Hourly Forecast Research Articles

Managing air quality: Predicting exceedances of legal limits for PM10 and O3 concentration using machine learning methods

AbstractAir pollution imposes great costs on productivity, safety and health of individuals and dictates necessity of a proactive air pollution management. This, in turn, requires powerful tools for air quality modeling. In this article we develop a two‐stage procedure for predicting exceedances of the EU legal limits for PM10 and O concentrations using hourly data. Within the first stage we deploy machine learning methods to produce accurate 24‐h‐ahead forecasts of hourly pollutant concentrations at seven specific locations in the cities of Augsburg and Munich, Germany. The best performance was shown by the Stochastic Gradient Boosting Model—an ensemble tree‐based method, especially convenient because of its computational efficiency and robustness to overfitting. Its predictive ability was largely superior to that reported by similar studies. In the second stage, the hourly forecasts were used to predict the exceedances of the EU daily limits for PM10 and O concentrations. For both pollutants we could achieve the average probability of exceedances detection above 80%, while keeping the probability of false alarms at a reasonably low level. Such satisfactory results show that our approach can be successfully applied to anticipate the shocks, which would allow authorities to manage them in the most effective manner.

Solar forecasting with hourly updated numerical weather prediction

Solar forecasters have hitherto been restricting the application of numerical weather prediction (NWP) to day-ahead forecasting scenarios. With the hourly updated NWP models, such as the National Oceanic and Atmospheric Administration’s Rapid Refresh (RAP) or High-Resolution Rapid Refresh (HRRR), it is theoretically possible to utilize NWP for hour-ahead solar forecasting. Nonetheless, for NWP-based hourly forecasts to be useful, they ought to be post-processed, because such forecasts are almost always affected by the inherent bias of dynamical weather models. In this regard, Kalman filtering is herein used as a post-processing tool. Using one year of RAP and HRRR forecasts and ground-based observations made at seven geographically diverse locations in the contiguous United States, it is demonstrated that the post-processed versions of hourly updated NWP forecasts are sometimes able to attain a higher accuracy than those from classic time-series families of models, not only at long-range, but also at short-range forecast horizons. Although these improvements are marginal in terms of squared loss (about 1%–2%), since NWP models have a very different forecast-generating mechanism (solving the governing equations of motion) from that of time series methods (extrapolating data), NWP-based forecasts can be expected to be less correlated with those forecasts from time series models. Consequently, one should find this diversity profoundly rewarding during forecast combination, for the combined forecasts are able to consistently result in smaller error than the best component forecasts.

Tropical cyclone forecast using NCMRWF Global (12 km) and regional (4 km) models

Countries adjoining the North Indian Ocean (NIO) region are among the world's worst affected areas by tropical cyclones (TCs). An increase in frequency and intensity of TCs affecting this basin is noticed in recent years. Timely and accurate prediction of a TC can lead to a decrease in damages to life and property caused by the cyclone. In recent times, the forecasts of TC tracks and intensity have improved due to advancements in the resolution, data assimilation techniques and physics of Numerical Weather Prediction (NWP) models and improvements in the model's initial condition. In this study, we have analysed the forecasts of Super Cyclone (SuCS) 'Amphan' that occurred over the Bay of Bengal during 16-21 May, 2020 obtained from the Global and Regional version of the NCMRWF Unified Model, i.e., NCUM-G and NCUM-R, respectively. The analysis of the track error shows that the initial position and 24-hour forecast errors in both the models are very close. However, the 48 to 72-hour forecast errors are much lower in the NCUM-R compared to NCUM-G. The percentage decrease in the track errors from global to regional model is about 12% in the 48-hour forecasts and 17% in the 72-hour forecasts. The landfall position error is less than 50 km in the predictions after 18th May, 2020 in NCUM-G. However, NCUM-R shows a lower landfall location error as compared to NCUM-G. Comparing the intensity forecasts from NCUM-G and NCUM-R shows that the 48 hrs forecast of minimum sea level pressure (SLP) based on initial conditions of 16th May, 2020 are very close to the observed intensity. The rapid intensification of Amphan is also well predicted by both the models, although with a delay. The cyclone structure shows that the 850 hPa vorticity core is much stronger in NCUM-R forecasts than NCUM-G. The warm-core system in the NCUM-R forecasts is also more organized than in NCUM-G. On the other hand, the maximum wind radius is greater in NCUM-G forecasts than NCUM-R.

