Global and local weather prediction

Understanding of the Earth system is a prerequisite to predicting its behavior, the latter being however of a more direct use to many components of society. In that context, the priorities within the laboratory deal with improving climate models, exploring new approaches to prediction across scales and global and local weather prediction. The NCAR highlight deals with the WRF/ARW. Furthermore, the laboratory presents two highlights on CCSM and on the international experimental endeavor THORPEX.

WRF/ARW [NAR Highlight w/Detail] - MMM
Data assimilation/ensembles - MMM
USWRP/STEP - MMM
Model physics - MMM

WRF/ARW

WFR/ARW real-time forecast with 12/4/1.33 km nested grids initialized at 00 UTC on 29 August 2006 depicting the simulated radar reflectivity for hurricane Ernesto within the 1.33 km and 4 km (horizontal grid size) moving nested grids embedded in the 12 km outer domain at 30 h (red domains), 56 h (green domains), and 75 h (blue domains); the dark brown dashed line reflects the forecast track with circles marking its location at 12 h intervals. These forecasts confirm that the WRF/ARW predictions for hurricane track and intensity are as or more accurate than those from current operational forecast models. The vortex-following moving nested grids allow very high resolution forecasts to be produced with great savings in computed time due to the reduced sizes needed for the moving nested grids. High resolution figure

The overall goal of the WRF Model project is to develop a next-generation mesoscale forecast model and data assimilation system that will advance both the understanding and prediction of mesoscale weather, and will accelerate the transfer of research advances into operations. The model is being developed as a collaborative effort among the NCAR MMM Division, NOAA's National Centers for Environmental Prediction (NCEP) and Earth System Research Laboratory (ESRL), the Department of Defense's Air Force Weather Agency (AFWA) and Naval Research Laboratory (NRL), the Center for the Analysis and Prediction of Storms (CAPS) at the University of Oklahoma, and the Federal Aviation Administration (FAA), along with the participation of numerous university scientists. The WRF model is intended to improve the forecast accuracy of significant weather features across scales ranging from cloud to synoptic, with priority emphasis on horizontal grids of 1 10 kilometers. The model incorporates advanced numerics and data assimilation techniques, multiple movable nesting capability, and improved physics, particularly for treatment of convection and mesoscale precipitation systems. It is well suited for a wide range of applications, from idealized research to operational forecasting, and has the flexibility to accommodate future enhancements.

The WRF project was instigated by MMM scientists nearly a decade ago in an attempt to promote closer ties between research and operational forecast model development efforts. During this period, the WRF model has been developed and has matured to become the dominant model used for both operational forecasting and community research in the United States, and WRF is also used extensively worldwide. MMM has played the lead role in the WRF project in developing the WRF software infrastructure and the Advanced Research WRF (ARW) dynamic core, and in maintaining and supporting the model for users in the research community. This project provides the primary modeling system used in support of the NCAR strategic priority to investigate the dynamics and predictability of weather systems on time scales of 0-48 h. In addition, it contributes to NCAR’s mission to provide and support state-of-the-art modeling systems for broad use in the research community.

Within MMM, WRF-project activities are distributed across three broad areas: 1) development and enhancement of WRF model capabilities to meet the needs of MMM and community-research objectives; 2) research to advance the understanding and prediction of high-impact weather systems; and 3) WRF-model support to the research community. The overall timeframe for this project is open ended as the WRF system will continue to evolve to
meet future requirements for advanced weather research and forecasting. However, specific timelines are dictated by deliverables in outside funded projects (e.g., AFWA, KMA, FAA) and commitments to provide modeling capabilities in support of MMM projects and the research community.

