A Decision-tree Approach to Seasonal Prediction of Extreme Precipitation in Eastern China

Seasonal prediction of extreme precipitation has long been a challenge especially for the East Asian Summer Monsoon region, where extreme rains are often disastrous for the human society and economy. This paper introduces a decision‐tree (DT) method for predicting extreme precipitation in the rainy season over South China in April–June (SC‐AMJ) and the North China Plain in July–August (NCP‐JA). A number of preceding climate indices are adopted as predictors. In both cases, the DT models involving ENSO and NAO indices exhibit the best performance with significant skills among those with other combinations of predictors and are superior to their linear counterpart, the binary logistic regression model. The physical mechanisms for the DT results are demonstrated by composite analyses of the same DT path samples. For SC‐AMJ, an extreme season can be determined mainly via two paths: the first follows a persistent negative NAO phase in February–March; the second goes with decaying El Nino. For NCP‐JA, an extreme season can also be traced via two paths: the first is featured by “non El Nino” and an extremely negative NAO phase in the preceding winter; the second follows a shift from El Nino in the preceding winter to La Nina in the early summer. Most of the mechanisms underlying the decision rules have been documented in previous studies, while some need further studies. The present results suggest that the decision‐tree approach takes advantage of discovering and incorporating various nonlinear relationships in the climate system, hence is of great potential for improving the prediction of seasonal extreme precipitation for given regions with increasing sample observations.