By John Smart, John A. McGinley, Steve Albers, Daniel Birkenheuer, Stephen A. Early, James Edwards, Peter A. Stamus, and Sher Wagoner
Figure 1. LAPS-MM5 zero-hour mean sea level pressure (mb) and cloud top height analyses for 0000 UTC 15 December 1999 in the European domain. Cloud top heights are color coded with scale included (feet x 1000). Analysis domain grid is 36 km.
IntroductionThe Local Analysis and Prediction System (LAPS) has come a long way since it was deployed at the Air Force Global Weather Center in 1995 (see FSL Forum, December 1996). As part of a Proof of Concept system, LAPS demonstrated high resolution two- and three-dimensional forecast products that helped improve weather prediction during the Bosnia peacekeeping effort. The success of the system for the Bosnia campaign led to a request for FSL to develop a worldwide-capable LAPS for the Air Force's operational Global Theater Weather Analysis and Prediction System (GTWAPS).
This article recaps some of the earlier work that FSL performed at the Air Weather Service [since, renamed the Air Force Weather Agency (AFWA)] and describes more recent LAPS improvements and developments.
Initial LAPS Installation at AFWA
FSL's role in the AFWA upgrades that improved the theater weather forecasting capability for Air Force operations worldwide began with the installation of LAPS. In collaboration with the Air Force Global Weather Center and Argonne National Laboratory, FSL developed a mesoscale data assimilation and initialization system, integrated LAPS (merging Air Force data formatting and ingest) into an existing parallelized operational mesoscale model (MM90), and developed a prototype data visualization system for displaying graphical data. LAPS was configured to cover the Bosnian theater, and both LAPS and MM5 out- put grids were displayed on the WFO-Advanced workstation that FSL had installed at the Global Weather Center.
Following the 1995 effort, AFWA requested a system that was resizable and relocatable without having to recompile the software. Work to convert the local-domain LAPS analysis to operate on either large or small domains with varying data densities evolved into a "renaissance" of LAPS. These changes rendered it able to relocate and resize "on the fly" and transformed it into a highly versatile system for global data assimilation. A recent example of LAPS-MM5 graphical output at AFWA is illustrated in Figure 1.
The upgrades and developments discussed here include MM5 initialization, LAPS software configuration and development, AFWA-LAPS local infrastructure, LAPS global data assimilation, new data ingest capability, redesign of software analyses, and the addition of a Four-Dimensional Data Assimilation (4DDA) capability that utilizes the MM5 output fields. Other LAPS changes include Y2K stability, software configuration management, and automated installation capabilities.
The primary use of LAPS assimilation at AFWA is to initialize the Fifth Generation Mesoscale Model (MM5), developed jointly by the Pennsylvania State University and the National Center for Atmospheric Research. AFWA takes the LAPS three-dimensional products (22 pressure levels at 50 mb increments) and interpolates them to the MM5 sigma coordinate (33 levels). LAPS is run with several time delays to maximize data density. The time delay feature allows for dynamic nudging of the first model forecasts to the corresponding analyses. AFWA is now running LAPS on a 3-hour cycle, corresponding to the MM5 initialization and forecast output requirements.
LAPS Software Configuration and Development
The LAPS software is routinely built each week for local operational runs that support FSL's WFO-Advanced, Regional Observation Cooperative (ROC), public, and Web products. About once a month this software is approved for public release and the tarfile is posted on the Web. [Tarfile is short for tape archive, a Unix utility that combines a group of files into a single file.] When new capabilities are thoroughly tested, the tarfile is approved for porting to AFWA. New software ported to AFWA is easily merged into the existing theaters (analysis domains) through the LAPS localization process. Next, the Harris Corporation (overseer of the LAPS software at AFWA) integrates the new code into the GTWAPS software configuration management system. This software is then recompiled, and new executables are quickly used in the operational theaters. AFWA is now using LAPS in 12 theaters around the globe.
