基于贝叶斯模型平均的全球洪水中长期预报评估研究

筛选全球5839个水文站逐日径流数据，采用超阈值采样法提取洪水发生频率及时间，将各季节最大日流量作为季节洪水量级，以优选的多个大尺度气候因子的最佳前置月份序列作为潜在预报因子，基于贝叶斯模型平均法构建全球尺度洪水中长期预报模型，并利用均方误差技术指数（MSESS）评价模型的预报效果。结果表明：全球范围内，洪水量级和频率模拟预报效果合格（0.6>MSESS>0.2）的水文站点占比分别为48%和28%；利用前置季节气候因子数据，驱动所构建的洪水中长期预报模型，有效预报了2020年鄱阳湖流域洪水量级将异常偏高。

Mid-long Term Global Flood Prediction Based on Bayesian Model Averaging

Flooding has taken a devastating societal and economic losses around the world, leading to numerous fatalities. Limted to the numbers and spatial distribution of hydrological stations, the flood prediction across the globe remains to be explored. Therefore, this study attributed seasonal flood magnitude/frequency to large-scale climate indices,including Arctic Oscillation (AO), North Atlantic Oscillation (NAO), East Atlantic (EA), East Atlantic/West Russia (EAWR), Pacific Decadal Oscillation (PDO), Atlantic Multidecadal Oscillation (AMO), Niño 3.4 SST and Niño 4 SST, by using hydrological data from 5839 gauging stations. The block maxima approach (BMA) and the peaks-over-threshold sampling technique (POT) were used to obtain flood magnitude and frequency based on these hydrological stations. Global assessmment indicated that flood magnitude and frequency are significantly correlated with large-scale climate indices at most stations (75%), and the influence of climate indices on flood generally has considerable hysteresis quality, with two-quarter lag for 31% of global stations. The climate indices showing statistically significantly correlated with floods are subsequently applied as predictors to build the global-scale season-ahead prediction models by Bayesian model averaging (BMA), with prediction performance evaluated by the mean squared error skill score (MSESS). The performance of simulation was qualified (0.6>MSESS>0.2) for 48% and 28% of the global hydrological stations for flood magnitude and frequency, respectively. Using the pre-season climate index, the flood forecasting models developed in this study perform well in forecast the flood anomalies over South China in 2020. The prediction based on Bayesian model averaging provides a new perspective for global flood prediction by identifying relevant climate indices, with valuable insights for flood risk management and mitigating flood hazards.