Operational Numerical Weather Prediction Research Articles

A modified deep learning weather prediction using cubed sphere for global precipitation

Deep learning (DL), a potent technology to develop Digital Twin (DT), for weather prediction using cubed spheres (DLWP-CS) was recently proposed to facilitate data-driven simulations of global weather fields. DLWP-CS is a temporal mapping algorithm wherein time-stepping is performed through U-NET. Although DLWP-CS has shown impressive results for fields, such as temperature and geopotential height, this technique is complicated and computationally challenging for a complex, non-linear field, such as precipitation, which depends on other prognostic environmental co-variables. To address this challenge, we modify the DLWP-CS and call our technique “modified DLWP-CS” (MDLWP-CS). In this study, we transform the architecture from a temporal to a spatio-temporal mapping (multivariate setup), wherein precursor(s) of precipitation can be used as input. As a proof of concept, as a first simple case, a 2-m surface air temperature is used to predict precipitation using MDLWP-CS. The model is trained using hourly ERA-5 reanalysis and the resulting experimental findings are compared to two benchmark models, viz, the linear regression and an operational numerical weather prediction model, which is the Global Forecast System (GFS). The fidelity of MDLWP-CS is much better compared to linear regression and the results are equivalent to GFS output in terms of daily precipitation prediction with 1 day lag. These results provide an encouraging framework for an efficient DT that can facilitate speedy, high fidelity precipitation predictions.

A hybrid data assimilation system based on machine learning

In the earth sciences, numerical weather prediction (NWP) is the primary method of predicting future weather conditions, and its accuracy is affected by the initial conditions. Data assimilation (DA) can provide high-precision initial conditions for NWP. The hybrid 4DVar-EnKF is currently an advanced DA method used by many operational NWP centres. However, it has two major shortcomings: The complex development and maintenance of the tangent linear and adjoint models and the empirical combination of the results of 4DVar and EnKF. In this paper, a new hybrid DA method based on machine learning (HDA-ML) is presented to overcome these drawbacks. In the new method, the tangent linear and adjoint models in the 4DVar part of the hybrid algorithm can be easily obtained by using a bilinear neural network to replace the forecast model, and a CNN model is adopted to fuse the analysis of 4DVar and EnKF to adaptively obtain the optimal coefficient of combination rather than the empirical coefficient as in the traditional hybrid DA method. The hybrid DA methods are compared with the Lorenz-96 model using the true values as labels. The experimental results show that HDA-ML improves the assimilation performance and significantly reduces the time cost. Furthermore, using observations instead of the true values as labels in the training system is more realistic. The results show comparable assimilation performance to that in the experiments with the true values used as the labels. The experimental results show that the new method has great potential for application to operational NWP systems.

A 3-D Cloud Detection Method for FY-4A GIIRS and Its Application in Operational Numerical Weather Prediction System

A 3-D Cloud Detection Method for FY-4A GIIRS and Its Application in Operational Numerical Weather Prediction System

Verification of operational numerical weather prediction model forecasts of precipitation using satellite rainfall estimates over Africa

AbstractRainfall is an important variable to be able to monitor and forecast across Africa, due to its impact on agriculture, food security, climate‐related diseases and public health. Numerical weather models (NWMs) are an important component of this work, due to their complete spatial coverage, high resolution and ability to forecast into the future. In this study, the spatio‐temporal skill of short‐term forecasts of rainfall across Africa from 2016 through 2018 is evaluated. Specifically, the European Centre for Medium‐Range Weather Forecasts (ECMWF) and the National Centers for Environmental Prediction‐Global Forecast System (NCEP‐GFS) forecast models are verified by Rainfall Estimates 2.0 (RFE2) and African Rainfall Climatology Version 2 (ARC2), which are fused products of satellite and in situ observations and are commonly used in analysis of African rainfall. Model rainfall forecasts show good consistency with the satellite rainfall observations in spatial distribution over Africa on the seasonal timescale. Evaluation metrics of daily and weekly forecasts show high spatial and seasonal variations over the African continent, including a strong link to the location of the inter‐tropical convergence zone (ITCZ) and topographically enhanced precipitation. The rainfall forecasts at 1 week aggregation time are improved against daily forecasts.

