QSCAT winds proposal

QSCAT winds: Can they improve mesoscale model predictions?

John S. Snook Ralph F. Milliff Peter A. Stamus

Colorado Research Associates (a division of NWRA) Boulder, CO 80301

1. INTRODUCTION

Operational forecasters rely heavily on numerical weather prediction (NWP), which has provided substantially increased forecast skill over the last two decades (Kalnay et al. 1998). NWP skill has progressed primarily in response to three factors: 1) increased computer power, 2) improvements to observing systems, and 3) scientific advances in numerical models and usage of available observations. Future advances in NWP forecast skill will likely result from continued improvements in these three areas. In the area of observations, the enhancement of observing systems, especially wind (Kalnay et al. 1998), in data sparse areas, such as over the world's oceans, has the potential to be a significant contributor toward increasing NWP skill.

The QuikSCAT (QSCAT) satellite mission (http://winds.jpl.nasa.gov/missions/quikscat) provides remotely sensed surface vector wind information over the world's oceans. QSCAT has been in "wind observing mode" for all but very brief interruptions since August 1999. Surface vector winds are retrieved for 25 km wind vector cells (WVC) in 1600 km swaths that span the sub-satellite ground track in its polar orbit orientation. Nominal global coverage of 92% of the ice-free global ocean is achieved every day.

Proper insertion of available observations into operational NWP systems is as important as the observational data themselves. Amalgamating diverse data sets which may contain large numbers of observations with differing observational frequencies and varying error characteristics into a single, coherent three-dimensional representation of the atmosphere can be a daunting task. The Local Analysis and Prediction System (LAPS) is designed to facilitate these data assimilation goals (McGinley et al. 1991, Albers et al. 1996, McGinley 2001). LAPS development is on-going at NOAA's Forecast Systems Laboratory. The system is capable of combining all available meteorological data sources, such as satellite, radar, and directly-sensed data, into a three-dimensional atmospheric analysis at high spatial and temporal resolutions. The system is particularly well suited to incorporate satellite information that typically contains numerous observational data at inconsistent observational frequencies and varying quality (Birkenheuer 1999). Horizontal grid spacing is user controllable allowing one to specify a high-resolution grid capable of retaining the high spatial quality of satellite data. LAPS, therefore, is a useful operational nowcasting tool, and also provides the capability to utilize a variety of high-resolution data sets for initialization of mesoscale NWP models.

Regional NWP efforts have expanded tremendously during the last decade (Mass and Kuo 1998, Snook 1998) primarily due to advancements in affordable computer workstations (Cotton et al. 1994). Mesoscale models that are designed to function on grid scales of less than 10 km have a rich development history in the research community. Two such models are the Penn State University-National Center for Atmospheric Research (NCAR) mesoscale model (MM5, Grell et al. 1995) and the Colorado State University Regional Atmospheric Modeling System (RAMS) model (Pielke et al. 1992). These models have been tuned to run efficiently on today's affordable workstations and also scale well in a parallel computer environment using multi-processor platforms. Numerous examples of operational mesoscale modeling efforts (e.g. Snook et al. 1998) have demonstrated the viability of running mesoscale models to support real-time forecasting activities. Most recently, a collaborative effort, between NCAR, the NOAA National Centers for Environmental Prediction (NCEP), the NOAA Forecast Systems Laboratory, and the Oklahoma State University, is developing a comprehensive mesoscale modeling system, called the Weather and Research Forecasting (WRF) model, designed to satisfy the future requirements of both the research and operational NWP communities. It will be necessary to incorporate the latest advances in observational platforms (e.g., QSCAT) and data assimilation techniques (e.g., LAPS) into these NWP systems to take full advantage of what these models have to offer.

2. PROGRESS TO DATE

Colorado Research Associates (CoRA) scientists are active participants on the NASA Ocean Vector Winds Science Team, and are very familiar with QSCAT data. QSCAT wind speed and direction are retrieved from composite radar backscatter returns, using empirically-based geophysical model functions. Wind speed accuracies are within 1 m/s, for winds between 3 and 20 m/s, based on comparisons with global distributions of meteorological buoy data. Wind direction retrievals are somewhat ambiguous, with as many as four wind direction retrievals fitting a given model function well enough to be retrieved. In the QSCAT mission standard surface vector wind product (available as the Level 2B product from NASA/JPL), wind direction ambiguities are resolved using a median filter algorithm. For these retrievals, wind direction accuracies are better than 25 degrees RMS with respect to the selected ambiguity, for winds in the range 3 to 20 m/s (see table at the end of Milliff et al. 2001 for a comparison of past, present, and planned scatterometer mission accuracies, coverages, etc.)