AFWA-LAPS Local Infrastructure
To ensure that new software runs properly, staff from FSL's Local Analysis and Prediction Branch developed a local infrastructure that collects raw (preanalysis) data and LAPS (postanalysis) products from AFWA's computer network. The preanalysis infrastructure collects the same raw data that are used in the AFWA-LAPS runs. Enough data is collected to run LAPS for two analysis times in two theaters, so that FSL can closely duplicate what is being done at AFWA and can further develop the local software for new and existing AFWA data. This capability has proven to be a very valuable tool in meeting AFWA's rapid development cycle. For the other half of the local infrastructure, the postanalysis product collection, the LAPS products from AFWA are collected and processed by the Web product generation process. This procedure allows many interested organizations to view AFWA-LAPS analysis fields accessible from the LAPS Website.
LAPS Global Data Assimilation
The earlier LAPS at AFWA generated fields for meteorological analysis, application algorithms, and model initialization. The LAPS assimilation scheme was built to adapt gracefully to varying amounts and types of data and to be ported to different geographic areas. Its processing consisted of data ingest, data reformatting to accepted standards, quality control, and objective analysis.
The redesigned LAPS system is dynamically adjustable for domain size and location for data assimilation and weather prediction on polar stereo, Lambert, and Mercator projections. LAPS currently assimilates and analyzes data on 12 operational GTWAPS domains, including Alaska, extended continental United States (CONUS), Europe, Southwest Asia, Central America, South America, and East Asia. The assimilation is performed on large and unusually coarse (for LAPS36-km) domains. The LAPS output provides the initial conditions for MM5 through a data gridding and interpolation process managed by AFWA staff. When LAPS is not running in FDDA, it uses the Navy Operational Global Atmospheric Prediction System (NOGAPS) or the Aviation (AVN) model forecasts (one degree latitude/longitude resolution) for the first-guess background.
New Data for LAPS at AFWA
FSL staff added new datasets to the AFWA-LAPS analyses during this past year. Some of the datasets have been used for years for in-house LAPS runs, while others are relatively new to LAPS. The schematic in Figure 2 illustrates the current ingest capability and data assimilation at AFWA. For example, the datasets include ACARS winds (automated aircraft reports), which have been used for some time at FSL, and satellite cloud-drift winds (Figure 3), which have been used only on an experimental basis at FSL. Temperature soundings derived from polar-orbiting satellites such as the Television and Infrared Operational Satellite (TIROS) and the Defense Meteorological Satellite Program (DMSP) have been added, as well as temperatures from ACARS. Both datasets are new to LAPS, though ACARS temperatures (Figure 4) have been available at FSL and on the Internet.
The satellite database at AFWA is extensive, with data from polar and geosynchronous satellites covering the entire globe. LAPS already uses data from the GOES-8 and GOES-10 geosynchronous operational environmental satellites. Unique at AFWA is other geosynchronous satellite data that became available to LAPS this year, including data from the Geosynchronous Meteorological Satellite (GMS) and the European Meteorological Satellite (METEOSAT). Both GMS and METEOSAT only have visible, water vapor, and the window IR channel (11.0 micron), but these data are of tremendous value for the LAPS cloud, moisture, and surface analyses. An example of METEOSAT brightness temperatures analyzed by LAPS for the European theater is shown in Figure 5.
A global data assimilation system must be able to use radiance measurements from polar-orbiting satellites. Our familiarization with the AFWA satellite database will allow LAPS to access these data in future projects and treat them using variational methods. The polar-orbiting satellites offer both radiometric and microwave radiance data, which will enhance the analysis of temperature and clouds.
Figure 2. Illustration of the current ingest capability and data assimilation at AFWA.
Figure 3. 500-hpa wind observations for 1800 UTC 27 July 1999 over LAPS-AFWA extended CONUS analysis domain. Plot includes RAOBS (lavender), ACARS (vertically displaced = purple, at the level = blue), and cloud drift winds (vertically displaced = brown, at the level = yellow).
Figure 4. Plot of LAPS 500-hpa temperature analysis (dashed orange, every 2oK), AVN model background (dashed green, every 2oK), derived satellite soundings (lavender, oK), RAOBS (blue, oK), and ACARS (red, oK) for 0000 UTC 28 July 1999.
Figure 5. Example of METEOSAT brightness temperatures (oK) analyzed by LAPS for the European theater for 1800 UTC 10 November 1999.
Improved LAPS Analyses
Wind Analysis New software changes are being implemented in the wind analysis to make the best use of the expanded datasets. The sheer number of observations necessitates a more efficient weighting and looping strategy (Figure 6). We are using large domains (up to 5000 km across) with a three-hour analysis cycle. Since the time window for observations is rather long (up to about three hours), we correct the wind observations for the time tendency.
