Risk assessment through multivariate analysis on the magnitude and occurrence date of daily storm events in the Shenzhen bay area

Multivariate storm frequency modeling would help to gain improved understanding of complex storm process, and provide useful information for regulating flooding risk. In this paper, the dependency between magnitudes and directional occurrence dates of storm events was investigated with copula models. Taking account of extreme storms under various return periods, the concurrence risks as well as temporal representations of the recurrence risk were evaluated by the exceedance probability levels. A case study was demonstrated through the selected storms out of a 53-year long daily rainfall time series in the Shenzhen bay area by using the annual maximum precipitation and peak-over-threshold sampling method. Based on multiple univariate probability functions, a mixed distribution was proposed to determine the most suitable marginal distribution of storm magnitudes and directional dates at three spatially separated stations. The goodness-of-fit tests and ordinary least squares criterion of root-mean-square error were then employed to determine the copula model from the Archimedean and Plackett copulas. According to the coincidence exceedance probabilities of 200, 100, 50 and 20 years return periods, results showed an increase in the conjunctive risk of extreme rainfall events for recent 30 years with regards to pair and triple station, implying the significance to further strengthen the regional flooding control measures. Besides, temporal variations of exceedance probabilities at each station revealed an increasingly concentrated risk during wet period and a shift of the risk peak to earlier dates during 1 year, which would indicate the flooding precautions should be prepared in advance.