This study evaluates the performance of the National Centers for Environmental Prediction Global Forecast System (GFS) against observations made by the U.S. Department of Energy Atmospheric Radiation Measurement (ARM) Program at the southern Great Plains site for the years 2001–04. The spatial and temporal scales of the observations are examined to search for an optimum approach for comparing grid-mean model forecasts with single-point observations. A single-column model (SCM) based upon the GFS was also used to aid in understanding certain forecast errors. The investigation is focused on the surface energy fluxes and clouds. Results show that the overall performance of the GFS model has been improving, although certain forecast errors remain. The model overestimated the daily maximum latent heat flux by 76 W m−2 and the daily maximum surface downward solar flux by 44 W m−2, and underestimated the daily maximum sensible heat flux by 44 W m−2. The model’s surface energy balance was reached by a cancellation of errors. For clouds, the GFS was able to capture the observed evolutions of cloud systems during major synoptic events. However, on average, the model largely underestimated cloud fraction in the lower and midtroposphere, especially for daytime nonprecipitating low clouds because shallow convection in the GFS does not produce clouds. Analyses of surface radiative fluxes revealed that the diurnal cycle of the model’s surface downward longwave flux (SDLW) was not in phase with that of the ARM-observed SDLW. SCM experiments showed that this error was caused by an inaccurate scaling factor, which was a function of ground skin temperature and was used to adjust the SDLW at each model time step to that computed by the model’s longwave radiative transfer routine once every 3 h. A method has been proposed to correct this error in the operational forecast model. It was also noticed that the SDLW biases changed from mostly negative in 2003 to slightly positive in 2004. This change was traced back to errors in the near-surface air temperature. In addition, the SDLW simulated with the newly implemented Rapid Radiative Transfer Model longwave routine in the GFS is usually 5–10 W m−2 larger than that simulated with the previous routine. The forecasts of surface downward shortwave flux (SDSW) were relatively accurate under clear-sky conditions. The errors in SDSW were primarily caused by inaccurate forecasts of cloud properties. Results from this study can be used as guidance for the further development of the GFS.
1 November 2008 in Journal of Applied Meteorology and Climatology
This study examines the dependence of surface albedo on solar zenith angle (SZA) over snow-free land surfaces using the intensive observations of surface shortwave fluxes made by the U.S. Department of Energy Atmospheric Radiation Measurement (ARM) Program and the National Oceanic and Atmospheric Administration Surface Radiation Budget Network (SURFRAD) in 1997–2005. Results are used to evaluate the National Centers for Environmental Prediction (NCEP) Global Forecast Systems (GFS) parameterization and several new parameterizations derived from the Moderate Resolution Imaging Spectroradiometer (MODIS) products. The influence of clouds on surface albedo and the albedo difference between morning and afternoon observations are also investigated. A new approach is taken to partition the observed upward flux so that the direct-beam and diffuse albedos can be separately computed. The study focused first on the ARM Southern Great Plains Central Facility site. It is found that the diffuse albedo prescribed in the NCEP GFS matched closely with the observations. The direct-beam albedo parameterized in the GFS is largely underestimated at all SZAs. The parameterizations derived from the MODIS product underestimated the direct-beam albedo at large SZAs and slightly overestimated it at small SZAs. Similar results are obtained from the analyses of observations at other stations. It is also found that the morning and afternoon dependencies of direct-beam albedo on SZA differ among the stations. Attempts are made to improve numerical model algorithms that parameterize the direct-beam albedo as a product of the direct-beam albedo at SZA = 60° (or the diffuse albedo), which varies with surface type or geographical location and/or season, and a function that depends only on SZA. A method is presented for computing the direct-beam albedos over these snow-free land points without referring to a particular land-cover classification scheme, which often differs from model to model.