NCAR continued to develop new capabilities for the Advanced Research WRF (ARW) model and to support its widespread use as a community resource. During the past year, 1,000 new users registered to download the model code, bringing the total number of registered users to over 4,200. Over half of these users are distributed over some 82 foreign countries. MMM organized and conducted the 7th Annual WRF Users Workshop in June for 203 participants from 101 different institutions, and conducted three user tutorials (one in India) to instruct new users in the use of the ARW model and WRF-VAR data assimilation system. MMM also convened international WRF workshops in Korea, Taiwan, and China to assist these groups in both research and operational applications with WRF. MMM manages the common repository for WRF code, assists community researchers in code development, coordinates the addition of new code to the repository, and oversees the integration and testing for new releases of the WRF code to the community. Last year, MMM released updated versions of WRF, V2.2.1 in November 2005 and V2.1.2 in January 2006, which contained a number of new physics packages and enhancements to existing physics, improved treatment of 2-way interacting nested grids, and modifications for nested regional climate applications. Operational forecast systems based on the WRF/ARW were implemented at NCEP (Short-Range Ensemble Forecasts) and AFWA (worldwide regional theatres) during the year, and operational WRF/ARW models are under development in Korea, India, Israel, and Taiwan.

More recent enhancements to the ARW system include numerous updates for new physics packages, as well as new capabilities, such as a new preprocessor (WPS) for model initialization, nudging techniques that relax the model fields toward prescribed data on the model grid (in collaboration with Penn State) or at the location of individual observations (in collaboration with RAL), and a new urban canopy model (also in collaboration with RAL). Moving nested grid techniques were refined and employed in real-time hurricane forecasts for the first time last year. To facilitate atmosphere/ocean model coupling needed for advanced hurricane forecasting, WRF has been integrated into the Earth System Modeling Framework (ESMF). Model performance in parallel computing has been enhanced and the WRF model was ported to the Microsoft Windows operating system

Last year, the ARW was adapted for use in the Antarctic Mesoscale Prediction System (AMPS) effort, and provided real-time NWP support for the United States Antarctic Program for the 2005–2006 field season (October–March). In addition, model performance was evaluated through real-time forecasts of landfalling hurricanes, and forecasts in support of the MIRAGE field program and FAA forecast demonstration experiments. Research continued in exploring global implementations for WRF, as well as evaluation of a global version in spherical coordinates developed and contributed by Cal Tech. All of the plans outlined in last year’s program plan were accomplished except for the preparation of the Cal Tech global version for community release, which has been delayed due to lack of resources.

In the coming year, MMM will continue to assist both domestic and international users of the WRF/ARW system, and will organize WRF workshops and tutorials to accommodate the expanding use of WRF for both research and operations. Current plans are to convene the 8th WRF Users Workshop in June 2007 and to conduct tutorials in January and July of next year. A major new code release (V2.2) is planned that will allow community access to all of the newly developed model features. The growing user base for the WRF system supported by MMM, together with the strong community participation in the users workshop and tutorials, reflect the value of MMM community support for the WRF system. In addition, MMM is negotiating to establish an external advisory board in cooperation with the DTC to provide community input on MMM code management and community-support activities.

New model capabilities will also be developed and tested. Gravity-wave absorption at the upper boundary will be addressed to improve forecasting over mountainous terrain. MMM will conduct systematic testing of a newly developed positive definite advection scheme to see if model biases in overpredicting precipitation at high resolution can be significantly reduced (by eliminating the current practice of zeroing out small negative values of water species that artificially increases precipitation), and new monotonic advection schemes will also be developed and tested. In existing cold-start forecasts, the early forecast period is degraded by adjustments to imbalances in the initial state. MMM is developing a new digital filter for WRF that should allow the model to spin-up convective activity more quickly and with less spurious noise. The impact of this filter on very short-term (0-6 h) cloud-scale NWP forecasts will be evaluated in the coming year. Comprehensive analyses of past real-time convection-resolving forecasts will also continue to understand the reasons for good and bad forecast results, and to test techniques for addressing model deficiencies. Model enhancements that demonstrate the potential for forecast improvements will be incorporated in MMM real-time forecast experiments and made available for both research and operational use.
This research is conducted as part of the Short-Term Explicit Prediction (STEP) Program (funded by USWRP) in collaboration with RAL and EOL.

