Probabilistic Quantitative Precipitation Forecasts Research Articles

Deep Learning of a 200-member Ensemble with a Limited Historical Training to Improve the Prediction of Extreme Precipitation Events

Abstract This study introduces a deep learning (DL) scheme to generate reliable and skillful probabilistic quantitative precipitation forecasts (PQPFs) in a postprocessing framework. Enhanced machine learning model architecture and training mechanisms are proposed to improve reliability and skill of PQPFs while permitting computationally efficient model fitting using a short training dataset. The methodology is applied to postprocessing of 24h accumulated PQPFs from an ensemble forecast system recently introduced by the Center for Western Weather and Water Extremes (CW3E) and for lead times from one to six days. The ensemble system was designed based on a high-resolution version of the Weather Research and Forecasting (WRF) model, named West-WRF, to produce a 200-member ensemble in near real-time (NRT) over the Western U.S. during the boreal cool seasons to support forecastinformed reservoir operations (FIRO) and studies of prediction of heavy-to-extreme events. Postprocessed PQPFs are compared with those from the raw West-WRF ensemble, the operational Global Ensemble Forecast System (GEFSv12) as well as the ensemble from the European Centre for Medium-Range Weather Forecasts (ECMWF). As an additional baseline, we provide PQPFs verification metrics from a recently developed neural network postprocessing scheme proposed by Ghazvinian et al. (2022). Results demonstrate that skill of postprocessed forecasts significantly outperform PQPFs and deterministic forecasts from raw ensembles and the recently developed algorithm. The resulting PQPFs broadly improve upon the reliability and skill of baselines in predicting heavy to extreme precipitation (e.g., > 75 mm) across all lead times while maintaining the spatial structure of the high-resolution raw ensemble.

Improving National Blend of Models Probabilistic Precipitation Forecasts Using Long Time Series of Reforecasts and Precipitation Reanalyses. Part II: Results

Abstract This second part of the series presents results from verifying a precipitation forecast calibration method discussed in the first part, based on quantile mapping (QM), weighting of sorted members, and dressing of the ensemble. NOAA’s Global Ensemble Forecast System, version 12 (GEFSv12), reforecasts were used in this study. The method was validated with preoperational GEFSv12 forecasts from December 2017 to November 2019. The method is proposed as an enhancement for GEFSv12 precipitation postprocessing in NOAA’s National Blend of Models. The first part described adaptations to the methodology to leverage the ∼20-yr GEFSv12 reforecast data. As shown here in this part, when compared with probabilistic quantitative precipitation forecasts from the raw ensemble, the adapted method produced downscaled, high-resolution forecasts that were significantly more reliable and skillful than raw ensemble-derived probabilities, especially at shorter lead times (i.e., <5 days) and for forecasts of events from light precipitation to >10 mm (6 h)−1. Cool-season events in the western United States were especially improved when the QM algorithm was applied, providing a statistical downscaling with realistic smaller-scale detail related to terrain features. The method provided less value added for forecasts of longer lead times and for the heaviest precipitation.

Performance of a Convective-Scale Ensemble Prediction System on 2017 Warm-Season Afternoon Thunderstorms over Taiwan

Abstract A 16-member convective-scale ensemble prediction system (CEPS) developed at the Central Weather Bureau (CWB) of Taiwan is evaluated for probability forecasts of convective precipitation. To address the issues of limited predictability of convective systems, the CEPS provides short-range forecasts using initial conditions from a rapid-updated ensemble data assimilation system. This study aims to identify the behavior of the CEPS forecasts, especially the impact of different ensemble configurations and forecast lead times. Warm-season afternoon thunderstorms (ATs) from 30 June to 4 July 2017 are selected. Since ATs usually occur between 1300 and 2000 LST, this study compares deterministic and probabilistic quantitative precipitation forecasts (QPFs) launched at 0500, 0800, and 1100 LST. This study demonstrates that initial and boundary perturbations (IBP) are crucial to ensure good spread–skill consistency over the 18-h forecasts. On top of IBP, additional model perturbations have insignificant impacts on upper-air and precipitation forecasts. The deterministic QPFs launched at 1100 LST outperform those launched at 0500 and 0800 LST, likely because the most-recent data assimilation analyses enhance the practical predictability. However, it cannot improve the probabilistic QPFs launched at 1100 LST due to inadequate ensemble spreads resulting from limited error growth time. This study points out the importance of sufficient initial condition uncertainty on short-range probabilistic forecasts to exploit the benefits of rapid-update data assimilation analyses. Significance Statement This study aims to understand the behavior of convective-scale short-range probabilistic forecasts in Taiwan and the surrounding area. Taiwan is influenced by diverse weather systems, including typhoons, mei-yu fronts, and local thunderstorms. During the past decade, there has been promising improvement in predicting mesoscale weather systems (e.g., typhoons and mei-yu fronts). However, it is still challenging to provide timely and accurate forecasts for rapid-evolving high-impact convection. This study provides a reference for the designation of convective-scale ensemble prediction systems; in particular, those with a goal to provide short-range probabilistic forecasts. While the findings cannot be extrapolated to all ensemble prediction systems, this study demonstrates that initial and boundary perturbations are the most important factors, while the model perturbation has an insignificant effect. This study suggests that in-depth studies are required to improve the convective-scale initial condition accuracy and uncertainty to provide reliable probabilistic forecasts within short lead times.