Stacking Machine Learning Models to Forecast Hourly and Daily Electricity Consumption of Household Using Internet of Things

The objective of this paper is to design an efficient electricity consumption forecasting model using stacking ensemble technique and Internet of Things (IoT). Two stage process is applied in this paper. In the first stage, fifteen forecasting models (Auto-ARIMA, Holt-Winter (Additive), Exponential, Facebook Prophet, Light Gradient Boosting, AdaBoost, Support Vector Regression, Decision Tree, Extra Tree, Random Forest, Elastic net, K-Nearest Neighbour’s, XGBoost, Linear Regression, Long Short Term Memory) are applied to forecast electricity consumption at an hourly and daily level. In the next stage, the best four models are selected and stacked. We have considered the dataset of energy consumption by electrical appliances per minute in a house over seven days. The models are evaluated using root mean square error (RMSE), mean absolute error (MAE), R-square, and mean absolute percentage error (MAPE). The results show that the extra tree performed better among all the algorithms, and stacking further improves performance. Elastic net and decision tree algorithms have taken less time as compared to other models applied in this study.

A Kilometer-Scale Coupled Atmosphere-Wave Forecasting System for the European Arctic

AbstractAccurately simulating the interactions between the components of a coupled Earth modelling system (atmosphere, sea-ice, and wave) on a kilometer-scale resolution is a new challenge in operational numerical weather prediction. It is difficult due to the complexity of interactive mechanisms, the limited accuracy of model components and scarcity of observations available for assessing relevant coupled processes. This study presents a newly developed convective-scale atmosphere-wave coupled forecasting system for the European Arctic. The HARMONIE-AROME configuration of the ALADIN-HIRLAM numerical weather prediction system is coupled to the spectral wave model WAVEWATCH III using the OASIS3 model coupling toolkit. We analyze the impact of representing the kilometer-scale atmosphere-wave interactions through coupled and uncoupled forecasts on a model domain with 2.5 km spatial resolution. In order to assess the coupled model’s accuracy and uncertainties we compare 48-hour model forecasts against satellite observational products such as Advanced Scatterometer 10 m wind speed, and altimeter based significant wave height. The fully coupled atmosphere-wave model results closely match both satellite-based wind speed and significant wave height observations as well as surface pressure and wind speed measurements from selected coastal station observation sites. Furthermore, the coupled model contains smaller standard deviation of errors in both 10m wind speed and significant wave height parameters when compared to the uncoupled model forecasts. Atmosphere and wave coupling reduces the short term forecast error variability of 10 m wind speed and significant wave height with the greatest benefit occurring for high wind and wave conditions.

Rainfall-Runoff Short-Term Forecasting Method Based on LSTM

A rainfall-runoff forecasting method based on Long Short Term Memory (LSTM) was proposed in this study, which can extract the trend characteristics of runoff time series data through introducing daily rainfall data collected at related upstream stations and making use of the advantages of LSTM in saving long-term sequence feature information and avoid vanishing gradient, and identify the nonlinear mapping relationship in between, thus establishing a short-term runoff forecasting model. In this study, 24-hour and 5-day short-term forecasting models were established based on the runoff data collected at Danba Hydrologic Station in Dadu River Basin and historical rainfall collected at three upstream stations (Xiaojin, Dawei, and Fubian River). The experimental results showed that the forecasting models performed well during the inspection period. In 24-hour forecasting, RMSE was 98.016 and MAE was 45.709, which were 73.993 and 32.699, respectively in 5-day forecasting, indicating better performance and increased forecasting accuracy than simple LSTM model.

