RIRW Verification

RIRW Verification: Model Version Comparison

• The probability of detection (POD) and false alarm rate (FAR) are 7% and 69% for H215 forecasts. Relaxing the intensity threshold to 25 kt in 24-h improves the POD to 14% and FAR to 60%.

RI Contingency tables

30kt in 24-h

OBS
Yes No
FCST Yes 81 1084
No 183 28605

25kt in 24-h

OBS
Yes No
FCST Yes 262 1588
No 389 27714









• The distributions of the difference of change in a-deck values in the last 24-h and change in b-deck values in the last 24-h (ABDEL) are examined and for H215 forecasts. The median of the hits (green boxes) are within the observation error (±5).



Click for larger image.

• A homogeneous comparison of HWRF-14 and H215 forecasts indicates HWRF-14 and H215 forecast hits (green) are closer to the zero line, indicating that the ABDEL is within the observation error. The false alarms (red boxes) for both the versions indicate that HWRF tends to over-predict the intensity chage by 15-20 kt compared to the observed change. However, when RI events are missed (Orange boxes), the intensity change is underestimated by 20-30 kt.



Click for larger image.

• A homogeneous comparsion indicates that the POD (FAR) of RI events for HWRF-14 forecasts are 4% (77%) compared to 7% (70%) for H215. Relaxing the intensity change threshold to 25 kt in 24-h increases the POD of HWRF-14 (H215) to 10% (14%).

H215 RI Contingency tables

H215 30kt in 24-h

OBS
Yes No
FCST Yes 72 169
No 973 27006

H215 25kt in 24-h

OBS
Yes No
FCST Yes 230 360
No 1421 26209










H214 RI Contingency tables

H214 30kt in 24-h

OBS
Yes No
FCST Yes 42 138
No 1003 27037

H214 25kt in 24-h

OBS
Yes No
FCST Yes 166 341
No 1485 26228