Understanding the impact of assimilating FORMOSAT‐7/COSMIC‐2 radio occultation refractivity on tropical cyclone genesis: Observing system simulation experiments using Hurricane Gordon (2006) as a case study

AbstractStudies have shown that assimilating the radio occultation (RO) observations, including those from the FORMOSAT‐3/COSMIC (constellation observing systems for meteorology, ionosphere, and climate) (FS3‐C), provides positive impacts on tropical cyclone (TC) forecasts. The FS3‐C's successor, the FORMOSAT‐7/COSMIC‐2 (FS7‐C2), provides denser spatial data coverage over the Tropics and Subtropics, where severe weather systems often occur. This study investigates the impact of FS7‐C2 refractivity profiles on the prediction of TC genesis. A quick observing system simulation experiment is conducted for the period when Hurricanes Helene and Gordon (2006) occurred over the North Atlantic Ocean using a regional ensemble data assimilation system. Though assimilating FS3‐C or FS7‐C2 ROs successfully reproduces Helene's development, assimilating FS7‐C2 ROs better captures the genesis and development of Gordon with abundant moisture and vorticity in Gordon's core region, providing conditions favorable for the development of deep convection. A minimum area‐mean total precipitable water vapor of 54 mm, as well as the existence of mid‐level cyclonic vorticity (e.g., 500 hPa), at the storm core region in the initial condition is required for forecasting Gordon's genesis. Also, the assimilation of FS7‐C2 ROs in our experiments reduces the 500 hPa geopotential error by 22% and improves probabilistic quantitative precipitation forecast compared with assimilating FS3‐C ROs. Two sensitivity tests are conducted to evaluate the impact of low‐level negatively biased FS7‐C2 RO profiles and the removal of FS7‐C2 data below 3 km on Gordon's genesis. The former test does not degrade Gordon's genesis forecast skills due to a dipole error correlation between the background ROs and the moisture field over an observed RO profile near Gordon. The latter test does degrade Gordon's forecast skills but is still better than the assimilation of FS3‐C ROs since the features of low‐level moisture and mid‐level vorticity are preserved to some extent.

Improving Probabilistic Quantitative Precipitation Forecasts Using Short Training Data through Artificial Neural Networks

Abstract Conventional statistical postprocessing techniques offer limited ability to improve the skills of probabilistic guidance for heavy precipitation. This paper introduces two artificial neural network (ANN)-based, geographically aware, and computationally efficient postprocessing schemes, namely, the ANN-multiclass (ANN-Mclass) and the ANN–censored, shifted gamma distribution (ANN-CSGD). Both schemes are implemented to postprocess Global Ensemble Forecast System (GEFS) forecasts to produce probabilistic quantitative precipitation forecasts (PQPFs) over the contiguous United States (CONUS) using a short (60 days), rolling training window. The performances of these schemes are assessed through a set of hindcast experiments, wherein postprocessed 24-h PQPFs from the two ANN schemes were compared against those produced using the benchmark quantile mapping algorithm for lead times ranging from 1 to 8 days. Outcomes of the hindcast experiments show that ANN schemes overall outperform the benchmark as well as the raw forecast over the CONUS in predicting probability of precipitation over a range of thresholds. The relative performance varies among geographic regions, with the two ANN schemes broadly improving upon quantile mapping over the central, south, and southeast, and slightly underperforming along the Pacific coast where skills of raw forecasts are the highest. Between the two schemes, the hybrid ANN-CSGD outperforms at higher rainfall thresholds (i.e., >50 mm day−1), though the outperformance comes at a slight expense of sharpness and spatial specificity. Collectively, these results confirm the ability of the ANN algorithms to produce skillful PQPFs with a limited training window and point to the prowess of the hybrid scheme for calibrating PQPFs for rare-to-extreme rainfall events.

