European Centre For Medium-Range Weather Forecasts Ensemble Research Articles

Physics-Based vs Data-Driven 24-Hour Probabilistic Forecasts of Precipitation for Northern Tropical Africa

Abstract Numerical weather prediction (NWP) models struggle to skillfully predict tropical precipitation occurrence and amount, calling for alternative approaches. For instance, it has been shown that fairly simple, purely data-driven logistic regression models for 24-h precipitation occurrence outperform both climatological and NWP forecasts for the West African summer monsoon. More complex neural network–based approaches, however, remain underdeveloped due to the non-Gaussian character of precipitation. In this study, we develop, apply, and evaluate a novel two-stage approach, where we train a U-Net convolutional neural network (CNN) model on gridded rainfall data to obtain a deterministic forecast and then apply the recently developed, nonparametric Easy Uncertainty Quantification (EasyUQ) approach to convert it into a probabilistic forecast. We evaluate CNN+EasyUQ for 1-day-ahead 24-h accumulated precipitation forecasts over northern tropical Africa for 2011–19, with the Integrated Multi-satellitE Retrievals for GPM (IMERG) data serving as ground truth. In the most comprehensive assessment to date, we compare CNN+EasyUQ to state-of-the-art physics-based and data-driven approaches such as monthly probabilistic climatology, raw and postprocessed ensemble forecasts from the European Centre for Medium-Range Weather Forecasts (ECMWF), and traditional statistical approaches that use up to 25 predictor variables from IMERG and the ERA5 reanalysis. Generally, statistical approaches perform about on par with postprocessed ECMWF ensemble forecasts. The CNN+EasyUQ approach, however, clearly outperforms all competitors in terms of both occurrence and amount. Hybrid methods that merge CNN+EasyUQ and physics-based forecasts show slight further improvement. Thus, the CNN+EasyUQ approach can likely improve operational probabilistic forecasts of rainfall in the tropics and potentially even beyond. Significance Statement Precipitation forecasts in the tropics remain a great challenge despite their enormous potential to create socioeconomic benefits in sectors such as food and energy production. Here, we develop a purely data-driven, machine learning–based prediction model that outperforms traditional, physics-based approaches to 1-day-ahead forecasts of rainfall occurrence and rainfall amount over northern tropical Africa in terms of both forecast skill and computational costs. A combined data-driven and physics-based (hybrid) approach yields further (slight) improvement in terms of forecast skill. These results suggest new avenues to more accurate and more resource-efficient operational precipitation forecasts in the Global South.

Deep Learning for Postprocessing Global Probabilistic Forecasts on Subseasonal Time Scales

Abstract Subseasonal weather forecasts are becoming increasingly important for a range of socioeconomic activities. However, the predictive ability of physical weather models is very limited on these time scales. We propose four postprocessing methods based on convolutional neural networks to improve subseasonal forecasts by correcting systematic errors of numerical weather prediction models. Our postprocessing models operate directly on spatial input fields and are therefore able to retain spatial relationships and to generate spatially homogeneous predictions. They produce global probabilistic tercile forecasts for biweekly aggregates of temperature and precipitation for weeks 3–4 and 5–6. In a case study based on a public forecasting challenge organized by the World Meteorological Organization, our postprocessing models outperform the bias-corrected forecasts from the European Centre for Medium-Range Weather Forecasts (ECMWF), and achieve improvements over climatological forecasts for all considered variables and lead times. We compare several model architectures and training modes and demonstrate that all approaches lead to skillful and well-calibrated probabilistic forecasts. The good calibration of the postprocessed forecasts emphasizes that our postprocessing models reliably quantify the forecast uncertainty based on deterministic input information in the form of ECMWF ensemble mean forecast fields only.

