Study Of Predictability And Bias Correction Methods Of Quantitative Precipitation Forecasts

Quantitative precipitation forecasts(QPFs)are one of the most difficult and challenging topics in the operational weather forecasting.Ensemble weather forecasting can represent the uncertainties of QPFs and generate probabilistic QPFs(PQPFs),which improve the potential value of meteorological applications.This study is based on ensemble forecasts,focusing on the predictability and bias correction methods of QPFs.The main content and conclusions are as follows:(1)Based on previous studies,this study establishes the ideal stochastic theoretical model for ensemble precipitation forecast,considering the skewed characteristics of precipitation distribution and subjecting to the constraint of the relationship between ensemble mean and ensemble spread from actual forecast data.The ideal stochastic model is used to theoretically study the predictability and error distribution of precipitation forecasts.A method to estimate the distribution of forecast errors using ensemble spread is proposed.Results show that the spread-skill relationship of ensemble precipitation forecasts depends on the selected metrics of ensemble spread and ensemble forecast error.Even the ideal ensemble prediction system(EPS)cannot depict the perfect linear correlation.As the ensemble size increases,the level of linear correlation gradually saturates when the number of ensemble members reaches 10 to20.Although a strong linear correlation between the ensemble spread and the single value of forecast errors does not exist,there is a perfect correlation between the ensemble spread and the statistical measures of forecast errors(e.g.standard deviation,quantile).Therefore,this study constructs probabilistic error forecasts by fitting the distribution of ensemble mean errors with the ensemble spread.In addition,systematic biases and changed ensemble spread are applied to the ideal model to achieve a nonideal model,which is used to theoretically study the error distribution of QPFs from the nonideal EPS.Results show that the ensemble mean from the ideal EPS is not better than that from the nonideal EPS for extreme precipitation events.The reason is that the ensemble mean can smooth the extreme precipitation forecasts and extreme precipitation samples are insufficient.Small positive systematic biases can improve the forecasts of heavy rain events while deteriorate the forecast performance for light rain events.Nevertheless,the ideal EPS has better probabilistic forecasts than the nonideal EPS because the probabilistic forecasts can better represent the uncertainty information of ensemble forecasts.Large systematic biases and ensemble spread will cause large probabilistic forecast errors.(2)The ensemble precipitation forecasts in the northern hemisphere(NH)during the summers(June to August)of 2008-2012 from six operational centers in the THORPEX Interactive Grand Global Ensemble(TIGGE)are comprehensively verified.The characteristics of different EPSs are summarized.The influence of EPS upgrade on the performance of 24-hour QPFs and PQPFs has been investigated.Considering the large meridional span,the ensemble forecast weighted verification system is designed to compare and verify the EPS performance in the NH midlatitudes and tropics scientifically.Moreover,considering the change of the verification samples before and after the EPS upgrade,a center without upgrade(China Meteorological Administration,CMA)is chosen as the benchmark to eliminate the influence of interannual variability on forecast skill.Results show that the systematic bias and ensemble spread have great influence on the EPS performance.The Canadian Meteorological Centre(CMC)is the most reliable(with the least probabilistic systematic bias).However,the increased ensemble spread after modifying the physics schemes in the EPS brings large probabilistic forecast errors.Generally,the European Centre for Medium-Range Weather Forecasts(ECMWF)performs best,while has poor performance for light precipitation events in the NH tropics due to large probabilistic systematic biases.CMA has large systematic biases in the NH tropics and causes the decreased forecast skill.The 0-24 hour Japan Meteorological Agency(JMA)EPS has remarkable moist biases in the NH tropics as this EPS employs moist singular vectors(SVs)for the entire tropics and the specific humidity is perturbed with a large amplitude.The United Kingdom Meteorological Office(UKMO),the US National Centers for Environmental Prediction(NCEP)and ECMWF significantly improved their forecast performance after EPS upgrade,while there is no significant improvement for JMA.(3)The study optimizes the decaying weight in the adaptive frequency matching method(AFMM),which is used by NCEP for the operational calibration of ensemble precipitation forecasts.The functional relationship between the decaying weight and time scale of forecast error is explored.The optimized AFMM has been implemented experimentally in China National Meteorological Center(NMC),Fujian Meteorological Bureau and Anhui Meteorological Bureau.To further improve the forecast performance of extreme precipitation events,the regional frequency matching method(RFMM)and the gridded frequency matching method(GFMM)are proposed based on the long-term training samples from the NCEP global ensemble forecast system(GEFS)reforecast dataset.The RFMM only pools more temporal training samples based on the AFMM,while the GFMM considers the spatial dependency of precipitation forecast biases.Results show that considering the local systematic biases for precipitation forecast is more important than simply pooling more training samples.In general,all the three frequency matching methods(FMMs)can calibrate the systematic biases in the raw forecasts.The RFMM only slightly improves the AFMM while the GFMM significantly improves the QPFs and PQPFs.