Tropical cyclone track forecasting using data assimilation of high resolution satellite derived winds

Tropical Cyclone (TC) Justin formed from a tropical depression in the Coral Sea, northeast of Australia, on March 7, 1997 and had a devastating affect in the area over the next three weeks until it finally dissipated during the last week of March. TC Justin was responsible for the loss of about 40 lives, mostly at sea or from severe flooding on the Australian mainland between Cairns and Townsville. In addition, property damage was extensive, in Australia and Papua New Guinea. Parts of its track were extremely difficult to predict operationally. In the first six days of its existence, Justin exhibited very little movement, whereas the operational numerical model guidance available during this period generally forecast a large movement of the TC center towards the coast. We have chosen several 72 hour forecasts at times when Justin threatened to make landfall but, in fact, recurved away from the Australian coastline. All operational models predicted landfall and evacuation procedures were implemented along parts of the northeast Australian coast.

In this study we examine the impact of hourly satellite-derived winds in continuous data assimilation schemes on one of the most significant and erratic tropical cyclones of the 1996/97 season. The forecasts were re-run using a combination of four dimensional variational data assimilation, a very high resolution model and high temporal and spatial resolution satellite derived winds. Much improved forecasts of Justin's track were obtained for these critical periods. Here, only track forecasts will be considered. Intensities will be addressed in future studies. The inclusion of hourly satellite- derived winds in two data assimilation schemes had a dramatic impact on the track forecasts of Justin. Two 72-hour forecasts of track position were performed, namely, from 00 UTC, 8 March and from 00 UTC, 10 March. The forecast from 00 UTC 8 March using both dynamic nudging and 4D-variational assimilation schemes verified extremely well with the best track, especially when compared to the control and the operational forecast. Similarly, both nudging and 4D-variational assimilation of the hourly satellite derived winds the forecast from 00 UTC 10 March was far superior to the operational, control and CLIPER forecasts.