Forecasting large hail and lightning using additive logistic regression models and the ECMWF reforecasts

Abstract. Additive logistic regression models for lightning (ARlig) and large hail (ARhail) were developed using convective parameters from the ERA5 reanalysis, hail reports from the European Severe Weather Database (ESWD), and lightning observations from the Met Office Arrival Time Difference network (ATDnet). The models yield the probability of lightning and large hail in a given timeframe over a particular grid point. To explore the value of this approach to medium-range forecasting, the models were applied to the European Centre for Medium Range Weather Forecasts (ECMWF) reforecasts to reconstruct probabilistic lightning and large hail forecasts for 11 ensemble members, from 2008 to 2019 and for lead times up to 228 h. The lightning and large hail models were based on different predictor parameters: most unstable convective available potential energy (CAPE), 925–500 hPa bulk shear, mixed layer mixing ratio, wet bulb zero height (for large hail), most unstable lifted index, mean relative humidity between 850 and 500 hPa, 1 hourly accumulated convective precipitation and specific humidity at 925 hPa (for lightning). First, we compared the lightning and hail ensemble forecasts for different lead times with observed lightning and hail focusing on a recent hail outbreak. Second, we evaluated the predictive skill of the model as a function of forecast lead time using the area under the ROC curve (AUC) as a validation score. This analysis showed that ARhail has a very high predictive skill (AUC > 0.95) for a lead time up to 60 h. ARhail retains a high predictive skill even for extended forecasts (AUC = 0.86 at 180 h lead time). Although ARlig exhibits a lower predictive skill than ARhail, lightning forecasts are also skilful both in the short term (AUC = 0.92 at 60 h) and in the medium range (AUC = 0.82 at 180 h). Finally, we compared the performance of the 4-dimensional hail model with that of composite parameters such as the significant hail parameter (SHP) or the product of CAPE and the 925–500 hPa bulk shear (CAPESHEAR). Results show that ARhail outperforms CAPESHEAR at all lead times and SHP at short-to-medium lead times. These findings suggests that the combination of additive logistic regression models and ECMWF ensemble forecasts can create highly skilful medium-range hail and lightning forecasts for Europe.

FuXi: a cascade machine learning forecasting system for 15-day global weather forecast

Over the past few years, the rapid development of machine learning (ML) models for weather forecasting has led to state-of-the-art ML models that have superior performance compared to the European Centre for Medium-Range Weather Forecasts (ECMWF)’s high-resolution forecast (HRES), which is widely considered as the world’s best physics-based weather forecasting system. Specifically, ML models have outperformed HRES in 10-day forecasts with a spatial resolution of 0.25∘. However, the challenge remains in mitigating the accumulation of forecast errors for longer effective forecasts, such as achieving comparable performance to the ECMWF ensemble in 15-day forecasts. Despite various efforts to reduce accumulation errors, such as implementing autoregressive multi-time step loss, relying on a single model has been found to be insufficient for achieving optimal performance in both short and long lead times. Therefore, we present FuXi, a cascaded ML weather forecasting system that provides 15-day global forecasts at a temporal resolution of 6 hours and a spatial resolution of 0.25∘. FuXi is developed using 39 years of the ECMWF ERA5 reanalysis dataset. The performance evaluation demonstrates that FuXi has forecast performance comparable to ECMWF ensemble mean (EM) in 15-day forecasts. FuXi surpasses the skillful forecast lead time achieved by ECMWF HRES by extending the lead time for Z500 from 9.25 to 10.5 days and for T2M from 10 to 14.5 days. Moreover, the FuXi ensemble is created by perturbing initial conditions and model parameters, enabling it to provide forecast uncertainty and demonstrating promising results when compared to the ECMWF ensemble.

Parametric Postprocessing of Dual-Resolution Precipitation Forecasts

Abstract All major weather centers issue ensemble forecasts, which differ both in ensemble size and spatial resolution, even while covering the same domain. These parameters directly determine both the forecast skill of the prediction and the computation cost. In the past few years, the plans for upgrading the configuration of the Integrated Forecast System of the European Centre for Medium-Range Weather Forecasts (ECMWF) from a single forecast with 9-km resolution and a 51-member ensemble with 18-km resolution induced an extensive study of the forecast skill of both raw and postprocessed dual-resolution predictions comprising ensemble members of different horizontal resolutions. We investigate the predictive performance of the censored shifted gamma (CSG) ensemble model output statistic (EMOS) approach for statistical postprocessing with the help of dual-resolution, 24-h, precipitation accumulation ensemble forecasts over Europe with various forecast horizons. We consider the operational 50-member ECMWF ensemble as of high resolution and extend it with a low-resolution (29-km grid), 200-member experimental forecast. The investigated dual-resolution combinations consist of subsets of these two forecast ensembles with equal computational cost, which is equivalent to the cost of the operational ensemble. Our case study verifies that, compared with the raw ensemble combinations, EMOS postprocessing results in a significant improvement in forecast skill and that skill is statistically indistinguishable between any of the analyzed mixtures of dual-resolution combinations. Furthermore, the semilocally trained CSG EMOS provides an efficient alternative to the state-of-the-art quantile mapping without requiring additional historical data.

