Numerical Weather Prediction Models Research Articles

Improvements in the land surface configuration to better simulate seasonal snow cover in the European Alps with the CNRM-AROME (cycle 46) convection-permitting regional climate model

Abstract. Snow cover modeling remains a major challenge in climate and numerical weather prediction (NWP) models even in recent versions of high-resolution coupled surface–atmosphere (i.e., at kilometer scale) regional models. Evaluation of recent climate simulations, carried out as part of the WCRP-CORDEX Flagship Pilot Study on Convection (FPSCONV) with the CNRM-AROME convection-permitting regional climate model at 2.5 km horizontal resolution, has highlighted significant snow cover biases, severely limiting its potential in mountain regions. These biases, which are also found in AROME numerical weather prediction (NWP) model results, have multiple causes, involving atmospheric processes and their influence on input data to the land surface models in addition to deficiencies of the land surface model itself. Here we present improved configurations of the SURFEX-ISBA land surface model used in CNRM-AROME. We thoroughly evaluated these configurations on their ability to represent seasonal snow cover across the European Alps. Our evaluation was based on coupled simulations spanning the winters of 2018–2019 and 2019–2020, which were compared against remote sensing data and in situ observations. More specifically, the study tests the influence of various changes in the land surface configuration, such as the use of multi-layer soil and snow schemes, the division of the energy balance calculation by surface type within a grid cell (multiple patches), and new physiographic databases and parameter adjustments. Our findings indicate that using only more detailed individual components in the surface model did not improve the representation of snow cover due to limitations in the approach used to account for partial snow cover within a grid cell. These limitations are addressed in further configurations that highlight the importance, even at kilometer resolution, of taking into account the main subgrid surface heterogeneities and improving representations of interactions between fractional snow cover and vegetation. Ultimately, we introduce a land surface configuration, which substantially improves the representation of seasonal snow cover in the European Alps in coupled CNRM-AROME simulations. This holds promising potential for the use of such model configurations in climate simulations and numerical weather prediction both for AROME and other high-resolution climate models.

Evaluation of forecasted wind speed at turbine hub height and wind ramps by five NWP models with observations from 262 wind farms over China

AbstractAccurate wind speed forecasts are essential for optimizing the efficiency of wind energy operations. Currently, there is limited research on nationwide assessment of predictive performance in multiple numerical weather prediction (NWP) models for wind speed at turbine hub height over China, especially concerning wind ramp events. Utilizing observed measurements from 262 wind farms, this study evaluated the performance of five NWP models in forecasting the mean state and spatiotemporal variations of wind speed as well as wind ramps. The results indicated that the European Center for Medium‐Range Weather Forecast Integrated Forecasting System (ECMWF–IFS) performed the best in forecasting climatological wind speed with a temporal correlation coefficient (TCC) of 0.74 and root mean square error (RMSE) of 2.34 m s−1. Although not widely utilized in China, the model from Meteo‐France (MF–ARPEGE) showed promising potential for wind energy forecasting with a TCC of 0.72 and RMSE of 2.45 m s−1. In terms of temporal variations of wind speed, all the models could reasonably predict the seasonal variations of wind speed, whereas only three NWP models were able to capture the characteristics of the observed diurnal variation. An error decomposition analysis further revealed that the primary source of predicted error for wind speed was the sequence error component (SEQU), indicating the model errors were mainly attributed from the temporal inconsistency between forecasts and observations. Furthermore, the occurrences of wind ramps were generally underestimated by NWP models, while this shortcoming can be partly overcome by improving the time resolution of NWP models.

At-scale Model Output Statistics in mountain environments (AtsMOS v1.0)

Abstract. This paper introduces the AtsMOS (At-scale Model Output Statistics) workflow, designed to enhance mountain meteorology predictions through the downscaling of coarse numerical weather predictions using local observational data. AtsMOS provides a modular, open-source toolkit for local and large-scale forecasting of various meteorological variables through modified model output statistics – and may be applied to data from a single station or an entire network. We demonstrate its effectiveness through an example application at the summit of Mt. Everest, where it improves the prediction of both meteorological variables (e.g. wind speed, temperature) and derivative variables (e.g. facial frostbite time) critical for mountaineering safety. As a bridge between numerical weather prediction models and ground observations, AtsMOS contributes to hazard mitigation, water resource management, and other weather-dependent issues in mountainous regions and beyond.