Figure 6. LAPS 300 hpa wind analysis (kts, standard wind barb convention; isotachs at 10-kt interval) for 0000 UTC 5 March 1999. Plotted are ACARS wind observations near the 300-hpa level that are used in the analysis.
An improved feature of the wind analysis is a smooth blending with the model background. This entails the subtraction of the model first-guess background from the observations prior to the analysis. For data-void regions, the weights are set so that the analysis increment approaches zero, and the analysis reverts to the model first guess. Since the weights are continuous and fairly smooth, obvious bulls'-eyes are avoided in these data-void regions. Any density or spatial distribution of data is thus accommodated in a graceful manner. An example of this is having thousands of ACARS winds over the continental United States, while no observations of any kind exist over the Atlantic Ocean. The analysis can also work with the extreme situations of having only one RAOB in the entire domain, or no data at all.
Some additional work needs to be done in assessing the quality of the data, especially the cloud-drift winds, in terms of both random and systematic errors. Weighting each observation according to the instrument error can be explored, along with the successive correction algorithm described below for the temperature analysis.
A wind analysis for the extended continental United States (CONUS, t2 on the Internet) domain is shown in Figure 6. An example of a wind analysis for a region of sparse observational data, the Antarctic analysis domain, is shown in Figure 7.
Figure 7. LAPS 500-hpa height (dm) and wind (barbs, kts, standard convention) analyses in the Antarctic domain at 0000 UTC 17 December 1999. The zero degree longitude points toward the top of the figure; the dateline (180o longitude) points toward the bottom. Wind observations from radiosondes (three of them) are plotted in green barbs (upper right side and near the center).
Temperature Analysis A significant redesign is underway for the temperature analysis to make the best use of the expanded datasets (Figure 4). Again, the huge volume of observations requires a more efficient weighting and looping strategy. Analyzing strong vertical temperature gradients with a combination of vertical temperature profiles (satellite and rawinsonde soundings) with point data (ACARS) calls for a new, more generic three- dimensional weighting algorithm. This contrasts with the previous scheme of using a simple vertical interpolation of soundings to each LAPS pressure level. The new algorithm involves a multiple-pass Barnes analysis with successive correction and a decreasing (three- dimensional) radius of influence with each analysis pass.
An important feature of the new algorithm is retaining (and even improving) the ability to handle a variable density of observations in different parts of the domain. We often see thousands of satellite observations of some parts of the domain, with a smaller number (or none at all) of ACARS temperatures and RAOBS. As with the wind analysis, the temperature analysis will analyze large-scale features where there is relatively sparse data and smaller-scale features where there is denser data to support them.
The temperature analysis blends smoothly with and incorporates structure from the model background (done in increment space similar to the wind analysis).
The treatment of observation error has become more of an issue, since we are combining satellite temperature soundings having a larger error with the other data sources that tend to be more accurate. Each successive correction iteration, occurring at a smaller scale, will tend to match the observations more closely. A criterion for stopping the iterations is when the root-mean- square (rms) fit of the analysis to the observations equals the composite rms instrument error. This is assumed to be 5 m s-1 if all the observations are satellite soundings, 1 m s-1 if all are other data sources, or an intermediate value if a mix of data exists. Thus, when the domain is dominated by satellite soundings, the iterations stop early and produce a large-scale analysis. In this case, spurious small-scale features are suppressed, and the large-scale smoothing of the observations tends to cancel out the large random instrument error. Conversely, if the data are deemed to be accurate, and there is any clustering of the observations in local areas, the iterations will continue, and will allow fine features to be resolved in those areas having a lot of data to support them.
Further investigation may be needed to see if any systematic errors are occurring along with the random errors in satellite temperatures. If so, we may need some additional analysis strategies. Weighting each temperature observation according to the instrument error can also be explored.
Moisture Analysis The 3-D moisture analysis has been available to AFWA since its initial installation at Offutt Air Force Base in 1997. The generic LAPS port allows AFWA to take advantage of traditional moisture analysis techniques, including cloud moisture, layer water vapor retrieval using satellite radiances, and the LAPS surface humidities for boundary layer enhancement in the 3-D moisture analysis.