Uncertainty growth and forecast reliability during extratropical cyclogenesis

Abstract. In global numerical weather prediction, the strongest contribution to ensemble variance growth over the first few days is at synoptic scales. Hence it is particularly important to ensure that this synoptic-scale variance is reliable. Here we focus on wintertime synoptic-scale growth in the North Atlantic storm track. In the 12 h background forecasts of the Ensemble of Data Assimilations (EDA) from the European Centre for Medium-Range Weather Forecasts (ECMWF), we find that initial variance growth at synoptic scales tends to be organized in particular flow situations, such as during the deepening of cyclones (cyclogenesis). Both baroclinic and diabatic aspects may be involved in the overall growth rate. However, evaluation of reliability through use of an extended error–spread equation indicates that the ECMWF ensemble forecast, which is initialized from the EDA but with additional singular vector perturbations, appears to have too much variance at a lead time of 2 d and that this over-spread is associated with cyclogenesis situations. Comparison of variance growth rates and reliability with other forecast systems within The International Grand Global Ensemble (TIGGE) archive indicates some sensitivity to the model or its initialization. For the ECMWF ensemble forecast, sensitivity experiments suggest that a large part of the total day-2 spread in cyclogenesis cases is associated with the growth of EDA uncertainty, but up to 25 % can be associated with the additional singular vector perturbations to the initial conditions and up to 25 % with the representation of model uncertainty. The sensitivities of spread to resolution, the explicit representation of convection, and the assimilation of local observations are also considered. The study raises the question of whether the EDA now successfully represents initial uncertainty (and the enhanced growth rates associated with cyclogenesis) to the extent that singular vector perturbations could be reduced in magnitude to improve storm track reliability. This would leave a more seamless forecast system, allowing short-range diagnostics to better help improve the model and model-uncertainty representation, which could be beneficial throughout the forecast range.

Monthly streamflow forecasting by machine learning methods using dynamic weather prediction model outputs over Iran

Seasonal hydrological forecasts play a critical role in water resources management. The Copernicus Climate Change Service (C3S) data store provides open access to monthly hydrological forecasts for up to six-months. This study aims to evaluate, for the first time, 1- to 3-month runoff forecasts using the European Centre for Medium-Range Weather Forecasts (ECMWF) ensembles of precipitation, runoff, and temperature in 1981–2015 period over a total of 30 s-level basins in Iran. We adopted the 5th, 50th and 95th ECMWF ensemble quantiles for each variable that represent low, medium and high probability of occurrence, respectively. Pearson correlation analysis (Pca), Recursive Feature Elimination (RFE) via random forest (RF) model, and Bayesian Networks (BN) feature selection algorithms were used in order to reduce input variable dimension and select potential predictors to be fed to the machine learning models. Multiple Linear Regression (MLR), Artificial Neural Networks (ANN), Support Vector Regression (SVR), Random Forest (RF), and eXtreme Gradient Boosting (XGBoost) machine learning models were used with Repeated K-Fold cross validation (rK-Fold CV) while model efficiency was evaluated using modified Kling-Gupta efficiency coefficient (KGE’), Nash-Sutcliffe Efficiency coefficient (NSE), and Normalized Root Mean Square Error (NRMSE). Results of this study revealed that C3S runoff ensembles have the highest impact on forecast accuracy of streamflow, followed by precipitation and temperature. Overall, model performance yield a best-to-worst ranking of ANN, XGBoost, RF, MLR, and SVR with KGE’ values of 0.70, 0.68, 0.66, 0.57, and 0.41, respectively. The predictive performance of all models decreased with lead times beyond 1-month, where ANN and XGBoost outperformed other models with KGE’ of 0.65 for 2-month lead time and 0.60 for 3-month lead time. The three superior models of XGBoost, ANN, and RF, were employed with RFE and BN FSAs most frequently across Iran’s 30 s level basins in all lead times. Almost all models in the arid central region of Iran showed the lowest performance while highest skills were achieved in the western regions of Iran. Finally, for all models and over all regions, the model performance reduced by increase in lead-time.

