Bob: Thanks again for supplying me with the 6 h GFDL intensity forecasts that I requested a couple of months ago. Below are the results of the verifications of the GFDL and inland wind decay model (IWDM) forecasts that were performed based on the forecasts that you provided. I have also included a description of the methodology that was used to perform the model verifications. Some of these results have been included in a paper that Mark and I have recently completed (but have not yet submitted) describing an empirical decay model for the New England region of the U.S. which we defined as the are north of 37 deg. lat. In fact, much of the methodology section presented below was taken from that paper. Let me know if you have any questions. Thanks again, John P.S. The 1995 maximum wind values were multiplied by 0.85 to obtain the the surface wind at 10 m as you suggested when you originally provided me with the data. Methodology: Landfalling TCs from 1995-1997 were employed to evaluate the GFDL and IWDM model performance. The landfall intensity and landfall time required to run the IWDM for these cases were determined using forecasts from the operational Geophysical Fluid Dynamics Laboratory (GFDL) hurricane model. Table 1 shows the error statistics for landfalling tropical cyclones from 1995-1997 that NHC determined were of tropical storm or greater intensity at landfall. The error statistics depicted in Table 1 were computed based upon the last GFDL forecast prior to landfall provided that this forecast occurred within 12 h of the time of landfall. The decision to verify only those forecasts made within 12 h of landfall was made because FEMA recommends that only inland wind forecasts made at the last NHC advisory time prior to landfall should be employed in the decision making process of emergency management officials (FEMA 1995). As was stated previously, both the landfall time and intensity required to run the IWDM were determined based upon the GFDL model forecasts. The landfall time was defined as the time at which the TC crossed the coastline based upon linear interpolation of the 6 hourly GFDL forecast positions. The landfall intensity was defined as either the NHC operational maximum sustained 1-min surface wind (MSSW) at the initial forecast time or the last 6 hourly GFDL forecast MSSW prior to landfall whichever occurred last. These procedures were used to allow for the fairest possible comparison between the two models. The statistics shown in Table 1 were obtained by verifying the GFDL and IWDM predicted (MSSW) every 6 h against that shown in NHC HURDAT file. When performing the verifications shown in Table 1, errors were only computed for verification times when the system was tropical and when both the GFDL forecast position and the NHC HURDAT positions were overland. The mean elapsed time after landfall presented in Table 1 is the mean time that the TCs in this sample remained overland. Table 1 indicates that both the GFDL and IWDM did a good job of predicting the decay of the TC winds of the storms in this sample. The table also shows that the GFDL model had smaller errors within the first 24 h after landfall while the IWDM had smaller errors for the period from 24-72 h after landfall. Interestingly, both models tended to underpredict the MSSW during the forecast period immediately after landfall while overpredicting the wind later in the forecast period. A paired t-test (Miller and Freund 1977) was performed to determine if the differences in the absolute errors (AE) shown in Table 1 were significant at the 95% level. It should be noted that no attempt was made to account for serial correlation when performing the significance tests. The null hypothesis of the test was that the AE were equal, while the alternate hypothesis was that AE were not equal. The results of the paired t test indicated that these differences were not statistically significant except for the period from 60-72 h when the IWDM had AE that were statistically smaller than the GFDL model. However, since the errors in Table 1 were determined based upon only 7 landfalling TCs, additional forecast verifications are required before any final conclusions can be reached about the relative accuracy of these models. Nevertheless, the above evaluations are encouraging and suggest that both models can be used to make reasonably accurate overland TC wind forecasts. Table 1. GFDL and IWDM error verification for a homogeneous sample of forecasts made for 7 TCs that made landfall in the U.S. south of 37 deg. N latitude from 1995-1997. The variables in the table are the: 1) no. of forecast (N), 2) minimum (Tmin) and maximum (Tmax) elapsed time after landfall, 3) mean elapsed time (avgdt), 4) average change in MSSW (avgdv), 5) average error (E) (predicted- observed), 6) average absolute error (AE), 7) root-mean square error (RMSE), 8) variance explained (r**2). Note: Erin(1995) made landfall twice. Missing values are denoted by -99.0 GFDL model error verification: N Tmin(h) Tmax(h) Avgdt(h) Avgdv(h) Avge(kt) Avgabe(kt) Rmse(kt) r**2 ----------------------------------------------------------------------- 8 0 0 0 -99.0 -3.8 3.8 5.6 -99.0 15 0 12 5.5 -27.7 -2.3 5.9 7.5 0.88 11 12 24 17.7 -41.8 1.1 6.4 7.7 0.93 6 24 36 30.0 -57.5 7.8 7.8 8.7 0.42 6 36 48 42.0 -60.0 7.2 7.2 7.9 0.06 6 48 60 54.0 -60.8 6.7 7.0 8.2 0.23 6 60 72 66.0 -61.7 9.7 9.7 10.0 0.06 50 0 72 28.6 -46.3 3.3 7.0 8.2 0.88 IWDM model error verification: N Tmin(h) Tmax(h) Avgdt(h) Avgdv(h) Avge(kt) Avgabe(kt) Rmse(kt) r**2 ----------------------------------------------------------------------- 8 0 0 0 -99.0 -3.8 3.8 5.6 -99.0 15 0 12 5.5 -27.7 -2.7 8.2 9.3 0.82 11 12 24 17.7 -41.8 -2.4 8.7 12.8 0.81 6 24 36 30.0 -57.5 2.0 6.7 6.8 0.65 6 36 48 42.0 -60.0 2.2 6.2 6.6 0.34 6 48 60 54.0 -60.8 4.5 5.5 5.9 0.60 6 60 72 66.0 -61.7 5.5 5.3 5.8 0.68 50 0 72 28.6 -46.3 0.3 7.2 9.0 0.86