3. Nowcaster evaluation from Sydney 2000 demonstration

The Nowcaster was deployed in Sydney Australia during the three-month Sydney 2000 Forecast Demonstration Project (FDP) that was sanctioned by the World Weather Research Program. This demonstration brought together state-of-the-art nowcasting systems from Australia, Great Britain, Canada and the United States.  J. Wilson, R. Roberts, T. Saxen, and C. Mueller were involved with an evaluation of the Nowcaster forecasts as part of this program. Briefly, the Nowcaster forecasts increase or decrease extrapolated radar reflectivity values based on the combined value of a number of forecast variables. These variables are based on scientific research and their relationship to storm evolution. In addition new areas of reflectivity may be generated (initiation). The reflectivity magnitude of initiated storms and the growth or dissipation rate of existing storms is a function of the magnitude of the summed influence of all the nowcast variables.  Figure C2 is an example of a 60 min reflectivity forecast. In this case the large brown and yellow areas in panel b represent initiation forecasts based on favorable boundary relative steering flow, boundary collision, vertical velocity and boundary-storm collision.

Figure C3 shows time series plots of the CSI for both the extrapolation only nowcasts (light gray line) and Nowcaster forecasts (dark line). The 30 min nowcasts are shown in the left hand column and the 60 min in the right hand column. The three cases represent convectively active days where there was a large difference between extrapolation and Nowcaster forecasts. First, it is apparent that there is considerable fall off in the statistical scores from 30 to 60 min. Second, the scores with the human inserted boundaries are superior to those from extrapolation only. The improvement is particularly noticeable during the periods in which the echo area was increasing. The improvement is a result of forecasting storm growth and initiation.

It is evident that for 60 min, the skill in nowcasting precise details of storm location, i.e. within a few kilometers and few minutes in accuracy, is very low. Supercells and large squall lines are possible exceptions during periods when they have reached a steady state. However, irrespective of the low skill scores at 1 km precision, there often appears to be significant skill in nowcasting the general area and intensity of convection that would provide considerable information to forecasters and aviation users.

 

 

 

[top of page]

 

4. Satellite Cloud Classification and Growth

R. Roberts and T. Saxen evaluated the contribution of satellite-based, feature-detection algorithms toward improving the overall performance of the 30 and 60 min Nowcaster forecasts.  The CloudClass and RateOfChange (time rate of change of IR temperature) algorithms ran in real-time at the White Sands Forecast Offices.

The impact of including the various satellite interest fields is shown in Figure C4 for a case from White Sands, NM.  A gust front produced by storms over the eastern range traverses the valley converging onto the slopes of the western range resulting in upslope flow and formation of new thunderstorms along the western range. The primary factors contributing to the thunderstorm forecasts include: characteristics of the convective boundary layer environment surrounding the gust front, terrain effects, detection of both cumulus clouds and clear sky, and cloud top cooling rates.  Two sets of forecasts are discussed here: 1) those produced using both satellite and radar information (red polygons) in real-time at White Sands and 2) those produced in post-analysis excluding satellite information in the Nowcaster system.  During real-time operations, radar information plus a favorable combination of CloudClass (cumulus clouds present), RateOfChange (values cooler than –8 deg/15 min having the greatest interest), and cloud top temperatures between 0 and –20 deg C must be met for a 30 min thunderstorm forecast to be generated. 

At the start of this event, only one storm echo is present in the valley between the mountain ranges and both methods produce identical forecasts for this storm to persist, indicating that this is primarily a radar-based forecast. The GOES Visible image at 18:15 shows cumulus clouds were definitely present in the vicinity of the outflow boundary and above the higher terrain; the CloudClass field (not shown) accurately classifies this region with cumulus clouds and clear sky.  However, the RateOfChange field shows no marked cooling of clouds except over the Sacramento mountains where mature storms exist and the cloud tops in the valley region have not reached subfreezing temperatures.  By 18:45, a marked change has occurred in the satellite imagery; the line of cumulus clouds in the valley have undergone significant development.  All along this line, temperature changes exceed –8 deg C/15 min, the minimum threshold used for vigorous thunderstorm development. As a result of the RateOfChange (ROC) information, the 30 min forecast polygons (red) at 18:45 and 18:51 extend well to the north and to the south of the radar-based forecast into areas of no radar cumulus echo aloft. These thunderstorm initiation forecasts verified well 30 min later (see panels at 19:16 and 19:20).

In complex terrain, it is often difficult for the Nowcaster to predict convection initiation as boundaries are not detectable in these regions. As the outflow boundary approaches the San Andres Mountains in Figure C4, new thunderstorms develop above the mountains by 19:28. The Nowcaster does surprisingly well in forecasting the new initiation over the mountains, starting at 19:10 with discrete cellular forecasts and evolving into forecasts at 19:28 for a line of thunderstorms that indeed occur by 19:58. Here the ROC field, along with CloudClass, cloud top temperatures and terrain information were the key information used in producing these forecasts. Clearly the inclusion of satellite information has provided added value and accuracy to thunderstorm initiation forecasts.

