ABSTRACT We apply a tropical cyclone (TC) track classification method to the subseasonal TC predictions issued by the European ensemble prediction model (EPS) on the Southwest Indian Ocean. This track typology is based on a subdivision of the basin into three areas. Each simulated storm track is assigned a three‐digit code, where each digit represents the area of genesis, the westernmost position, and the easternmost position, respectively. To account for the intensity bias in the simulated TCs, we conduct sensitivity tests that result in lowering the tropical storm wind threshold to 29kt and filtering out systems with a lifetime maximum intensity lower than 34kt. The model skill is evaluated against the performance of a 1‐month moving climatology, to validate its ability to capture intra‐seasonal variations in TC activity and favoured track typology. Results show an overestimation of occurrence probabilities for all track types, with the central part of the basin and the Mozambique Channel being the regions most affected. These limitations confine the raw model's skill to the second week (W2, i.e., +7 to +14 days) of forecast only. However, when evaluating the model on its capacity to assign a track type to TC genesis known to be valid at the basin scale, the EPS exhibits a 20% performance gain for W2 and a 5% gain at W3 and W4, compared to the moving climatology. These results demonstrate that when TC forecasters consider an EPS genesis prediction to be reliable, they can leverage the corresponding track type predictions to characterise TC risk more precisely, even a month in advance. Furthermore, aggregating track types based on their likelihood of impacting inhabited areas within the basin further enhances predictive skill. An impact‐based forecasting product is derived from this work and will be evaluated in operations by the TC forecasters in La Reunion.
Colomb et al. (Sat,) studied this question.