A Rare Tropical Cyclone Associated With Southwest Monsoon Over the Northern Part of the South China Sea—Tropical Storm Maliksi

AbstractThis study examines the characteristics and development of Tropical Storm Maliksi, which is a special case of tropical cyclone developed in the northwestern part of the South China Sea during a southwest monsoon outbreak. Detailed analyses were conducted using observational data and forecast products. Surface observations, radar wind profilers, aircraft data, and satellite products were used to evaluate Maliksi's wind structure, revealing multiple circulation centers and gale force winds. Vertical wind profiles, warm core structure, wind waves, and the influence of sea temperatures and salinity on Maliksi's intensification were investigated. Regarding forecasting, AI‐based models outperformed conventional numerical weather prediction (NWP) models in predicting Maliksi's initial development, though both struggled to capture the persistence of strong winds as the system moved inland. High‐resolution NWP simulations were employed to examine terrain‐induced wind variability around Hong Kong International Airport, revealing the mountain wake effect and uneven wind and turbulence distribution. These findings provide insights into the challenges of forecasting and monitoring such tropical cyclones, and highlight the need for enhanced observational platforms and forecasting tools along coastlines vulnerable to these systems.

Weather Prediction with Feature Selection and Random Forest

Accurate weather prediction is crucial for various sectors, including agriculture, disaster management, and aviation. Traditional weather prediction relies heavily on numerical weather prediction (NWP) models that simulate atmospheric processes using mathematical equations. While these methods have been the backbone of forecasting for decades, they often require significant computational resources and may need help capturing localized weather events. In contrast, machine learning models can quickly analyze large datasets to identify patterns and relationships that traditional methods might miss. This paper employs a Random Forest model to investigate the impact of feature selection on weather prediction accuracy. This paper identifies those critical to model performance by systematically excluding individual features, offering valuable insights into how ML can enhance traditional forecasting techniques. The results underscore the importance of features such as “Present_Tmin” and “LDAPS_LH”, which significantly influence model performance as measured by Root Mean Square Error (RMSE). This study contributes to a better understanding of feature selection in developing accurate weather prediction models. It lays the groundwork for future research to improve prediction accuracy through advanced techniques and expanded datasets.

Recommended coupling to global meteorological fields for long-term tracer simulations with WRF-GHG

Abstract. Atmospheric transport models are often used to simulate the distribution of greenhouse gases (GHGs). This can be in the context of forward modeling of tracer transport using surface–atmosphere fluxes or flux estimation through inverse modeling, whereby atmospheric tracer measurements are used in combination with simulated transport. In both of these contexts, transport errors can bias the results and should therefore be minimized. Here, we analyze transport uncertainties in the commonly used Weather Research and Forecasting (WRF) model coupled with the greenhouse gas module (WRF-GHG), enabling passive tracer transport simulation of CO2 and CH4. As a mesoscale numerical weather prediction model, WRF's transport is constrained by global meteorological fields via initialization and at the lateral boundaries of the domain of interest. These global fields were generated by assimilating various meteorological data to increase the accuracy of modeled fields. However, in limited-domain models like WRF, the winds in the center of the domain can deviate considerably from these driving fields. As the accuracy of the wind speed and direction is critical to the prediction of tracer transport, maintaining a close link to the observations across the simulation domain is desired. On the other hand, a link that is too close to the global meteorological fields can degrade performance at smaller spatial scales that are better represented by the mesoscale model. In this work, we evaluated the performance of strategies for keeping WRF's meteorology compatible with meteorological observations. To avoid the complexity of assimilating meteorological observations directly, two main strategies of coupling WRF-GHG with ERA5 meteorological reanalysis data were tested over a 2-month-long simulation over the European domain: (a) restarting the model daily with fresh initial conditions (ICs) from ERA5 and (b) nudging the atmospheric winds, temperatures, and moisture to those of ERA5 continuously throughout the simulation period, using WRF's built-in four-dimensional data assimilation (FDDA) in grid-nudging mode. Meteorological variables and simulated mole fractions of CO2 and CH4 were compared against observations to assess the performance of the different strategies. We also compared planetary boundary layer height (PBLH) with radiosonde-derived estimates. Either nudging or daily restarts similarly improved the meteorology and GHG transport in our simulations, with a small advantage of using both methods in combination. However, notable differences in soil moisture were found that accumulated over the course of the simulation when not using frequent restarts. The soil moisture drift had an impact on the simulated PBLH, presumably via changing the Bowen ratio. This is partially mitigated through nudging without requiring daily restarts, although not entirely alleviated. Soil moisture drift did not have a noticeable impact on GHG performance in our case, likely because it was dominated by other errors. However, since the PBLH is critical for accurately simulating GHG transport, we recommend transport model setups that tie soil moisture to observations. Our method of frequently re-initializing simulations with meteorological reanalysis fields proved suitable for this purpose.

