Convection-Allowing Ensemble Forecasts Research Articles

Using machine learning to predict convection-allowing ensemble forecast skill: Evaluation with the NSSL Warn-on-Forecast System

Abstract Forecasters routinely calibrate their confidence in model forecasts. Ensembles inherently estimate forecast confidence, but are often underdispersive, and ensemble spread does not strongly correlate with ensemble-mean error. The misalignment between ensemble spread and skill motivates new methods for “forecasting forecast skill” so that forecasters can better utilize ensemble guidance. We have trained logistic regression and random forest models to predict the skill of composite reflectivity forecasts from the NSSL Warn-on-Forecast System (WoFS), a 3-km ensemble that generates rapidly updating forecast guidance for 0-6-h lead times. The forecast skill predictions are valid at 1-h, 2-h, or 3-h lead times within localized regions determined by the observed storm locations at analysis time. We use WoFS analysis and forecast output and NSSL Multi-Radar / Multi-Sensor composite reflectivity for 106 cases from the 2017-2021 NOAA Hazardous Weather Testbed Spring Forecasting Experiments. We frame the prediction task as a multi-classification problem, where the forecast skill labels are determined by averaging the extended Fractions Skill Scores (eFSS) for several reflectivity thresholds and verification neighborhoods, then converting to one of three classes based on where the average eFSS ranks within the entire dataset: POOR (bottom 20%), FAIR (middle 60%), or GOOD (top 20%). Initial machine learning (ML) models are trained on 323 predictors; reducing to 10 or 15 predictors in the final models only modestly reduces skill. The final models substantially outperform carefully developed persistence- and spread-based models, and are reasonably explainable. The results suggest that ML can be a valuable tool for guiding user confidence in convection-allowing (and larger-scale) ensemble forecasts.

Impacts of initial condition perturbation blending in 10- and 40-member convection-allowing ensemble forecasts

Abstract A series of convection-allowing 36-hour ensemble forecasts during the 2018 Spring season are used to better understand the impacts of ensemble configuration and blending different sources of initial condition (IC) perturbation. Ten- and 40-member ensemble configurations are initialized with the multi-scale IC perturbations generated as a product of convective-scale data assimilation (MULTI), and initialized with the MULTI IC perturbations blended with IC perturbations downscaled from coarser resolution ensembles (BLEND). The forecast performance of both precipitation and non-precipitation variables is consistently improved by the larger ensemble size. The benefit of the larger ensemble is largely, but not entirely, due to compensating for under-dispersion in the fixed-physics ensemble configuration. A consistent improvement in precipitation forecast skill results from blending in the 10-member ensemble configuration, corresponding to a reduction in the ensemble calibration error (i.e., reliability component of Brier Score). In the 40-member ensemble configuration, the advantage of blending is limited to the ∼18-22 hour lead times at all precipitation thresholds, and the ∼35-36 hour lead times at the lowest threshold, both corresponding to an improved resolution component of the Brier Score. The advantage of blending in the 40-member ensemble during the diurnal convection maximum of ∼18-22 hour lead times is primarily due to cases with relatively weak synoptic scale forcing while advantages at later lead times beyond ∼30 hours lead time are most prominent on cases with relatively strong synoptic scale forcing. The impacts of blending and ensemble configuration on forecasts of non-precipitation variables is generally consistent with the impacts on the precipitation forecasts.

