GIBBS Winds

Using a Bayesian Hierarchical Model (BHM), Tim Hoar has generated "Gibbs Winds" for the Tropics in the Indian and Western Pacific Oceans: 24° S - 24° N, 52° - 180° E, December 25, 1999 to February 29, 2003. For more details, see Tim's BHM web pages.

In summary, the model is organized as follows:

Data Stage Distributions:

• QSCAT surface winds; 25km resolution in 1600 km-wide swaths (8° gaps at Eq.)
• NCEP-FNL; analysis winds on regular 1° grid, 4-times daily, deficient power spectral density (PSD) at scales shorter than 500 km.

• Propagating large-scale modes of the equatorial &beta -plane; Wheeler and Kiladis (1999).
• Nested wavelet bases &rarr power-law relation, PSD ~ k -5/3 ; Wikle et al (1999).

• 0.5° res; 23.5° N - 23.5° S, 60° E - 180 ° at 00,06,12,18 UTC, Dec 1999 - Mar 2003.
• 50-member ensemble of independent surface wind field realizations selected from the Posterior.

Here are a few plots to illustrate how the model works.

Fig. 1: Probabilty Distribution

o emphasize the probability distribution of surface winds, this graphic shows a 7 day time series of the histograms of zonal velocity at a single point in the domain (starting 5 January 2002, near 150° E on the equator). Each day, red distributions correspond to 0600 UTC, gold to 1200 UTC, green to 1800 UTC, and blue to midnight UTC. The local satellite overpasses are nearest the gold and blue distributions. The uncertainties are lowest then, so the histograms are most peaked around single values then (see the range and means projected on the "roof" of the diagram). Data for the intervening times are taken from the NCEP analyses and they are more uncertain on the small spatial scales of our grid. This is reflected in broader distributions in the zonal wind histograms that are red and green. Realize now that there are histogram time series like these for each grid point, and for the meridional wind component as well.

(Click on the image for an enlarged plot in a pop-up window, .. then click on another image to switch to that plot in the pop-up window.)

Fig. 2: Two Members of the 50-member set for Nov 10, 2000, 00hr

Shown are members no. 0 and 9, for U, Divergence, and Standard Deviation of all members for U. (.. you might need to enlarge your display window to see U, Div, and St.Dev. in one row!). The color bar for St.Dev. has been chosen so that the QSCAT data swaths can easily be identified by low uncertainty (blue) in regions of ascending or descending satellite swaths.

Fig. 3: Mean of all 50 members on Nov 10, 2000, for 00hr, 06hr, 12hr, and 18hr.

Shown are Means of U, Divergence, and Standard Deviation of U (.. you might need to enlarge your display window to see U, Div, and St.Dev. in one row!). The color bar for St.Dev. has been chosen to emphasize low uncertainty (blue) in regions of QSCAT coverage (ascending or descending swaths).