High-Resolution Wind Speed Estimates for the Eastern Mediterranean Basin: A Statistical Comparison Against Coastal Meteorological Observations

Wind speed (and direction) estimated from numerical weather prediction (NWP) models is essential to wind energy applications, especially in the absence of reliable fine scale spatio-temporal wind information. This study evaluates four high-resolution wind speed numerical datasets (UERRA MESCAN-SURFEX, CERRA, COSMO-REA6, and NEWA) against in situ observations from coastal meteorological stations in the eastern Mediterranean basin. The evaluation is based on statistical comparisons of long-term wind speed data from 2009 to 2018 and involves an in-depth statistical comparison as well as a preliminary wind power density assessment at or near the meteorological station locations. The results show that while all datasets provide valuable insights into regional wind variability, there are notable differences in model performance. COSMO-REA6 and UERRA exhibit higher variability in wind speed but tend to underestimate extreme values, particularly in the southern coastal areas, whereas CERRA and NEWA provided closer fits to observed wind speeds, with CERRA showing the highest correlation at most stations. NEWA data, where available, overestimate average wind speeds but capture extreme values well. The comparison reveals that while all datasets provide valuable insights into the spatial and temporal variability of wind resources, their performance varies by location and season, emphasizing the need for the careful selection and potential calibration of these models for accurate wind energy assessments. The study provides essential groundwork for leveraging these datasets in planning and optimizing offshore wind energy projects, contributing to the region’s transition to renewable energy sources.

RUFCO: a deep-learning framework to post-process subseasonal precipitation accumulation forecasts

Abstract Post-processing is a critical step in attaining calibrated and reliable probabilistic forecast output from numerical weather prediction models. A novel deep-learning framework is proposed to post-process 20 years of 7- and 14-day precipitation accumulation reforecasts from the Global Ensemble Forecast System at subseasonal timescales (week 1, week 2, and combined weeks 3-4 forecasts) over the contiguous United States. The network builds upon previous studies and is a combination of three parallel-trained components suitable for subseasonal prediction. The first is a ResUnet architecture which learns non-linear relationships between binned observed precipitation and input images of weather and geographical variables. The second conditions the network to the month-of-year via a Feature-wise Linear Modulation (FiLM) layer. The third helps the network learn when to revert the forecast to that of climatology. The RUFCO (named for its components ResUnet, FiLM, and Climatological-Offramp) forecasts are compared against raw and climatological forecasts as well as those from a state-of-the-art distributional regression post-processing model, “CSGD,” and a simple bias-corrected model. At week 1, every method exhibited a competitive advantage over climatological forecasts. At week 2, RUFCO generated forecasts with statistically significant improvement over climatology at 82-94% of the domain, beating CSGD’s coverage of 76-90% of the grid points. At week 3, RUFCO’s skillful coverage was 65-85% while CSGD’s dropped to only 12-37%. At the longer lead times, RUFCO achieved the highest domain-averaged skill scores across seasons. However, the network tends to “smooth” forecast skill making it less competitive with CSGD in limited areas with strongly spatially-varying biases.

