Research of a Naive Bayesian method-based potential forecast model for short-duration heavy rainfall during summer in Zhejiang Province

Short-duration heavy rainfall (SDHR) is one of the types of severe convective weather. Located on the coast of East China, Zhejiang Province experiences frequent SDHR events during summer. To improve the forecasting capability for SDHR in Zhejiang, this study utilizes hour rain gauge data from national and regional meteorological stations in Zhejiang Province, ERA5 reanalysis and fine-grid numerical forecast products during June to August from 2010 to 2023. First, the distribution characteristics of environmental parameters prior to the occurrence of SDHR are analyzed. Then, the forecast predictors are optimized using point-biserial correlation analysis. Finally, a potential forecast model for summer short-duration heavy rainfall in Zhejiang Province is established based on the naive Bayesian method, and the model is evaluated under different weather patterns. The results show that: (1) Convective available potential energy (CAPE), precipitable water (PWAT), lower-level relative humidity (RH₇₀₀ and RH₈₅₀), lower-level specific humidity (q₇₀₀ and q₈₅₀), and mid-to-lower level vertical wind shear (SHR₀₃ and SHR₀₆) are optimal predictors for SDHR in Zhejiang Province. Before the occurrence of SDHR, the median values of CAPE, PWAT, RH₇₀₀, RH₈₅₀, q₇₀₀, q₈₅₀, SHR₀₃ and SHR₀₆ are 738 J·kg^-1, 56.7 mm, 79.8%, 86.1%, 9.5 g·kg^-1, 14.8 g·kg^-1, 6 m·s^-1 and 7 m·s^-1, respectively. (2) The model was tested and evaluated for short-duration heavy rainfall events in Zhejiang Province from June to August 2023. The P_OD at lead times of 3 h, 6 h, and 12 h is 94.9%, 91.6% and 90.5% respectively, and the T_S are 58.4%, 64.0% and 58.2%, indicating that the potential forecast model has reasonable applicability for forecasting SDHR events. (3) For the Meiyu-front-type, subtropical-high-edge-type, and upper-level-trough-type short-duration heavy rainfall events in the summer of 2023, the model achieves a P_OD of 95% at all lead times, with a M_AR below 8% for these three types. For the subtropical-high-controlled-type heavy rainfall event, the P_OD reaches 80%. These results demonstrate that the model provides useful indications for the potential forecasting of SDHR under different weather patterns and has good application value.