Improving the Short-Range Precipitation Forecast of Numerical Weather Prediction through a Deep Learning-Based Mask Approach

Due to various technical issues,existing numerical weather prediction(NWP)models often perform poorly at forecasting rainfall in the first several hours.To correct the bias of an NWP model and improve the accuracy of short-range precipitation forecasting,we propose a deep learning-based approach called UNet Mask,which combines NWP forecasts with the output of a convolutional neural network called UNet.The UNet Mask involves training the UNet on historical data from the NWP model and gridded rainfall observations for 6-hour precipitation forecasting.The overlap of the UNet output and the NWP forecasts at the same rainfall threshold yields a mask.The UNet Mask blends the UNet output and the NWP forecasts by taking the maximum between them and passing through the mask,which provides the corrected 6-hour rainfall forecasts.We evaluated UNet Mask on a test set and in real-time verification.The results showed that UNet Mask outperforms the NWP model in 6-hour precipitation prediction by reducing the FAR and improving CSI scores.Sensitivity tests also showed that different small rainfall thresholds applied to the UNet and the NWP model have different effects on UNet Mask's forecast performance.This study shows that UNet Mask is a promising approach for improving rainfall forecasting of NWP models.

A Short-Range Quantitative Precipitation Forecast Algorithm Using Back-Propagation Neural Network Approach

A back-propagation neural network （BPNN） was used to establish relationships between the shortrange （0-3-h） rainfall and the predictors ranging from extrapolative forecasts of radar reflectivity, satelliteestimated cloud-top temperature, lightning strike rates, and Nested Grid Model （NGM） outputs. Quan- titative precipitation forecasts （QPF） and the probabilities of categorical precipitation were obtained. Results of the BPNN algorithm were compared to the results obtained from the multiple linear regression algorithm for an independent dataset from the 1999 warm season over the continental United States. A sample forecast was made over the southeastern United States. Results showed that the BPNN categorical rainfall forecasts agreed well with Stage Ⅲ observations in terms of the size and shape of the area of rainfall. The BPNN tended to over-forecast the spatial extent of heavier rainfall amounts, but the positioning of the areas with rainfall ≥25.4 mm was still generally accurate. It appeared that the BPNN and linear regression approaches produce forecasts of very similar quality, although in some respects BPNN slightly outperformed the regression.

Quantitative Precipitation Forecast Experiment Based on Basic NWP Variables Using Deep Learning

The quantitative precipitation forecast(QPF)performance by numerical weather prediction(NWP)methods depends fundamentally on the adopted physical parameterization schemes(PS).However,due to the complexity of the physical mechanisms of precipitation processes,the uncertainties of PSs result in a lower QPF performance than their prediction of the basic meteorological variables such as air temperature,wind,geopotential height,and humidity.This study proposes a deep learning model named QPFNet,which uses basic meteorological variables in the ERA5 dataset by fitting a non-linear mapping relationship between the basic variables and precipitation.Basic variables forecasted by the highest-resolution model(HRES)of the European Centre for Medium-Range Weather Forecasts(ECMWF)were fed into QPFNet to forecast precipitation.Evaluation results show that QPFNet achieved better QPF performance than ECMWF HRES itself.The threat score for 3-h accumulated precipitation with depths of 0.1,3,10,and 20 mm increased by 19.7%,15.2%,43.2%,and 87.1%,respectively,indicating the proposed performance QPFNet improved with increasing levels of precipitation.The sensitivities of these meteorological variables for QPF in different pressure layers were analyzed based on the output of the QPFNet,and its performance limitations are also discussed.Using DL to extract features from basic meteorological variables can provide an important reference for QPF,and avoid some uncertainties of PSs.