The Lateral Boundary Perturbations Growth and Their Dependence on the Forcing Types of Severe Convection in Convection-Allowing Ensemble Forecasts

The application of lateral boundary perturbations (LBPs) helps to restore dispersion in convection-allowing ensemble forecasts (CAEFs). However, the applicability of LBPs remains unclear because of the differences between convection systems. Short-range (24 h) ensemble forecasts are carried out to explore this issue with a strong-forcing (SF) case and a weak-forcing (WF) case in East China. The dependence of LBPs on the forcing types of severe convection is investigated regarding the forecast error growth caused by the lateral boundary conditions (LBCs). The results show that the LBPs mainly influence the SF case rather than the WF case, especially after a 12-h forecast. The large-scale errors dominate in the SF case because the change in the synoptic-scale system affects the forecast error evolution. In contrast, the large-scale errors are mainly derived from the upscaling of the small-scale errors in the WF case, indicating that using LBPs is only insufficient in such a case. In sensitivity experiments that vary the magnitude of LBPs from 10% to 150% of its original value, CAEFs demonstrate more sensitive to LBPs in the SF case than in the WF case, indicating that the WF case has intrinsically limited predictability. Overall, LBPs are more suitable for the SF case, while additional perturbations from other sources are required for CAEFs in the WF case because of the limits of intrinsic predictability.

Short-Term Convection-Allowing Ensemble Precipitation Forecast Sensitivity to Resolution of Initial Condition Perturbations and Central Initial States

Abstract Nine sets of 36-h, 10-member, convection-allowing ensemble (CAE) forecasts with 3-km horizontal grid spacing were produced over the conterminous United States for a 4-week period. These CAEs had identical configurations except for their initial conditions (ICs), which were constructed to isolate CAE forecast sensitivity to resolution of IC perturbations and central initial states about which IC perturbations were centered. The IC perturbations and central initial states were provided by limited-area ensemble Kalman filter (EnKF) analyses with both 15- and 3-km horizontal grid spacings, as well as from NCEP’s Global Forecast System (GFS) and Global Ensemble Forecast System. Given fixed-resolution IC perturbations, reducing horizontal grid spacing of central initial states improved ∼1–12-h precipitation forecasts. Conversely, for constant-resolution central initial states, reducing horizontal grid spacing of IC perturbations led to comparatively smaller short-term forecast improvements or none at all. Overall, all CAEs initially centered on 3-km EnKF mean analyses produced objectively better ∼1–12-h precipitation forecasts than CAEs initially centered on GFS or 15-km EnKF mean analyses regardless of IC perturbation resolution, strongly suggesting it is more important for central initial states to possess fine-scale structures than IC perturbations for short-term CAE forecasting applications, although fine-scale perturbations could potentially be critical for data assimilation purposes. These findings have important implications for future operational CAE forecast systems and suggest CAE IC development efforts focus on producing the best possible high-resolution deterministic analyses that can serve as central initial states for CAEs. Significance Statement Ensembles of weather model forecasts are composed of different “members” that, when combined, can produce probabilities that specific weather events will occur. Ensemble forecasts begin from specified atmospheric states, called initial conditions. For ensembles where initial conditions differ across members, the initial conditions can be viewed as a set of small perturbations added to a central state provided by a single model field. Our study suggests it is more important to increase horizontal resolution of the central state than resolution of the perturbations when initializing ensemble forecasts with 3-km horizontal grid spacing. These findings suggest a potential for computational savings and a streamlined process for improving high-resolution ensemble initial conditions.

Sensitivity of Forecast Uncertainty to Different Microphysics Schemes within a Convection-Allowing Ensemble during SoWMEX-IOP8

Abstract A mesoscale convective system that occurred in southwestern Taiwan on 15 June 2008 is simulated using convection-allowing ensemble forecasts to investigate the forecast uncertainty associated with four microphysics schemes—the Goddard Cumulus Ensemble (GCE), Morrison (MOR), WRF single-moment 6-class (WSM6), and WRF double-moment 6-class (WDM6) schemes. First, the essential features of the convective structure, hydrometeor distribution, and microphysical tendencies for the different microphysics schemes are presented through deterministic forecasts. Second, ensemble forecasts with the same initial conditions are employed to estimate the forecast uncertainty produced by the different ensembles with the fixed microphysics scheme. GCE has the largest spread in most state variables due to its most efficient phase conversion between water species. By contrast, MOR results in the least spread. WSM6 and WDM6 have similar vertical spread structures due to their similar ice-phase formulas. However, WDM6 produces more ensemble spread than WSM6 does below the melting layer, resulting from its double-moment treatment of warm rain processes. The model simulations with the four microphysics schemes demonstrate upscale error growth through spectrum analysis of the root-mean difference total energy (RMDTE). The RMDTE results reveal that the GCE and WDM6 schemes are more sensitive to initial condition uncertainty, whereas the MOR and WSM6 schemes are relatively less sensitive to that for this event. Overall, the diabatic heating–cooling processes connect the convective-scale cloud microphysical processes to the large-scale dynamical and thermodynamical fields, and they significantly affect the forecast error signatures in the multiscale weather system.