MMM will continue to evaluate alternative approaches in seeking the optimal numerical framework for a global version of WRF that will be well suited for future use in the regime of a “global cloud model.” MMM plans to interact with CGD to explore the opportunities to collaborate in designing a common modeling system that would meet the broader objectives for an earth-system model. The grid structure on the globe and numerical integration techniques are key areas that will require careful consideration before significant code development begins. In the coming year, MMM will explore the Yin-Yang grid (two overlapping grids the cover the sphere like panels on a tennis ball) to determine the feasibility of maintaining conservation and higher order numerics for this grid system. This work will require augmented software-engineering support to assist in coding test modules and evaluating alternative techniques in existing models.

The WRF modeling system will provide great value to the weather and climate research communities. It is a high-performance computational model that scales well from small single processor calculations to massive petascale applications. In addition to being a state-of–the-art weather forecast model, it has also been adapted to atmospheric chemistry and air quality modeling, and to nested regional climate applications. The successful migration of WRF into operational forecasting will provide significant economic and societal benefits. Maintaining and supporting WRF for widespread community use greatly leverages resources by allowing researchers access to a sophisticated modeling system without a large resource investment and the opportunity to contribute to and benefit from the advancement of a common modeling system.

The WRF project within MMM is supported by NCAR/NSF base funds through the MMM Division budget and USWRP resources, from the NSF Office of Polar Programs for Antarctic forecasting, and from outside funded projects with AFWA, KMA (Korea), and the FAA.

return to top

Data assimilation / ensembles

	2-day forecasts of the track of typhoon Haitang beginning from 00Z 16 July 2005. A 6-h assimilation window with the prototype WRF 4D-Var provides a forecast of landfall on Taiwan that is noticeably better than forecasts from operational analyses or from the WRF 3D-Var analysis. High resolution figure

Data assimilation is the process of combining observations and a previous forecast to provide a gridded estimate of the atmospheric state at a certain time. These estimates can then be used as initial conditions for subsequent forecasts or as tools to analyze and understand the atmosphere. While much progress has been made at global scales, data assimilation for scales of less than a few hundred kilometers (the “mesoscale,” where most severe and damaging weather occurs) remains a significant open problem in atmospheric science. Mesoscale data assimilation is especially challenging for two reasons. First, mesoscale motions are intimately coupled to complex physical processes such as those involving moisture, cloud and rain or interaction with the land or ocean surface. These processes are difficult to represent accurately in numerical models. They also lead to distinct and strongly nonlinear dynamics at the mesoscale, so that balances between mass and wind, which pertain at large scales in the atmosphere and underlie global data assimilation schemes, are questionable at the mesoscale. Second, observations that are plentiful (e.g., Doppler radar measurements of wind and reflectivity) involve only a subset of atmospheric variables, while observing platforms that measure all relevant variables (i.e., radiosondes) are sparse and resolve mesoscale motions poorly. To overcome these difficulties, there has been substantial effort within ESSL/MMM to advance mesoscale data assimilation.

One component of these efforts has been to build, and support for the community, a data assimilation facility for WRF based on variational assimilation techniques, known as WRF-Var. The WRF-Var system grew out of similar efforts for earlier versions of NCAR’s MM5 mesoscale model. At present, the community system is used at numerous universities in the U.S. and abroad and is the basis for operational assimilation system development at the Korea Meteorological Administration (KMA), the Air Force Weather Agency (AFWA), the Central Weather Bureau (CWB), the Beijing Meteorological Bureau (BMB), the National Center for Medium Range Weather Forecasting (NCMRWF), and the India Meteorological Department (IMD). The WRF-Var system supports the assimilation of a wide variety of both conventional and remotely sensed observations, including the direct assimilation of microwave radiances from the AMSU and SSMI instruments.