Several studies have shown that wavenumber spectra computed from scatterometer winds contain realistic power over synoptic to sub-mesoscale spatial length scales (e.g., see Freilich and Chelton, 1986; Wikle et al. 1999). Moreover, these are length scales over which surface winds from weather-center analyses are known to be deficient (e.g., see Chin et al. 1998). For example, spectral power corresponding to spatial scales on the order of a few hundred kilometers in the NCEP FNL winds is more than an order of magnitude lower than in spectra based on QSCAT for the same time period (personal communication Jan Morzel, 2002).

Realistic high wavenumber variability in QSCAT winds is supportive of realistic mesoscale surface convergences and divergences in the surface wind fields. Our pilot experiment, described below, is initialized at 0300 UTC with surface winds for 15 April 2001. Figure 1 demonstrates the surface convergence differences between QSCAT (Fig. 1a) and NCEP FNL (Fig. 1b) for the strong low pressure system in the northeastern Pacific at this time. The NCEP FNL (Fig. 1b) surface winds have been tri-linearly interpolated to the QSCAT WVC locations for this comparison. Convergence patterns are stronger in the QSCAT winds for bands at the leading edge of the low pressure system as it impinges upon the West Coast of the U.S.

The assimilation of surface winds from scatterometer observations into global numerical weather prediction models has reached operational status at ECMWF (since the ERS-1,2 systems) and NCEP (starting recently with QSCAT data). Atlas and Hoffman (2000) and Atlas et al. (2001) review issues and results in maximizing the forecast impact of these data in the global AGCM context. Applications of scatterometer winds in atmospheric mesoscale models are less well developed. While data assimilation experiments have been implemented (e.g., Majumdar et al. 2002), these typically involve process studies for specific meteorological phenomena of interest (e.g., hurricanes). Near real time QSCAT data are made available to NOAA marine forecast offices for nowcasts. These data are produced for public consumption by the NOAA Marine Observing Systems Team.

A complete LAPS is fully operational at CoRA. The National Weather Service (NWS) utilizes a system called NOAAPort that broadcasts all operational NWS meteorological data and forecasts through a geostationary satellite link to all NWS Forecast Offices around the country. An on-site NOAAPort receiver allows CoRA to ingest all publicly available NWS meteorological products in real-time. The operational CoRA-LAPS is integrated with the NOAAPort feed to create hourly, three-dimensional analyses of the atmosphere. CoRA-LAPS currently supports three separate operational activities: 1) FORETELL - a system that provides winter maintenance decision support to the Iowa Department of Transportation (DOT) (Snook 2001, www.foretell.com), 2) RATIS - a system that provides winter maintenance decision support to the three northern New England DOT's (www.foretell.com), and 3) Foresight Weather - a system that provides high-resolution weather forecasts tailored for the utility industry (www.fswx.com). LAPS uses either an 8 or 10-km horizontal grid spacing with 21 vertical isobaric levels with 50 hPa increments for these projects. Each application uses a customized domain designed to meet the requirements of the end-user - regional domains for FORETELL and RATIS, a national domain for Foresight Weather. In addition to atmospheric state variables, CoRA-LAPS generates derived variables such as vorticity, divergence, atmospheric stability, and wind chill.

Each of the above operational activities utilizes the MM5 mesoscale model to generate real-time predictions. The FORETELL and RATIS systems produce 30 hour forecasts four times per day. The Foresight Weather project uses a cluster of 12 IBM 44P computers each with four processors to create two daily 51 hour forecasts and one daily 11 day forecast. CoRA has developed a variety of visualization techniques, including web-based, Vis5d, and GrADS, to present forecast products to the end-user. Comprehensive validation schemes have been implemented to monitor NWP forecast quality. CoRA also has experience using the RAMS and WRF models (Poulos et al. 2002, Meyers et al. 2002).