Error Growth Dynamics within Convection-Allowing Ensemble Forecasts over Central U.S. Regions for Days of Active Convection

AbstractError growth is investigated based on convection-allowing ensemble forecasts starting from 0000 UTC for 13 active convection events over central to eastern U.S. regions from spring 2018. The analysis domain is divided into the northwest (NW), northeast (NE), southeast (SE), and southwest (SW) quadrants (subregions). Total difference energy and its decompositions are used to measure and analyze error growth at and across scales. Special attention is paid to the dominant types of convection with respect to their forcing mechanisms in the four subregions and the associated difference in precipitation diurnal cycles. The discussions on the average behaviors of error growth in each region are supplemented by four representative cases. Results show that the meso-γ-scale error growth is directly linked to precipitation diurnal cycle while meso-α-scale error growth has a strong link to large-scale forcing. Upscale error growth is evident in all regions/cases but up-amplitude growth within its own scale plays different roles in different regions/cases. When large-scale flow is important (as in the NE region), precipitation is strongly modulated by the large-scale forcing and becomes more organized with time, and upscale transfer of forecast error is stronger. On the other hand, when local instability plays more dominant roles (as in the SE region), precipitation is overall less organized and has the weakest diurnal variations. Its associated errors at the γ and β scale can reach their peaks sooner and meso-α-scale error tends to rely more on growth of error with its own scale. Small-scale forecast errors are directly impacted by convective activities and have a short response time to convection while increasingly larger-scale errors have longer response times and delayed phase within the diurnal cycle.

Initial Conditions for Convection-Allowing Ensembles over the Conterminous United States

AbstractFive sets of 48-h, 10-member, convection-allowing ensemble (CAE) forecasts with 3-km horizontal grid spacing were systematically evaluated over the conterminous United States with a focus on precipitation across 31 cases. The various CAEs solely differed by their initial condition perturbations (ICPs) and central initial states. CAEs initially centered about deterministic Global Forecast System (GFS) analyses were unequivocally better than those initially centered about ensemble mean analyses produced by a limited-area single-physics, single-dynamics 15-km continuously cycling ensemble Kalman filter (EnKF), strongly suggesting relative superiority of the GFS analyses. Additionally, CAEs with flow-dependent ICPs derived from either the EnKF or multimodel 3-h forecasts from the Short-Range Ensemble Forecast (SREF) system had higher fractions skill scores than CAEs with randomly generated mesoscale ICPs. Conversely, due to insufficient spread, CAEs with EnKF ICPs had worse reliability, discrimination, and dispersion than those with random and SREF ICPs. However, members in the CAE with SREF ICPs undesirably clustered by dynamic core represented in the ICPs, and CAEs with random ICPs had poor spinup characteristics. Collectively, these results indicate that continuously cycled EnKF mean analyses were suboptimal for CAE initialization purposes and suggest that further work to improve limited-area continuously cycling EnKFs over large regional domains is warranted. Additionally, the deleterious aspects of using both multimodel and random ICPs suggest efforts toward improving spread in CAEs with single-physics, single-dynamics, flow-dependent ICPs should continue.