Simulation of the Neutral Atmospheric Flow Using Multiscale Modeling: Comparative Studies for SimpleFoam and Fluent Solver

The reproduced planetary boundary layer (PBL) wind is commonly applied in downscaled simulations using commercial CFD codes with Reynolds-averaged Navier–Stokes (RANS) turbulence modeling. When using the turbulent inlets calculated by numerical weather prediction models (NWP), adjustments of the turbulence eddy viscosity closures and wall function formulations are concerned with maintaining the fully developed wind profiles specified at the inlet of CFD domains. The impact of these related configurations is worth discussing in engineering applications, especially when commercial codes restrict the internal modifications. This study evaluates the numerical performances of open-source OpenFOAM 2.3.0 and commercial Fluent 17.2 codes as supplementary scientific comparisons. This contribution focuses on the modified turbulence closures to incorporate turbulent profiles produced from mesoscale PBL parameterizations and the modified wall treatments relating to the roughness length. The near-ground flow features are evaluated by selecting the flat terrains and the classical Askervein benchmark case. The improvement in near-ground wind flow under the downscaled framework shows satisfactory performance in the open-source CFD platform. This contributes to engineers realizing the micro-siting of wind turbines and quantifying terrain-induced speed-up phenomena under the scope of wind-resistant design.

Weather Prediction with Machine Learning

Weather forecasting is the technical process of predicting atmospheric conditions at a location. Many scientists have also been trying to predict the weather, both formally and informally, for years. Weather forecasting is done by collecting data from specific locations with different climate attributes. Traditional numerical weather prediction models have made significant progress, but there are still limitations in their accuracy, especially in detecting local weather events. In recent years, machine learning techniques have become a powerful tool for improving weather forecasts by exploiting the large amounts of data and complex patterns inherent in atmospheric systems. We propose a machine- learning model that uses historical data to train the model. The model is then used to predict weather with better accuracy than traditional models.

Role of midtropospheric heating and surface forcing on monsoon intraseasonal transitions

AbstractMonsoon intraseasonal transitions are intrinsically chaotic; hence, their predictability is limited. In the past the majority of studies were conducted solely keeping in account of dynamic prospective like boreal summer intraseasonal oscillation or Madden–Julian oscillation; however its thermodynamic aspect got limited attention. The current study looks at the internal thermodynamic mechanisms pertaining to monsoon intraseasonal transition using two independent sets of reanalysis data (European Centre for Medium‐range Weather Forecasts Reanalysis version 5 and Indian Monsoon Data Assimilation and Analysis), along with the observed cloud type from Met Office Integrated Data Archive System Land and Marine surface station data. A lead–lag composite analysis shows that the midtropospheric warming by diabatic heating is a thermal control to restrict the deepening of convective cloud (cumulonimbus) as the wet spell progresses. This midtroposhperic heating cuts off the formation of deep convective clouds and results in the formation of relatively shallow clouds, such as stratocumulus and stratus. On the other hand, surface forcing during dry spells humidifies the troposphere by detrainment from the shallow convection with the formation of cumulus and cumulus congestus clouds. This facilitates the preconditioning of the atmosphere for deep convection and results in the onset of wet spells by the formation of cumulonimbus clouds. The comprehensive approach used in this work may further help in identifying the limitation of the predictability in monsoon transition period from the numerical weather prediction models.

Studies on Heavy Precipitation in Portugal: A Systematic Review

This systematic review, based on an adaptation of the Preferred Reporting Items for Systematic Reviews and Meta-Analyses (PRISMA) statement from 2020, focuses on studies of the atmospheric mechanisms underlying extreme precipitation events in mainland Portugal, as well as observed trends and projections. The 54 selected articles cover the period from 2000 to 2024, in which the most used keywords are “portugal” and “extreme precipitation”. Of the 54, 23 analyse trends and climate projections of precipitation events, confirming a decrease in total annual precipitation, especially in autumn and spring, accompanied by an increase in the frequency and intensity of extreme precipitation events in autumn, spring and winter. Several articles (twelve) analyse the relationship between synoptic-scale circulation and heavy precipitation, using an atmospheric circulation types approach. Others (two) establish the link with teleconnection patterns, namely the North Atlantic Oscillation (NAO), and still others (three) explore the role of atmospheric rivers. Additionally, five articles focus on evaluating databases and Numerical Weather Prediction (NWP) models, and nine articles focus on precipitation-related extreme weather events, such as tornadoes, hail and lightning activity. Despite significant advances in the study of extreme precipitation events in Portugal, there is still a lack of studies on hourly or sub-hourly scales, which is critical to understanding mesoscale, short-lived events. Several studies show NWP models still have limitations in simulating extreme precipitation events, especially in complex orography areas. Therefore, a better understanding of such events is fundamental to promoting continuous improvements in operational weather forecasting and contributing to more reliable forecasts of such events in the future.

