AbstractAccurate prediction of spring drought in China is helpful toward reducing associated agricultural losses. In this study, spring drought is defined as the three‐month Standardized Precipitation Evapotranspiration Index (SPEI) ending in May. Based on the year‐to‐year increment and downscaling method, three single‐predictor prediction models (P1 model, P2 model, and P3 model) and two multi‐predictor models for spring drought at 677 stations in China are developed for the period 1983–2020. As physical processes affecting spring drought in the China region, the tropical Pacific–Indian Ocean sea surface temperature (SST) in winter, the Davis Strait–Barents sea‐ice concentration (SIC) in winter, and the 500‐hPa spring vertical velocity, predicted by the Climate Forecast System, version 2 (CFSv2), are considered in the prediction models. The prediction skill of the downscaling models for the spring SPEI is measured by cross‐validation for the period 1983–2020. The CFSv2 model only shows convincing prediction skill for the spring SPEI at 35 of 677 stations. However, among the 677 stations, the temporal correlation coefficient between the observed and predicted spring SPEI at 621 stations for P1 model, 598 stations for P2 model, 545 stations for P3 model, 674 stations for the statistical downscaling model (SD model), and 675 stations for the hybrid downscaling dynamical–statistical prediction model (HD model) exceeds the 95% confidence level. Therefore, compared to the CFSv2 model, the prediction skill for spring drought is improved by the single‐predictor and multi‐predictor models. The prediction skill of HD model for spring drought, which combines preceding observational predictors and the simultaneous predictor of the CFSv2 model, is higher than that of SD model. In addition, the severe drought that occurred in Northeast and North China in spring 2017 can be successfully predicted by HD model.
Read full abstract