3. PILOT EXPERIMENT

Abundant and accurate initialization data for the surface wind field from QSCAT might play a significant positive role in improving NWP forecast skill. A preliminary sensitivity experiment was designed to demonstrate the ability to evaluate QSCAT winds in a mesoscale model at CoRA, and to provide a cursory look at the potential for increased mesoscale NWP skill using QSCAT winds. The experiment design utilized three NWP forecasts using a commonly configured mesoscale model each with a different initial condition: 1) NWS analysis from NCEP, 2) LAPS high-resolution analysis derived from NWS data, and 3) LAPS analysis using NWS data plus QSCAT wind data. A case study day of 15 April 2001 was selected because QSCAT wind data was readily available for this date and observational data showed a significant eastern Pacific cyclone. A common domain was selected that covered the eastern Pacific and western two-thirds of the United States (Fig. 2). The domain covers a large area of collected QSCAT winds and the full cyclone area, and additionally covers a large portion of the U.S. to allow for the evaluation of any forecast improvement over land. The domain grid uses a horizontal grid spacing of 25-km, roughly equivalent to the observational spacing of the QSCAT winds, for a total area of 6150 km by 3650 km (247 x 147 grid points).

Experiment (1) used data from the NCEP global Aviation (AVN) model initialized at 0000 UTC 15 April 2001. The AVN grids were available on a 1.25 X 1.25 degree latitude-longitude grid and were spatially interpolated to the experiment grid domain. The AVN 0000 UTC analysis grid and the 6-hour forecast grid were temporally interpolated to the experiment initial time of 0300 UTC. For experiment (2), LAPS was configured to assimilate data directly to the 25-km experiment grid domain. LAPS analyses were generated using a spatially- and temporally-interpolated AVN atmospheric prediction from the 0000 UTC initialized run as a background field and all available NOAAPort observations (primarily surface METAR reports) from 0300 UTC were assimilated into the analysis. 0300 UTC was selected because a satellite pass within a half hour of this time provided the maximum amount of QSCAT wind observations within the model domain (Fig. 2). LAPS generated a similar analysis for experiment (3) except that all QSCAT winds within +/- 30 minutes of 0300 UTC were incorporated. The LAPS surface analysis has internal diagnostics that include a check of the analysis against the current dataset (Snook et al. 1998). The analyses are interpolated to each observation location using a bicubic spline, and analysis minus observation differences are collected. Table 1 shows the results of this dependent verification for the 0300 UTC 15 Apr 2001 LAPS wind speed analyses. For the "without QSCAT" analysis, the LAPS wind speeds showed a slight fast bias (0.47 kts) with a mean absolute error (MAE) of 1.72 kts. This was a significant improvement over the background wind speeds from the AVN model at these 762, mostly land-based, observation locations. The LAPS analysis with QSCAT observations showed a further reduction in both bias and MAE. The LAPS analyzed speeds were slightly slower than the observations (-0.17 kts), while the MAE was well within the expected sensor error of around 1 m/s. Note the difference in the bias and MAE for the AVN background in the with and without QSCAT analyses. There is, of course, no change in the AVN background fields. The difference is due to the 6000+ QSCAT observations--the AVN fields over the ocean are not as good relative to the QSCAT observations as they are over land relative to the surface-based observations. The LAPS analysis was able to blend the AVN background with the QSCAT (and other) observations to correct for these differences.

Forecasts using the MM5 (version 3.5) mesoscale model were generated for each experiment using their respective initial conditions. With the exception of the initial condition, the model configuration was the same for all three experiments. Details of the model configuration are summarized in Table 2. The forecast model horizontal grid is equivalent to LAPS (i.e., 247 x 147 grid points with a 25-km grid increment) and 25 sigma-pressure (terrain following) levels were used in the vertical. Forecasts were generated out to 69 hours and used AVN forecasts, available at 6-hour increments, from the 0000 UTC initialized run as forecast lateral boundary conditions. MM5 model physics were selected to complement the grid-scale resolution and the spring-season meteorology. The model is nonhydrostatic and the highest available level of moisture complexity (Reisner et al. 1998) was employed. The MM5 Reisner2 scheme uses predictive equations for cloud water, rain, cloud ice, snow, graupel, and number concentration of cloud ice.