Calibration and Quantitative Forecast of Extreme Daily Precipitation Using the Extreme Forecast Index (EFI)

Based on the extreme forecast index (EFI) of ECMWF, the “observed” EFI (OEFI) of observation is defined and the EFI is calibrated. Then the EFI equivalent percentile (EFIEP) and EFI equivalent quantile (EFIEQ) are designed to forecast the daily extreme precipitation quantitatively. The formulation indicates that the EFIEP is correlated not only to the EFI but also to the proportion of no precipitation. This characteristic is prominent as two areas with nearly same EFIs but different proportions of no precipitation. Cases study shows that the EFIEP can forecast reliable percentile of daily precipitation and 100% percentiles are forecasted for over max extreme events. The EFIEQ is a considerable tool for quantitative precipitation forecast (QPF). Compared to the probabilistic forecast of ensemble prediction system (EPS), it is quantitative and synthesizes the advantage of extreme precipitation location forecast of EPS. Using the observations of 2311 stations of China in 2016 to verify the EFIEP and EFIEQ, the results show that the forecast biases are around 1. The threat scores (TS) for 20 years return period events are about 0.21 and 0.07 for 36 and 180 hours lead times respectively. The equivalent threat scores (ETS) are all larger than 0 and nearly equal to the TS. The TS for heavy rainfall are 0.23 and 0.07 for 36 and 180 lead times respectively. The scores are better than those of high resolution deterministic model (HRDet) and show significant forecast skills for quantitative forecast of extreme daily precipitation.

Bayesian Processor of Output for Probabilistic Quantitative Precipitation Forecast over Central and West Africa

The main goal of this work is a feasibility study for the Bayesian Processor of Output (BPO) method applied to tropical convective precipitation regimes over Central and West Africa. The study uses outputs from the Weather Research and Forecasting (WRF) model to develop and test the BPO technique. The model ran from June 01 to September 30 of 2010 and 2011. The BPO method is applied in each grid point and then in each climatic zone. Prior (climatic) distribution function is estimated from the Tropical Rainfall Measuring Mission (TRMM) data for the period 2002-2011. Many distribution functions have been tested for the fitting. Weibull distribution is found to be a suitable fitting function as shown by goodness of fit (gof) test in both cases. The rain pattern increases with the value of the probability p. BPO method noticeably improves the distribution of precipitation as shown by the spatial correlation coefficients. It better detects certain observed maxima compared to the raw WRF outputs. Posterior distribution (forecasting) functions allow for a simulated rainfall amount, to deduce the probability that observed rainfall falls above a given threshold. The probability of observing rainfall above a given threshold increases with simulated rainfall amounts.

A SIMILARITY SCHEME FOR QUANTITATIVE FORECAST OF PRECIPITATION OF TYPHOONS

A quantitative scheme is put forward in our work of forecasting the storm rainfall of typhoons for specific sites.Using the initial parameters,weather situations and physical quantities as well as numerical weather prediction products,the scheme constructs multivariate,objective and similarity criteria for environmental factors for the time between the current and forthcoming moment within the domain of forecast.Through defining a non-linear similarity index,this work presents a comprehensive assessment of the similarity between historical samples of typhoons and those being forecast in terms of continuous dynamic states under the multivariate criteria in order to identify similar samples.The historical rainfall records of the similar samples are used to run weighted summarization of the similarity index to determine site-specific and quantitative forecasts of future typhoon rainfall.Samples resembling the typhoon being forecast are selected by defining a non-linear similarity index composed of multiple criteria.Trial tests have demonstrated that this scheme has positive prediction skill.

Short-Term Precipitation Forecasting Rolling Update Correction Technology Based on Optimal Fusion Correction

In order to improve the availability of regional model precipitation forecast, this project intends to use the measured heavy rainfall data of dense automatic stations to carry out historical precipitation in the high resolution: the Severe Weather Automatic Nowcast System (SWAN) quantitative precipitation forecast and the High-Resolution Rapid Refresh (HRRR) regional numerical model precipitation forecast in short-term nowcasting aging. Based on the error analysis, the grid fusion technology is used to establish the measured rainfall, HRRR regional model precipitation forecast, and optical flow radar quantitative precipitation forecast (QPF) three-source fusion correction scheme, comprehensively integrate the revised forecasting effect, adjust the fusion correction parameters, establish an optimal correction plan, generate a frozen rolling update revised product based on measured dense data and short-term forecast, and put it into business operation, and perform real-time effect rolling test evaluation on the forecast product.

