Geoscience Reference
In-Depth Information
exhibit moderate probabilities of detection (0.4 to 0.7) with very low false
alarm rates (0.1-0.2) mainly for forecast probability thresholds of 10-40%
(Fig. 2a). Although the false alarm rate never exceeds 0.2 even at extended
lead-times, VarEPS forecasts made over seven days in advance tend to have
low POD once forecast probability thresholds increase beyond 20%. If a 96-hr
time window is used, the probability of detection of the VarEPS on average
increases by about 0.2 and the false alarm rate by about 0.1 for forecasts of TC
genesis made less than seven days in advance (Fig. 2b). Forecasts greater than
seven days in advance benefit even more from the increase in time window, as
the POD for a ten-day lead-time forecast increases on an average by about
0.15, and the FAR increases by only 0.05 for a forecast probability threshold
of 10-30%.
We now evaluate how well the VarEPS forecasts perform for TC track
prior to genesis. The average ensemble mean track error at a lead-time of 24
hrs is 69 n mi with a 50% interval of 42-86 n mi, and at 120 hrs it is 241 n mi
with a 50% interval of 116-324 n mi for all pre-genesis VarEPS forecasts
during the period 2007-2010 (Fig. 3). From a lead-time of 24 hrs to 240 hrs,
the mean ensemble track error growth is nearly linear at 41 n mi per day, so
that by a lead-time of 240 hrs, the total mean ensemble error is 409 n mi with
a 50% interval of 184-504 n mi. In addition, the ensemble track error distribution
becomes increasingly non-Gaussian as forecast lead-times increases. The
implication is that beyond 72 hrs, the mean ensemble track error grows larger
than the maximum likelihood of the pre-genesis track error distribution. To
place these pre-genesis track errors in perspective, the Indian Meteorological
Department's post-genesis track errors at a lead-time of 48 hrs and 72 hrs
typically average 162 n mi and 270 n mi, respectively. This is equivalent to the
VarEPS ensemble mean
pre-genesis
forecasts at lead-times of 90 and 138 hours,
respectively.
3.2 Post-genesis Tropical Cyclone Forecasts
Figure 4a compares the VarEPS control and ensemble mean post-genesis
forecasts with other forecasting agencies including the JTWC, the U.S. Navy's
version of the GFDL (GFDN), the United Kingdom Meteorological Office's
global model (UKMET), the National Center for Environmental Prediction's
Global Forecast System model (GFS) and the U.S. Navy's NOGAPS model.
Since these forecasts were obtained through the U.S. Navy's Automated Tropical
Cyclone Forecasting System, most of the forecast guidance is limited to 72
hrs. Although the VarEPS control and ensemble mean forecasts on an average
begin with the largest initial track error, 12 hrs later and beyond the VarEPS
control and ensemble mean exhibit the lowest track errors among all other
model forecasts. In addition, the VarEPS control and ensemble mean on an
average exhibit slightly lower track errors than the JTWC through a lead-time
of 72 hrs. Relative to the next best performing forecast model, the VarEPS
Search WWH ::
Custom Search