Towards a 2DEnVar surface data assimilation approach within the convective scale numerical weather prediction model AROME‐France

AbstractSurface data assimilation (DA) plays a key role in the initialisation of soil variables in coupled surface–atmosphere numerical weather prediction (NWP) models. Météo‐France, as many other NWP centres, uses an operational surface DA method based on an optimal interpolation (OI), implemented more than 30 years ago. The current surface DA system implemented at Météo‐France in the operational limited‐area AROME‐France model operates in two steps. First, a two‐dimensional OI method analyses the diagnostic screen‐level parameters, which are then used to initialise the prognostic soil variables according to a one‐dimensional OI method. This article presents the implementation and the results of a new two‐dimensional ensemble‐based variational (2DEnVar) system, using the object‐oriented prediction system framework, for the diagnostic screen‐level variable analyses (2‐m temperature and 2‐m relative humidity) within the AROME‐France model. These analysed variables are then used to initialise the soil temperature and soil moisture of the first and second soil layers. A two‐dimensional variational system is initially implemented and compared with an equivalent system that uses an OI surface analysis. The comparison of 2‐m temperature increments shows that the two systems behave in a similar way. From this two‐dimensional variational approach, a 2DEnVar scheme is developed and compared with the OI screen‐level analysis used operationally (reference system). The results are encouraging even though a degradation in forecast quality is noticed at short ranges for 2‐m temperature and 2‐m relative humidity. This is mainly due to a low spread near the surface in the ensemble‐based data assimilation that is used to compute background‐error covariances in 2DEnVar. Sensitivity tests have been performed to improve the initial version of the 2DEnVar scheme. A positive and significant impact is noticed when a multiplicative inflation is applied to the background perturbations coming from the ensemble‐based data assimilation for the computation of 2‐m temperature and 2‐m relative humidity background‐error covariances. Finally, when comparing the inflated 2DEnVar system with the reference, the two systems behave similarly. This suggests that the 2DEnVar scheme, based on ensemble statistics, is capable of properly reconstructing background‐error structures derived from the empirical correlations described in OI.

Numerical Weather Prediction of Sea Surface Temperature in South China Sea Using Attention-Based Context Fusion Network

Numerical weather prediction of sea surface temperature (SST) is crucial for regional operational forecasts. Deep learning offers an alternative approach to traditional numerical general circulation models for numerical weather prediction. In our previous work, we developed a sophisticated deep learning model known as the Attention-based Context Fusion Network (ACFN). This model integrates an attention mechanism with a convolutional neural network framework. In this study, we applied the ACFN model to the South China Sea to evaluate its performance in predicting SST. The results indicate that for a 1-day lead time, the ACFN model achieves a Mean Absolute Error of 0.215 °C and a coefficient of determination (R2) of 0.972. In addition, in situ buoy data were utilized to validate the forecast results. The Mean Absolute Error for forecasts using these data increased to 0.500 °C for a 1-day lead time, with a corresponding R2 of 0.590. Comparative analyses show that the ACFN model surpasses traditional models such as ConvLSTM and PredRNN in terms of accuracy and reliability.

Development of Vertical Radar Reflectivity Profiles Based on Lightning Density Using the Geostationary Lightning Mapper Dataset in the Subtropical Region of Brazil

The study aims to develop vertical radar reflectivity profiles based on lightning density data from the Geostationary Lightning Mapper (GLM) on the GOES-16 satellite in the subtropical region of Brazil. The primary objective is to improve the assimilation of lightning data in numerical weather prediction models. The methodology involves the analysis of polarimetric radar data from Chapecó-SC and Jaraguari-MS, spanning from January 2019 to December 2023, and their correlation with lightning data from the GLM. Radar reflectivity profiles were created for different lightning density classes, categorized into six classes based on geometric progression. Results show a significant relationship between lightning activity and radar reflectivity, with distinct profiles for convective and stratiform events. These findings demonstrate the potential of using GLM data to enhance short-term weather forecasting, particularly for severe weather events. The study concludes that the integration of GLM data into weather models can lead to more accurate predictions of intense precipitation events, contributing to better preparedness and response strategies.

