Frequently Asked Questions (FAQ)

Climatology | Impacts | Extremes| TCs affecting WA towns | Forecast Accuracy
Frequently Asked Questions | Annual summaries

Definitions | Characteristics and Formation | Climatology | Extremes | Forecasting | More Information

Forecasting

2. How accurate are cyclone warnings from the Bureau of Meteorology?

Tropical cyclones have historically had a reputation for being unpredictable. Much effort has been dedicated to improving the forecasting skill in both location and intensity. The Bureau of Meteorology routinely issues forecasts of cyclone location and intensity at 12, 24 and 48 hour time-steps. All official forecasts are verified by comparison with the best track, the official estimate of the location and intensity of a tropical cyclone. A best track is prepared for every tropical cyclone, after the fact, using all available data.

Cyclones vary considerably in their predictability. Some exhibit rapid changes in intensity or change course, speed up or slow down, primarily in response to changes in the surrounding environment. Cyclone Lena (1993), for example, was moving to the west but made a U-turn and returned close to its original path. The 24-hour forecast error for Lena was 258 km, more than double the 1999/2000 - 2003/04 average. Cyclone Gwenda (1999) intensified from a category 2 to category 5 cyclone in 24 hours but subsequently weakened from category 5 to category 2 in less than 24 hours before crossing the coast a few hours later. Others follow a steadier course and more even development cycle such as Vance (1999) making forecasts considerably easier. Also cyclones at the category 1 stage are typically difficult to locate as the centre may not be apparent from satellite imagery, compared to stronger systems that have a well-defined eye. Those systems that markedly change their course or intensity close to the coast present the greatest challenge to forecasters and decision-makers in the community. Community awareness is much higher when a cyclone develops well offshore prior to crossing compared to one that rapidly develops near the coast.

The track of Lena (1993) compared to the steadier track of Vance (1999).

The graph below shows the yearly accuracy of position and forecast accuracy for cyclones off the northwest coast (105 - 125°E) from 1984/85 to 2003/04.. The five-year average (1999/2000 to 2003/04) accuracy has been 27 km for the initial position, 73 km at 12 hours, 121 km at 24 hours and 182 km at 48 hours. Today a 48 hour forecast is as accurate as those issued for a 24 hour prediction ten years ago.

Graph of cyclone position and forecast accuracy from 1984 to 2004 for forecasts issued by the Perth TCWC.
Note: there were no cyclones in the 1987/88 season.

Improvements in the forecast position accuracy are due to a combination of more accurate computer models, improved monitoring technology, such as satellites and a better understanding of the physics of tropical cyclone movement. The use of multiple computer models has increased the forecast skill. The map below compared the actual track of Monty (2004 with forecast tracks from several different computer models sourced from around the world available on one day. The spread of the model tracks provides an estimate of uncertainty in the forecast. Each model is different and no one particular model is always better than the others. The accuracy of each computer model also changes from one model run to the next. A forecast track made from the ensemble of selected computer models provides the basis of the official forecast.

The official track of TC Monty (2004) shown in maroon, compared to forecast tracks from selected computer models.

Although the skill in track forecast has improved greatly, there has been much slower progress in intensity forecasting. This remains the focus of intensive study around the world.

Go to the US Atlantic Oceanographic and Meteorological Laboratory's Hurricane FAQ page to see the accuracy of the US National Hurricane Center for tropical cyclones (hurricanes) in the north Atlantic Ocean.