A CRPS-Based Spatial Technique for the Verification of Ensemble Precipitation Forecasts

Traditional precipitation skill scores are affected by the well-known"double penalty"problem caused by the slight spatial or temporal mismatches between forecasts and observations.The fuzzy(neighborhood)method has been proposed for deterministic simulations and shown some ability to solve this problem.The increasing resolution of ensemble forecasts of precipitation means that they now have similar problems as deterministic forecasts.We developed an ensemble precipitation verification skill score,i.e.,the Spatial Continuous Ranked Probability Score(SCRPS),and used it to extend spatial verification from deterministic into ensemble forecasts.The SCRPS is a spatial technique based on the Continuous Ranked Probability Score(CRPS)and the fuzzy method.A fast binomial random variation generator was used to obtain random indexes based on the climatological mean observed frequency,which were then used in the reference score to calculate the skill score of the SCRPS.The verification results obtained using daily forecast products from the ECMWF ensemble forecasts and quantitative precipitation estimation products from the OPERA datasets during June-August 2018 shows that the spatial score is not affected by the number of ensemble forecast members and that a consistent assessment can be obtained.The score can reflect the performance of ensemble forecasts in modeling precipitation and thus can be widely used.

The Impact of Assimilating Radar-estimated Rain Rates on Simulation of Precipitation in the 17-18 July 1996 Chicago Floods

Rainfall prediction remains one of the most challenging problems in weather forecasting. In order to improve high-resolution quantitative precipitation forecasts （QPF）, a new procedure for assimilating rainfall rate derived from radar composite reflectivity has been proposed and tested in a numerical simulation of the Chicago floods of 17–18 July 1996. The methodology is based on the one-dimensional variation scheme （1DVAR） assimilation approach introduced by Fillion and Errico but applied here using the Kain-Fritsch convective parameterization scheme （KF CPS）. The novel feature of this work is the continuous assimilation of radar estimated rain rate over a three hour period, rather than a single assimilation at the initial （analysis） time. Most of the characteristics of this precipitation event, including the propagation, regeneration of mesoscale convective systems, the frontal boundary across the Midwest and the evolution of the low-level jet are better captured in the simulation as the radar-estimated precipitation rate is assimilated. The results indicate that precipitation assimilation during the early stage can improve the simulated mesoscale feature of the convection system and shorten the spin-up time significantly. Comparison of precipitation forecasts between the experiments with and without the 1DVAR indicates that the 1DVAR scheme has a positive impact on the QPF up to 36 hours in terms of the bias and bias equalized threat scores.

BMA probability quantitative precipitation forecasting of land-falling typhoons in south-east China

The probability of quantitative precipitation forecast(PQPF)of three Bayesian Model Averaging(BMA)models based on three raw super ensemble prediction schemes(i.e.,A,B,and C)are established,which through calibration of their parameters using 1-3 day precipitation ensemble prediction systems(EPSs)from the China Meteorological Administration(CMA),the European Centre for Medium-Range Weather Forecasts(ECMWF)and the National Centers for Environmental Prediction(NCEP)and observation during land-falling of three typhoons in south-east China in 2013.The comparison of PQPF shows that the performance is better in the BMA than that in raw ensemble forecasts.On average,the mean absolute error(MAE)of 1 day lead time forecast is reduced by 12.4%,and its continuous ranked probability score(CRPS)of 1-3 day lead time forecast is reduced by 26.2%,respectively.Although the amount of precipitation prediction by the BMA tends to be underestimated,but in view of the perspective of probability prediction,the probability of covering the observed precipitation by the effective forecast ranges of the BMA are increased,which is of great significance for the early warning of torrential rain and secondary disasters induced by it.

