Beth Ebert

Group: Weather and Environmental Prediction

Beth Ebert
Beth Ebert

Contact Details:

  • Fax: 9669 4660
  • Phone: 03 9669 4688
  • Address:
    Bureau of Meteorology
    GPO Box 1289
    Melbourne, 3001

Personal Data:


  • B.S. Atmospheric Sciences, University of California, 1981
  • M.S. Meteorology, University of Wisconsin, Madison, 1984
  • Ph.D. Meteorology, University of Wisconsin, Madison, 1987

Interests and Research:

Validation of satellite precipitation estimates:

Rainfall products from operational satellite precipitation algorithms are easily obtainable via the web or FTP, and are being used for many diverse meteorological, climate, hydrological, agricultural, and other applications. It is therefore important to have an idea of their accuracy and expected error characteristics. The Australian gridded rain gauge analysis is being used to intercompare and validate several operational and semi-operational satellite precipitation algorithms on daily and longer time scales. These include GPCP products, as well as several algorithms from NASA GSFC, NOAA NESDIS, NOAA CPC, Naval Research Laboratory, UC Irvine, and U. Bristol. A few NWP models are included for comparison. The validation results are updated on a daily basis, and results are displayed on the SatRainVal web site. A paper by Ebert et al. 2007 describes the performance of satellite precipitation estimates over Australia, the United States, and Western Europe.

 Verification of quantitative precipitation forecasts:

In order to improve the accuracy of quantitative precipitation forecasts (QPFs) from the Bureau's numerical weather prediction (NWP) models, it is first necessary to understand the strengths and weaknesses of the models. QPFs from the Bureau's regional and global NWP models, as well as NWP models from several operational centres overseas, are verified on a daily basis against the operational rainfall analyses. Results show that the models have good skill in the southern part of Australia, but less skill in the tropical north. As part of the QPF verification process, I developed an interactive verification tool called RAINVAL that produces maps, time series, and statistics to show the skill of the QPFs when compared to the rainfall analysis and to each other. It is currently used to monitor the daily rainfall forecasts over Australia from a large number of NWP models. It is also being used in the Bureau to assess new versions of the regional and global models, as well as rainfall output from overseas models.

One approach to verifying QPFs is to focus on individual weather systems (contiguous rain areas, or CRAs) as opposed to point-by-point domain rainfall verifications. The QPF error for an individual rain system has components due to incorrect location, incorrect magnitude, and incorrect pattern (shape). By using pattern translation and minimisation of the squared difference between forecast and observed rainfall, the displacement of the forecast can be determined. Overlaying the two patterns, the remaining error can partitioned into contributions from magnitude and shape errors. This verification strategy can reveal systematic errors in NWP QPFs, which in turn can be related to weaknesses in model dynamics, rainfall parameterisations, surface topography, etc. The forecast for the rain event inself can be classified as a "hit", "miss", etc., using this technique. Details may be found in Ebert and McBride (2000). The IDL code can be downloaded by clicking here.

Neighborhood (a.k.a. fuzzy) verification approaches reward closeness by relaxing the requirement for exact matches between forecasts and observations. The key to the neighborhood approach is the use of a spatial window or neighborhood surrounding the forecast and/or observed points. The treatment of the points within the window may include averaging (upscaling), thresholding, or generation of a PDF, allowing a variety of continuous, categorical, and probabilistic verification metrics to be employed. The size of this neighborhood is varied to provide verification results at multiple scales, thus allowing the user to determine at which scales the forecast has useful skill. Other windows could be included to represent closeness in time, closeness in intensity, and/or closeness in some other important aspect. Ebert (2008) describes a framework for neighborhood verification that incorporates several neighborhood verification methods appearing in the literature over recent years. The IDL code can be downloaded by clicking here.

The WWRP has held Forecast Demonstration Projects during the Sydney 2000 and Beijing 2008 Olympics games to demonstrate the skill and utility of advanced nowcasting systems. In Sydney we found that the systems indeed showed significant skill, but by and large did not perform better than simple extrapolation (Ebert et al., 2004). For the Beijing FDP I developed a real-time nowcast verification system that was used by forecasters and scientists. In additional to the "usual" visual and statistical verification, some of the advanced verification techniques that have been developed in recent years were also included.

I am a member of the WWRP/WGNE Joint Working Group on Forecast Verification Research established in 2003. One of our activities is to maintain a Forecast Verification Web Page that describes both the standard and the newer diagnostic verification methods. It addresses various issues in verification such as grid box vs. point verification, confidence intervals on the verification results, pooling vs. stratifying results, etc., and has a section for FAQ. It also has a few downloadable verification datasets that can be used to test verification methods.

 Poor man's ensemble:

We run a poor man's ensemble (PME) forecast for rainfall, in which QPFs from several operational NWP models are combined to give deterministic and probabilistic rainfall forecasts. This approach is cheap and efficient, and gives deterministic forecasts that are more accurate, on average, than any one of the component models. In particular, the location of the rain system is much much improved using the poor man's ensemble. It also gives useful probabilistic forecasts out to several days. In fact, the poor man's ensemble had greater probabilistic skill than the 51-member ECMWF Ensemble Prediction System out to 48 h. The details of this study are given in Ebert (2001, 2002). Forecasts from the PME can be viewed by clicking here.

Positions and Committees:

  • World Weather Research Program (WWRP) Joint Scientific Committee, 2011-present
  • Australian National Fire Danger Rating Science Sub-group, 2010-present
  • WGNE/WGCM Climate Metrics Panel, 2009-present
  • AMS Committee on Probability and Statistics, 2007-2009
  • WWRP Beijing 2008 FDP Team, 2005-2009
  • GIFS-TIGGE Working Group, 2005-2011
  • GPM GV Steering Committee, 2003-2010
  • WWRP/WGNE Joint Working Group on Verification (JWGV), 2003-present; co-chair 2008-present
  • International Precipitation Working Group (IPWG), 2002-present
  • WWRP Sydney 2000 FDP Verification Team, 2001-2002
  • Science Advisory Team, Global Precipitation Climatology Project (GPCP), 1998-2001
  • Atmospheric and Oceanic Sciences Committee of the Australian Academy of Science, 1997-2000

Publications (1999-current):

Please note that the Disclaimers for the Bureau of Meteorology and CSIRO apply also to this site.
index.php last modified Monday, 02nd May, 2011 @ 01:29UTC | Site Map | | © 2014