A novel hybrid artificial neural network - Parametric scheme for postprocessing medium-range precipitation forecasts

Many present-day statistical schemes for postprocessing weather forecasts, in particular precipitation forecasts, rely on calibration using prescribed statistical models to relate forecast statistics to distributional parameters. The efficacy of such schemes is often constrained not only by prescribed predictor-predictand relation, but also by arbitrary choices of temporal window and lead time range for training. To address this limitation, we propose an end-to-end, computationally efficient hybrid postprocessing scheme capable of producing full predictive distributions of precipitation accumulation without explicit stratification of forecast-observation pairs by forecast lead time and season. The proposed framework uses the censored, shifted gamma distribution (CSGD) as the predictive distribution but uses an artificial neural network (ANN) to estimate the distributional parameters of CSGD through a unified approach. This approach, referred to as ANN-CSGD, allows for simultaneous estimation of distributional parameters over multiple lead times and seasons in a single model by incorporating the latter variables as predictors to the ANN. We test our proposed ANN-CSGD model for postprocessing of ensemble mean forecasts of 24-h precipitation totals over selected river basins in California, at one- to seven-day lead times, from the Global Ensemble Forecast System (GEFS). The probabilistic quantitative precipitation forecasts (PQPFs) from the ANN-CSGD, are more skillful overall than those from the benchmark CSGD and the Mixed-type meta-Gaussian distribution (MMGD) models. The ANN-CSGD PQPFs highly improve the performance of those from CSGD in predicting the probability of precipitation (PoP) and are also much sharper and reliable at higher precipitation thresholds. We demonstrate how the hybrid approach, by using the entire available training data and its modified formulation, efficiently represents interactions between GEFS forecasts and season/lead times, thus leading to enhanced predictive performance.

The adaptability of typical precipitation ensemble prediction systems in the Huaihe River basin, China

Evaluating the adaptability of precipitation forecasting is of great importance for regional flood control and drought warnings. This study conducted evaluations using the 1–9 days cumulative precipitation forecast data of five typical operational global ensemble prediction systems (EPSs) from TIGGE (i.e., The Observing System Research and Predictability Experiment Interactive Grand Global Ensemble) and the observed daily precipitation data of 40 meteorological stations over the Huaihe River basin (HB). A series of verification metrics is used to evaluate the performances of quantitative precipitation forecasts (QPFs) and probabilistic quantitative precipitation forecasts (PQPFs) from the five EPSs from April to December 2015 in terms of overall performance, different precipitation thresholds, different lead times and the spatial distribution over the HB. The adaptability of the multimodel superensemble integrated from the five EPSs by the Bayesian model average is also examined during the main flood season. The results show that (1) the forecast quality of the China Meteorological Administration EPS is the worst for all lead times, which may relate to its having the fewest ensemble members. The European Centre for Medium-Range Weather Forecasts (ECMWF) EPS performs the best in terms of QPF and PQPF qualities for longer lead times because ECMWF has the largest ensemble members. (2) All EPSs have better discrimination at low thresholds, indicating the reference value for drought warnings. ECMWF is expected to obtain the best PQPF skill for a large threshold through postprocessing; (3) due to the differences in climates in the North and South of the basin, QPF and PQPF qualities are better in the northern HB than in the southern HB; (4) except for climate, the PQPF skill is also influenced by precipitation type, while the QPF accuracy is affected by terrain. The PQPF is good at forecasting the precipitation caused by ocean effects but not by mountain topography. The QPF accuracy decreases in mountainous areas; and (5) the multimodel superensemble has little effect on PQPF skill improvement but can improve QPF accuracy when raw EPSs have significantly different QPF accuracies.