QpefBD:A Benchmark Dataset Applied to Machine Learning for Minute-Scale Quantitative Precipitation Estimation and Forecasting

Nowcasts of strong convective precipitation and radar-based quantitative precipitation estimations have always been hot yet challenging issues in meteorological sciences.Data-driven machine learning,especially deep learning,provides a new technical approach for the quantitative estimation and forecasting of precipitation.A high-quality,large-sample,and labeled training dataset is critical for the successful application of machine-learning technology to a specific field.The present study develops a benchmark dataset that can be applied to machine learning for minutescale quantitative precipitation estimation and forecasting(QpefBD),containing 231,978 samples of 3185 heavy precipitation events that occurred in 6 provinces of central and eastern China from April to October 2016-2018.Each individual sample consists of 8 products of weather radars at 6-min intervals within the time window of the corresponding event and products of 27 physical quantities at hourly intervals that describe the atmospheric dynamic and thermodynamic conditions.Two data labels,i.e.,ground precipitation intensity and areal coverage of heavy precipitation at 6-min intervals,are also included.The present study describes the basic components of the dataset and data processing and provides metrics for the evaluation of model performance on precipitation estimation and forecasting.Based on these evaluation metrics,some simple and commonly used methods are applied to evaluate precipitation estimates and forecasts.The results can serve as the benchmark reference for the performance evaluation of machine learning models using this dataset.This paper also gives some suggestions and scenarios of the QpefBD application.We believe that the application of this benchmark dataset will promote interdisciplinary collaboration between meteorological sciences and artificial intelligence sciences,providing a new way for the identification and forecast of heavy precipitation.

基于QPE和QPF的遗传神经网络洪水预报试验

以湖北省清江上游水布垭控制流域为例,利用分组Z-I关系并结合地面雨量站资料对雷达估算降水进行校准,计算出流域实况平均面雨量;再利用遗传算法和神经网络相结合的方法建立订正AREM预报降水的模型;最后,将订正前后的AREM预报降水输入新安江水文模型进行洪水预报试验。结果表明:订正后AREM预报降水能明显提高过程的累计降水量预报精度,平均相对误差减小幅度在60%以上,对逐小时过程降水预报精度也有一定提高,但与实况相比仍有一定差距;订正前后AREM预报降水的洪水预报试验的确定性系数的场次平均从-32.6%提高到64.38%,洪峰相对误差从39%减小到25.04%,确定性系数的提高效果优于洪峰相对误差,整体上洪水预报精度有所提高。

一种基于分数技巧评分定义的降水预报跳跃指数及其应用

基于CMA-BJ系统提供的2021年8月9日2套降水预报结果(业务数值预报模式和同化试验结果)和2022年6月4日、2023年9月17日的业务数值预报模式结果,结合定性分析,利用4种客观评价指标(不确定度、均方根误差、不一致指数和基于分数技巧评分(FSS)定义的预报跳跃指数)对该系统降水预报不一致特征进行了定量评估。3次降水过程的分析结果显示:预报跳跃指数不仅可以识别出2021年8月9日和2022年6月4日业务数值预报模式结果中降水量预报明显减小的3个预报时次,而且对于降水过程预报相对稳定的个例(2021年8月9日同化试验和2023年9月17日业务预报结果),随着预报时次逐渐临近最新预报,该指数整体呈现波动上升或者数值较大、波动较小的特征,表明15个连续降水预报特征逐渐与最新预报趋于一致或者大体相似,与定性分析结果相对吻合。其他3种指数对于降水预报不一致问题的表征存在不足,不确定度和均方根误差显著受到预报降水量的影响,同时不确定度不能反映预报不一致的时间特征,不一致指数随预报时次逐时滚动变化较大,确定的预报不一致时次较多,与定性分析结果存在明显偏差。

湿Q矢量释用技术在登陆华东台风定量降水预报(QPF)中的应用研究

利用湿Q矢量对数值预报模式输出的风场、温度场、温度露点场进行动力释用,并考虑地形强迫作用,得到一个独立于数值模式直接预报输出降水场的释用预报降水场即湿Q矢量释用(Q^MVIP)技术。结合2010—2014年汛期(6—9月)登陆我国华东14个历史台风降水实况资料以及华东区域中尺度模式(基于WRF V3.1)(简称WRF模式)预报产品,统计检验分析了Q^MVIP技术对登陆台风降水的定量预报能力。结果表明,Q^MVIP技术较WRF模式明显改善了对25.0 mm·(24 h)^(-1)以上及50.0 mm·(24 h)^(-1)以上降水的定量预报能力。进一步结合"菲特"台风(2013)登陆前后所引发的24 h累积降水场进行比较分析发现,Q^MVIP技术对台风暴雨落区、强度的反映能力均优于WRF模式。这表明,湿Q矢量释用技术可以在一定程度上弥补现有数值预报模式对登陆台风定量降水预报(QPF)能力的不足。