Figure 3 shows objective validation of temperature, dew point, and wind through comparison of surface observations with MM5 forecasts through 44 hours for each experiment. Mean absolute and rms differences are computed by subtracting each observation from the MM5 forecast. All available surface observations within the MM5 model domain are used which generally ranged between 700 and 800 locations. MM5 forecasts were interpolated to each observation location using a piecewise bicubic spline technique. Forecast model temperatures are adjusted for differences between low-level model height and surface observation elevation by using a standard lapse rate of -6.5 K/km. The results for temperature and dew point suggest that experiments 2 and 3 show an improvement in forecast quality when compared to experiment 1 through about the first 24 forecast hours. Beyond 24 hours, the forecasts are nearly identical suggesting that the forecast lateral boundary condition, which is the same for all three experiments, has become an important influence over the limited area domain. Also, the improvement in wind forecasts is only observed in the initial condition. Interestingly, forecasts 2 and 3 are nearly identical indicating that all noted improvements are due to the high-resolution LAPS analysis and essentially no improvement results from the introduction of QSCAT winds into the LAPS analysis. This result may be related to the location of the land-based surface observations which are geographically displaced away from the strongest influence of the over-water-based QSCAT wind observations.

QSCAT wind observations were available at approximately 12 hour increments following model initial time. Figure 4 shows objective validation of wind compared to just the QSCAT observations. Observational data were available at the 11, 24, and 36 hour forecast times with the number of locations ranging from just over 5000 to nearly 7700. The rms difference validation suggests only a very small improvement in the model forecast skill when the initial condition included the QSCAT winds based on the rms error validation. While improvement to prediction skill beyond 24 hours may not be expected because of the forecast lateral boundary condition constraint, the results at 11 hours are, at least for this case study, disappointing. More case study and operational applications will be necessary to determine if this case is an anomaly.

Qualitative validation results of forecast precipitation are, however, much more encouraging. Figure 5 shows the 12-hour forecast of mean sea-level pressure and accumulated precipitation for the previous three hours for each of the three experiments. Experiment one shows a narrow elongated band of precipitation with a maximum accumulation of about 5 mm in the three hour period. Experiments 2 and 3 indicate a broader and less elongated band, and the QSCAT inclusive run (experiment 3) shows greater maximum precipitation accumulation of greater than 7.5 mm. Twelve hours later, the 24-hour forecast (Fig. 6) shows a similar single rain band structure for the control run (experiment 1), while experiments 2 and 3 indicate a more banded structure with the most structure noted in the QSCAT run (experiment 3). This type of banded precipitation structure is often observed with well developed low pressure systems, such as in this case study, over the Pacific, and SSM/I satellite imagery suggests that this system did indeed contain multiple bands of precipitation.

By the time of the 36-hour forecast, all three runs are indicating multiple precipitation bands (Fig. 7). The QSCAT run (experiment 3), however, shows more bands with greater accumulated precipitation when compared to the other two experiments. The greatest differences between runs is noted beyond the 36-hour forecast. As the Pacific cyclone makes landfall, the accumulated precipitation tends to diminish in the control and no-QSCAT winds runs (experiments 1 and 2), while the multiple banded precipitation structure continues in the QSCAT run (experiment 3) (Fig. 8 and Fig. 9). The differences at the time of the 60-hour forecast are quite dramatic, with only small, isolated areas of precipitation noted in the non-QSCAT runs and rather intense, elongated bands of precipitation making landfall in the QSCAT run. These results are very interesting in that the three experiments showed very little difference in the predictions of temperature, moisture, and wind, and yet, very significant differences are noted in the accumulated precipitation even after nearly three days of forecast. Accurate model precipitation forecasts along the U.S. West Coast are difficult because of the sparcity of data over the Pacific. This case study suggests that QSCAT wind data has the potential to improve model precipitation forecasts over oceanic and coastal regions.

4. ISSUES FOR PROPOSED WORK

The results to date, while tantalizing, are not sufficient to identify a signal in the LAPS-MM5 response to initialization with QSCAT winds. We are encouraged to pursue this project in a carefully designed experimental setup to quantify QSCAT initialization impacts as functions of: model output variable, season, synoptic setting, forecast lead-time, etc.