Chemistry‐Weather Interacted Model System GRAPES_Meso5.1/CUACE CW V1.0: Development, Evaluation and Application in Better Haze/Fog Prediction in China

AbstractThe Chinese Meteorology Administration chemistry model Chinese Unified Atmospheric Chemistry Environment (CUACE) is online integrated into the mesoscale operational numerical weather prediction (NWP) model (GRAPES_Meso5.1) and aerosol‐cloud‐radiation interaction is achieved to establish the first version (V1) of chemistry‐weather (CW) interacted model GRAPES‐Meso5.1/CUACE CW V1. The most polluted winter 2016–2017 is selected to study the meteorology impacts on haze/fog prediction, the impact of aerosol‐radiation, aerosol‐cloud and CW interaction (ARI, ACI, CWI) on haze/fog prediction and NWP. Single way model without CWI displays reasonable PM2.5 and visibility prediction in general. However, modeled PM2.5 peaks are underestimated and visibility valleys are overestimated during haze/fog pollution, the underestimation of relative humidity (RH) contributes major to this misestimation; CWI model cut the negative bias of PM2.5 peaks and the positive bias of visibility valleys. The improvement of 5 and 3 km low visibility by CWI during severe haze/fog period is more obvious than that of 10 km, which just compensates for the largest deficiency in low visibility prediction related with severe haze/fog by single way model; The NWP including sea level pressures, RH, temperature, wind speed are also improved by CWI from surface to upper troposphere; ARI contributes larger to the predicted PM2.5,visibility and NWP improvement than that of ACI, their relative contributions varies with model vertical height and the overlapping condition of cloud and aerosols. Due to the joint contribution of RH and PM2.5, CWI's improving on visibility is larger than PM2.5. This study illustrates the importance of including CWI in air quality prediction model.

Downscaling of Historical Wind Fields over Switzerland Using Generative Adversarial Networks

Abstract Near-surface wind is difficult to estimate using global numerical weather and climate models, because airflow is strongly modified by underlying topography, especially that of a country such as Switzerland. In this article, we use a statistical approach based on deep learning and a high-resolution digital elevation model to spatially downscale hourly near-surface wind fields at coarse resolution from ERA5 reanalysis from their original 25-km grid to a 1.1-km grid. A 1.1-km-resolution wind dataset for 2016–20 from the operational numerical weather prediction model COSMO-1 of the national weather service MeteoSwiss is used to train and validate our model, a generative adversarial network (GAN) with gradient penalized Wasserstein loss aided by transfer learning. The results are realistic-looking high-resolution historical maps of gridded hourly wind fields over Switzerland and very good and robust predictions of the aggregated wind speed distribution. Regionally averaged image-specific metrics show a clear improvement in prediction relative to ERA5, with skill measures generally better for locations over the flatter Swiss Plateau than for Alpine regions. The downscaled wind fields demonstrate higher-resolution, physically plausible orographic effects, such as ridge acceleration and sheltering, that are not resolved in the original ERA5 fields. Significance Statement Statistical downscaling, which increases the resolution of atmospheric fields, is widely used to refine the outputs of global reanalysis and climate models, most commonly for temperature and precipitation. Near-surface winds are strongly modified by the underlying topography, generating local flow conditions that can be very difficult to estimate. This study develops a deep learning model that uses local topographic information to spatially downscale hourly near-surface winds from their original 25-km resolution to a 1.1-km grid over Switzerland. Our model produces realistic high-resolution gridded wind fields with expected orographic effects but performs better in flatter regions than in mountains. These downscaled fields are useful for impact assessment and decision-making in regions where global reanalysis data at coarse resolution may be the only products available.