Impact of Soil Moisture Uncertainty on Summertime Short-range Ensemble Forecasts

To investigate the impact of soil moisture uncertainty on summertime short-range ensemble forecasts（SREFs）, a fivemember SREF experiment with perturbed initial soil moisture（ISM） was performed over a northern China domain in summertime from July to August 2014. Five soil moisture analyses from three different operational/research centers were used as the ISM for the ensemble. The ISM perturbation produced notable ensemble spread in near-surface variables and atmospheric variables below 800 h Pa, and produced skillful ensemble-mean 24-h accumulated precipitation（APCP24） forecasts that outperformed any single ensemble member. Compared with a second SREF experiment with mixed microphysics parameterization options, the ISM-perturbed ensemble produced comparable ensemble spread in APCP24 forecasts, and had better Brier scores and resolution in probabilistic APCP24 forecasts for 10-mm, 25-mm and 50-mm thresholds. The ISM-perturbed ensemble produced obviously larger ensemble spread in near-surface variables. It was, however, still under-dispersed, indicating that perturbing ISM alone may not be adequate in representing all the uncertainty at the near-surface level, indicating further SREF studies are needed to better represent the uncertainties in land surface processes and their coupling with the atmosphere.

CMA-BJ V2.0系统华北地区降水预报性能评估

利用CMA-BJ V2.0系统在2021年汛期(6—9月)华北地区预报的平均日降水量和24 h内逐时降水量,评估不同水平分辨率(3 km和9 km)在降水量、有效降水时次占比、降水强度、降水日变化等方面的预报性能。结果表明:9 km和3 km分辨率预报均可较好地反映降水量和落区,捕捉平均日降水量大于8 mm的降水区域分布特征,但降水量级的预报较观测偏大;对小时降水量和有效降水时次占比日变化的预报与观测基本一致,但对傍晚的峰值预报偏强,且多个时段空报,同时高估了小时降水量。与9 km分辨率预报相比,3 km分辨率预报对有效降水时次占比随累积降水量的变化趋势与观测更接近,对小时有效降水时次占比日变化、峰谷值出现时间的预报也与观测更接近。9 km分辨率预报对弱降水过程的预报能力更优,而3 km分辨率预报对强降水过程的预报能力更优。

Evaluation of TIGGE Daily Accumulated Precipitation Forecasts over the Qu River Basin, China

Quantitative precipitation forecasts(QPFs)provided by three operational global ensemble prediction systems(EPSs)from the THORPEX(The Observing System Research and Predictability Experiment)Interactive Grand Global Ensemble(TIGGE)archive were evaluated over the Qu River basin,China during the plum rain and typhoon seasons of 2009–13.Two post-processing methods,the ensemble model output statistics based on censored shifted gamma distribution(CSGD-EMOS)and quantile mapping(QM),were used to reduce bias and to improve the QPFs.The results were evaluated by using three incremental precipitation thresholds and multiple verification metrics.It is demonstrated that QPFs from NCEP and ECMWF presented similarly skillful forecasts,although the ECMWF QPFs performed more satisfactorily in the typhoon season and the NCEP QPFs were better in the plum rain season.Most of the verification metrics showed evident seasonal discriminations,with more satisfactory behavior in the plum rain season.Lighter precipitation tended to be overestimated,but heavier precipitation was always underestimated.The post-processed QPFs showed a significant improvement from the raw forecasts and the effects of post-processing varied with the lead time,precipitation threshold,and EPS.Precipitation was better corrected at longer lead times and higher thresholds.CSGD-EMOS was more effective for probabilistic metrics and the root-mean-square error.QM had a greater effect on removing bias according to bias and categorical metrics,but was unable to warrant reliabilities.In general,raw forecasts can provide acceptable QPFs eight days in advance.After post-processing,the useful forecasts can be significantly extended beyond 10 days,showing promising prospects for flood forecasting.