Over the past three years, a team within ESSL/MMM has worked to develop a four-dimensional variational (4DVar) capability within WRF-Var. Assimilation with 4DVar is the emerging standard for global numerical weather prediction, but has been applied operationally for a limited-area mesoscale model only at the Japanese Meteorological Agency. In 4DVar, one seeks the model solution that minimizes the weighted differences over a certain time interval between the solution and available observations. This minimization allows dynamical information from the forecast model to modify how the observations affect the state estimate in space and time and for unobserved variables. Experiments with a prototype 4DVar system for WRF-Var are underway and results are shown in the figure. In the next year, ESSL/MMM scientists plan to enhance the 4DVar capability system through the inclusion of additional physics in WRF’s linear and adjoint models (which are used in the minimization) and the addition of constraints that will limit spurious inertia-gravity waves in the analysis. The 4DVar system will be released to the community sometime before 2009 and is planned for operations at AFWA in 2008. Support for this work includes NSF, USWRP, AFWA, KMA, CAA, NASA, CRIEPI, BMB, and USAID.

return to top

USWRP/STEP

One hour and six hour forecasts of radar reflectivity with WRF, valid at 0100 UTC and 0600 UTC, using (ab) conventional data only, or (cd) also including radar assimilation, both after a three hour data assimilation cycle. The radar reflectivity observations from WSR-88D radar network at (e) 0100 UTC and (f) 0600 UTC 13 June 2002 are shown for verification; the color bar for the reflectivity fields is shown on the bottom of the figure; and, the simulated wind barbs (a full barb represents 5 m/s) are overlapped in (a)-(d). The addition of radar information clearly improves both one hour and six hour forecasts in this case. High resolution figure

The ShortTerm Explicit Prediction (STEP) Program is a newly established acrossLaboratory NCAR activity to improve the short-term forecasting of high-impact weather such as severe thunderstorms (heavy rain, tornados, downburst, flash flood, lightning and hail), winter storms (snow, freezing rain and drizzle), and hurricanes. It is motivated by the knowledge that improvements in shortterm forecasts of such weather can have significant societal and economic benefits, including: (1) reduced fatalities and injuries due to weather hazards; (2) reduced private, public, and industrial property damage; and (3) improved efficiency and savings for industry, transportation, and agriculture. It is also motivated by the fact that there has been significant advancement in a number of fields that are required to make progress in this area. These include the ability to observe the fourdimensional structure of the atmosphere, the development of new data assimilation techniques, such as 3DVar/4DVar and the Ensemble Kalman Filter (EnKF), to effectively use these new data sources, and the continuing development of numerical modeling systems, such as the Advanced Research Weather Research and Forecasting model (WRFARW), that can be run at fine enough grid resolutions to properly represent the physical processes critical to the production of such hazardous weather. The program includes research into basic understanding of high-impact weather systems, development of forecast techniques, realtime testing of forecast systems, verification, and interaction with users. It is a collaborative effort incorporating national and international scientists, engineers, and operational personnel from universities, government institutions and the private sector.

