Close window
Close window
This section of the new site isn't ready yet. We've brought you back to the current site.

Tropical cyclone forecast verification - northern hemisphere 2017

1. Introduction

A summary of tropical cyclone activity in the Northern Hemisphere for the 2017 season is presented below together with an assessment of the performance of the Met Office global model in predicting the track and intensity of these tropical cyclones.

Tropical cyclones are experienced in the North Pacific, North Atlantic and North Indian Oceans and nearby tropical seas. For the purpose of tropical cyclone verification the northern hemisphere is divided in to four basins; the North-West Pacific (west of 180° E to the Malay Peninsula), the North-East Pacific (east of 180° E), the North Atlantic and the North Indian Ocean (west of the Malay Peninsula). Mean error statistics for each basin are presented together with a table of statistics for the whole Northern Hemisphere. Verification is performed at 12-hour intervals up to forecast time T+168, although statistics are only presented at 24-hour intervals in this report. The global model resolution in operation was 0.234375° x 0.15625° x 70 levels up to 11 July 2017. This is equivalent to a horizontal resolution of 26 km × 17 km at the equator. Thereafter the resolution was 0.140625° x 0.09375° x 70 levels. This is equivalent to a horizontal resolution of 16 km × 10 km at the equator.

Tropical cyclone forecast verification - measures of error

Advisory positions received in real time from RSMC Tokyo, JTWC Hawaii, NHC Miami and CPHC Honolulu are used as verifying observations of storm location. Best track data from these centres will be obtained once they become available. Past experience shows that use of best track rather than real time data usually only makes minor differences to seasonal error statistics. Some mean error statistics for last season are also included for the purposes of a comparison. Forecast tracks are only verified when a depression reaches tropical storm status.

2. Tropical cyclone activity

Tropical cyclone activity
NWP NEP NAT NI TOTAL
Tropical depressions (<34 knots) 7 (7) 2 (1) 1 (1) 0 (0) 10 (9)
Tropical storms (34-63 knots) 14 (11) 9 (8) 7 (8) 2 (4) 32 (31)
Hurricanes/typhoons (>63 knots) 12 (13) 9 (13) 10 (7) 2 (1) 33 (34)
Total 33 (31) 20 (22) 18 (16)  4 (5) 75 (74)

Categories assigned based on 1-minute averaged winds

Basin name abbreviations:
NWP : North-west Pacific (west of 180° E)
NEP : North-east Pacific (east of 180° E)
NAT : North Atlantic
NI : North Indian (west of Malay Peninsula)

The number in brackets indicates the figure for the 2016 season.

3. Summary of all northern hemisphere storms

3.1 North-west Pacific basin storms

Mean Error Statistics
T+0 T+24 T+48 T+72 T+96 T+120 T+144 T+168
Possibly verified 215 164 121 87 63 47 34 26
Detection rate (%) 100 99 98 99 100 94 94 92
AT error (km) 1 -13 -39 -47 -27 -19 -153 -393
CT error (km) -7 -11 -21 -32 -80 -97 -39 57
Track skill (%) ***** 70 76 73 ***** ***** ***** *****
* 2016 skill (%) ***** 65 71 69 ***** ***** ***** *****
DPE (km) 39 71 111 188 301 393 447 651
* 2016 DPE (km) 41 72 133 222 296 422 568 841
Central pressure bias (mb) 0.5 -1.7 -7.4 -10.7 -13.0 -18.1 -20.6 -28.7

* DPE and skill for all north-west Pacific storms in 2016 season

Plot of the observed tracks of all storms in the north-west Pacific basin
Track forecast errors in the north-west Pacific basin
Track forecast skill in the north-west Pacific basin

2017 saw a similar levels of activity to 2016. Track forecast errors were mostly lower than last season's values. Skill against CLIPER was again at its highest values ever achieved. There was a strong bias (as measured by central pressure) which increased with lead time.

3.2 North-east Pacific basin storms

Table of mean error statistics
T+0 T+24 T+48 T+72 T+96 T+120 T+144 T+168
Possibly verified 129 96 74 55 41 28 20 12
Detection rate (%) 100 99 100 98 98 89 75 67
AT error (km) 1 12 43 58 106 220 291 366
CT error (km) -7 -6 -32 -71 -95 -173 -183 -316
Track skill (%) ***** 39 53 51 ***** ***** ***** *****
* 2016 skill (%) ***** 66 75 66 ***** ***** ***** *****
DPE (km) 36 74 123 181 284 378 464 546
* 2016 DPE (km) 35 75 115 163 213 281 313 414
Central pressure bias (mb) -0.1 0.1 -1.3 -5.5 -5.8 -4.2 -1.9 -1.5

* DPE and skill for all north-east Pacific storms in 2016 season

Plot of the observed tracks of all storms in the north-east Pacific basin
Track forecast errors in the north-east Pacific basin
Track forecast skill in the north-east Pacific basin

There were fewer hurricanes and shorter lived tropical cyclones than in 2016. Track forecast errors were higher than last season at all but the shortest lead times. Skill scores against CLIPER were lower than the last two seasons. The intensity bias (as measured by central pressure) was generally very small.