一种联合历史和实时资料的定量降水短时预报技术研究

本文提出一种基于百分位映射,使用实时和历史资料联合订正模式定量降水短时订正预报技术,并与仅使用历史资料或实时资料的订正效果进行对比。结果表明:三种方案均能有效订正模式原始降水预报偏差,提高0~12 h降水预报准确率。对于晴雨预报,采用联合订正方案,预报效果最优。在0~7 h预报时效内,仅采用实时资料的订正方案准确率明显优于仅采用历史资料的订正方案,在8~12 h预报时效内,后者准确率稍高。所有预报时效内,仅采用实时资料的订正方案降水预报范围较仅采用历史资料的方案略偏大。对于较强降水预报,采用联合订正方案准确率为三种方案最优,仅采用实时资料的方案预报准确率虽优于仅采用历史资料的方案,但预报范围及量级较实况明显偏大。

浙江梅汛期暴雨预报的客观订正方案对比分析

基于2019—2021年浙江省自动站观测资料和多模式预报数据,分析了各模式对梅汛期暴雨预报的综合表现,并采用12组降水订正方案开展了2020年和2021年浙江省梅汛期降水预报的客观订正,对比了各订正方案对模式暴雨预报的改进效果。结果表明:ECMWF、CMA-SH9和CMA-MESO梅汛期暴雨预报表现优于NCEP-GFS和CMA-GFS,且频率偏差关系稳定,可联合用于开展多模式预报客观订正;由于逐年梅汛期暴雨特征差异大,频率匹配算法无法对预报进行有效订正;最优评分法(OTS)能显著提升ECMWF模式暴雨预报TS评分,但空报率有所增加;对ECMWF降水预报经OTS量级订正后再开展基于集合平均的概率匹配订正,能明显改善以大雨带稳定性降水为主的梅汛期暴雨预报质量,但对于对流性较强的梅汛期暴雨过程订正效果不佳;优选预报成员的各类多模式融合算法均能够有效改进对流性较强的梅汛期暴雨过程预报质量,包括多模式平均、自适应集成和时滞集合预报在2020年和2021年均有明显正技巧;对各模式降水预报经OTS订正后再开展集成预报能够进一步提高梅汛期暴雨预报质量,且对稳定性暴雨和对流性暴雨过程均有较好的订正能力,其中经多模式时滞集合分级订正算法集成OTS量级订正预报表现最优。

ON THE SENSITIVITY OF PRECIPITATION FORECASTS TO THE MOIST PHYSICS AND THE HORIZONTAL RESOLUTION OF NUMERICAL MODEL

The impacts of the enhanced model's moist physics and horizontal resolution upon the QPFs (quantitative precipitation forecasts)are investigated by applying the HIRLAM(high resolution limited area model)to the summer heavy-rain cases in China.The performance of the control run, for which a 0.5°×0.5°grid spacing and a traditional“grid-box supersaturation removal+Kuo type convective paramerization”are used as the moist physics,is compared with that of the sensitivity runs with an enhanced model's moist physics(Sundqvist scheme)and an increased horizontal resolution(0.25°×0.25°),respectively.The results show: (1)The enhanced moist physics scheme(Sundqvist scheme),by introducing the cloud water content as an additional prognostic variable and taking into account briefly of the microphysics involved in the cloud-rain conversion,does bring improvements in the model's QPFs.Although the deteriorated QPFs also occur occasionally,the improvements are found in the majority of the cases,indicating the great potential for the improvement of QPFs by enhancing the model's moist physics. (2)By increasing the model's horizontal resolution from 0.5°×0.5°,which is already quite high compared with that of the conventional atmospheric soundings,to 0.25°×0.25°without the simultaneous enhancement in model physics and objective analysis,the improvements in QPFs are very limited.With higher resolution,although slight amelioration in locating the rainfall centers and in resolving some finer structures of precipitation pattern are made,the number of the mis- predicted fine structures in rainfall field increases with the enhanced model resolution as well.