Understanding the Predictability within Convection-Allowing Ensemble Forecasts in East China: Meteorological Sensitivity, Forecast Error Growth and Associated Precipitation Uncertainties Across Spatial Scales

This study investigates the practical predictability of two simulated mesoscale convective systems (MCS1 and MCS2) within a state-of-the-art convection-allowing ensemble forecast system. The two MCSs are both controlled by the synoptic Meiyu-front but differ in mesoscale orographic forcing. An observation system simulation experiment (OSSE) setup is first built, which includes flow-dependent multiple-scale initial and lateral boundary perturbations and a 12 h 30-member ensemble forecast is thereby created. In combination with the difference total energy, the decorrelation scale and the ensemble sensitivity analysis, both forecast error evolution, precipitation uncertainties and meteorological sensitivity that describe the practical predictability are assessed. The results show large variabilities of precipitation forecasts among ensemble members, indicative of the practical predictability limit. The study of forecast error evolution shows that the error energy in the MCS1 region in which the convection is blocked by the Dabie Mountains exhibits a simultaneous peak pattern for all spatial scales at around 6 h due to strong moist convection. On the other hand, when large-scale flow plays a more important role, the forecast error energy in the MCS2 region exhibits a stepwise increase with increasing spatial scale. As a result of error energy growth, the precipitation uncertainties evolve from small scales and gradually transfer to larger scales, implying a strong relationship between error growth and precipitation across spatial scales, thus explaining the great precipitation variability within ensemble members. These results suggest the additional forcing brought by the Dabie Mountains could regulate the predictability of Meiyu-frontal convection, which calls for a targeted perturbation design in convection-allowing ensemble forecast systems with respect to different forcing mechanisms.

A Comparison of Methods to Sample Model Errors for Convection-Allowing Ensemble Forecasts in the Setting of Multiscale Initial Conditions Produced by the GSI-Based EnVar Assimilation System

Abstract A gridpoint statistical interpolation (GSI)-based hybrid ensemble–variational (EnVar) scheme was extended for convective scales—including radar reflectivity assimilation—and implemented in real-time spring forecasting experiments. This study compares methods to address model error during the forecast under the context of multiscale initial condition error sampling provided by the EnVar system. A total of 10 retrospective cases were used to explore the optimal design of convection-allowing ensemble forecasts. In addition to single-model single-physics (SMSP) configurations, ensemble forecast experiments compared multimodel (MM) and multiphysics (MP) approaches. Stochastic physics was also applied to MP for further comparison. Neighborhood-based verification of precipitation and composite reflectivity showed each of these model error techniques to be superior to SMSP configurations. Comparisons of MM and MP approaches had mixed findings. The MM approach had better overall skill in heavy-precipitation forecasts; however, MP ensembles had better skill for light (2.54 mm) precipitation and reduced ensemble mean error of other diagnostic fields, particularly near the surface. The MM experiment had the largest spread in precipitation, and for most hours in other fields; however, rank histograms and spaghetti contours showed significant clustering of the ensemble distribution. MP plus stochastic physics was able to significantly increase spread with time to be competitive with MM by the end of the forecast. The results generally suggest that an MM approach is best for early forecast lead times up to 6–12 h, while a combination of MP and stochastic physics approaches is preferred for forecasts beyond 6–12 h.

Medium-Range Convection-Allowing Ensemble Forecasts with a Variable-Resolution Global Model