Global Distributions of Tropospheric and Stratospheric Gravity Wave Momentum Fluxes Resolved by the 9-km ECMWF Experiments

Abstract Based on 20-day control forecasts by the 9-km Integrated Forecasting System (IFS) at the European Centre for Medium-Range Weather Forecasts (ECMWF) for selected periods of summer and winter events, this study investigates global distributions of gravity wave momentum fluxes resolved by the highest-resolution-ever global operational numerical weather prediction model. Two supplementary datasets, including 18-km ECMWF IFS experiments and the 30-km ERA5, are included for comparison. In the stratosphere, there is a clear dominance of westward momentum fluxes over the winter extratropics with strong baroclinic instability, while eastward momentum fluxes are found in the summer tropics. However, meridional momentum fluxes, locally as important as the above zonal counterpart, show different behaviors of global distribution characteristics, with northward and southward momentum fluxes alternating with each other especially at lower altitudes. Both events illustrate conclusive evidence that stronger stratospheric fluxes are found in the ECMWF forecast with finer resolution, and that ERA5 datasets have the weakest signals in general, regardless of whether regridding is applied. In the troposphere, probability distributions of vertical motion perturbations are highly asymmetric with more strong positive signals especially over latitudes covering heavy rainfall, likely caused by convective forcing. With the aid of precipitation accumulation, a simple filtering method is proposed in an attempt to eliminate those tropospheric asymmetries by convective forcing, before calculating tropospheric wave-induced fluxes. Furthermore, this research demonstrates promising findings that the proposed filtering method could help in reducing the potential uncertainties with respect to estimating tropospheric wave-induced fluxes. Finally, absolute momentum flux distributions with proposed approaches are presented, for further assessment in the future.

Grid refinement in ICON v2.6.4

Abstract. This article describes the implementation of grid refinement in the ICOsahedral Nonhydrostatic (ICON) modeling system. It basically follows the classical two-way nesting approach known from widely used mesoscale models like MM5 or WRF, but it differs in the way feedback from fine grids to coarser grids is applied. Moreover, the ICON implementation supports vertical nesting in the sense that the upper boundary of a nested domain may be lower than that of its parent domain. Compared to the well-established implementations on quadrilateral grids, new methods had to be developed for interpolating the lateral boundary conditions from the parent domain to the child domain(s). These are based on radial basis functions (RBFs) and partly apply direct reconstruction of the prognostic variables at the required grid points, whereas gradient-based extrapolation from parent to child grid points is used in other cases. The runtime flow control is written such that limited-area domains can be processed identically to nested domains except for the lateral boundary data supply. To demonstrate the functionality and quality of the grid nesting in ICON, idealized tests based on the Jablonowski–Williamson test case (Jablonowski and Williamson, 2006) and the Schär mountain wave test case (Schär et al., 2002) are presented. The results show that the numerical disturbances induced at the nest boundaries are small enough to be negligible for real applications. This is confirmed by experiments closely following the configuration used for operational numerical weather prediction at DWD, which demonstrate that a regional refinement over Europe has a significant positive impact on the forecast quality in the Northern Hemisphere.

Variational quantum solutions to the advection–diffusion equation for applications in fluid dynamics

Constraints in power consumption and computational power limit the skill of operational numerical weather prediction by classical computing methods. Quantum computing could potentially address both of these challenges. Herein, we present one method to perform fluid dynamics calculations that takes advantage of quantum computing. This hybrid quantum–classical method, which combines several algorithms, scales logarithmically with the dimension of the vector space and quadratically with the number of nonzero terms in the linear combination of unitary operators that specifies the linear operator describing the system of interest. As a demonstration, we apply our method to solve the advection–diffusion equation for a small system using IBM quantum computers. We find that reliable solutions of the equation can be obtained on even the noisy quantum computers available today. This and other methods that exploit quantum computers could replace some of our traditional methods in numerical weather prediction as quantum hardware continues to improve.

Diagnostics of Tropical Variability for Numerical Weather Forecasts

Abstract Tropical precipitation and circulation are often coupled and span a vast spectrum of scales from a few to several thousands of kilometers and from hours to weeks. Current operational numerical weather prediction (NWP) models struggle with representing the full range of scales of tropical phenomena. Synoptic to planetary scales are of particular importance because improved skill in the representation of tropical larger-scale features such as convectively coupled equatorial waves (CCEWs) has the potential to reduce forecast error propagation from the tropics to the midlatitudes. Here we introduce diagnostics from a recently developed tropical variability diagnostics toolbox, where we focus on two recent versions of NOAA’s Unified Forecast System (UFS): operational GFSv15 forecasts and experimental GFSv16 forecasts from April to October 2020. The diagnostics include space–time coherence spectra to identify preferred scales of coupling between circulation and precipitation, pattern correlations of Hovmöller diagrams to assess model skill in zonal propagation of precipitating features, CCEW skill assessment, plus a diagnostic aimed at evaluating moisture–convection coupling in the tropics. Results show that the GFSv16 forecasts are slightly more realistic than GFSv15 in their coherence between precipitation and model dynamics at synoptic to planetary scales, with modest improvements in moisture convection coupling. However, this slightly improved performance does not necessarily translate to improvements in traditional precipitation skill scores. The results highlight the utility of these diagnostics in the pursuit of better understanding of NWP model performance in the tropics, while also demonstrating the challenges in translating model advancements into improved skill.