The new element of LAPS devised specifically for AFWA is the incorporation of radiosonde observation (RAOB) data in the moisture scheme. Prior to the AFWA project, there had been no direct requirement to use radiosonde data in LAPS, since in real- time application, RAOBs generally miss the data cutoff time for hourly analysis. AFWA operational requirements permit a later data cutoff time, and as a result, RAOBs are a viable new data source for LAPS application. Since the RAOBs were added to the moisture analysis in 1998, their application has been studied and improved.
RAOB insertion, along with the other serial adjustments to background moisture, occurs after input quality control corrects supersaturation or negative specific humidity and the field has been adjusted for boundary layer moisture. Furthermore, the RAOB insertion step precedes the GOES moisture adjustment (a multichannel variational step) and the final analysis adjustment for cloud saturation. The reason for placing the RAOB adjustment prior to the GOES variational step is to take advantage of the in- situ RAOB measurements to improve the analysis field, providing a better first guess for the variational satellite moisture adjustment. By placing the cloud enhancement after the RAOB adjustment (following the satellite step), we further guarantee that the cloudy areas are indeed saturated.
The RAOB insertion follows the boundary layer adjustment. The LAPS ".snd" file contains all relevant RAOB data used in this analysis. Simultaneously, the RAOB data are located within the LAPS domain, and the appropriate weighting function for each grid point is based on the distance between the RAOB location and LAPS grid point. These weights are applied later. The RAOB influence is assessed at each grid point in the domain and compared to a threshold corresponding to a nearest distance of 2880 km. If the distance to the nearest RAOB exceeds this value at any location in the domain, the RAOB data insertion is skipped for that cycle. This is done to prevent cases in which singular RAOBs or dense clusters of RAOBs cause artifacts in the analyzed fields and also reveal RAOB location. An AFWA requirement is that the imaging of any analysis must not reveal the location of confidential assets. From experience, the best way to prevent this is to guarantee critical data density.
Next the RAOB data are vertically interpolated to the LAPS pressure levels, and then differences are computed between the RAOB and LAPS field at each level. This set of differences is then assumed to represent the mean bias and the variance in the differences that relate to random (Gaussian) error at each level. Thus, the mean difference and standard deviation are computed for each layer and used to determine outliers using a tolerance of plus or minus two-sigma difference of the mean bias.
Data passing this criterion at specific levels are then used for scaling purposes. A scaling ratio (RAOB/gridpoint value) is computed for each accepted RAOB location and LAPS level. This scaling function is then analyzed over the domain using the weights that were previously determined. The resulting scaling field is then applied at each grid point.
Using a scaling field has important benefit over the more traditional direct bias adjustment. Direct bias adjustment works well for small domains, such as the Front Range LAPS. However, when domains extend to thousands of kilometers, as configured by AFWA, adjusting to differences can run into serious problems, especially when allowing a RAOB to influence grid points that may be many thousands of kilometers away. We observed that direct difference adjustments could result in the generation of negative moisture if a reduction bias was applied to a very low moisture region under the influence of a remote RAOB. The scaling solution is immune to this problem, since a 50% reduction in moisture at a RAOB location could be applied to a more remote location. Even if the moisture at a remote location were extremely low, it would only be halved. Furthermore, this adjustment would not only be slight in absolute moisture (when the initial amount was low), but would never be negative. The scaling approach also is not prone to the generation of artifacts that draw attention to RAOB locations (such as bulls' eyes) in the analyses.
Four-Dimensional Data Assimilation
The Air Force project has taken LAPS development into the challenging area of 4DDA. Here, MM5 mod-el outputs are used as the background first guess for data assimilation. These analyses, in turn, are used to initialize subsequent MM5 model runs. AFWA uses a parallel version of the MM5 local-scale model for 4DDA.
