Evaluation Of Quantitative Precipitation Forecasts In The Middle And Lower Reaches Of The Yangtze River During The Summer Of 2014

Precipitation forecasts have always been the emphases of operational weather forecasting.With the rapid development of numerical weather forecasting(NWP)models,the qu,antitative precipitation forecasts(QPFs)from model outputs are critical for operational forecasters to produce weather forecasts.This study aimed to comprehensively evaluate QPFs in the mid and lower reaches of the Yangtze River during the summer of 2014,using the traditional statistical scores and a spatial verification technique MODE(Method for Object-based Diagnostic Evaluations).The evaluation focuses on the performances of the QPFs of the Global Forecast System(GFS)model at global scale and the WRF-NJU model at convection resolved scale.Possible causes of WRF-NJU model errors are also diagnosed and analyzed through a case study.In addition,the advantage and applicability of MODE applied to models of different resolutions are discussed.Firstly,the evaluation results of the GFS indicate the model can forecast the synoptic situation well in the first 60h periods and accurately depict the generation and dissipation of the main summer precipitation events in the study area.However,discrepancies between the model output and the observation exist in the amount of average precipitation.More specifically,evaluation scores show that the GFS have an overall good forecasting capability of small rainfall despite larger false alarms than misses of forecast.On the contrast,the deficiency of GFS performance emerges in higher precipitation thresholds,exhibiting serious misses of the forecast compared with observation.Moreover,the forecasts within the 24h have obvious superiority in almost each test score,providing relatively high reference value for forecasters.The results based on MODE show the GFS have good ability to depict the general form and orientation of the rain reign at synoptic scale,but there exists the shape,displacement and size errors of the rainfall area.Secondly,the comparison of WRF-NJU and the GFS indicates the former has no advantage in the forecasts of average precipitation at large scale.However,it has higher temporal and spatial resolution and quite reasonably describes the diurnal variation characteristic of precipitation in the study area.Furthermore,using MODE to investigate the forecast performance of the mesoscale system at different scales.The number of meso-P scale objects significantly exceeds that of meso-a scale objects.The forecast intensity of the objects at both scales are stronger than the observation,while the error characteristics of the shape,location and orientation of the rainfall area differs between the two scales.Overall,WRF-NJU has good forecasting capability of large scale synoptic process despite of overforecasting.However,it shows deficiencies in forecasting convective precipitation system featuring serious missing of forecasts.The performance of the model forecast varies with the extension of validation time.Because of the effect of spin-up process,the model performance reaches peak at 3h and then decreases with growing forecast errors,the deterioration of forecast being particularly evident after 33h.Finally,possible causes of the forecasts errors of WRF-NJU are analyzed through the evaluation of a heavy precipitation case from 3-5 July 2014.Results indicate the forecast of high-level circulation is basically accurate,while error existing in the intensity and location of the low-level vortex systems and affect the forecast of precipitation location.In addition,the favorable instability condition are not fully represented in the model forecast,reflecting insufficient capability of the model to depict changes of precipitation system and needs further optimization.Comparison between the results of traditional test scores and MODE method indicates that the verification of MODE resembles the ETS score and there is no essential distinction for evaluation of the model with relatively coarse resolution.However,MODE not only provides more meaningful information which is physical based,but also makes up for the problems using traditional evaluation for NWP models with high resolution,manifesting its critical role in the verification of fine scale model forecasts.Moreover,MODE can be useful for the evaluation of synoptic systems.