Long Range Tropical Cyclone Forecasts User Guide
Comparing ACCESS-GE2 and S1
Direct comparison between GE2 and S1 was carried out for the 2017-18 cyclone season for 4-10 day forecasts. It showed that GE2 was more skilful at these shorter leadtimes (see below). Therefore it is recommended that GE2 products are used for guidance at shorter leadtimes, with S1 products used to provide guidance at longer timeframes only (i.e days 8-14 and onwards).

However, the S1 forecasts (which are made independently of GE2) at shorter lead times could be used as a qualitative 'multi-model ensemble'. If both GE2 and S1 are forecasting similar storm formation at shorter leadtimes the user can have better confidence in the forecasts.

SH Raw w/ 14 m/s threshold Calibrated
ACCESS-GE2 0.227 0.379 N/A
ACCESS-S1 0.215 0.302 0.302

SH-SMALL Raw w/ 14 m/s threshold Calibrated
ACCESS-GE2 0.196 0.331 N/A
ACCESS-S1 0.185 0.276 0.260


SH ACCESS-GE2 w/threshold


SH-SMALL ACCESS-GE2 w/threshold

SH ACCESS-S1 w/threshold



SH-SMALL ACCESS-S1 w/threshold


Based on these results and those of the S1 hindcast (available at Skill and Analysis Page ), the following suggestions are provided to users.

Raw tracks
Forecast storm tracks are provided for both ACCESS-GE2 (for 4-10 days forecasts) and ACCESS-S1 (for week 1-4 forecasts). In both instances, tracks that do not reach a 14 m/s 850 hPa wind-speed threshold during their lifetime are plotted using a dotted lined (see below).

This product gives insights into likely storm directions. Tracks from each ensemble member are plotted in a different colour. This product has good skill for GE2 for days 4-10. Skill for ACCESS-GE2 is good for the 2nd-week forecast (days 8-14). Skill is limited for the 3rd-week forecast (days 15-21) as spatial accuracy is poor at this leadtime. It should not be used for guidance beyond 21 days.
Strike probabilities
Strike probabilities are also provided for both ACCESS-GE2 (for 4-10 days forecasts) and ACCESS-S1 (for week 1-4 forecasts). These compute the number of forecast storm tracks that pass within a certain distance of each other. This number is then expressed as a percentage of the overall model ensemble size (e.g. for ACCESS-S1, if 15 storms passed within 300 km of each other at a given location, the strike probability at that point = 15/33 = 45%). Only tracks that meet the windspeed threshold are used to compute strike probabilities.

Since this product is essentially a different visualisation of the raw tracks (with wind-speed threshold applied) the guidance for its use is the same.

The GE2 forecasts use a distance of 120 km which is consistent with the original definition of Hurricane strike probabilities developed by the US National Weather Service. To better account for the uncertainty of longer-range forecasts, the S1 forecasts use a distance of 300 km (which is consistent with that used by the ECMWF for their multi-week cyclone forecasts).

Calibrated probabilities
Calibrated probabilities are only available for ACCESS-S1. These calibrated probabilities have been computed using the S1 hindcast (valid from 1990-2012) in the Southern Hemisphere only. They are available in two spatial resolutions.

Both products have skill out to week four (22-28 day forecasts) however skill is better at longer leadtimes for the coarser product (computed on overlapping 15x20 degree regions).

Raw probabilities for a given forecast are plotted in addition to the calibrated probabilities to show the effect of the calibration, which usually reduces the chance of storm formation. By default, the OWZ tracking scheme when applied to ACCESS-S1 produces too many false alarms, especially at longer lead times.