Implementation of 4DDA into LAPS operations presented new problems requiring quick resolutions. For instance, the first attempt to transfer moisture between the model and analysis was conveyed through relative humidity. The legacy of this unsuccessful approach emanated from the pre-4DDA application, in which the moisture analysis was simply initialized once each cycle from a given model background. If the relative humidity to specific humidity algorithm disagreed slightly between the model and LAPS (i.e., due to different reference temperatures or functions describing the conversion), no long-term consequence followed, since it was generally slight and unnoticed and could easily be corrected by the analysis. In a 4DDA context, however, when moisture was relayed back and forth between the model and analysis during each cycle, any inconsistencies between relative humidity to specific humidity conversion caused moisture to either build up or to decrease over time due to the closed, integrated, and cumulative effects of 4DDA. Once the relative humidity to specific humidity interface was standardized, this problem was solved. It is now recommended that all moisture transfer between analysis and model be done in a specific humidity or mixing ratio, and not relative humidity, in order to minimize the likelihood of any internal artificial buildup or builddown of moisture in the system.
Another problem being studied is the breakdown in the use of surface humidity for boundary layer moisture tracking in the 4DDA system. For reasons still unclear, when surface moisture is mixed into the lowest tropospheric levels of the 3-D moisture analysis, it eventually dominates the column. This is not seen in 3DDA. However, in 4DDA it may be similar to the mechanism stated above regarding the nonconservative moisture effects that occur when the relative humidity moisture transfer was not set up equitably.
Summary and Future WorkThe mission statement of FSL is to research, develop, and transfer new weather technologies into operational centers. The Local Analysis and Prediction Branch has met this challenge with LAPS at AFWA and for other operational systems, such as AWIPS/WFO. The LAPS system adapts to numerous computer platforms, is capable of ingesting several different types of data or the same data in several different formats, and is capable of relocating and resizing the grid on the fly for any location on earth. (Figure 8 shows an example of the LAPS-MM5 48-hour wind forecast for Hurricane Jose at 0000 UTC 22 October 1999.)
The LAPS system continues to be work in progress, as we have described in this article. Currently, AFWA uses limited capability available in the LAPS system. For example, profiler and radar data are presently unavailable to LAPS at AFWA, and the use of satellite data is limited due to their very large domains extending beyond the limits of a single geosynchronous satellite field of view. However, much has been learned during the collaboration between staff at FSL and AFWA. We work as a team to create a system that best meets AFWA's theater weather forecasting needs. This includes frequent trips and good communication to further expand the new data ingests, tune the analysis to these new (and existing) data, and to adapt the analysis to their operational needs. This project gives FSL staff yet another opportunity to "grow" LAPS technology in the rapidly developing operational center.
Figure 8. An MM5 48-hour wind (barbs, standard convention, and color coded) forecast valid at 0000 UTC 22 October 1999 for Hurricane Jose. The domain is a 12-km grid spacing nested within the LAPS-MM5 36-km analysis domain.
A critical part of technology transfer is to ensure that new systems improve upon the existing system. To accomplish this task, a well-planned and reliable verification system must exist so that decisions can be made unambiguously. To this end, the Global Theater Weather Analysis and Prediction System program is further defining the requirements and methodology. In addition, LAPS staff are refining text output with statistics that allow users and developers to assess the quality of the analyses. Work will continue on the verification procedures required at AFWA. The verification process must address all parts of the system, from data ingest, assimilation, model initialization, and forecasts.
Finally, our AFWA interaction opens the door for FSL to provide a future critical service: shadowing operational systems. This would offer a testbed for rapid prototyping and testing, a backup in times of emergency, and a quick response for analyzing other operational problems. The acquisition of high-performance computing systems and staff with expertise to develop and/or support variational analysis software, four-dimensional data assimilation, research of model ensembles, and involvement with the Weather Research and Forecast model will position FSL to meet a wide spectrum of operational needs in the coming years.
Acknowledgments The dedicated staff at Offutt Air Force Base have made possible the porting, installation, and running of the LAPS software. In particular, we thank Mr. Roy Peck and Mr. Chuck Meier with Harris Corp., and Mr. Mark Surmeier, Captain Robert Williams and the staff at AFWA. We also appreciate the assistance of Mr. Gordon Brooks of AFWA in supplying the graphics for Figures 1 and 8.
[Editor's Note: More information on the topics covered in this article is available on the LAPS Website, including references to published papers.]
(John Smart manages the LAPS-AFWA project and works as a researcher in the Local Analysis and Prediction Branch, headed by John A. McGinley. He can be reached by e-mail, firstname.lastname@example.org, or by phone at (303)497-6590.)