Abstract Two sets of global, 132-h (5.5-day), 10-member ensemble forecasts were produced with the Model for Prediction Across Scales (MPAS) for 35 cases in April and May 2017. One MPAS ensemble had a quasi-uniform 15-km mesh while the other employed a variable-resolution mesh with 3-km cell spacing over the conterminous United States (CONUS) that smoothly relaxed to 15 km over the rest of the globe. Precipitation forecasts from both MPAS ensembles were objectively verified over the central and eastern CONUS to assess the potential benefits of configuring MPAS with a 3-km mesh refinement region for medium-range forecasts. In addition, forecasts from NCEP’s operational Global Ensemble Forecast System were evaluated and served as a baseline against which to compare the experimental MPAS ensembles. The 3-km MPAS ensemble most faithfully reproduced the observed diurnal cycle of precipitation throughout the 132-h forecasts and had superior precipitation skill and reliability over the first 48 h. However, after 48 h, the three ensembles had more similar spread, reliability, and skill, and differences between probabilistic precipitation forecasts derived from the 3- and 15-km MPAS ensembles were typically statistically insignificant. Nonetheless, despite fewer benefits of increased resolution for spatial placement after 48 h, 3-km ensemble members explicitly provided potentially valuable guidance regarding convective mode throughout the 132-h forecasts while the other ensembles did not. Collectively, these results suggest both strengths and limitations of medium-range high-resolution ensemble forecasts and reveal pathways for future investigations to improve understanding of high-resolution global ensembles with variable-resolution meshes.

Assessing the Skill of Convection-Allowing Ensemble Forecasts of Precipitation by Optimization of Spatial-Temporal Neighborhoods

The current neighborhood probability (NP) method mainly considers the spatial displacement error in high-resolution precipitation forecasts, but the problem of the forecast time exceeding or lagging behind the observed field has not been properly solved. Therefore, a temporal factor was introduced into the NP method in this paper, and precipitation forecasts were evaluated in different spatial-temporal neighborhoods based on the improved NP method and fractions skill score (FSS), combined with the relative operating characteristic (ROC) curve. The results indicated that the forecasting accuracy of the ensemble forecast was higher than the control forecast. The neighborhood ensemble probability (NEP) and probability matched mean (PMM) methods were superior to the traditional ensemble mean (EM) method in forecasting heavy rainfall, which compensated for the limitations of the heavy rainfall forecasting ability of EM. For such squall line processes, a spatial scale of 15–45 km neighborhood radius could effectively rectify the displacement error of precipitation. There was a corresponding relationship between temporal scale and rainfall intensity in convective-scale precipitation forecast, so the temporal uncertainty of different levels of precipitation could be captured by different temporal scales. The spatial and temporal scales had interdependent influences on precipitation forecast effects, which could be affected by the intrinsic spatial-temporal scale of convective-scale weather systems. The improved NP method could simultaneously reflect the spatial and temporal uncertainties of convective-scale precipitation forecasts in high-resolution models, achieving a comprehensive assessment of spatial-temporal scale and providing probabilistic forecast results that match different levels of precipitation.

Prediction of Lake-Effect Snow Using Convection-Allowing Ensemble Forecasts and Regional Data Assimilation

Abstract Lake-effect snow (LES) is a cold-season mesoscale convective phenomenon that can lead to significant snowfall rates and accumulations in the Great Lakes region of the United States. While limited-area numerical weather prediction models have shown skill in prediction of warm-season convective storms, forecasting the sharp nature of LES precipitation timing, intensity, and location is difficult because of model error and initial and boundary condition uncertainties. Ensemble forecasting can incorporate and quantify some sources of forecast error, but ensemble design must be considered. This study examines the relative contributions of forecast uncertainties to LES forecast error using a regional convection-allowing data assimilation and ensemble prediction system. Ensembles are developed using various methods of perturbations to simulate a long-lived and high-precipitation LES event in December 2013, and forecast performance is evaluated using observations including those from the Ontario Winter Lake-Effect Systems (OWLeS) campaign. Model lateral boundary conditions corresponding to weather conditions beyond the Great Lakes region play an influential role in LES precipitation forecasts and their uncertainty, as evidenced by ensemble spread, particularly at lead times beyond one day. A strong forecast dependence on regional initial conditions was shown using data assimilation. This sensitivity impacts the timing and intensity of predicted precipitation, as well as band location and orientation assessed with an object-based verification approach, giving insight into the time scales of practical predictability of LES. Overall, an assimilation-cycling convection-allowing ensemble prediction system could improve future lake-effect snow precipitation forecasts and analyses and can help quantify and understand sources of forecast uncertainty.