Evaluation of an Application of Probabilistic Quantitative Precipitation Forecasts for Flood Forecasting

Probabilistic streamflow forecasts using precipitation derived from ensemble-based Probabilistic Quantitative Precipitation Forecasts (PQPFs) are examined. The PQPFs provide rainfall amounts associated with probabilities of exceedance for all grid points, which are averaged to the watershed scale for input to the operational Sacramento Soil Moisture Accounting hydrologic model to generate probabilistic streamflow predictions. The technique was tested using both the High-Resolution Rapid Refresh Ensemble (HRRRE) and the High-Resolution Ensemble Forecast version 2.0 (HREF) for 11 river basins across the upper Midwest for 109 cases. The resulting discharges associated with low probability of exceedance values were too large; no events were observed having discharges above the 10% exceedance value predicted from the technique applied to both ensembles, and no events were observed having discharges above the 25% exceedance value from the HREF-based forecast. The large differences are due to using the same precipitation exceedance value at all points; it is unlikely that all watershed points would experience the heavy rainfall associated with the 5% probability of exceedance. The technique likely can be improved through calibration of the basin-average precipitation forecasts based on typical distributions of precipitation within the convective systems that dominate warm-season precipitation events or calibration of the resulting probabilistic discharge forecasts.

A Nonhomogeneous Regression-Based Statistical Postprocessing Scheme for Generating Probabilistic Quantitative Precipitation Forecast

AbstractThis paper introduces a new, two-part scheme for postprocessing single-valued precipitation forecast to create probabilistic quantitative precipitation forecast (PQPF). This scheme, herein referred to as the mixed-type nonhomogeneous regression (MNHR), combines the use of logistic regression for estimating rainfall intermittency and nonhomogeneous regression for estimation of additional parameters of the conditional distribution. The performance of MNHR is evaluated relative to operational mixed-type meta-Gaussian distribution (MMGD) and the censored, shifted gamma distribution (CSGD) in postprocessing Global Ensemble Forecast System (GEFS) reforecasts averaged over 25 watersheds in the American River basin in California. The results point to superior performance of MNHR relative to MMGD and CSGD in terms of the skill of postprocessed PQPFs at 24- and 96-h accumulation windows. In addition, it is observed that the performance of CSGD tends to trail behind MNHR and MMGD at least for the 24-h window, though the performance differences tend to narrow at higher forecast amounts and longer lead times. Our analyses suggest that CSGD’s underperformance arises partly from its tendency to inflate the shift parameter estimates, which is pronounced over the study site possibly because of infrequent rainfall occurrence. By contrast, MNHR’s use of logistic regression helps avoid such bias, and its formulation of conditional distribution addresses the lack of skewness of MMGD for higher forecast amounts. Moreover, MHNR-based PQPF exhibits both superior calibration and relatively high sharpness at short lead times and on an unconditional sense, whereas it features lower sharpness relative to the other two suites when conditioned on higher forecast amount. This trade-off between calibration andconditionalsharpness warrants further research.

Science and Prediction of Heavy Rainfall over China: Research Progress since the Reform and Opening-Up of New China

This paper reviews the major progress on development of the science and prediction of heavy rainfall over China since the beginning of the reform and opening-up of new China (roughly between 1980 and 2019). The progress of research on the physical mechanisms of heavy rainfall over China is summarized from three perspectives: 1) the relevant synoptic weather systems, 2) heavy rainfall in major sub-regions of China, and 3) heavy rainfall induced by typhoons. The development and application of forecasting techniques for heavy rainfall are summarized in terms of numerical weather prediction techniques and objective forecasting methods. Greatly aided by the rapid progress in meteorological observing technology and substantial improvement in electronic computing, studies of heavy rainfall in China have advanced to investigating the evolution of heavy-rain-producing storms and observational analysis of the cloud microphysical features. A deeper and more systematic understanding of the synoptic systems of importance to the production of heavy rainfall has also been developed. Operational forecast of heavy rainfall in China has changed from subjective weather event forecasts to a combination of both subjective and objective quantitative precipitation forecasts, and is now advancing toward probabilistic quantitative precipitation forecasts with the provision of forecast uncertainty information.