Skill, Predictability, and Cluster Analysis of Atlantic Tropical Storms and Hurricanes in the ECMWF Monthly Forecasts

Abstract In this paper we analyze Atlantic Ocean hurricane activity in the European Centre for Medium-Range Weather Forecasts (ECMWF) monthly hindcasts for the period 1998–2017. The main climatological characteristics of Atlantic tropical cyclone (TC) activity are considered at different lead times and across the entire ECMWF ensemble using three diagnostic variables: the number of tropical cyclones, the number of hurricanes, and the accumulated cyclone energy. The impacts of changing horizontal resolution and stochastic parameterization are clear in these diagnostic variables. The model skill scores for the number of tropical cyclones and accumulated cyclone energy by lead time are also computed. Using cluster analysis, we compare the characteristics of the forecast TC tracks with observations. Although four of the ECMWF clusters have similar characteristics to observed ones, one of the ECMWF clusters does not have a corresponding one in observations. We consider the predictability of each of these clusters, as well the modulation of their frequency by climate modes, such as the El Niño–Southern Oscillation and the Madden–Julian oscillation, taking advantage of the very large sample size of TC datasets in these hindcasts.

More accuracy with less precision

AbstractReducing the numerical precision of the forecast model of the Integrated Forecasting System (IFS) of the European Centre for Medium‐Range Weather Forecasts (ECMWF) from double to single precision results in significant computational savings without negatively affecting forecast accuracy. The computational savings allow to increase the vertical resolution of the operational ensemble forecasts from 91 to 137 levels earlier than anticipated and before the next upgrade of ECMWF's high‐performance computing facility. This upgrade to 137 levels harmonises the vertical resolution of the medium‐range deterministic forecasts and the medium‐range and extended‐range ensemble forecasts. Increasing the vertical resolution of the ensemble forecasts substantially improves forecast skill for all lead times as well as the mean of the model climate. ECMWF's ensemble and deterministic forecasts will run operationally at single precision from IFS model cycle 47R2 onwards.

Revision of the Stochastically Perturbed Parametrisations model uncertainty scheme in the Integrated Forecasting System

AbstractThe Stochastically Perturbed Parametrisations scheme (SPP) represents model uncertainty in numerical weather prediction by introducing stochastic perturbations into the physical parametrisation schemes. The perturbations are constructed in such a way that the internal consistency of the physical parametrisation schemes is preserved. We developed a revised version of SPP for the Integrated Forecasting System of the European Centre for Medium‐Range Weather Forecasts (ECMWF). The revised version introduces perturbations to additional quantities and modifies the probability distributions sampled by the scheme. Medium‐range ensemble forecasts with the revised SPP are considerably more skilful than ensemble forecasts with the original implementation of SPP. The revised version of SPP is similar, in terms of forecast skill, to the Stochastically Perturbed Parametrisation Tendency scheme (SPPT), which is currently used to represent model uncertainty in the operational ECMWF ensemble forecasts.

Improving Wind Forecasts in the Lower Stratosphere by Distilling an Analog Ensemble Into a Deep Neural Network

AbstractWe discuss improving forecasts of winds in the lower stratosphere using machine learning to postprocess the output of the European Centre for Medium‐Range Weather Forecasts (ECMWF) Integrated Forecast System. We postprocess global three‐dimensional predictions and demonstrate distilling the analog ensemble (AnEn) method into a deep neural network, which reduces postprocessing latency to near zero maintaining increased forecast skill. This approach reduces the error with respect to ECMWF high‐resolution deterministic prediction between 2–15% for wind speed and 15–25% for direction and is on par with ECMWF ensemble (ENS) forecast skill to hour 60. Verifying with Loon data from stratospheric balloons, AnEn has 20% lower error than ENS for wind speed and 15% for wind direction, despite significantly lower real‐time computational cost to ENS. Similar performance patterns are reported for probabilistic predictions, with larger improvements of AnEn with respect to ENS. We also demonstrate that AnEn generates a calibrated probabilistic forecast.