Comparison of the Morrison and WDM6 Microphysics Schemes in the WRF Model for a Convective Precipitation Event in Guangdong, China, Through the Analysis of Polarimetric Radar Data

Numerical weather prediction (NWP) models are indispensable for studying severe convective weather events. Research demonstrates that the outcomes of convective precipitation simulations are profoundly influenced by the choice between single or double-moment schemes for ice precipitation particles and the categorization of rimed ice. The advancement of dual-polarization radar has enriched the comparative validation of these simulations. This study simulated a convective event in Guangdong, China, from May 7 to 8, 2017, employing two bulk microphysical schemes (Morrison and WDM6) in the WRF v4.2 model. Each scheme was divided into two versions: one representing rimed ice particles as graupel (Mor_G, WDM6_G) and the other as hail (Mor_H, WDM6_H). The simulation results indicated negligible differences between the rimed ice set as graupel or hail particles, for both schemes. However, the Morrison schemes (Mor_G, Mor_H) depicted a more accurate raindrop size distribution below the 0 °C height level. A further analysis suggested that disparities between the Morrison and WDM6 schemes could be attributed to the intercept parameter (N0) setting for snow and graupel/hail in WDM6 scheme. The prescribed snow and graupel/hail N0 of WDM6 scheme might influence the melting processes, leading to a higher number concentration but a reduced mass-weighted diameter of raindrops. Reducing the intercept parameter for snow and graupel/hail in the WDM6 scheme could potentially enhance the simulation of convective precipitation. Conversely, the increase in N0 might deteriorate the precipitation simulation performance of the WDM6_G scheme, whereas the WDM6_H scheme exhibits minimal sensitivity to such changes.

Addressing challenges associated with training sample sizes for effective calibration of NWP precipitation forecasts

Abstract Effective calibration of precipitation forecasts produced by numerical weather prediction (NWP) models faces challenges associated with the training sample size. Newly-operationalized NWP models may only accumulate a small number of forecasts and thus may limit robust parameter inference in forecast calibration. It is necessary to investigate how the performance of forecast calibration changes with the amount of training data, to determine an effective training sample size. In this study, we thoroughly investigate the impacts of training sample size on precipitation forecast calibration based on the Seasonally Coherent Calibration (SCC) model across Australia. Overall, the performance of the model tends to stabilize in most parts of Australia when raw forecasts of 10 months or longer are used for parameter inference. Whether the training dataset cover wet months substantially affects forecast calibration. The findings of this study are critical for understanding the impacts of training sample size on forecast calibration, and will provide implications for future forecast calibration and the generation of hindcasts.

Validation of OceanSat-3 sea surface winds for their utilization in the NCMRWF NWP assimilation systems

The Indian Space Research Organization (ISRO) launched OceanSat-3, the third-generation satellite in the OceanSat series, carrying a Ku-band scatterometer in November 2022. The novelty of the OceanSat-3 scatterometer (OSCAT-3) is its high-resolution sea surface wind products generated at 12.5 km in addition to 25 km. This study aims to check the quality of OSCAT-3 winds for their possible assimilation in the National Centre for Medium Range Weather Forecasting (NCMRWF) operational Numerical Weather Prediction (NWP) models. The newly available high-resolution OSCAT-3 wind products are validated against in-situ buoy winds and NCMRWF NWP model equivalents for six months from July to December 2023. The validation results of OSCAT-3 winds are also compared with those of the Advanced Scatterometer (ASCAT) onboard the Meteorological Operational Satellite-C (MetOp-C) using the same method. Validation of OSCAT-3 winds against buoy winds shows that the Root Mean Square Errors (RMSEs) in wind speed (direction) is less than 2 m/s (15°) over the Tropics (20° N − 20° S) and slightly higher than 2 m/s (20°) over the Northern Hemisphere (20° N − 90° N), which is closer to the satellite’s mission specifications of 1.8 m/s and 20°. The RMSE in the wind speed/direction in both the scatterometer winds are less than 2 m/s (20°) over the three different geographical regions: the Tropics, Northern Hemisphere, and Southern Hemisphere, but with lower errors for ASCAT winds. The better agreement of ASCAT winds with the NWP equivalents can be attributed partially to the routine assimilation of ASCAT winds in the NCMRWF operational NWP models and also due to the frequency of operation, C-band compared to the OSCAT-3 operated in the Ku band. Since there are three distinct datasets, scatterometer winds, buoy winds, and the NWP equivalents, triple collocation is also carried out in this study. Triple collocation also shows that the errors in OSCAT-3 winds are within the mission goal. The findings from this study indicate that the OSCAT-3 vector winds are of similar quality as its predecessors (OSCAT-2 and SCATSAT-1) and can be assimilated operationally in the NCMRWF NWP models as well as in other global operational NWP models.