EFSO at Different Geographical Locations Verified with Observing System Experiments

AbstractAn ensemble-based forecast sensitivity to observations (EFSO) diagnosis has been implemented in an atmospheric general circulation model–ensemble Kalman filter data assimilation system to estimate the impacts of specific observations from the quasi-operational global observing system on weekly short-range forecasts. It was examined whether EFSO reasonably approximates the impacts of a subset of observations from specific geographical locations for 6-h forecasts, and how long the 6-h observation impacts can be retained during the 7-day forecast period. The reference for these forecasts was obtained from 12 data-denial experiments in each of which a subset of three radiosonde observations launched from a geographical location was excluded. The 12 locations were selected from three latitudinal bands comprising (i) four Arctic regions, (ii) four midlatitude regions in the Northern Hemisphere, and (iii) four tropical regions during the Northern Hemisphere winter of 2015/16. The estimated winter-averaged EFSO-derived observation impacts well corresponded to the 6-h observation impacts obtained by the data denials and EFSO could reasonably estimate the observation impacts by the data denials on short-range (from 6 h to 2 day) forecasts. Furthermore, during the medium-range (4–7 day) forecasts, it was found that the Arctic observations tend to seed the broadest impacts and their short-range observation impacts could be projected to beneficial impacts in Arctic and midlatitude North American areas. The midlatitude area was located just downstream of dynamical propagation from the Arctic toward the midlatitudes. Results obtained by repeated Arctic data-denial experiments were found to be generally common to those from the non-repeated experiments.

Evaluating the Impact of Planetary Boundary Layer, Land Surface Model, and Microphysics Parameterization Schemes on Cold Cloud Objects in Simulated GOES‐16 Brightness Temperatures

AbstractInfrared brightness temperatures (BTs) from the Geostationary Observing Environmental Satellite‐16 Advanced Baseline Imager are used to examine the ability of several microphysics and planetary boundary layer (PBL) schemes, as well as land surface models (LSM) and surface layers, to simulate upper‐level clouds. Six parameterization configurations were evaluated. Cloud objects are identified using the Method for Object‐Based Diagnostic Evaluation (MODE) and analyzed using the object‐based threat score, mean‐error distance, and pixel‐based metrics including the mean absolute error and mean bias error (MBE) for matched objects where the displacement between objects has been removed. Objects are identified using either a fixed BT threshold of 235 K or the 6.5th percentile of BTs for each model configuration. Analysis of the MODE‐identified cloud objects shows that, compared to a configuration with the Thompson microphysics scheme, Mellor‐Yamanda‐Nakanishi‐Niino (MYNN) PBL, Global Forecasting System (GFS) surface layer, and Noah LSM, the configuration employing the National Severe Storms Laboratory microphysics produced more cloud objects with higher BTs. Changing the PBL from MYNN to Shin‐Hong or Eddy‐Diffusivity Mass‐Flux also resulted in a slightly lower accuracy, though these changes result in configurations which more accurately reproduced the number of observation cloud objects and slightly reduced the high MBE. Changing the LSM from Noah to RUC reduces forecast accuracy by producing too many cloud objects with too low BTs. As the forecast hour increases, this accuracy reduction increases at a greater rate than occurred when changing the microphysics or PBL scheme and is further enhanced when using the MYNN surface layer rather than the GFS.

GPS-ZTD data assimilation and its impact on wintertime haze prediction over North China Plain using WRF 3DVAR and CMAQ modeling system.