The primary activity for STEP this year was the development of a comprehensive research plan to help coordinate research efforts across the institution (see http://www.mmm.ucar.edu/events/STEP1WS/index.htm). However, a wide range of activities that are central to STEP research goals, and that involve MMM, RAL and EOL scientists, have been ongoing over the past year(s) (see the research plan on the website for a complete listing). Within MMM, WRFARW development efforts continue to be a critical component of STEP (see section 12-A ), offering a range of new capabilities, from improved numerics and physics to new data assimilation systems that can address the shortterm forecast problem. Experimental realtime explicit convective forecasts to 36h at 4 km grid resolution over most of the central and eastern U.S. (June 1 through July 31, 2006) continue to demonstrate enhanced capability at forecasting convective mode, propagation, precipitation episodes, and diurnal tendencies as compared to coarser resolution operational simulations using convective parameterizations. Sensitivity testing has also been completed for a variety of physics options (e.g., planetary boundary layer (PBL), microphysics, and land surface) for specific case studies, to help characterize particular capabilities and/or biases. For instance, a systematic bias was identified in the representation of boundary layer moisture with the YSU PBL scheme which led to significant reductions in convective instability 24 h into the forecasts. Work is ongoing to identify the source of this bias. An MMM/ASP colloquium on “The challenges of convective forecasting” was held from 1121 July 2006 to help motivate 24 PhD students from 16 universities (5 international) as to the wide range of research challenges associated with shortterm (036h) convective forecasting (see http://www.asp.ucar.edu/colloquium/2006/convectiveforecast/
index.jsp for more information). Analysis of the WRFARW realtime forecasts this upcoming year will emphasize a better characterization of the modes of forecast failure, and, in particular, the various sources and scales of errors in the model initial state and their impact on forecast accuracy. Comparisons between forecasts that were initialized at both 00 UTC (36h forecast) and 12 UTC (18h forecast) this year will also be completed, to consider the potential benefits and pitfalls of initializing suorecasts at differing points within the diurnal cycle.

One of the critical component the STEP program is the development of advanced data assimilation systems for WRF that can assimilate highresolution observations. Assimilation of radial velocity and reflectivity from multiple Doppler radars were explored, using WRF 3DVar and WRF EnKF assimilation systems and a 4DVar system of a cloudscale model (VDRAS), in a number of case studies. The objective is to improve shortterm forecasting of high-impact weather by initializing the numerical models at the cloudresolving resolution. From these studies it was found that the inclusion of highresolution radar observations significantly reduced the model spinup problem, hence improved forecasting skills in the short range, especially in the first 6 hours (e.g., see attached figure). In the coming year, the data assimilation effort will focus on the testing on multiple cases for each technique. The assimilation of radar refractivity will also be performed by some of the assimilation systems. Efforts at developing new techniques for verifying high-resolution forecast guidance have also progressed. During the past year, the first papers were published summarizing a newly developed object-based verification system. Development of the Method of Objectbased Diagnostic Evaluation (MODE) has continued through collaborations between MMM and RAL, extending to more sophisiticated techniques for matching forecast and observed features. At present, an intercomparison between several object based methods, involving collaborators outside NCAR, is underway and is based on precipitation and radar datasets collected during the NOAA/NSSL spring program in 2005. Future development of verification will focus on this intercomparison to understand the strengths and weaknesses of different approaches.

Other efforts planned for this year include further coordination and focusing of the program (a workshop is scheduled for 4 Oct 2006), continued planning for forecasting demonstration efforts for the Beijing Olympics for 2007/2008, and initial planning for a mesoscale forecast experiment associated with the DC3 field program in 2009 or 2010. Sponsors of this work include NSF/NCAR base, NSFUSWRP, FAA, NOAA/NWS, the Army, and the Beijing Meteorological Bureau.

return to top

Model physics

The Weather Research and Forecasting (WRF) Model is being used in an increasingly wider set of applications as computing power improves. WRF was developed as a community mesoscale model for numerical weather prediction, case study, and idealized simulations, and as a tool for related applications such as air quality research and forecasting. Some examples of newer applications that have resulted from improved computing resources are real-time cloud-resolving forecasting, including moving-nest hurricane forecasting, and nested regional climate modeling in the whole tropical belt (see other sections). With these applications come new priorities in physics development to enable better hurricane and regional climate modeling. These priorities fit with several of ESSL's priorities, including those of weather prediction and simulation across scales. Furthermore the aim of providing the university research community with a relevant up-to-date modeling system is met by continually updating the model to make use of the new capabilities in the current computing era, and improvements in model physics form one critical aspect of this development.

WRF already has a large set of physics options designed for its range of uses, from fast physics packages for operational uses, to more complex packages for scientific studies. The table shown summarizes the current WRF physics options available to the ARW dynamical core as of its last release (Version 2.1.2) in January 2006.

return to top