Spatial prediction of outdoor thermal comfort using an operational NWP model for city and regional scales: A case study in Korea

The outdoor thermal environment has key implications for human health, and numerical simulation models have been used to predict thermal comfort. The potential of a numerical weather prediction (NWP)-based thermal environment prediction model has not yet been comprehensively explored. Using a predicted mean vote-based thermal comfort index (i.e., perceived temperature [PT]) in South Korea, this study developed and evaluated a thermal environment prediction model using an operational NWP model of the Korea Meteorological Administration. The suitability of four bias-adjustment methods to reduce bias in PT prediction was examined, applying the outputs of the NWP model using in-situ data for 2018–2019. Bias-adjusted PT prediction performance was assessed for 2020 based on PT and thermo-physiological stress categories for South Korea. The predicted PT successfully represented PT spatial information. Mean absolute errors (MAEs) were 2.17 and 1.15 °C for hourly and daily PT, and the bias-adjustment method improved prediction accuracy by 18 and 36%, respectively, based on the MAE. PT prediction via NWP modeling can be further improved by reducing the statistical error in surface solar radiation modeling. The proposed scheme can be applied in urban heat mitigation, developing early warning systems for thermal health risks, energy consumption prediction, and tourism planning.

GNSS Storm Nowcasting Demonstrator for Bulgaria

Global Navigation Satellite System (GNSS) is an established atmospheric monitoring technique delivering water vapour data in near-real time with a latency of 90 min for operational Numerical Weather Prediction in Europe within the GNSS water vapour service (E-GVAP). The advancement of GNSS processing made the quality of real-time GNSS tropospheric products comparable to near-real-time solutions. In addition, they can be provided with a temporal resolution of 5 min and latency of 10 min, suitable for severe weather nowcasting. This paper exploits the added value of sub-hourly real-time GNSS tropospheric products for the nowcasting of convective storms in Bulgaria. A convective Storm Demonstrator (Storm Demo) is build using real-time GNSS tropospheric products and Instability Indices to derive site-specific threshold values in support of public weather and hail suppression services. The Storm Demo targets the development of service featuring GNSS products for two regions with hail suppression operations in Bulgaria, where thunderstorms and hail events occur between May and September, with a peak in July. The Storm Demo real-time Precise Point Positioning processing is conducted with the G-Nut software with a temporal resolution of 15 min for 12 ground-based GNSS stations in Bulgaria. Real-time data evaluation is done using reprocessed products and the achieved precision is below 9 mm, which is within the nowcasting requirements of the World Meteorologic Organisation. For the period May–September 2021, the seasonal classification function for thunderstorm nowcasting is computed and evaluated. The probability of thunderstorm detection is 83%, with a false alarm ration of 38%. The added value of the high temporal resolution of the GNSS tropospheric gradients is investigated for a storm case on 24–30 August 2021. Real-time tropospheric products and classification functions are integrated and updated in real-time on a publicly accessible geoportal.

Radar Reflectivity–Based Model Initialization Using Specified Latent Heating (Radar-LHI) within a Diabatic Digital Filter or Pre-Forecast Integration