Severe haze frequently hits the North China Plain (NCP), especially in winter during recent years. Meteorological factors affect aerosol formation and its optical properties, and accurate meteorological fields are imperative for accurate aerosol simulations. The impacts of Global Positioning System Zenith Total Delay (GPS-ZTD) data assimilation on meteorology and aerosol simulations were evaluated in this study using the WRF-CMAQ (the Weather Research and Forecasting model and Community Multiscale Air Quality) modelling system over the NCP during 01-31 December 2019. After bias correction, GSP-ZTD data were assimilated into the WRF model using the 3DVAR technique. Two sensitivity tests (CTR and ZTD) were conducted. The WRF model had generally acceptable performance for surface and upper air meteorological variables, PM2.5 and visibility. From the aspect of BIAS, STDE, RMSE, and R, the assimilation of ZTD data improved the underestimation of ground relative humidity (RH). The improvement was more pronounced in the first 18 forecast hours. The mean RH BIAS decreased by 8%. Surface pressure was also improved in ZTD. The influence of ZTD data assimilation on ground temperature and wind tended to be neutral. The BIAS of ZTD decreased by 3% after data assimilation while STED or RMSE increased slightly. After ZTD data assimilation, the PM2.5 underestimation decreased by 3.4% over NCP. And station mean BIAS or RMSE of PM2.5 decreased at more than 70% stations. After ZTD data assimilation, the visibility overestimation was reduced by 2.5%. And more than 81% stations over had lower visibility BIAS or RMSE. Station mean PM2.5 mass concentration increased by 1.5% in ZTD. The primary aerosol species increased by approximately 1%, and most secondary aerosol species increased by greater than 2% affected by both aerosol physical and chemical process. Although the improvement of PM2.5 seems marginal from the perspective of regional or temporal average, the contribution of ZTD data assimilation on specific pollution episodes at specific stations can be great. The improvement of PM2.5 troughs was in the range of 1-5 μg/m3, while the overestimation of PM2.5 peaks was reduced by few up to dozens μg/m3. This will contribute to the extreme value prediction during pollution episode.

Forecasting Seizure Likelihood With Wearable Technology.

The unpredictability of epileptic seizures exposes people with epilepsy to potential physical harm, restricts day-to-day activities, and impacts mental well-being. Accurate seizure forecasters would reduce the uncertainty associated with seizures but need to be feasible and accessible in the long-term. Wearable devices are perfect candidates to develop non-invasive, accessible forecasts but are yet to be investigated in long-term studies. We hypothesized that machine learning models could utilize heart rate as a biomarker for well-established cycles of seizures and epileptic activity, in addition to other wearable signals, to forecast high and low risk seizure periods. This feasibility study tracked participants' (n = 11) heart rates, sleep, and step counts using wearable smartwatches and seizure occurrence using smartphone seizure diaries for at least 6 months (mean = 14.6 months, SD = 3.8 months). Eligible participants had a diagnosis of refractory epilepsy and reported at least 20 seizures (mean = 135, SD = 123) during the recording period. An ensembled machine learning and neural network model estimated seizure risk either daily or hourly, with retraining occurring on a weekly basis as additional data was collected. Performance was evaluated retrospectively against a rate-matched random forecast using the area under the receiver operating curve. A pseudo-prospective evaluation was also conducted on a held-out dataset. Of the 11 participants, seizures were predicted above chance in all (100%) participants using an hourly forecast and in ten (91%) participants using a daily forecast. The average time spent in high risk (prediction time) before a seizure occurred was 37 min in the hourly forecast and 3 days in the daily forecast. Cyclic features added the most predictive value to the forecasts, particularly circadian and multiday heart rate cycles. Wearable devices can be used to produce patient-specific seizure forecasts, particularly when biomarkers of seizure and epileptic activity cycles are utilized.

Feed- Forward Neural Network based Day Ahead Nodal Pricing

An electricity locational marginal pricing prediction normally recognized by 24-hour day-ahead nodal price forecast. In this paper first collected all physical and technical data i.e. availability of generation and their cost characteristics, real and reactive demands at various buses, transmission capacity availability at various conditions like peak and off-peak conditions. All these input data are used as input for computation of optimal power flow. The nodal prices are calculated with AC-DC optimal power flow methodology for IEEE 30 bus system. The resulted optimal real electricity bus voltages, nodal prices, reactive and real demands, angles have been given as inputs to Artificial Neural Network (ANN) for predict day ahead nodal prices.