Evaluation of the Consistency of ECMWF Ensemble Forecasts

AbstractAn expected benefit of ensemble forecasts is that a sequence of consecutive forecasts valid for the same time will be more consistent than an equivalent sequence of individual forecasts. Inconsistent (jumpy) forecasts can cause users to lose confidence in the forecasting system. We present a first systematic, objective evaluation of the consistency of the European Centre for Medium‐Range Weather Forecasts (ECMWF) ensemble using a measure of forecast divergence that takes account of the full ensemble distribution. Focusing on forecasts of the North Atlantic Oscillation and European Blocking regimes up to 2 weeks ahead, we identify occasional large inconsistency between successive runs, with the largest jumps tending to occur at 7–9 days lead. However, care is needed in the interpretation of ensemble jumpiness. An apparent clear flip‐flop in a single index may hide a more complex predictability issue which may be better understood by examining the ensemble evolution in phase space.

An Investigation of Large Along-Track Errors in Extratropical Transitioning North Atlantic Tropical Cyclones in the ECMWF Ensemble

Abstract The synoptic evolution and mechanisms for the largest medium-range (72–120 h) along-track errors of tropical cyclones (TC) are investigated. The mean along-track errors (ATEs) of the 51-member European Centre for Medium-Range Weather Forecasts (ECMWF) ensemble are evaluated for 393 forecasts (85 TCs) during the 2008 to 2016 North Atlantic seasons. The 27 unique forecasts within the upper quintile of most negative ATEs (i.e., slow bias greater than 500 km by 72 h) are inherently fast-moving TCs that undergo extratropical transition as they recurve and interact with a 300-hPa upstream trough and a downstream ridge. Both the trough and ridge are underamplified by only 5–10 m ~60 h before the time of largest ATE. The height errors then grow rapidly due to underpredicted 300–200-hPa potential vorticity advection by both the nondivergent wind and the irrotational wind from the TC’s outflow. Both wind components are underpredicted and result in weak biases in the trough’s developing potential vorticity gradient and associated jet streak. The underamplification of the upstream trough is exacerbated by underpredicted 700-hPa cold advection extending from beneath the trough into the TC at 48–36 h before the largest ATE. Standardized differences are consistent with the mean errors and reveal that weaker divergent outflow is driven by underpredicted near-TC precipitation, which corresponds to underpredicted 700-hPa moisture fluxes near the TC at ~108 h before the largest ATE. The ensemble member ATEs at 72–120 h generally show little correlation with their ATEs before 36 h, suggesting that initial position uncertainty is not the primary source of ATE variability later in the forecast.

Evaluating U.S. East Coast Winter Storms in a Multimodel Ensemble Using EOF and Clustering Approaches

AbstractEmpirical orthogonal function (EOF) and fuzzy clustering tools were applied to generate and validate scenarios in operational ensemble prediction systems (EPSs) for U.S. East Coast winter storms. The National Centers for Environmental Prediction (NCEP), European Centre for Medium-Range Weather Forecasts (ECMWF), and Canadian Meteorological Centre (CMC) EPSs were validated in their ability to capture the analysis scenarios for historical East Coast cyclone cases at lead times of 1–9 days. The ECMWF ensemble has the best performance for the medium- to extended-range forecasts. During this time frame, NCEP and CMC did not perform as well, but a combination of the two models helps reduce the missing rate and alleviates the underdispersion. All ensembles are underdispersed at all ranges, with combined ensembles being less underdispersed than the individual EPSs. The number of outside-of-envelope cases increases with lead time. For a majority of the cases beyond the short range, the verifying analysis does not lie within the ensemble mean group of the multimodel ensemble or within the same direction indicated by any of the individual model means, suggesting that all possible scenarios need to be taken into account. Using the EOF patterns to validate the cyclone properties, the NCEP model tends to show less intensity and displacement biases during 1–3-day lead time, while the ECMWF model has the smallest biases during 4–6 days. Nevertheless, the ECMWF forecast position tends to be biased toward the southwest of the other two models and the analysis.

