Real-Time Four-Dimensional Data Assimilation (RTFDDA) and Forecasting Advances


Many applications, including military tests and operations, renewable energy assessment and prediction, power grid weather safety, weather-related emergency response etc., desire rapid-update precision weather information for local areas of interests. RTFDDA (Real-time Four-Dimensional Data Assimilation and forecasting system) is a mesoscale numerical weather modeling system that has been developed to meet the critical weather needs of these regional and local applications. RTFDDA is built upon the WRF (Weather Research and Forecasting) model and is designed to effectively and efficiently assimilate diverse available weather observations into WRF and provide best possible weather information service to a target application. An important feature of RTFDDA is that it allows for smooth and uninterrupted assimilation of diverse weather observations and produces physically consistent and dynamically balanced 4D weather analyses and “cloud/precipitation “spun-up” predictions. In the past 17 years, RTFDDA has been applied to over 50 weather-critical applications across the US and the international regions. The applications include military testing and operations, regional operational NWP (numerical weather prediction), dispersion and transportation emergency response, urban meteorology, energy, water resource and flood prediction, etc.   

Figure 1. RTFDDA and its extension for regional microclimatology dynamical downscaling, ensemble prediction and LES/VLES NWP of microscale flows.
Figure 1. RTFDDA and its extension for regional microclimatology dynamical downscaling, ensemble prediction and LES/VLES NWP of microscale flows.

RTFDDA is continuously improved with respect to its data assimilation schemes, new data source, dynamical configuration and physical parameterizations to advance the RTFDDA system itself and improve its accuracy and capabilities. In the last 10 years, RTFDDA has evolved from a single mesoscale deterministic analysis and forecasting system to a modeling capability suite that integrates ensemble prediction technology (Ensemble-RTFDDA), regional climate downscaling with four-dimensional data assimilation (Climate-FDDA) and microscale NWP with a refined LES model grid at 10s to 100s meters of grid intervals (RTFDDA-LES) (Fig. 1). As a result, RTFDDA is now capable of producing customized high-resolution and ultra-high-resolution precision weather analysis and forecasts, probabilistic weather forecasts, and multi-year/multi-decadal microclimatology analyses for a given target region. In addition, the WRF-Chem model has been assessed and is being added to RTFDDA for forecasting sand and dust storms.     

RTFDDA currently integrates the following data assimilation technologies: Newtonian relaxation based “observation-nudging” and “analysis-nudging” FDDA schemes, the community WRFDA, GSI, DART-EnKF data assimilation modules, a four-dimensional relaxation ensemble Kalman filter (4D-REKF) FDDA scheme and a hydrometeor-latent-heat-nudging (HLHN) radar data assimilation schemes. The technologies are configured to formulate hybrid data assimilation to provide optimal modeling according to the application requirements.  


Advances in WRF FDDA

In order to assimilate increasing number and types of diverse observations and support the 4-dimensional relaxation ensemble Kalman filter data assimilation scheme, WRF model is modified to refactor the observation-nudging data structures. The new data structure significantly improves the computing efficiency and memory usage of the FDDA scheme and avoids the long-existing data-handling bottleneck of the legacy FDDA data structure. Meanwhile, this data structure is more compatible with those employed by the other NCAR data assimilation tools such as DART and GSI, which eases the hybrid deployment of different data assimilation schemes. This new data structure is critical for conducting Doppler radar radial velocity data assimilation with the advanced 4D-REKF technologies.  

Weather radars detects detailed hydrometeor information inside convection systems. Radar data assimilation (RDA) has been an important measure to improve mesoscale numerical predictions of precipitation systems and the accompanying severe weather phenomena. A significant effort has been put into refining the RTFDDA-RDA scheme. Studies have been carried out to assess the impact of two key experimental parameters, the time windows and weights and the nudging coefficients, in the hydrometeor and latent heat nudging (HLHN) scheme, the core engine of RTFDDA RDA. This research greatly improved our understanding of the complex digesting processes of the RDA innovations and led to a new design of different latent heat controls for stratiform and convective clouds. This work dramatically improves the RTFDDA-RDA performance for 0 – 6 hr convection forecast.