Optimization of Meteorological Elements and Icing Prediction Based on Satellite-ground Data Fusion

Icing prediction and early warning depend on icing prediction model and weather element forecast. In order to improve the accuracy of weather forecast, we test and analyze the forecast improvement of the temperature, wind speed and precipitation using numerical model and data fusion and assimilation of satellite-ground data. Through comparative analysis, data fusion and assimilation can significantly improve the forecast level of temperature and precipitation. The accuracy of 72-hour forecast does not decrease significantly compared with that of 24-hour forecast. Therefore, data fusion and assimilation using satellite-ground data has a better guiding significance for the prediction and early warning of 24-72 hour ice cover.

Visualization and Quantitative Assessment to Verify Precipitation Forecasts Associated with Mid-Latitude Cyclones across the Eastern United States

Analysis was conducted to verify forecast against observation precipitation associated with mid-latitude cyclones over the Eastern US in winter and spring 2013 using Geographic Information Systems (GIS). The forecast data are day two 24-hour Quantitative Precipitation Forecasts (QPF) produced by the Global Forecast System (GFS) model. The analysis methods produced categorical geographic error maps of hits, misses and false alarms in spatial relation to the mid-latitude cyclones and traditional verification scores for each day. A hypothesis test was also performed to determine if the GFS mean forecast precipitation over the study area is significantly different from the mean observed precipitation during mid-latitude cyclones. The spatial verification maps, as an analytical and visualization tool, provided evidences on geographical relationship between correct predictions (hits and correct negatives) and incorrect predictions (misses and false alarms). Working together with quantitative scores and hypothesis test, spatial verification maps reveal that the GFS model has a tendency to over forecast precipitation coverage associated with mid-latitude cyclones over the Eastern US and often moves the mid-latitude cyclones too fast.

An application of the multi-physical method of determining error of WRF models to simulate the track and intensity Usagi storm in 2018

This paper presents the test results of the WRF model error determination methods simulating the trajectory and intensity of storm Usagi in 2018. The study conducted three experiments: (1) The Combination of 11 options physical model, 21 composites, no increase in error correlation (MP); (2) Using a set of physical model, 21 composite components, multiplier growth factor l = 6.5 (MI); (3) Using a set of physical model, 21 compositions, no increase in error correlation (PF). Test results show that the multi-physics (MP) method has quite well simulated the intensity as well as the moving direction of the northern cold high pressure in the active Usagi storm area. As a result, The 2018 - Usagi 's trajectory and intensity is simulated in MP test better than in MI test and PF test. Specifically, at the 48-hour forecast term, the orbital prediction error of the MP test is below 350 km which is lower than the two tests (MI and PF), The orbital error in the MP test at the forecast term of 60 and 72 hours is 3-6% reduction in compared with the PF test, and in compared with the MI test, the orbital predictive error of the MP test decreased from 5% to 10% at the 12 hour to 72 hours forecast period. In terms of intensity, absolute error of Pmin (Vmax) in MP test always has lower value than two MI and PF tests. In particular, the absolute error of Vmax in the MP test decreased from 30-40% in compared to the other two trials at all forecasting terms, especially at the forecast term longer than 2 days. Thus, the multi-physics method can be a potential application of determining the error for the model to simulate the trajectory and intensity of storms affecting Vietnam.

Analysis on the Causes of a Weather Forecast Mistake in Yan Liang Airport

Using the conventional weather maps and the T639 model numerical forecast products provided by Micaps system, we re-analyzed and diagnosed the precipitation weather process at Yan Liang Airport on May 26, 2018, and analyzed the reasons for the error in the 24-hour weather forecast on May 25. The analysis shows that the upper trough lines are distributed in the north and south, and the plateau low trough on the south side develops into a low vortex. With better water vapor conditions, the Changjiang-Huaihe shear line develops vigorously. When the colder air from the north-west meets the warm and humid airflow of the Changjiang-Huaihe shear line, the low vortex replenishes sufficient water vapor, and the floating dust brings abundant condensation nuclei to facilitate precipitation. During the whole process, the weather system moved slower than expected, but the weather process produced was more intense.

