• Buizza, R., 2010: The value of a variable resolution approach to numerical weather prediction. Mon. Wea. Rev., 138, 10261042.

All Time Past Year Past 30 Days
Abstract Views 0 0 0
Full Text Views 11 11 0
PDF Downloads 1 1 0

Comments on “The Value of a Variable Resolution Approach to Numerical Weather Prediction”

View More View Less
  • 1 Meteorological Developmental Laboratory, Office of Science and Technology, NOAA/National Weather Service, Silver Spring, Maryland
© Get Permissions
Full access

Corresponding author address: Bob Glahn, Meteorological Development Laboratory, Office of Science and Technology, NOAA/National Weather Service, 1325 East-West Highway, Silver Spring, MD 20910. E-mail: harry.glahn@noaa.gov

The original article that was the subject of this comment/reply can be found at http://journals.ametsoc.org/doi/abs/10.1175/2009MWR3077.1.

Corresponding author address: Bob Glahn, Meteorological Development Laboratory, Office of Science and Technology, NOAA/National Weather Service, 1325 East-West Highway, Silver Spring, MD 20910. E-mail: harry.glahn@noaa.gov

The original article that was the subject of this comment/reply can be found at http://journals.ametsoc.org/doi/abs/10.1175/2009MWR3077.1.

Buizza (2010) presents a tremendous amount of information relating to the question of the relative accuracy of a model run at different horizontal resolutions. Specifically, the model run by the European Centre for Medium-Range Weather Forecasts (ECMWF) was run at various resolutions out to 8 days [this model was operational at ECMWF between 5 June and 6 November 2007 (Buizza)]. Full details of these integrations are contained in Buizza, including the initializations. Whereas the ECMWF runs the model out to 15 days with 51 members, only 5 members were used, and the integrations only went to 8 days because of the very large amount of computer time required (Buizza, p. 1027).

As stated by Buizza:

“The key question that is addressed in this work is which of two comparable-cost configurations gives the best medium-range forecasts: a constant resolution configuration or one with a variable resolution, higher in the earlier forecast range and lower afterward?”

Six model configurations were used. Accuracy and skill were assessed by a number of scores over the Northern Hemisphere for a winter season (1 December 2007–28 February 2008) based on two assumptions of “ground truth.” One of the model configurations was a high T799 resolution that was used as ground truth, a method Buizza calls the idealized model error (IME) approach. The other ground truth was ECMWF analyzes, which Buizza calls “the realistic scenario.” I will focus on only two of the configurations used for comparison, the two that Buizza uses for his major conclusion.

One was a T319 model run out to 8 days, and the other was a higher-resolution T399 model run to 3 days, then extended to 8 days with a T255 model (called VAR3 by Buizza). The conclusion reached was the following:

“Results have indicated that VAR3 forecasts are more accurate than T319 forecasts for the whole forecast range. VAR3 forecasts are definitely more accurate in the IME case, when forecasts are verified against T799 control forecasts, but less so when forecasts are verified against ECMWF T799 analyses.”

Buizza states the IME results are statistically significant to the 5% level for up to day 8, but the differences in the 2 configurations were not significant at that level when compared to ECMWF analyses. The presented results of the forecasts out to 8 days are evidence of this conclusion. However, the computer time required for VAR3 is about 25% more than is required by the constant T319 model for an 8-day integration (Buizza’s Fig. 1 and the statement on p. 1027). This seems to violate the “comparable cost” criterion stated in the key question being addressed. In other words, this comparison does not seem to be of comparable cost, and therefore cannot answer the question posed. Buizza states that the computer times required for the configurations to be run to 15 days would be very comparable, VAR3 requiring 2% less than the T319. However, no results are presented for the 15-day runs. While the forecasts to 8 days may be better for VAR3, the lower-resolution T255 would be run to 15 days, and the forecasts might deteriorate so that the VAR3 would produce worse results than the T319 for the latter part of the 8–15-day period. In fact, because the differences were rather small in the realistic case at 8 days, one might really expect that. Because the time for running to 15 days is so close for the 2 configurations, it is tempting to believe that the experiment was designed to be run to 15 days, but the computer time required did not allow it.

Model resolution is a very important issue, especially when a compromise must be made with a number of members of an ensemble, vertical resolution, physics packages, etc., for an operational model suite to fit into the available computer time. Buizza has addressed one aspect of this issue. Unfortunately, his results do not answer the key question.

REFERENCE

Buizza, R., 2010: The value of a variable resolution approach to numerical weather prediction. Mon. Wea. Rev., 138, 10261042.

Save