Dependence of Probabilistic Quantitative Precipitation Forecast Performance on Typhoon Characteristics and Forecast Track Error in Taiwan

Abstract This study investigates the probabilistic quantitative precipitation forecast (PQPF) performance of typhoons that affected Taiwan during 2011–16. In this period, a total of 19 typhoons with a land warning issued by the Central Weather Bureau (CWB) are analyzed. The PQPF is calculated using the ensemble precipitation forecast data from the Taiwan Cooperative Precipitation Ensemble Forecast Experiment (TAPEX), and the verification data, verification thresholds, and typhoon characteristics are obtained from the CWB. The overall PQPF performance of TAPEX has an acceptable reliability and discrimination ability, and the higher probability error is distributed at the mountainous area of Taiwan. The PQPF performance is significantly influenced by typhoon characteristics (e.g., typhoon tracks, sizes, and forward speeds). The PQPFs for westward-moving, large, or slow typhoons have higher reliability and discrimination ability, and lower-probability error than those for northward-moving, small, or fast typhoons, except for similar reliability between fast and slow typhoons. Because northward-moving or small typhoons have larger forecast track error, and their PQPF performance is sensitive to the accuracy of the forecast track, a higher probability error occurs than that for westward-moving or large typhoons. Furthermore, because there is no difference in track error between fast and slow typhoons, the larger track spread for slow typhoons increases the rainfall forecast spread and reduces the probability error. The orientation of Taiwan’s topography and the topographic effect also influence and increase the distribution and value of probability error for northward-moving, small, or fast typhoons. In summary, forecast track characteristics are influenced by typhoon characteristics and further affect the PQPF performance.

Probabilistic Precipitation Forecast Postprocessing Using Quantile Mapping and Rank-Weighted Best-Member Dressing

Abstract Hamill et al. described a multimodel ensemble precipitation postprocessing algorithm that is used operationally by the U.S. National Weather Service (NWS). This article describes further changes that produce improved, reliable, and skillful probabilistic quantitative precipitation forecasts (PQPFs) for single or multimodel prediction systems. For multimodel systems, final probabilities are produced through the linear combination of PQPFs from the constituent models. The new methodology is applied to each prediction system. Prior to adjustment of the forecasts, parametric cumulative distribution functions (CDFs) of model and analyzed climatologies are generated using the previous 60 days’ forecasts and analyses and supplemental locations. The CDFs, which can be stored with minimal disk space, are then used for quantile mapping to correct state-dependent bias for each member. In this stage, the ensemble is also enlarged using a stencil of forecast values from the 5 × 5 surrounding grid points. Different weights and dressing distributions are assigned to the sorted, quantile-mapped members, with generally larger weights for outlying members and broader dressing distributions for members with heavier precipitation. Probability distributions are generated from the weighted sum of the dressing distributions. The NWS Global Ensemble Forecast System (GEFS), the Canadian Meteorological Centre (CMC) global ensemble, and the European Centre for Medium-Range Weather Forecasts (ECMWF) ensemble forecast data are postprocessed for April–June 2016. Single prediction system postprocessed forecasts are generally reliable and skillful. Multimodel PQPFs are roughly as skillful as the ECMWF system alone. Postprocessed guidance was generally more skillful than guidance using the Gamma distribution approach of Scheuerer and Hamill, with coefficients generated from data pooled across the United States.

Comparison of Probabilistic Quantitative Precipitation Forecasts from Two Postprocessing Mechanisms

Abstract This article compares the skill of medium-range probabilistic quantitative precipitation forecasts (PQPFs) generated via two postprocessing mechanisms: 1) the mixed-type meta-Gaussian distribution (MMGD) model and 2) the censored shifted Gamma distribution (CSGD) model. MMGD derives the PQPF by conditioning on the mean of raw ensemble forecasts. CSGD, on the other hand, is a regression-based mechanism that estimates PQPF from a prescribed distribution by adjusting the climatological distribution according to the mean, spread, and probability of precipitation (POP) of raw ensemble forecasts. Each mechanism is applied to the reforecast of the Global Ensemble Forecast System (GEFS) to yield a postprocessed PQPF over lead times between 24 and 72 h. The outcome of an evaluation experiment over the mid-Atlantic region of the United States indicates that the CSGD approach broadly outperforms the MMGD in terms of both the ensemble mean and the reliability of distribution, although the performance gap tends to be narrow, and at times mixed, at higher precipitation thresholds (>5 mm). Analysis of a rare storm event demonstrates the superior reliability and sharpness of the CSGD PQPF and underscores the issue of overforecasting by the MMGD PQPF. This work suggests that the CSGD’s incorporation of ensemble spread and POP does help enhance its skill, particularly for light forecast amounts, but CSGD’s model structure and its use of optimization in parameter estimation likely play a more determining role in its outperformance.