Abstract A technique for model initialization using three-dimensional radar reflectivity data has been developed and applied within the NOAA 13-km Rapid Refresh (RAP) and 3-km High-Resolution Rapid Refresh (HRRR) regional forecast systems. This technique enabled the first assimilation of radar reflectivity data for operational NOAA forecast models, critical especially for more accurate short-range prediction of convective storms. For the RAP, the technique uses a diabatic digital filter initialization (DFI) procedure originally deployed to control initial inertial gravity wave noise. Within the forward-model integration portion of diabatic DFI, temperature tendencies obtained from the model cloud/precipitation processes are replaced by specified latent heating–based temperature tendencies derived from the three-dimensional radar reflectivity data, where available. To further refine initial conditions for the convection-allowing HRRR model, a similar procedure is used in the HRRR, but without DFI. Both of these procedures, together called the “Radar-LHI” (latent heating initialization) technique, have been essential for initialization of ongoing precipitation systems, especially convective systems, within all NOAA operational versions of the 13-km RAP and 3-km HRRR models extending through the latest implementation upgrade at NCEP in 2020. Application of the latent heat–derived temperature tendency induces a vertical circulation with low-level convergence and upper-level divergence in precipitation systems. Retrospective tests of the Radar-LHI technique show significant improvement in short-range (0–6 h) precipitation system forecasts, as revealed by reflectivity verification scores. Results presented document the impact on HRRR reflectivity forecasts of the radar reflectivity initialization technique applied to the RAP alone, HRRR alone, and both the RAP and HRRR. Significance Statement The large forecast uncertainty of convective situations, even at short lead times, coupled with the hazardous weather they produce, makes convective storm prediction one of the most significant short-range forecast challenges confronting the operational numerical weather prediction community. Prediction of heavy precipitation events also requires accurate initialization of precipitation systems. An innovative assimilation technique using radar reflectivity data to initialize NOAA operational weather prediction models is described. This technique, which uses latent heating specified from radar reflectivity (and can accommodate lightning data and other convection/precipitation indicators), was first implemented in 2009 at NOAA/NCEP and continues to be used in 2022 in the NCEP-operational RAP and HRRR models, making it a backbone of the NOAA rapidly updated numerical weather prediction capability.

Verification and Validation of the COSMIC-2 Excess Phase and Bending Angle Algorithms for Data Quality Assurance at STAR

In recent years, Global Navigation Satellite System (GNSS) radio occultation (RO) has become a critical observation system for global operational numerical weather prediction. Constellation Observing System for Meteorology, Ionosphere, Climate (COSMIC) 2 (COSMIC-2) has been a backbone RO mission for NOAA. NOAA also began to purchase RO data from commercial sources in 2020. To ensure the consistent quality of RO data from different sources, NOAA Center for Satellite Applications and Research (STAR) has developed capabilities to process all available RO data from different missions. This paper describes the STAR RO processing systems which convert the pseudo-range and carrier phase observations to excess phases and bending angles (BAs). We compared our COSMIC-2 data products with those processed by the University Corporation for Atmospheric Research (UCAR) COSMIC Data Analysis and Archive Center (CDAAC). We processed more than twelve thousand COSMIC-2 occultation profiles. Our results show that the excess phase difference between UCAR and STAR is within a few centimeters at high altitudes, although the difference increases towards the lower atmosphere. The BA profiles derived from the excess phase are consistent with UCAR. The mean relative BA differences at impact height from 10 to 30 km are less than 0.1% for GLObal NAvigation Satellite System (GLONASS) L2C signals and Global Positioning System (GPS) L2C and L2P signals. The standard deviations are 1.15%, 1.15%, and 1.32% for GLONASS L2C signal and for GPS L2C and L2P signals, respectively. The BA profiles agree with those derived from European Center for Medium-range Weather Forecast (ECMWF) reanalysis version 5 (ERA5). The Signal-to-Noise-Ratio (SNR) plays an essential role in the processing. The STAR BA profiles with higher L1 SNRs (L1 at 80 km) tend to yield more consistent results than those from UCAR, with a negligible difference and a smaller deviation than lower SNR profiles. Profiles with lower SNR values tend to show a more significant standard deviation towards the surface during the open-loop stage in the lower troposphere than those of higher SNR. We also found that the different COSMIC-2 clock solutions could contribute to the significant relative BA difference at high altitudes; however, it has little effect on the lower troposphere comparisons given larger BA values.