Short-range solar radiation forecasts over Sweden

Abstract. In this article the performance for short-range solar radiation forecasts by the global deterministic and ensemble models from the European Centre for Medium-Range Weather Forecasts (ECMWF) is compared with an ensemble of the regional mesoscale model HARMONIE-AROME used by the national meteorological services in Sweden, Norway and Finland. Note however that only the control members and the ensemble means are included in the comparison. The models resolution differs considerably with 18 km for the ECMWF ensemble, 9 km for the ECMWF deterministic model, and 2.5 km for the HARMONIE-AROME ensemble. The models share the same radiation code. It turns out that they all underestimate systematically the Direct Normal Irradiance (DNI) for clear-sky conditions. Except for this shortcoming, the HARMONIE-AROME ensemble model shows the best agreement with the distribution of observed Global Horizontal Irradiance (GHI) and DNI values. During mid-day the HARMONIE-AROME ensemble mean performs best. The control member of the HARMONIE-AROME ensemble also scores better than the global deterministic ECMWF model. This is an interesting result since mesoscale models have so far not shown good results when compared to the ECMWF models. Three days with clear, mixed and cloudy skies are used to illustrate the possible added value of a probabilistic forecast. It is shown that in these cases the mesoscale ensemble could provide decision support to a grid operator in terms of forecasts of both the amount of solar power and its probabilities.

Comparative Assessment of Two Objective Forecast Models for Cases of Persistent Extreme Precipitation Events in the Yangtze–Huai River Valley in Summer 2016

Abstract Two persistent extreme precipitation events (PEPEs) that caused severe flooding in the Yangtze–Huai River valley in summer 2016 presented a significant challenge to operational forecasters. To provide forecasters with useful references, the capacity of two objective forecast models in predicting these two PEPEs is investigated. The objective models include a numerical weather prediction (NWP) model from the European Centre for Medium-Range Weather Forecasts (ECMWF), and a statistical downscaling model, the Key Influential Systems Based Analog Model (KISAM). Results show that the ECMWF ensemble provides a skillful spectrum of solutions for determining the location of the daily heavy precipitation (≥25 mm day−1) during the PEPEs, despite its general underestimation of heavy precipitation. For lead times longer than 3 days, KISAM outperforms the ensemble mean and nearly one-half or more of all the ensemble members of ECMWF. Moreover, at longer lead times, KISAM generally performs better in reproducing the meridional location of accumulated rainfall over the two PEPEs compared to the ECMWF ensemble mean and the control run. Further verification of the vertical velocity that affects the production of heavy rainfall in ECMWF and KISAM implies the quality of the depiction of ascending motion during the PEPEs has a dominating influence on the models’ performance in predicting the meridional location of the PEPEs at all lead times. The superiority of KISAM indicates that statistical downscaling techniques are effective in alleviating the deficiency of global NWP models for PEPE forecasts in the medium range of 4–10 days.

Improving Predictions of Precipitation Type at the Surface: Description and Verification of Two New Products from the ECMWF Ensemble

Abstract The medium-range ensemble (ENS) from the European Centre for Medium-Range Weather Forecasts (ECMWF) Integrated Forecasting System (IFS) is used to create two new products intended to face the challenges of winter precipitation-type forecasting. The products themselves are a map product that represents which precipitation type is most likely whenever the probability of precipitation is >50% (also including information on lower probability outcomes) and a meteogram product, showing the temporal evolution of the instantaneous precipitation-type probabilities for a specific location, classified into three categories of precipitation rate. A minimum precipitation rate is also used to distinguish dry from precipitating conditions, setting this value according to type, in order to try to enforce a zero frequency bias for all precipitation types. The new products differ from other ECMWF products in three important respects: first, the input variable is discretized, rather than continuous; second, the postprocessing increases the output information content; and, third, the map-based product condenses information into a more accessible format. The verification of both products was developed using four months’ worth of 3-hourly observations of present weather from manual surface synoptic observation (SYNOPs) in Europe during the 2016/17 winter period. This verification shows that the IFS is highly skillful when forecasting rain and snow, but only moderately skillful for freezing rain and rain and snow mixed, while the ability to predict the occurrence of ice pellets is negligible. Typical outputs are also illustrated via a freezing-rain case study, showing interesting changes with lead time.

