advanced

Estimation of PBL Structure Using a Simple Model, Surface Observations, and an Ensemble Filter

by Joshua Hacker, RAL

This work is aimed at determining the potential for surface observations to provide information about the overlying atmosphere, particularly in the planetary boundary layer (PBL).  Surface observations have proven difficult to assimilate in the past.  But with mesoscale data assimilation and forecasting applications becoming common, surface observations offer a dense, robust, and inexpensive source of data that fills gaps not accurately observed by the synoptic balloon network or satellites.  An hypothesis is tested: Surface observations are assimilated more effectively via an ensemble filter than other assimilation approaches because the filter takes advantage of flow-dependent covariance information and does not impose additional dynamic balance constraints.

In collaboration with D. Rostkier-Edelstein (Israel Institute for Biological Research), a simple column model and an ensemble filter are used to quantify the information available in surface observations over the Southern Great Plains (SGP).   The model was originally developed by M. Pagowski (NOAA/ESRL), and contains the suite of soil, surface layer, and PBL parameterization schemes available in the Weather Research and Forecast (WRF) model.  RAL scientists completed the work necessary to initialize and force the column from a variety of sources, and couple the model to the Data Assimilation Research Testbed (DART, developed in IMAGE) software.  External forcing, including radiative fluxes at the earth’s surface and background geostrophic winds, is taken from the real-time WRF forecasts run in ESSL/MMM.  In these experiments, ensembles of column model realizations are forced by random samples of the 3D WRF forecasts so that no information about specific flow scenarios exists in the background ensemble.  Data assimilation with the ensemble filter is done in the DART framework.  Surface observations taken at the Atmospheric Radiation Measurement (ARM) Central Facility near Lamont, OK, are assimilated hourly in independent runs spanning the period May-July 2003. 

Verification against independent rawinsonde profiles shows that the surface observations provide substantial information about the state of the overlying atmosphere (Fig. 1).  Temperature, water vapor mixing ratio, and winds in the PBL can be specified with reasonable accuracy from only surface observations, without the benefit of any relevant 3D background information.  The depth of influence of the surface observations depends on the local time of day, with the greatest error reduction occurring coincident with the convective daytime PBL when the connection between the surface and the PBL state is strongest.  But the recursive nature of the ensemble filter allows for accuracy through a deep layer well into the night, despite a weaker nighttime impact of the surface observations.

These results are significant because they represent the potential to provide additional information about the state of the PBL in regions that are otherwise sparsely observed.  The additional information can be used in a 3-D mesoscale assimilation system, directly by forecasters, or as input to transport and dispersion models that can make use of multiple profiles.  Successful assimilation into this simple model suggests that 3-D ensemble assimilation approaches may prove effective in the PBL, where other approaches are often inadequate.  Forecasters concerned with, for example, nowcasting storm initiation can benefit from a “virtual profile” produced at every surface observing station, significantly increasing temporal and spatial data density compared to the rawinsonde and profiler networks.  Finally, transport and dispersion models may benefit from multiple profiles specifying the meteorology in the PBL, as opposed to one or none typically available in the model domain.  In all cases, uncertainty information is available directly from the ensemble of analyses.

In these experiments, surface observations were surprisingly effective at improving state estimates of the PBL.  The potential for application development, using ensemble assimilation in a column, has raised the priority level of this work.  RAL will continue to pursue research on fundamental issues such as the role of errors in the model and estimation model parameters such as soil moisture availability. The model forcing component will also be improved to include background information relevant to a current flow scenario rather than from a climatology.  Engineering effort will be required to make the column and assimilation system rapidly deployable to arbitrary locations worldwide.  This work has been funded by the Department of Defense and the National Science Foundation.  Additional funding from DOD sponsors is being sought to bring this concept to the level of a useful application.

Director's Message | Contents