Correction Algorithm Of Short-Term Precipitation Ensemble Forecasts Based On Spatio-Temporal PLS Regression

The study of corrected algorithms for short-term precipitation ensemble forecasts is one of the popular crossover research directions in the field of statistical machine learning and meteorological.Most of the current correction algorithms for short-term precipitation ensemble forecasts do not take into account the spatial correlation and extreme imbalance of the precipitation data.In this thesis,we propose a corrected algorithm based on Spatio-Temporal Partial Least Squares for short-term precipitation ensemble forecasts,which takes into account the spatial correlation and extreme imbalance of precipitation data and applies it to actual meteorological data.Firstly,the raw global ensemble forecasts data ECMWF from the European Center for Medium-Range Weather Forecasts are downscaled on both temporal and spatial scales and converted to hourly forecasts of 0.05° × 0.05°.Secondly,two indicators are constructed to characterize the spatial correlation between precipitation observation sites,and the imbalanced data set is processed by sample weighting method.Finally,a partial least squares regression algorithm is used to model the processed data set to obtain the corrected refined forecasts.The actual case study shows that the proposed algorithm effectively corrects the ECMWF ensemble forecasts for most of the stations in Lanzhou and Tianshui cities,with significant improvement compared with the corrected results of frequency matching and quantile mapping algorithms,and with the mean value of the ECMWF ensemble forecasts and the CMA of the China Meteorological Administration’s guidance forecasts.It is of reference significance in the study of meteorological forecast corrections.