60-min forecast.

4. Stability

Use of RUC-II Analyses and Forecasts in 0-2 h Nowcasts of Deep Convection

RAP scientists made significant progress on the use of information from RUC-II analyses and forecasts to assess the potential for development and growth or dissipation of deep convection over 0-2 h. An experimental algorithm was implemented in real time (August 2002) and its performance has been subjectively evaluated during the final third of the 2002 warm season. Central to the algorithm is the determination, from operational RUC output, of the depth of lower-tropopheric thermodynamic conditions that are favorable for deep convection. Previous research using archived test cases found that environments with deep layers of significant convective available potential energy (CAPE) and minimal convective inhibition (CIN) were more susceptible to the upscale growth and persistence of deep convection. RUC analyses and forecasts provide the best means currently available to monitor important evolutions in these thermodynamic stability quantities on an hourly basis. In the algorithm, CAPE and CIN are estimated through a combination of 1-h RUC forecasts and 1-h extrapolated analyses. Fuzzy logic is used to incorporate information on other RUC-derived environmental thermodynamic and kinematic parameters important to convective initiation and evolution, including vertical wind shear, relative humidity, and differential advections into the algorithm. In particular, these additional parameters guide the modification of threshold values of CAPE and CIN favorable to support convection in differing meteorological situations.

Preliminary evaluation of algorithm performance indicated that the "depth of instability" diagnostic is particularly effective in nowcasting growth or decay of mesoscale convective systems (MCS) that often occur overnight until slightly after sunrise over the central United States. In these situations extremes of CAPE and CIN tend to be more concentrated than during the day, when surface heating promotes broader areas of favorable CAPE and CIN. Moreover, the forcing for nocturnal deep convection tends to occur on the mesoscale and is often situated above the planetary boundary layer (PBL), in contrast to the class of PBL-based convection that is most frequent in the afternoon and is more often associated with fine-scale surface boundaries. An example of the use of the "depth of instability" algorithm in anticipating the 2-h evolution of mesoscale convection is shown in Figure 2. Here, the algorithm performed favorably in terms growth and decay of areas of convection associated with an MCS, when compared with a 3-h operational RUC forecast of quantitative precipitation that relies more heavily on the performance of a convective parameterization scheme (note that the stability diagnostic relies on a combination of analyses and shorter term forecasts). Future work will involve further quantification of algorithm performance through a more thorough examination of 2002 cases run in research mode. We anticipate that such an analysis will both elucidate algorithm performance compared with other available means of nowcasting (e.g., model-based QPF) and facilitate tuning of the algorithm. The eventual incorporation of this processed RUC thermodynamic stability information into the auto-nowcaster is being planned.

Figure 2: NEXRAD vertically integrated liquid (colorfill) valid at 0900 UTC 22 August 2002 with overlayed contour field of (top) forecasted precipitation rate valid at the same time (0900 UTC) from a RUC-II operational model run initialized 3 h earlier and (bottom) number of 25-mb-deep thermodynamically unstable layers (in contour intervals of 4) from algorithm output 2 h earlier.

5. Utility of Satellite toward 0-2hr forecasts

Several satellite-based algorithms were run in realtime during the RCWF demonstration over the ANC domain. Upstream processes included: 1) correction for parallax in the satellite data, 2) advection of cloud features, and 3) thresholding of VIS and IR data to remove unwanted features (very cold cloud tops, etc) prior to identification of cloud growth areas. Feature detection algorithms run in realtime included: 1) classification of cloud type, 2) cloud growth based on IR temperature rate of change (ROC), and 3) clouds undergoing change of phase (using the satellite-derived reflectance field).

Cumulus cloud growth was evident on satellite during several of the weather events during the RCWF ANC demonstration. Growth of clouds was not always tied to boundary layer triggering, but instead was frequently influenced by the dynamics associated with synoptic scale features. For example, on 14 June, growth of clouds occurred along several cloud lines oriented perpendicular to a low pressure center located north of the ANC domain in central Michigan. These linear bands of clouds spaced at regular intervals evolved into lines of thunderstorms and tracked counter-clockwise around the low pressure center. Although growth of the cloud lines were observed in the ROC field, this information alone, when surface triggering is not detectable, is not sufficient to trigger storm initiation nowcasts in the ANC system.

Since many of the events this summer were influenced by synoptic scale forces, all events where thunderstorm initiation was observable on satellite will be reviewed in the next few months. This will include an analysis of the real-time performance of the above algorithms in detecting important precursor features. We will also examine ways to improve the use of the satellite information in the ANC initiation nowcasts, particularly when growth appears to be decoupled from surface forcing.

 

[TOP]