 

[top of page]

 

5. Variational Doppler Radar Analysis System (VDRAS)

In the past year A. Crook and J. Sun have run the VDRAS system in the Denver region as well as in Sydney Australia as part of the Sydney 2000 Forecast Demonstration Project.  This year the VDRAS code was ported from the DEC Alpha architecture used in previous years, to the Pentium architecture.  For the realtime application this year, a Pentium II, 1GHz, dual processor running Linux was used.  The two processors allowed two versions of VDRAS to run, a standard version and an experimental version for testing various changes.

VDRAS was run throughout the summer and assimilated data from the KFTG (Front Range) WSR-88D and the Denver TDWR radars. Due to the proximity of the two radars, the dual Doppler coverage was very limited.  However, using data from two radars is still beneficial as it increases the coverage of essentially single Doppler information. A number of adjustments were made to the code including the use of the previous analysis as a first guess and changing the amount of filtering applied.  Due to interruptions in the delivery of wideband WSR-88D data this summer, it was difficult to perform a reliable assessment of the system.  However, there was a number of excellent convergence line cases that were examined in post analysis.

One of these case, July 11, 2001, is shown in Figure C5. Figure C5(a) depicts the VDRAS analysis with both KFTG and TDWR radars while Figure C5(b) shows the analysis using data from only KFTG. The analysis with two radars clearly shows stronger convergence, especially for the stationary convergence line in the northeast portion of the domain. The convergence corresponding to a developing convective cell northwest of the main gust front.  This convergence is barely shown by the analysis using KFTG radar data alone.

Efforts have also been made to provide short-term forecasts of the low-level wind using the VDRAS analyses as initial conditions. Preliminary results from a few gust front cases show that the numerical prediction is quite robust.  However, the forecast speed of the gust front is generally slower than observed which appears to be due to the fact that the full negative temperature anomaly is not being retrieved.  Efforts are being made to add a forcing term associated with evaporative cooling to the numerical model which should aid in the thermodynamic retrieval.

 

Figure C5.  VDRAS low-level wind field (white vectors) superpositioned on the convergence field, (a) depicts the VDRAS analysis with both KFTG and TDWR radars while (b) shows the analysis using data from only KFTG.

[top of page]

 

6. Research and algorithm development for use in NCWF

This years work on the National-scale included; (a) on going support for the AWC - NCWF algorithm reported in section N.4, (b) a research component that provided information about the utility and methodology for using the RUC data in NCWF and Nowcaster 0-3hr growth and decay algorithms, (c) continued storm climatology of the CONUS, and (d) algorithm development work towards NCWF growth and decay .

Trier examined Rapid Update Cycle (RUC-II) model output to assess how this information could be used in improving short-range (0-3 h) forecasts of deep convection over the continental United States. The particular focus was on use of RUC soundings of temperature, moisture, and winds and supporting fields (e.g., vertical motion, temperature and moisture advections) to anticipate the initiation of deep convection, which remains one of the most difficult aspects of the convective forecasting problem. Specific questions examined using RUC analysis output included:  (i) Is deep convection locally possible within 3 h?; (ii) If deep convection is possible, how will it be organized (e.g., as short-lived isolated cells, as short-lived lines, as long-lived lines, etc.)? (iii) For situation in which deep convection is likely, when will it be initiated? The RUC analyses, alone, were found to provide valuable guidance in answering the first two questions. The analysis indicate potential utility of RUC analysis soundings in anticipating convective development and the likely characteristics of its organization on 0-3 h timescales. Of particular importance is the information on the trends in thermodynamic stability that may be gleaned from these soundings. Diagnostics of thermodynamic stability such as CAPE and CIN need to be calculated for multiple layers, instead of for only surface-based air parcels, which is the common approach in diagnostics provided with standard operational forecasts.

D. Meganhardt, N. Rehak, G. Cunning and C. Mueller made significant progress in the development and implementation of a large-scale, day-time multi-cellular storm growth algorithm to build on the NCWF storm extrapolation.   The algorithm will run during the diurnal growth cycle (15 to 22 UTC). Example output from the algorithm are shown in Figure C6.  The RUC data are used to determine locations of large-scale forcing. These data are processed using two different algorithms.  The first utilizes fuzzy logic to determine the broad locations of large scale forcing based on the RUC surface equivalent potential temperature gradient, convergence, and vorticity fields.  The second algorithm tracks the large-scale boundary or front and provides motion vectors and the orientation of the line.  In the region of synoptic forcing, small storms are grown and clumped along the orientation of the boundary based on trending of radar and satellite data.  The satellite data will be used to help identify where lines of developing cumulus are located to provide input into the extent of growth.  The clumped regions are extrapolated based on the motion of the boundary. In areas outside of the region of large-scale forcing the extrapolation forecast from NCWF is used.  Plans are to run this algorithm off-line during the summer of 2002 in order to test it on a variety of cases and provide statistical evaluation.

 

[top of page]