3.3 North Atlantic basin storms

Table of mean error statistics
T+0 T+24 T+48 T+72 T+96 T+120 T+144 T+168
Possibly verified 183 146 119 97 81 69 62 52
Detection rate (%) 100 100 100 100 100 99 97 88
AT error (km) 0 -15 -45 -85 -132 -156 -212 -188
CT error (km) 1 -7 -6 -8 -25 -29 -47 -40
Track skill (%) ***** 74 77 75 ***** ***** ***** *****
* 2016 skill (%) ***** 66 76 76 ***** ***** ***** *****
DPE (km) 25 51 94 154 238 330 451 564
* 2016 DPE (km) 29 70 105 152 239 353 438 536
Central pressure bias (mb) 2.9 4.8 3.0 -0.9 -5.5 -8.2 -13.9 -24.0

* DPE and skill for all North Atlantic storms in 2016 season

Plot of the observed tracks of all storms in the North Atlantic basin
Track forecast errors in the North Atlantic basin
Track forecast skill in the North Atlantic basin

2017 saw greater activity than 2016. In particular there were more intense and long-lived hurricanes. Track forecast errors were at their lowest ever at short lead times and near to last season's values at longer lead times. Skill scores against CLIPER were either close to or higher than last season's values. A large strong bias (as measured by central pressure) was apparent at longer lead times.

3.4 North Indian Basin Storms

Table of mean error statistics
T+0 T+24 T+48 T+72 T+96 T+120 T+144 T+168
Possibly verified 19 12 7 4 2 0 0 0
Detection rate (%) 100 100 100 100 100 ***** ***** *****
AT error (km) 17 28 84 138 144 ***** ***** *****
CT error (km) 12 28 84 138 144 ***** ***** *****
Track skill (%) ***** 35 52 70 ***** ***** ***** *****
* 2016 skill (%) ***** 58 69 59 ***** ***** ***** *****
DPE (km) 52 92 118 147 149 ***** ***** *****
* 2016 DPE (km) 41 87 125 184 202 ***** ***** *****
Central pressure bias (mb) -4.3 -7.7 -12.3 -10.5 -1.5 ***** ***** *****

* DPE and skill for all North Indian storms in 2016 season

Plot of the observed tracks of all storms in the North Indian basin
Track forecast errors in the North Indian basin
Track forecast skill in the North Indian basin

Cyclone activity was a little lower than 2016, although more tropical storms reached hurricane strength. Track forecast errors were similar to last season. There was a strong bias in forecasts (as measured by central pressure).

3.5 Combined statistics for whole northern hemisphere

Table of mean error statistics
T+0 T+24 T+48 T+72 T+96 T+120 T+144 T+168
Possibly verified 546 418 321 243 187 144 116 90
Detection rate (%) 100 100 99 99 99 95 92 87
AT error (km) 1 -7 -20 -36 -42 -43 -124 -194
CT error (km) -4 -7 -17 -30 -59 -77 -63 -39
Track skill (%) ***** 66 72 71 ***** ***** ***** *****
* 2016 skill (%) ***** 65 74 70 ***** ***** ***** *****
DPE (km) 34 65 108 172 268 359 452 589
* 2016 DPE (km) 36 73 119 180 246 346 422 557
Central pressure bias (mb) 1.0 0.8 -2.2 -5.6 -8.1 -10.6 -14.2 -23.2

* DPE and skill for all northern hemisphere storms in 2016 season

There was a similar number of tropical cyclones in 2017 than in 2016, but they were generally stronger and longer lived.

Track forecast errors were the lowest ever achieved up to 72 hours and close to or above the values of the last few seasons at longer lead times. Skill scores against CLIPER were similar to last season's values. Cross track errors were further below zero than the last two seasons indicating an equatorwards bias. There was a slight slow bias.  Detection percentages were high except at the longest lead times.

Track forecast errors for the whole northern hemisphere
Track forecast skill for the whole northern hemisphere
Cross-track errors for the whole northern hemisphere
Along-track errors for the whole northern hemisphere

There was a continued downwards trend in the five-year running mean of track forecast errors.

Five-year running mean track forecast errors for the whole northern hemisphere

Tropical cyclones were much stronger in the model than last season, probably due to the increase in model resolution. This resulted in virtually no intensity bias at short lead times and a significant strong bias at longer lead times as measured by central pressure. Even 10m winds, which have always been too weak in the past, had a slow bias which reduced to zero by long lead times.

Central pressure forecast bias for the whole northern hemisphere
Peak 10m wind forecast bias for the whole northern hemisphere

4. Further tropical cyclone information

The Met Office tropical cyclone web pages contain information on tropical cyclone forecasting at the Met Office. Monthly updates of tropical cyclone activity and forecasts are made, together with observed and forecast  information of recent storms, track and intensity prediction error statistics, lists of names and real time tropical cyclone forecast guidance.

For further information on tropical cyclone forecasting email the Met Office.

Last updated:

Follow us on

Facebook, Twitter, YouTube, Instagram, Snapchat, or LinkedIn Facebook Follow @metoffice on Twitter YouTube Instagram Snapchat LinkedIn