Statistical Evaluation of the Temperature Forecast Error in the Lower‐Level Troposphere on Short‐Range Timescales Induced by Aerosol Variability

AbstractThis study statistically evaluated the aerosol impact on the temperature error in the lower‐level troposphere in short‐range numerical weather prediction (NWP). The Global Ensemble Forecast System version 12 (GEFSv12) reforecast exhibited large‐temperature errors in high‐loading areas (North India, Africa, South America, and China). In 1‐day GEFSv12 forecasts, the largest average temperature error occurred in the aerosol optical depth (AOD) peak month, and the daily error distribution corresponded to the daily AOD distribution. Even though the temperature error in the 1‐day operational forecasts was smaller than that in the GEFSv12 forecasts, the forecast uncertainties in the operational forecasts were comparable to those in 3‐day GEFSv12 forecasts over high‐loading areas. The daily temperature errors in all NWP models exhibited a correlation coefficient of ∼0.5–0.6 for the AOD over Central Africa and northern South America and ∼0.3–0.6 for AOD anomalies over China and northern South America. These results indicated that the interannual aerosol variability contributed 25–36% to errors, and the daily variability contributed 10%–36% to temperature errors in 3‐day forecasts. Although the correlation was low, aerosol impacts also emerged in North India and Central Africa. Partial correlation and composite analysis suggested that the direct effect mainly influenced temperature forecast errors over northern South America, whereas both direct and indirect effects influenced temperature errors over China. Model intercomparison revealed that operational NWP models could experience common forecast errors associated with aerosols in high‐loading areas.

Advantage of 30‐s‐Updating Numerical Weather Prediction With a Phased‐Array Weather Radar Over Operational Nowcast for a Convective Precipitation System

AbstractConvective precipitation systems in the summer often cause sudden heavy precipitation and largely affect various human activities, but the rapid evolution limits our predicting capability. Phased‐array weather radars (PAWRs) with a high spatiotemporal resolution are useful for observing such precipitation system. A recently developed numerical weather prediction (NWP) system assimilates PAWR observations with a 500‐m mesh NWP model. It initiates 30‐min extended forecasts every 30 s, much more frequently than the operational NWP and nowcasting systems. This study investigates the benefits of the 30‐s‐updating NWP system in a single but representative convective precipitation event in which a convective cloud developed within 10 min, and its evolution was not well predicted by operational precipitation nowcasting. The rapidly updating NWP system successfully predicts the evolution of the convective cloud. Assimilating the PAWR observations every 30 s continuously modifies the moisture and dynamical fields and improves the forecast accuracy consistently.

Land–Snow Data Assimilation Including a Moderately Coupled Initialization Method Applied to NWP

Abstract Initialization methods are needed for geophysical components of Earth system prediction models. These methods are needed from medium-range to decadal predictions and also for short-range Earth system forecasts in support of safety (e.g., severe weather), economic (e.g., energy), and other applications. Strongly coupled land–atmosphere data assimilation (SCDA), producing balanced initial conditions across the land–atmosphere components, has not yet been introduced to operational numerical weather prediction (NWP) systems. Most NWP systems have evolved separate data assimilation (DA) procedures for the atmosphere versus land/snow system components. This separated method has been classified as a weakly coupled DA system (WCDA). In the NOAA operational short-range weather models, a moderately coupled land–snow–atmosphere assimilation method (MCLDA) has been implemented, a step forward from WCDA toward SCDA. The atmosphere and land (including snow) variables are both updated within the DA using the same set of observations (aircraft, radiosonde, satellite radiances, surface, etc.). Using this assimilation method, land surface state variables have cycled continuously for 6 years since 2015 for the 3-km NOAA HRRR model and with CONUS cycling since 1997. Month-long experiments were conducted with and without MCLDA for both winter and summer seasons using the 13-km Rapid Refresh model with atmosphere (50 levels), soil (9 levels), and snow (up to 2 layers if present) on the same horizontal grid. Improvements were evident for 2-m temperature for all times of day out to 6–12 h for both seasons but stronger in winter. Better temperature forecasts were also shown in the 1000–900-hPa layer corresponding roughly to the boundary layer. Significance Statement Accuracy of weather models depends on accurate initial conditions for soil temperature and moisture as well as for the atmosphere itself. This paper describes a moderately coupled data assimilation method that modifies soil conditions based on forecast error corrections indicated by atmospheric observations. This method has been tested for a month-long period in summer and winter and shown to consistently improve short-range forecasts of 2-m temperature and moisture. This coupled data assimilation method is used already in NOAA operational short-range models to improve its prediction skill for clouds, convective storms, and general weather conditions.