Using Convection-Allowing Ensembles to Understand the Predictability of an Extreme Rainfall Event

AbstractThis research uses convection-allowing ensemble forecasts to address aspects of the predictability of an extreme rainfall event that occurred in south-central Texas on 25 May 2013, which was poorly predicted by operational and experimental numerical models and caused a flash flood in San Antonio that resulted in three fatalities. Most members of the ensemble had large errors in the location and magnitude of the heavy rainfall, but one member approximately reproduced the observed rainfall distribution. On a regional scale a flow-dependent diurnal cycle in ensemble spread growth is observed with large growth associated with afternoon convection, but the growth rate then reduced after convection dissipates the next morning rather than continuing to grow. Experiments that vary the magnitude of the perturbations to the initial and lateral boundary conditions reveal flow dependencies on the scales responsible for the ensemble growth and the degree to which practical (i.e., deficiencies in observing systems and numerical models) and intrinsic predictability limits (i.e., moist convective dynamic error growth) affect a particular convective event. Specifically, it was found that large-scale atmospheric forcing tends to dominate the ensemble spread evolution, but small-scale error growth can be of near-equal importance in certain convective scenarios where interaction across scales is prevalent and essential to the local precipitation processes. In a similar manner, aspects of the “upscale error growth” and “downscale error cascade” conceptual models are seen in the experiments, but neither completely explains the spread characteristics seen in the simulations.

Predictability and Dynamics of Warm-Core Mesoscale Vortex Formation with the 8 May 2009 “Super Derecho” Event

The predictability and dynamics of the warm-core mesovortex associated with the northern flank of the 8 May 2009 “super derecho” event are examined by coupling the Advanced Research Weather Research and Forecasting Model with the ensemble adjustment Kalman filter implementation within the Data Assimilation Research Testbed facility. Cycled analysis started at 1200 UTC 2 May 2009, with observations assimilated every 6 h until 1200 UTC 7 May 2009, at which time a 50-member ensemble of 36-h convection-allowing ensemble forecasts were launched. The ensemble forecasts all simulated a mesoscale convective system, but only 7 out of 50 members produced a warm-core mesovortex-like feature similar in intensity to the observed mesovortex. Ensemble sensitivity and composite analyses were conducted to analyze the environmental differences between ensemble members. A more amplified upstream upper-level trough near the time of observed convection initiation is associated with a stronger simulated mesovortex. The amplification of the trough results in increases in the magnitudes of the low-level jet and thermal gradient. Consequently, more moisture is transported poleward into western Kansas, leading to earlier convection initiation in ensemble members with the strongest mesovortices. A circulation budget is performed on the ensemble members with the strongest (member 10) and weakest (member 5) time-averaged circulations. The ascending front-to-rear flow, descending rear-to-front flow, and divergent low-level flow of an MCS are more prominent in member 10, which is hypothesized to allow for the convergence of more background cyclonic absolute vorticity and, thus, facilitating the development of a stronger mesovortex.

Impact of a Stochastic Kinetic Energy Backscatter Scheme on Warm Season Convection-Allowing Ensemble Forecasts

The efficacy of a stochastic kinetic energy backscatter (SKEB) scheme to improve convection-allowing probabilistic forecasts was studied. While SKEB has been explored for coarse, convection-parameterizing models, studies of SKEB for convective scales are limited. Three ensembles were compared. The SKMP ensemble used mixed physics with the SKEB scheme, whereas the MP ensemble was configured identically but without using the SKEB scheme. The SK ensemble used the SKEB scheme with no physics diversity. The experiment covered May 2013 over the central United States on a 4-km Weather Research and Forecasting (WRF) Model domain. The SKEB scheme was successful in increasing the spread in all fields verified, especially mid- and upper-tropospheric fields. Additionally, the rmse of the ensemble mean was maintained or reduced, in some cases significantly. Rank histograms in the SKMP ensemble were flatter than those in the MP ensemble, indicating the SKEB scheme produces a less underdispersive forecast distribution. Some improvement was seen in probabilistic precipitation forecasts, particularly when examining Brier scores. Verification against surface observations agree with verification against Rapid Refresh (RAP) model analyses, showing that probabilistic forecasts for 2-m temperature, 2-m dewpoint, and 10-m winds were also improved using the SKEB scheme. The SK ensemble gave competitive forecasts for some fields. The SK ensemble had reduced spread compared to the MP ensemble at the surface due to the lack of physics diversity. These results suggest the potential utility of mixed physics plus the SKEB scheme in the design of convection-allowing ensemble forecasts.