The impacts of background error covariance on particulate matter assimilation and forecast: An ideal case study with a modal aerosol model over China

Accurate estimation of background error covariance (BEC) is the key to successful data assimilation (DA). In aerosol three-dimensional variational (3DVAR) DA, the National Meteorological Center (NMC) method is typically applied to estimate BEC, which uses the difference between forecasts of dissimilar lengths valid at the common time. The difference will be considerably small when the underestimation of the aerosol is caused by the lack of emissions or the missing of chemical progress, which makes the aerosol concentration field too difficult to constrain. In this study, a modified module for adjusting the BEC of individual aerosol species was developed in the Gridpoint Statistical Interpolation (GSI) 3DVAR system. This module was mainly utilized to expand the standard deviations and the horizontal length scales of BEC for the specified aerosol components by multiplying an adjustment factor. The results of the impacts of BEC on PM10 24-hour forecast indicated that the horizontal length scales take a relatively more important role than the standard deviations. The horizontal length scales affect the influence sphere more significantly, which might be crucial for the longer length forecast. Moreover, the larger and the wider differences of the aerosol initial conditions produced by DA, the longer duration of DA benefits. Using the original BEC, the 24-hour forecast of PM10 reduced fractional error by 13%, while using the modified BEC in DA can decline fractional error by 29%. More work needs to be conducted to investigate how to modify the aerosol BEC in 3DVAR, or how to generate a suitable BEC, which is crucial for aerosol forecast and analysis, especially during the aerosol-polluted period.

NEURAL NETWORK FORECASTING OF ENERGY CONSUMPTION OF A METALLURGICAL ENTERPRISE

The subject of the research is the methods of constructing and training neural networks as a nonlinear modeling apparatus for solving the problem of predicting the energy consumption of metallurgical enterprises. The purpose of this work is to develop a model for forecasting the consumption of the power system of a metallurgical enterprise and its experimental testing on the data available for research of PJSC "Dneprospetsstal". The following tasks have been solved: analysis of the time series of power consumption; building a model with the help of which data on electricity consumption for a historical period is processed; building the most accurate forecast of the actual amount of electricity for the day ahead; assessment of the forecast quality. Methods used: time series analysis, neural network modeling, short-term forecasting of energy consumption in the metallurgical industry. The results obtained: to develop a model for predicting the energy consumption of a metallurgical enterprise based on artificial neural networks, the MATLAB complex with the Neural Network Toolbox was chosen. When conducting experiments, based on the available statistical data of a metallurgical enterprise, a selection of architectures and algorithms for learning neural networks was carried out. The best results were shown by the feedforward and backpropagation network, architecture with nonlinear autoregressive and learning algorithms: Levenberg-Marquard nonlinear optimization, Bayesian Regularization method and conjugate gradient method. Another approach, deep learning, is also considered, namely the neural network with long short-term memory LSTM and the adam learning algorithm. Such a deep neural network allows you to process large amounts of input information in a short time and build dependencies with uninformative input information. The LSTM network turned out to be the most effective among the considered neural networks, for which the indicator of the maximum prediction error had the minimum value. Conclusions: analysis of forecasting results using the developed models showed that the chosen approach with experimentally selected architectures and learning algorithms meets the necessary requirements for forecast accuracy when developing a forecasting model based on artificial neural networks. The use of models will allow automating high-precision operational hourly forecasting of energy consumption in market conditions.
 Keywords: energy consumption; forecasting; artificial neural network; time series.

Assimilation of MWHS-2/FY-3C 183 GHz Channels Using a Dynamic Emissivity Retrieval and Its Impacts on Precipitation Forecasts: A Southwest Vortex Case

The dynamic emissivity retrieved from window channels of the microwave humidity sounder II (MWHS-2) onboard the China Meteorological Administration’s FengYun (FY)-3C polar orbiting satellite can provide more realistic emissivity over lands and potentially improve the numerical weather prediction (NWP) forecasts. However, whether the assimilation with the dynamic emissivity works for the precipitation forecasts over the complex geography is less investigated. In this paper, a typical precipitating case generated by the Southwest Vortex is selected and the Weather Research and Forecasting data assimilation (WRFDA) system is applied to examine the impacts of assimilating MWHS-2/FY-3C with the uses of the emissivity atlas and the dynamic emissivity on the forecasts. The results indicate that the use of the dynamic emissivity retrieved from the 89 GHz channel of MWHS-2/FY-3C apparently increases the used data number for assimilation and does improve the initial fields and the 24-hour forecasts (from 0000 UTC 24 June 2016 to 0000 UTC 25 June 2016) of precipitation distribution and intensity except for the rainfall over 100 mm. But these positive impacts are not evidently better than those with the emissivity atlas. In general, these results still suggest that the future use of the dynamic emissivity in the assimilation over the complex terrain is promising.

