|
|
||
|
|
Thomas Hopson
Tom Hopson's work over the last 8 months focused on examining how well dynamically-generated ensemble weather forecasts can forecast their own potential uncertainty. In particular, he examined the relationship between ensemble spread and forecast skill: what verification measures currently exist to test this relationship, how well do they work and where do they fail, and what other alternatives are there? He also studied how well a statistical heteroscedastic error model (i.e. error that depends on the magnitude of the forecast variable) derived solely from hindcasts compares with dynamically-generated ensembles in forecasting "forecasting uncertainty". In particular, he showed how the real test of the utility of a dynamical ensemble forecast is that it should at least be able to beat this far less expensive simple statistical error model. The other part of his work dealt with operational flood forecasts of the Brahmaputra and Ganges Rivers as they flow into Bangladesh, at forecast lead-times of 1-day out to 6-months. He is part of a USAID-sponsored project that provides this information near-real-time to the country of Bangladesh as part of their nationwide flood forecasting efforts. These forecasts went "operational" in May (see Figure 1 and Figure 2 below; and also http://cfab.eas.gatech.edu/shortterm). In addition to producing these forecasts, as part of this work, he also developed a new statistical technique to improve the skill and reliability of both the driving ensemble weather forecasts from the European Centre for Medium-Range Weather Forecasts, and the ensemble river discharge forecasts.
Figure 1: Brahmaputra River operational 7- to 10-day ensemble discharge forecasts for May 1 to October 6, 2006. The horizontal black dashed line is the "critical" flood discharge for this river, and the black dotted line is the observed discharge for this period.
Figure 2: Using the ensemble discharge forecasts from Figure 1, confidence bounds on the forecasts can be generated. Shown here are 50% (yellow) and 95% (red) confidence bounds.
Funding Sources This research is supported by the National Science Foundation.
|
|