Eastern U.S. Verification of Ensemble Precipitation Forecasts

Abstract The quality of ensemble precipitation forecasts across the eastern United States is investigated, specifically, version 2 of the National Centers for Environmental Prediction (NCEP) Global Ensemble Forecast System Reforecast (GEFSRv2) and Short Range Ensemble Forecast (SREF) system, as well as NCEP’s Weather Prediction Center probabilistic quantitative precipitation forecast (WPC-PQPF) guidance. The forecasts are verified using multisensor precipitation estimates and various metrics conditioned upon seasonality, precipitation threshold, lead time, and spatial aggregation scale. The forecasts are verified, over the geographic domain of each of the four eastern River Forecasts Centers (RFCs) in the United States, by considering first 1) the three systems or guidance, using a common period of analysis (2012–13) for lead times from 1 to 3 days, and then 2) GEFSRv2 alone, using a longer period (2004–13) and lead times from 1 to 16 days. The verification results indicate that, across the eastern United States, precipitation forecast bias decreases and the skill and reliability improve as the spatial aggregation scale increases; however, all the forecasts exhibit some underforecasting bias. The skill of the forecasts is appreciably better in the cool season than in the warm one. The WPC-PQPFs tend to be superior, in terms of the correlation coefficient, relative mean error, reliability, and forecast skill scores, than both GEFSRv2 and SREF, but the performance varies with the RFC and lead time. Based on GEFSRv2, medium-range precipitation forecasts tend to have skill up to approximately day 7 relative to sampled climatology.

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.

Short-Term Probabilistic Forecasts of the 31 May 2013 Oklahoma Tornado and Flash Flood Event Using a Continuous-Update-Cycle Storm-Scale Ensemble System

AbstractA continuous-update-cycle storm-scale ensemble data assimilation (DA) and prediction system using the ARW model and DART software is used to generate retrospective 0–6-h ensemble forecasts of the 31 May 2013 tornado and flash flood event over central Oklahoma, with a focus on the prediction of heavy rainfall. Results indicate that the model-predicted probabilities of strong low-level mesocyclones correspond well with the locations of observed mesocyclones and with the observed damage track. The ensemble-mean quantitative precipitation forecast (QPF) from the radar DA experiments match NCEP’s stage IV analyses reasonably well in terms of location and amount of rainfall, particularly during the 0–3-h forecast period. In contrast, significant displacement errors and lower rainfall totals are evident in a control experiment that withholds radar data during the DA. The ensemble-derived probabilistic QPF (PQPF) from the radar DA experiment is more skillful than the PQPF from the no_radar experiment, based on visual inspection and probabilistic verification metrics. A novel object-based storm-tracking algorithm provides additional insight, suggesting that explicit assimilation and 1–2-h prediction of the dominant supercell is remarkably skillful in the radar experiment. The skill in both experiments is substantially higher during the 0–3-h forecast period than in the 3–6-h period. Furthermore, the difference in skill between the two forecasts decreases sharply during the latter period, indicating that the impact of radar DA is greatest during early forecast hours. Overall, the results demonstrate the potential for a frequently updated, high-resolution ensemble system to extend probabilistic low-level mesocyclone and flash flood forecast lead times and improve accuracy of convective precipitation nowcasting.

Bayesian Model Averaging with Stratified Sampling for Probabilistic Quantitative Precipitation Forecasting in Northern China during Summer 2010

Abstract To study the impact of training sample heterogeneity on the performance of Bayesian model averaging (BMA), two BMA experiments were performed on probabilistic quantitative precipitation forecasts (PQPFs) in the northern China region in July and August of 2010 generated from an 11-member short-range ensemble forecasting system. One experiment, as in many conventional BMA studies, used an overall training sample that consisted of all available cases in the training period, while the second experiment used stratified sampling BMA by first dividing all available training cases into subsamples according to their ensemble spread, and then performing BMA on each subsample. The results showed that ensemble spread is a good criterion to divide ensemble precipitation cases into subsamples, and that the subsamples have different statistical properties. Pooling the subsamples together forms a heterogeneous overall sample. Conventional BMA is incapable of interpreting heterogeneous samples, and produces unreliable PQPF. It underestimates the forecast probability at high-threshold PQPF and local rainfall maxima in BMA percentile forecasts. BMA with stratified sampling according to ensemble spread overcomes the problem reasonably well, producing sharper predictive probability density functions and BMA percentile forecasts, and more reliable PQPF than the conventional BMA approach. The continuous ranked probability scores, Brier skill scores, and reliability diagrams of the two BMA experiments were examined for all available forecast days, along with a logistic regression experiment. Stratified sampling BMA outperformed the raw ensemble and conventional BMA in all verifications, and also showed better skill than logistic regression in low-threshold forecasts.

