HWRF HNR2/HR20 Forecast Comparison

HNR2/HR20 Forecast Comparison Executive Summary

    • The DTC conducted its second extensive test of a HWRF configuration (HD33), demonstrating that a robust testing environment, functionally-similar to EMC’s, is available.

    • Over 600 HWRF runs for the Eastern North Pacific and North Atlantic basins for the 2010 season were conducted in order to establish a benchmark of the community code (HD33) and to compare the forecasts against a counterpart set produced at EMC (H21A).

    • Track errors for HD33 increase linearly with time from near zero at initialization time to 280 nm at the 5-day forecast in both basins.

    • Absolute intensity errors increase sharply in the first 6-h of forecast and then grow slowly out to 3-days, after which they remain virtually unchanged.

    • A negative intensity bias is noted for HD33 in the Pacific basin after the second day of forecasting, while in the Atlantic there is no statistically significant bias.

    • The HD33 forecast storm size is larger than the observed one, and continuously grows in size, for the 34-, 50-, and 64-kt wind radii in the Atlantic basin. In the Eastern North Pacific, the forecast size is over predicted at the onset, but decreases with forecast lead time.

    • The worst track and absolute intensity forecasts (outliers) were identified so that forecast improvements for these poorly performing cases can be addressed in the future.

    • While an exact match between the HD33 and H21A forecasts was not expected due to differences in computational platform and a few other minor setup differences noted in Section 3, a large number of statistically significant differences in track, intensity, and structure were found between the two sets.

    • Diagnostic investigations conducted after the test revealed that the differences were caused by a coding error in the convective parameterization. This bug behaved differently in different computational platforms. After correcting this bug, a small sample of forecasts was rerun and indicated that HD33 and H21A results were much closer.

    • Model output files have been archived and are available to the community for future studies. Forecast maps and verification graphics, along with this report and additional information are available in this website.