It remains to separate MM5 internal variability (e.g., in response to any perturbation of initial or boundary conditions) from the variability in the response due specifically to initialization with QSCAT winds. Weisse et al. (2000) demonstrate signal vs. noise separation of this kind in limited-area atmospheric model response to surface stress conditions. Adapting their procedures, an ensemble of initialization experiments must be run in the control case to identify the "noise" of the MM5 response to small perturbations, unrelated to initialization with QSCAT. Similarly, an ensemble of forecasts based on realistic perturbations of QSCAT initializations must be run as well to identify the model variability about the test case. The MM5 response to QSCAT initialization can then be quantified in terms of the differences in the variabilities of the control ensemble versus the test case ensemble. Comparisons of this kind can be made for a variety of synoptic conditions, over seasons, etc.

The variabilities (control vs. test case) are best placed in the context of the model output where we might expect the largest signal. Our pilot experiment identifies a potential signal in the model output having to do with precipitation; notably at lead times in excess of 12 hours. We can explore various sources of precipitation in the MM5 model and parameterizations; e.g., convective, vs resolved precipitation. Optimum forecast model grid resolution should also be evaluated.

Once the signal is identified and quantified in an optimal subset of model output variables, we can consider the signal realism. We are beginning to examine remote sensing observations from space that are relevant to the precipitation differences for the time period of our pilot experiment. Rain rate and cloud water concentrations are derived from SSM/I data and made available at www.ssmi.com (Wentz and Spencer 1998). Figure 10 depicts the rain-rate and cloud water about 16 hours into the pilot experiment forecast. These variables are directly comparable to MM5 model outputs. Land-based comparisons can be achieved using NWS point observations and operational radar data that is ingested in real-time through the CoRA NOAAPort system. Quantitative precipitation forecasts can be compared to observations and radar derived accumulated precipitation. Areal extent and banded structure of the precipitation forecasts can be compared to NWS radar reflectivity data.

Additional research questions raised by the pilot experiment include: how representative of measured responses in the real atmosphere (e.g., to small scale surface convergence) are the responses exhibited by MM5 given initialization with QSCAT winds? Do the precipitation differences make meteorological sense, or can they be directly tied to artifacts of a model parameterization? In the case of unrealistic dependence on model parameterization, how might the parameterization be improved given our results?

5. REFERENCES

Albers, S., J. McGinley, D. Birkenheuer, and J. Smart 1996: The Local Analysis and Prediction System (LAPS): Analyses of clouds, precipitation, and temperature. Wea. Forecasting, 11, 273-287.

Atlas, R. and R. N. Hoffman, 2000: The use of satellite surface wind data to improve weather analysis and forecasting. In Satellites, Oceanography and Society, D. Halpern (Ed.), Elsevier Oceanography Series, No. 63. Elsevier Sci. Ltd., Oxford.

Atlas, R., S. C. Bloom, J. Ardizzone, E. Brin, J. Terry, and T-W. Yu, 2001: The impact of QuikScat on weather analysis and forecasting. Preprints, 18th Conf. on Weather Analysis and Forecasting, Ft. Lauderdale, FL, Amer. Meteor. Soc, 172-175.

Birkenheuer, D., 1999: The effect of using digital satellite imagery in the LAPS Moisture Analysis. Wea. Forecasting, 14, 782-788.

Chin, T. M., R. F. Milliff and W. G. Large, 1998: Basin-scale high-wavenumber sea surface wind fields from a multiresolution analysis of scatterometer data. J. Atmos. Ocean Tech., 15, 741-763.

Cotton, W. R., G. Thompson, and P. W. Mielke Jr., 1994: Realtime mesoscale prediction on workstations. Bull. Amer. Meteor. Soc., 75, 349-362.

Davies, H. C., 1976: A lateral boundary formulation for multi-level prediction models. Quart. J. Roy. Meteor. Soc., 102, 405-418.

Freilich, M. H. and D. B. Chelton, 1986: Wavenumber spectra of Pacific winds measured by the Seasat scatterometer. J. Phys. Oceanog., 16, 741-757.

Grell, G. A., 1993: Prognostic evaluation of assumptions used by cumulus parameterizations. Mon. Wea. Rev., 121, 764-787.