Gridded probabilistic weather forecasts with an analog ensemble

This study extends the analog ensemble (AnEn) technique to generate probabilistic forecasts of 10 m wind speed over a grid. The AnEn has been widely used to generate short‐term probabilistic predictions of meteorological variables, air quality, wind and solar power, and to effectively downscale reanalysis fields. It is based on a historical dataset including observations paired with corresponding deterministic predictions. For each forecast lead time and location, the AnEn is generated using the observations corresponding to the past deterministic predictions that are more similar to the current forecast. So far, the AnEn has been used to generate predictions at specific locations with available observations. Here, it is extended over a two‐dimensional grid, where each grid point is treated independently, using meteorological analysis instead of observations. The forecast range is extended up to 6 days ahead to evaluate the performance of the AnEn in both the short and medium range. The proposed approach allows to extend the AnEn to applications requiring gridded probabilistic products that are often generated with dynamical ensemble models. The AnEn forecasts are generated using the European Centre for Medium‐Range Weather Forecasts (ECMWF) deterministic and analysis model. Also, the ECMWF Ensemble Prediction System (EPS) is used for comparison. Given that the AnEn predictions are generated independently at any location and lead time, the resulting spatial and temporal correlation may be degraded by noise. A reordering technique called Schaake Shuffle (SS) is then applied to the ensemble members to recover the spatio‐temporal correlation. The AnEn outperforms a calibrated version of EPS for the first two days of the prediction. During the third forecast day, the AnEn remains competitive with the calibrated EPS, while the latter is more skilful from 72 to 144 h ahead. The AnEn uses about one sixth of the computational resources necessary to generate the real‐time EPS prediction.

Wind power application research on the fusion of the determination and ensemble prediction

Abstract. The fused product of wind speed for the wind farm is designed through the use of wind speed products of ensemble prediction from the European Centre for Medium-Range Weather Forecasts (ECMWF) and professional numerical model products on wind power based on Mesoscale Model5 (MM5) and Beijing Rapid Update Cycle (BJ-RUC), which are suitable for short-term wind power forecasting and electric dispatch. The single-valued forecast is formed by calculating the different ensemble statistics of the Bayesian probabilistic forecasting representing the uncertainty of ECMWF ensemble prediction. Using autoregressive integrated moving average (ARIMA) model to improve the time resolution of the single-valued forecast, and based on the Bayesian model averaging (BMA) and the deterministic numerical model prediction, the optimal wind speed forecasting curve and the confidence interval are provided. The result shows that the fusion forecast has made obvious improvement to the accuracy relative to the existing numerical forecasting products. Compared with the 0–24 h existing deterministic forecast in the validation period, the mean absolute error (MAE) is decreased by 24.3 % and the correlation coefficient (R) is increased by 12.5 %. In comparison with the ECMWF ensemble forecast, the MAE is reduced by 11.7 %, and R is increased 14.5 %. Additionally, MAE did not increase with the prolongation of the forecast ahead.

Ensemble superparameterization versus stochastic parameterization: A comparison of model uncertainty representation in tropical weather prediction

AbstractStochastic schemes to represent model uncertainty in the European Centre for Medium‐Range Weather Forecasts (ECMWF) ensemble prediction system has helped improve its probabilistic forecast skill over the past decade by both improving its reliability and reducing the ensemble mean error. The largest uncertainties in the model arise from the model physics parameterizations. In the tropics, the parameterization of moist convection presents a major challenge for the accurate prediction of weather and climate. Superparameterization is a promising alternative strategy for including the effects of moist convection through explicit turbulent fluxes calculated from a cloud‐resolving model (CRM) embedded within a global climate model (GCM). In this paper, we compare the impact of initial random perturbations in embedded CRMs, within the ECMWF ensemble prediction system, with stochastically perturbed physical tendency (SPPT) scheme as a way to represent model uncertainty in medium‐range tropical weather forecasts. We especially focus on forecasts of tropical convection and dynamics during MJO events in October–November 2011. These are well‐studied events for MJO dynamics as they were also heavily observed during the DYNAMO field campaign. We show that a multiscale ensemble modeling approach helps improve forecasts of certain aspects of tropical convection during the MJO events, while it also tends to deteriorate certain large‐scale dynamic fields with respect to stochastically perturbed physical tendencies approach that is used operationally at ECMWF.