The history of UK weather forecasting: the changing role of the central guidance forecaster. Part 2: the birth of operational numerical weather prediction

In the second paper in this series, we review how, following the advent of the computer, highly simplified numerical weather prediction models were developed during the 1950s and 1960s, examples of which are shown. Although models provided guidance on the possible behaviour of synoptic systems for up to 2–3 days ahead in time, forecasters still faced the major challenge of inferring the distribution of weather. Satellite imagery proved a valuable addition to the forecasters' toolkit.

Development of model output statistics based on the least absolute shrinkage and selection operator regression for forecasting next‐day maximum temperature in South Korea

AbstractRegression models for model output statistics (MOS) based on least absolute shrinkage and selection operator methods were developed to forecast next‐day maximum surface air temperature (TMAX) during the warm season in South Korea. The forecast fields from the operational numerical weather prediction (NWP) system of the Korean Meteorological Administration for global and local forecasts and the observed TMAX data in 225 observation stations were used as input variables for the MOS. The training period was July and August (JA) from 2015 to 2018, and the regression models were tested using data from JA 2019. As a result of hindcasting for the test period, the MOS models performed significantly better for next‐day TMAX forecasting over South Korea than the numerical models during JA 2019. The mean TMAX errors were reduced by over 1°C in MOSs compared to those in the numerical models. However, the TMAX forecast performance was generally lower in the higher‐resolution NWP Local Data Assimilation and Prediction System (LDAPS)‐based MOS (LMOS) than in the lower‐resolution NWP Global Data Assimilation and Prediction System (GDAPS)‐based MOS. This pattern was dominant when LDAPS simulated the TMAX more accurately than average. In particular, the random TMAX error of LDAPS was larger than that of GDAPS during the training period, and a positive random error of TMAX was magnified in LMOS. Because the other predictors forecasted from LDAPS can be associated with lower TMAX forecast performance of LMOS, in addition to TMAX effects as a predictor, a new MOS was developed using both LDAPS and GDAPS outputs. The forecast accuracy was improved by up to 0.3°C when the forecast fields from the GDAPS substituted several LMOS predictors, even though TMAX was the primary predictor for LMOS.

A new global ocean ensemble system at the Met Office: Assessing the impact of hybrid data assimilation and inflation settings

AbstractWe have developed a global ocean and sea‐ice ensemble forecasting system based on the operational forecasting ocean assimilation model (FOAM) system run at the Met Office. The ocean model Nucleus for European Modelling of the Ocean (NEMO) and the community ice code (CICE) sea‐ice model are run at 1/4 resolution and the system assimilates data using a three‐dimensional variational assimilation (3DVar) version of NEMOVAR. This data assimilation (DA) system can perform hybrid ensemble/variational assimilation. A 36‐member ensemble of hybrid ensemble variational assimilation systems with perturbed observations (values and locations) has been set up, with each member forced at the surface by a separate member of the Met Office Global–Regional Ensemble Prediction System (MOGREPS‐G). The unperturbed member is forced by atmospheric fields from the Met Office operational numerical weather prediction (NWP) deterministic system. The system includes stochastic model perturbations and a relaxation to prior spread (RTPS) inflation scheme. A control run of the system using an ensemble of 3DVars is shown to be generally reliable for Sea‐Level Anomaly (SLA), temperature, and salinity (the ensemble spread being a good representation of the uncertainty in the ensemble mean), although the ensemble is underspread in eddying regions. The ensemble mean gives a 4% reduction in error in SLA compared with the deterministic 3DVar system currently used operationally. The system was tested with different weights for the ensemble component of the hybrid background‐error covariance matrix and different inflation factors. The best results, in terms of short‐range forecast error and ensemble reliability statistics, were obtained with hybrid three‐dimensional ensemble variational DA (3DEnVar). The RTPS inflation scheme is shown to be beneficial in producing an appropriate ensemble spread in response to hybrid DA. 3DEnVar with an ensemble hybrid weight of 0.8 leads to a reduction of 20% (5%) in the ensemble mean error for SLA (profile temperature and salinity) compared with an ensemble of standard 3DVars.