Severe Weather Prediction Using Storm Surrogates from an Ensemble Forecasting System

Abstract Probabilistic severe weather forecasts for days 1 and 2 were produced using 30-member convection-allowing ensemble forecasts initialized by an ensemble Kalman filter data assimilation system during a 32-day period coinciding with the Mesoscale Predictability Experiment. The forecasts were generated by smoothing the locations where model output indicated extreme values of updraft helicity, a surrogate for rotating thunderstorms in model output. The day 1 surrogate severe probability forecasts (SSPFs) produced skillful and reliable predictions of severe weather during this period, after an appropriate calibration of the smoothing kernel. The ensemble SSPFs exceeded the skill of SSPFs derived from two benchmark deterministic forecasts, with the largest differences occurring on the mesoscale, while all SSPFs produced similar forecasts on synoptic scales. While the deterministic SSPFs often overforecasted high probabilities, the ensemble improved the reliability of these probabilities, at the expense of producing fewer high-probability values. For the day 2 period, the SSPFs provided competitive guidance compared to the day 1 forecasts, although additional smoothing was needed to produce the same level of skill, reducing the forecast sharpness. Results were similar using 10 ensemble members, suggesting value exists when running a smaller ensemble if computational resources are limited. Finally, the SSPFs were compared to severe weather risk areas identified in Storm Prediction Center (SPC) convective outlooks. The SSPF skill was comparable to the SPC outlook skill in identifying regions where severe weather would occur, although performance varied on a day-to-day basis.

A Real-Time Convection-Allowing Ensemble Prediction System Initialized by Mesoscale Ensemble Kalman Filter Analyses

Abstract In May and June 2013, the National Center for Atmospheric Research produced real-time 48-h convection-allowing ensemble forecasts at 3-km horizontal grid spacing using the Weather Research and Forecasting (WRF) Model in support of the Mesoscale Predictability Experiment field program. The ensemble forecasts were initialized twice daily at 0000 and 1200 UTC from analysis members of a continuously cycling, limited-area, mesoscale (15 km) ensemble Kalman filter (EnKF) data assimilation system and evaluated with a focus on precipitation and severe weather guidance. Deterministic WRF Model forecasts initialized from GFS analyses were also examined. Subjectively, the ensemble forecasts often produced areas of intense convection over regions where severe weather was observed. Objective statistics confirmed these subjective impressions and indicated that the ensemble was skillful at predicting precipitation and severe weather events. Forecasts initialized at 1200 UTC were more skillful regarding precipitation and severe weather placement than forecasts initialized 12 h earlier at 0000 UTC, and the ensemble forecasts were typically more skillful than GFS-initialized forecasts. At times, 0000 UTC GFS-initialized forecasts had temporal distributions of domain-average rainfall closer to observations than EnKF-initialized forecasts. However, particularly when GFS analyses initialized WRF Model forecasts, 1200 UTC forecasts produced more rainfall during the first diurnal maximum than 0000 UTC forecasts. This behavior was mostly attributed to WRF Model initialization of clouds and moist physical processes. The success of these real-time ensemble forecasts demonstrates the feasibility of using limited-area continuously cycling EnKFs as a method to initialize convection-allowing ensemble forecasts, and future real-time high-resolution ensemble development leveraging EnKFs seems justified.