Plan for CAPS real time forecast support for IHOP

For further information, contact Ming Xue (mxue@ou.edu) or Keith Brewster (kbrewster@ou.edu).

CAPS has received a 3-year NSF grant to participate in IHOP, with a focus on the optimal utilization and impact of water vapor and other high resolution observations in storm-scale QPF. Particular emphasis will be given to the sensitivity of storm-scale QPF to initial conditions near the time of convective initiation, and to boundary conditions as well. Initial conditions for the storm-resolving ARPS will be obtained using 3DVAR schemes, and sensitivity will be studied using adjoint as well as traditional forward sensitivity methods. The estimation of background error covariances, via ensemble techniques applied at the meso- and storm scales, is another important component of the research. The study will take advantage of data collected during IHOP for both model initialization and verification, and also will examine the impact of routine and special data sets on forecast quality.  The project summary can be found at http://twister.ou.edu/IHOPAbstract.html.

To support the real time operations of IHOP, and to establish data sets for extensive post-field program analysis, CAPS will run its Advanced Regional Prediction System (ARPS) daily, at a variety of resolutions, during the entire field campaign. The real-time data also will be made available to the UCAR Joint Office of Science Support (JOSS) for field experiment archives.

A Pentium-IV cluster supercomputer with 256 of the latest-generation processors, and a powerful SMP with 32 processors, will be available at the University of Oklahoma for use by CAPS in creating real time analyses and forecasts.  CAPS plans to create forecasts on three grids covering relatively large areas and contingent upon final benchmark tests, with resolutions of 18, 6 and 2-km (see Fig.3). The 303x203 point, 18-km grid will cover the continental US, with forecasts starting at 12 UTC and extending for 48 hours. The 273x243 point, 6-km grid covering ~1600x1440 km² will be nested inside the 18 km grid, initialized at the same time, and run for 24 hours, ending at 12 UTC (7 am CDT). A 512x512 point, 2-km grid covering 1024x1024 km², centered over the IHOP domain, will be nested within the 6-km grid and be initiated at 15 UTC or 10 am CDT and extend for 15 hours, ending at 06UTC or 1am CDT.  The goal of the medium and high-resolution forecasts is to capture convective initiation in support of operations planning, and for verification purposes.  If sufficient resources exist, CAPS intends to run additional high-resolution (2-km) 12-hour forecasts starting at 00UTC, i.e., when convection is most likely to be present in the IHOP domain.  The goal of these forecasts is to study model initialization using radar observations of deep convection and to evaluate the performance of various retrieval and diabatic initialization procedures and resulting forecasts that may include MCS components extending into the early morning hours. Such procedures will also be employed on the 15 UTC 2-km grid, although convection will probably be less frequent in the morning. All grids will have 63 vertical levels, reaching a height of 20 km.  In addition to forecasts, CAPS will generate hourly high-resolution (6-km) analyses using ARPS Data Analysis System (ADAS). The analyses will incorporate WSR-88D Level-II and Level-III radar data for use in estimating water and temperature fields.  Wind fields will be adjusted by the analysis using radial velocity data and possibly single-Doppler velocity retrieval techniques. In all cases, the analyzed data will be incorporated into the model via incremental analysis updating (IAU).  Details of this plan are contingent upon achieving expected performance targets on the new supercomputers.

Fig 3. The 18, 6 and 2 km resolution ARPS model grids to be run by CAPS during IHOP.  The 2 km grid covers areas that climatologically favor convective initiation and MCS events, and also encompasses the ARM-CART region (shown in green).

Forecast products will be posted on the web in a manner similar to the current real-time ARPS forecasts (c.f. http://www.caps.ou.edu/wx). Special graphics products, focused on moisture, can be added to the graphics suite. It is anticipated that the IHOP Operations Center will have web access for viewing these data. Certain customization will be done based on input from the IHOP Operations Center and SPC.  If desired, CAPS also can generate NetCDF or GEMPAK grids that can be viewed on AWIPS via the D2D package or a GEMPAK viewer.  Figure 4 shows the projected time line for the four forecasts.  The forecast cycle spawned at 12 UTC, as described below, will occur once daily instead of using continuous cycling as in the FSL modeling approach.

Fig. 4. Timeline of proposed CAPS forecast at three resolutions.

18-km grid: 1 hour data collection, 10 minute analysis, 90 minute forecast.  Thus, the entire 48h forecast will be completed at 1440 UTC. Graphics will be generated as the model output is written to disk, therefore the web posting will be completed a few minutes after the forecast ends. This grid will use a 6-hour ETA forecast as analysis background, all available rawinsonde, profiler, standard surface observations and Oklahoma Mesonet data, and both visible and IR satellite data.

6-km grid: Analysis will be performed at 1315 UTC and completed by 1335 UTC; thus, this grid is spawned as the 18-km grid forecast is being completed, with the boundary conditions for this run coming from the 18-km forecast.  Forecast begins at 1345 UTC and is completed by 1545 UTC.  Preprocessing of Level-II radar data begins as soon as the data arrive, which typically happens within a minute after each radar volume, and a few minutes after data collection time for Level-III (NIDS).

2-km grid: This model run will start from a 15 UTC analysis, using the 3-hour 6-km forecast as the background.  The 6-km forecast will provide the boundary conditions.  The 2-km run will again incorporate Level-II and Level-III (NIDS) radar data.  Attempts will be made to perform single-Doppler velocity retrievals on Level-II data from Amarillo, Dodge City, Wichita, Tulsa, and Oklahoma City, Cheyenne, Denver, Goodland, and possibly Topeka radars and analyze the retrieved winds into the 2-km initial condition. Because of limited manpower available for the real-time effort in the current project, the latter effort will need to leverage projects with similar goals. Depending on the progress of these other projects, real time velocity and thermodynamic retrieval may or may not happen. We expect that the 2-km forecast will be completed by 18 UTC.

2-km grid (convection present): The proposed forecast, contingent upon available computer resources, will start from a 00 UTC analysis, using the 9-hour 6-km forecast as the background.  The 6-km forecast will provide the boundary conditions.  The 2-km run will include Level-II and Level-III (NIDS) radar data as in the 15 UTC run.  More serious attempts will be made to perform single-Doppler velocity and thermodynamic retrievals using Level-II radar data and analyze the retrieved quantities into the 2-km initial condition. We expect this run to be completed by 06 UTC.