A decision tree approach to identify predictors of extreme rainfall events - A case study for the Fiji Islands

Extreme rainfall events often lead to excessive river flows and severe flooding for Pacific Island nations. Fiji, in particular, is often exposed to extreme rainfall events and associated flooding, with significant impacts on properties, infrastructure, agriculture, and the tourism sector. While these occurrences are often associated with tropical cyclones (TCs), the specific characteristics of TCs that produce extreme rainfall are not well understood. In particular, TC intensity does not appear to be a useful guide in predicting rainfall, since weaker TCs are capable of producing large rainfall compared to more intense systems. Therefore, other TC characteristics, in particular TC track morphology and background climate conditions, may provide more useful insights into what drives TC related extreme rainfall. This study aimed to address this problem by developing a decision tree to identify the most important predictors of TC related extreme rainfall (i.e., 95th percentile) for Fiji. TC attributes considered include; TC duration, the average moving speed of TCs, the minimum distance of TCs from land, seasonality, intensity (wind speed) and the geometry of TCs (i.e., geographical location, shape and length via cluster and sinuosity analyses of TC tracks). In addition, potential predictors based on the phases of Indo-Pacific climate modes were input to the decision tree to represent large scale background conditions. It was found that a TC's minimum distance from land was the most important influence on extreme rainfall, followed by TC cluster grouping, seasonality and duration. The application of this model could result in improved TC risk evaluations and could be used by forecasters and decision-makers on mitigating TC impacts over the Fiji Islands.