Random Forests’ Postprocessing Capability of Enhancing Predictive Skill on Subseasonal Time Scales—A Flow-Dependent View on Central European Winter Weather

Abstract Weather predictions 2–4 weeks in advance, called the subseasonal time scale, are highly relevant for socioeconomic decision-makers. Unfortunately, the skill of numerical weather prediction models at this time scale is generally low. Here, we use probabilistic random forest (RF)-based machine learning models to postprocess the subseasonal to seasonal (S2S) reforecasts of the European Centre for Medium-Range Weather Forecasts (ECMWF). We show that these models are able to improve the forecasts slightly in a 20-winter mean at lead times of 14, 21, and 28 days for wintertime central European mean 2-m temperatures compared to the lead-time-dependent mean bias-corrected ECMWF’s S2S reforecasts and RF-based models using only reanalysis data as input. Predictions of the occurrence of cold wave days are improved at lead times of 21 and 28 days. Thereby, forecasts of continuous temperatures show a better skill than forecasts of binary occurrences of cold wave days. Furthermore, we analyze if the skill depends on the large-scale flow configuration of the atmosphere at initialization, as represented by weather regimes (WRs). We find that the WR at the start of the forecast influences the skill and its evolution across lead times. These results can be used to assess the conditional improvement of forecasts initialized during one WR in comparison to forecasts initialized during another WR. Significance Statement Forecasts of winter temperatures and cold waves 2–4 weeks in advance done by numerical weather prediction (NWP) models are often unsatisfactory due to the chaotic characteristics of the atmosphere and limited predictive skill at this time range. Here, we use statistical methods, belonging to the so-called machine learning (ML) models, to improve forecast quality by postprocessing predictions of a state-of-the-art NWP model. We compare the forecasts of the NWP and ML models considering different weather regimes (WRs), which represent the large-scale atmospheric circulation such as the typical westerly winds in Europe. We find that the ML models generally yield better temperature forecasts for 14, 21, and 28 days in advance and better forecasts of cold wave days 21 and 28 days in advance. The quality of forecasts depends on the WR present at the forecast start. This information can be used to assess the conditional improvement of forecasts.

Comparison of Physically Based and Empirical Modeling of Nighttime Spatial Temperature Variability during a Heatwave in and around a City

Abstract The numerical weather prediction model HARMONIE-AROME and a multiple linear regression model (referred to in this article as the TURCLIM model after the local climate observation network) were used to model surface air temperature for 25–31 July 2018 in the City of Turku, Finland, to study their performance in urban areas and surrounding rural areas. The 0200 LT (local standard time) temperatures modeled by the HARMONIE-AROME and TURCLIM models were compared to each other and against the observed temperatures to find the model best suited for modeling the urban heat island effect and other spatial temperature variabilities during heatwaves. Observed temperatures were collected from 74 sites, representing both rural and urban environments. Both models were able to reproduce the spatial nighttime temperature variation. However, HARMONIE-AROME modeled temperatures were systematically warmer than the observed temperatures in stable conditions. Spatial differences between the models were mostly related to the physiographic characteristics: for the urban areas, HARMONIE-AROME modeled on average 1.4°C higher temperatures than the TURCLIM model, while for other land-cover types, the average difference was 0.51°C at maximum. The TURCLIM model performed well when the explanatory variables were able to incorporate enough information on the surrounding physiography. Respectively, systematic cold or warm bias occurred in the areas in which the thermophysically relevant physiography was lacking or was only partly captured by the model. Significance Statement As more and more people are living in an urban environment, the demand for accurate urban climate modeling is growing. This study aims to understand the differences between the numerical weather prediction and multiple linear regression modeling and their limitations in modeling surface air temperature in subkilometer scale. The case study shows that models are capable of predicting the spatial variation of 0200 LT nighttime temperature during a heatwave in a high-latitude coastal city. Both models are therefore valuable assets for city planners who need accurate information about the impacts of the physiography on the urban climate. The results indicate that to improve the performance of the models, more accurate physiographic description and higher spatial resolution of the models are needed.