Another major advance with RTFDDA is a development of a lightning data assimilation (LDA) capability. Total lightning strikes are highly correlated with the total graupel mass in convection systems. An algorithm was developed to reconstruct graupel amount (specific mixing ratios) for three-dimensional model grid points based on total lightning observations. The reconstructed graupel was then assimilated into WRF based on the RTFDDA HLHN scheme. Case studies with severe convection events occurred in both the central plains, US and a southern China region demonstrated that RTFDDA-LDA is capable of significantly improving 0 – 3 hr lightning and precipitation forecast of convective storms.    

Figure 2. Verification of RTFDDA-RDA forecasts of a squall line event in Guangdong Province, China. Left panels: observed radar reflectivity; Middle: RTFDDA 1 and 3 hour forecasts without RDA; and Right: RTFDDA-RDA 1 and 3 hour forecasts.
Figure 2. Verification of RTFDDA-RDA forecasts of a squall line event in Guangdong Province, China. Left panels: observed radar reflectivity; Middle: RTFDDA 1 and 3 hour forecasts without RDA; and Right: RTFDDA-RDA 1 and 3 hour forecasts.  

Observation quality control (QC) and data selection algorithms are a vital component of the RTFDDA model system. An online data QC scheme is developed for RTFDDA. The scheme not only detects bad observations, but also adequately scales representativeness errors of observations caused by either improper siting of measuring instruments or limited resolvable weather for model grids. The scheme also takes account of observation-model inconsistency due to the differences between model terrain heights and observation station elevations, which allows assimilation of more surface observations into RTFDDA. Another work that have been completed in this year is related the fact that locally excessively dense observations can harm a modeling system. A data-thinning module was developed that can be employed to thin observation density according to given model grid sizes. Finally, an adaptive influence radius scheme is developed that permits shorter influence distance for dense observation areas and longer for the other.   

Sand and dust forecasting with WRF-Chem

Dust forecast capabilities have been developed with RTFDDA based on the fully coupled WRF-Chem. RTFDDA data assimilation of weather data is able to provide accurate weather environment for modeling dust emission and transportation in WRF-Chem and effectively improves the simulation of dust spin-up processes and thus the forecast accuracy. RTFDDA-dust forecasting systems have been deployed for the Mid-East countries including Saudi Arabia and Israel.


Research and development efforts will be carried out to further improve all major system components of RTFDDA-RDA. In particular, the new community WRF achievements, and the nudging-based FDDA scheme shall be merged and integrated for new deployments. All operational RTFDDA systems will be upgraded to WRF Version 3.9.1 in FY2018.  A plan has been set up to develop a RTFDDA framework based on the MPAS model.      

The main research on the RTFDDA data assimilation scheme will focus on the RTFDDA-GSI-HLHN hybrid radar data assimilation system, 4D-REKF four-dimensional data assimilation scheme for radar data assimilation, sand and dust simulation and prediction capability, and statistical model output processing techniques. For radar data assimilation, GSI and 4D-REKF data assimilation schemes will be assessed and compared for assimilation of Doppler radar radial velocities. A sensitivity study will be conducted with the assimilation of radar reflectivity measurements with HLHN. Specifically, the impact of data frequencies between six minutes and one hour, and more accurate latent heat derivation from the radar reflectivity will be studied. RTFDDA-RDA will be evaluated for 0 – 6h severe convection forecast radar reflectivity data assimilation scheme based on a high-resolution (2.5km grid) RTFDDA system that will span the contiguous US (CONUS) domain. In addition, algorithms for assimilating polarimetric radar products in combination of lightning measurements will be developed for RTFDDA-RDA. 4D-REKF is an advanced FDDA capability that combines the advantages of Newtonian relaxation based “observation nudging” and the advanced ensemble Kalman filter. 4D-REKF using flow-dependent data assimilation weights generated with dynamical ensemble forecasts and historical forecast analogs will also be studied.