Analysis of the Relative Operating Characteristic and Economic Value Using the LAPS Ensemble Prediction System in Taiwan

Abstract Measurement of the usefulness of numerical weather prediction considers not only the forecast quality but also the possible economic value (EV) in the daily decision-making process of users. Discrimination ability of an ensemble prediction system (EPS) can be assessed by the relative operating characteristic (ROC), which is closely related to the EV provided by the same forecast system. Focusing on short-range probabilistic quantitative precipitation forecasts (PQPFs) for typhoons, this study demonstrates the consistent and strongly related characteristics of ROC and EV based on the Local Analysis and Prediction System (LAPS) EPS operated at the Central Weather Bureau in Taiwan. Sensitivity experiments including the effect of terrain, calibration, and forecast uncertainties on ROC and EV show that the potential EV provided by a forecast system is mainly determined by the discrimination ability of the same system. The ROC and maximum EV (EVmax) of an EPS are insensitive to calibration, but the optimal probability threshold to achieve the EVmax becomes more reliable after calibration. In addition, the LAPS ensemble probabilistic forecasts outperform deterministic forecasts in respect to both ROC and EV, and such an advantage grows with increasing precipitation intensity. Also, even without explicitly knowing the cost–loss ratio, one can still optimize decision-making and obtain the EVmax by using ensemble probabilistic forecasts.

Verification of Quantitative Precipitation Reforecasts over the Southeastern United States

Abstract NOAA’s second-generation reforecasts are approximately consistent with the operational version of the 2012 NOAA Global Ensemble Forecast System (GEFS). The reforecasts allow verification to be performed across a multidecadal time period using a static model, in contrast to verifications performed using an ever-evolving operational modeling system. This contribution examines three commonly used verification metrics for reforecasts of precipitation over the southeastern United States: equitable threat score, bias, and ranked probability skill score. Analysis of the verification metrics highlights the variation in the ability of the GEFS to predict precipitation across amount, season, forecast lead time, and location. Beyond day 5.5, there is little useful skill in quantitative precipitation forecasts (QPFs) or probabilistic QPFs. For lighter precipitation thresholds [e.g., 5 and 10 mm (24 h)−1], use of the ensemble mean adds about 10% to the forecast skill over the deterministic control. QPFs have increased in accuracy from 1985 to 2013, likely due to improvements in observations. Results of this investigation are a first step toward using the reforecast database to distinguish weather regimes that the GEFS typically predicts well from those regimes that the GEFS typically predicts poorly.

Machine Learning Enhancement of Storm-Scale Ensemble Probabilistic Quantitative Precipitation Forecasts

Abstract Probabilistic quantitative precipitation forecasts challenge meteorologists due to the wide variability of precipitation amounts over small areas and their dependence on conditions at multiple spatial and temporal scales. Ensembles of convection-allowing numerical weather prediction models offer a way to produce improved precipitation forecasts and estimates of the forecast uncertainty. These models allow for the prediction of individual convective storms on the model grid, but they often displace the storms in space, time, and intensity, which results in added uncertainty. Machine learning methods can produce calibrated probabilistic forecasts from the raw ensemble data that correct for systemic biases in the ensemble precipitation forecast and incorporate additional uncertainty information from aggregations of the ensemble members and additional model variables. This study utilizes the 2010 Center for Analysis and Prediction of Storms Storm-Scale Ensemble Forecast system and the National Severe Storms Laboratory National Mosaic & Multi-Sensor Quantitative Precipitation Estimate as input data for training logistic regressions and random forests to produce a calibrated probabilistic quantitative precipitation forecast. The reliability and discrimination of the forecasts are compared through verification statistics and a case study.