Grell, G. A., J. Dudhia, and D. R. Stauffer, 1995: A description of the fifth-generation Penn State NCAR Mesoscale Model (MM5). NCAR Technical Note TN-398+STR. 122 pp.

Kalnay E., S. J. Lord, and R. D. McPherson, 1998: Maturity of operational numerical weather prediction: Medium range. Bull. Amer. Meteor. Soc., 79, 2753-2769.

Klemp, J. B. and D. R. Durran, 1983: An upper boundary condition permitting internal gravity wave radiation in numerical mesoscale models. Mon. Wea. Rev., 111, 430-444.

Majumdar, S. J., S. S. Chen, J. Tenerelli, and R. Foster, 2002: Using 4D-VAR to assimilate satellite data into MM5 Hurricane Floyd simulations. Paper presented 2002 Amer. Met. Soc. annual meeting.

Mass, C. F. and Y.-H. Kuo, 1998: Regional real-time numerical weather prediction: Current status and future potential. Bull. Amer. Meteor. Soc., 79, 253-263.

McGinley, J. A., 2001: Toward a surface data continuum: Use of the Kalman filter to create a continuous, quality controlled surface data set. Preprints, 18th Conf. on Weather Analysis and Forecasting, Ft. Lauderdale, FL, Amer. Meteor. Soc, 127-131.

McGinley, J. A., S. Albers, and P. Stamus, 1991: Validation of a composite convective index as defined by a real-time local analysis system. Wea. Forecasting, 6, 337-356.

Meyers, M. P., J. S. Snook, D. A. Wesley, and G. S. Poulos, 2002: A Rocky Mountain storm - Part II: The Forest Blowdown - observations, dynamics and modeling. Wea. Forecasting. (in review)

Milliff, R. F., M. H. Freilich, W. T. Liu, R. Atlas, and W. G. Large, 2001: Global ocean surface vector wind observations from space, in Observing the Oceans in the 21st Century. C. J. Koblinsky and N. R. Smith (eds), GODAE Project Office and Bureau of Meteorology, Melbourne, 102-119.

Pielke, R. A., W. R. Cotton, R. L. Walko, C. J. Tremback, W. A. Lyons, L. D. Grasso, M. E. Nicholls, M. D. Moran, D. A. Wesley, T. J. Lee, and J. H. Copeland, 1992: A comprehensive meteorological modeling system - RAMS. Meteor. Atmos. Phys., 49, 69-91.

Poulos, G. S., D. A. Wesley, J. S. Snook, and M. P. Meyers, 2002: A Rocky Mountain storm - Part I: The Blizzard - observations, dynamics and modeling. Wea. Forecasting. (accepted)

Reisner, J., R. J. Rasmussen, and R. T. Bruintjes, 1998: Explicit forecasting of supercooled liquid water in winter storms using the MM5 mesoscale model. Quart. J. Roy. Meteor. Soc., 124B, 1071-1107.

Snook, J. S., 2001: Foretell: An operational forecasting system designed for the surface transportation community. Preprints, 14th Conference on Numerical Weather Prediction, Ft. Lauderdale, FL, Amer. Meteor. Soc, J21-J24.

Snook, J. S., 1998: Comments on "Regional real-time numerical weather prediction: Current status and future potential". Bull. Amer. Meteor. Soc., 79, 2747-2748.

Snook, J. S., P. A. Stamus, J. Edwards, Z. Christidis, J. A. McGinley, 1998: Local-domain mesoscale analysis and forecast model support for the 1996 Centennial Olympic Games. Wea. Forecasting, 13, 138-150.

Weisse, R., H. Heyen and H. von Storch, 2000: Sensitivity of a regional atmospheric model to a sea state-dependent roughness and the need for ensemble calculations. Mon. Wea. Rev., 128, 3631-3642.

Wentz, F. J., R. W. Spencer, 1998: SSM/I rain retrievals within an unified all-weather ocean algorithm. J. Atmos. Sci., 55, 1613-1627.

Wikle, C. K., R. F. Milliff, W. G. Large, 1999: Surface wind variability on spatial scales from 1 to 1000 km observed during TOGA COARE. J. Atmos. Sci., 56, 2222-2231.