Evaluating Forecast Performance and Sensitivity to the GSI EnKF Data Assimilation Configuration for the 28–29 May 2017 Mesoscale Convective System Case

AbstractIn an effort to improve radar data assimilation configurations for potential operational implementation, GSI EnKF data assimilation experiments based on the operational system employed by the Center for Analysis and Prediction of Storms (CAPS) real-time Spring Forecast Experiments are performed. These experiments are followed by 6-h forecasts for an MCS on 28–29 May 2017. Configurations examined include data thinning, covariance localization radii and inflation, observation error settings, and data assimilation frequency for radar observations. The results show experiments that assimilate radar observations more frequently (i.e., 5–10 min) are initially better at suppressing spurious convection. However, assimilating observations every 5 min causes spurious convection to become more widespread with time, and modestly degrades forecast skill through the remainder of the forecast window. Ensembles that assimilate more observations with less thinning of data or use a larger horizontal covariance localization radius for radar data predict fewer spurious storms and better predict the location of observed storms. Optimized data thinning and horizontal covariance localization radii have positive impacts on forecast skill during the first forecast hour that are quickly lost due to the growth of forecast error. Forecast skill is less sensitive to the ensemble spread inflation factors and observation errors tested during this study. These results provide guidance toward optimizing the configuration of the GSI EnKF system. Among the DA configurations tested, the one employed by the CAPS Spring Forecast Experiment produces the most skilled forecasts while remaining computationally efficient for real-time use.

C‐LAEF: Convection‐permitting Limited‐Area Ensemble Forecasting system

AbstractC‐LAEF (Convection‐permitting Limited‐Area Ensemble Forecasting) has been developed at the Austrian national weather service ZAMG (Zentralanstalt für Meteorologie und Geodynamik) and has been running operationally at the European Centre for Medium Range Weather Forecasts (ECMWF) supercomputer since November 2019. It includes (a) an ensemble 3D variational blending technique to deal with atmospheric initial uncertainties, (b) an ensemble of land surface data assimilation to account for uncertainties in the initial land surface conditions, (c) a hybrid stochastic physics perturbation scheme to treat model uncertainties in the different physics parametrization schemes, and (d) a coupling with the global ensemble system IFS‐ENS (Integrated Forecasting System‐ENSemble) to consider uncertainties in the lateral boundary conditions. C‐LAEF has a horizontal resolution of 2.5 km and consists of 16 perturbed members plus one unperturbed control run. It runs four times a day and provides probabilistic forecasts up to 60 hr on a domain covering the whole Alpine area. This article describes the C‐LAEF system in detail and evaluates the relative contributions of the different perturbation techniques. The ensemble variational blending technique and the ensemble surface data assimilation provide additional spread in the first forecast hours, while the hybrid stochastically perturbed parametrization scheme improves the performance of C‐LAEF during the whole forecast range. The performance of C‐LAEF is evaluated extensively over one summer and one winter period and compared with its mesoscale counterpart ALADIN‐LAEF (Aire Limitee Adaptation dynamique Developpement InterNational – Limited‐Area Ensemble Forecasting) and the IFS‐ENS. State‐of‐the‐art probabilistic measures indicate that C‐LAEF is able to outperform ALADIN‐LAEF for all considered upper‐air variables. At the surface, C‐LAEF outperforms ALADIN‐LAEF and IFS‐ENS according to most conventional measures, particularly for wind and precipitation. C‐LAEF benefits from the higher resolution and the explicit treatment of deep convection and can provide more accurate probabilistic information for weather warnings.