A new dynamical core, known as the Finite-Volume Cubed-Sphere (FV3) and developed at both NASA and NOAA, is used in NOAA's Global Forecast System (GFS) and in limited-area models for regional weather and air quality applications. NOAA has also upgraded the operational FV3GFS to version 16 (GFSv16), which includes a number of significant developmental advances to the model configuration, data assimilation, and underlying model physics, particularly for atmospheric composition to weather feedback. Concurrent with the GFSv16 upgrade, we couple the GFSv16 with the Community Multiscale Air Quality (CMAQ) model to form an advanced version of the National Air Quality Forecasting Capability (NAQFC) that will continue to protect human and ecosystem health in the US. Here we describe the development of the FV3GFSv16 coupling with a “state-of-the-science” CMAQ model version 5.3.1. The GFS–CMAQ coupling is made possible by the seminal version of the NOAA-EPA Atmosphere–Chemistry Coupler (NACC), which became a major piece of the next operational NAQFC system (i.e., NACC-CMAQ) on 20 July 2021. NACC-CMAQ has a number of scientific advancements that include satellite-based data acquisition technology to improve land cover and soil characteristics and inline wildfire smoke and dust predictions that are vital to predictions of fine particulate matter (PM2.5) concentrations during hazardous events affecting society, ecosystems, and human health. The GFS-driven NACC-CMAQ model has significantly different meteorological and chemical predictions compared to the previous operational NAQFC, where evaluation of NACC-CMAQ shows generally improved near-surface ozone and PM2.5 predictions and diurnal patterns, both of which are extended to a 72 h (3 d) forecast with this system.
7 November 2022 in Geoscientific Model Development
The latest operational National Air Quality Forecast Capability (NAQFC) has been advanced to use the Community Multiscale Air Quality (CMAQ) model (version 5.3.1) with the CB6r3 (Carbon Bond 6 revision 3) AERO7 (version 7 of the aerosol module) chemical mechanism and is driven by the Finite-Volume Cubed-Sphere (FV3) Global Forecast System, version 16 (GFSv16). This update has been accomplished via the development of the meteorological preprocessor, NOAA-EPA Atmosphere–Chemistry Coupler (NACC), adapted from the existing Meteorology–Chemistry Interface Processor (MCIP). Differing from the typically used Weather Research and Forecasting (WRF) CMAQ system in the air quality research community, the interpolation-based NACC can use various meteorological outputs to drive the CMAQ model (e.g., FV3-GFSv16), even though they are on different grids. In this study, we compare and evaluate GFSv16-CMAQ and WRFv4.0.3-CMAQ using observations over the contiguous United States (CONUS) in summer 2019 that have been verified with surface meteorological and AIRNow observations. During this period, the Fire Influence on Regional to Global Environments and Air Quality (FIREX-AQ) field campaign was performed, and we compare the two models with airborne measurements from the NASA DC-8 aircraft. The GFS-CMAQ and WRF-CMAQ systems show similar performance overall with some differences for certain events, species and regions. The GFSv16 meteorology tends to have a stronger diurnal variability in the planetary boundary layer height (higher during daytime and lower at night) than WRF over the US Pacific coast, and it also predicted lower nighttime 10 m winds. In summer 2019, the GFS-CMAQ system showed better surface ozone (O3) than WRF-CMAQ at night over the CONUS domain; however, the models' fine particulate matter (PM2.5) predictions showed mixed verification results: GFS-CMAQ yielded better mean biases but poorer correlations over the Pacific coast. These results indicate that using global GFSv16 meteorology with NACC to directly drive CMAQ via interpolation is feasible and yields reasonable results compared to the commonly used WRF approach.
A cold air outbreak (CAO) is an extreme weather phenomenon that has significant social and economic impacts over a large region of the midlatitudes. However, the dynamical mechanism of the occurrence and evolution of CAO events, particularly the role of the stratosphere, is not well understood. Through an analysis of one extreme CAO episode that occurred on 27–31 January 2019 across much of the US Midwest, this study examined its thermodynamic structure and the impact of stratospheric downward transport using the single-field-view (SFOV) satellite products (with a spatial resolution of ~14 km at nadir) from the Cross-track Infrared Sounder (CrIS) onboard Suomi National Polar-Orbiting Partnership (SNPP) in conjunction with MERRA-2 and ERA-5 reanalysis products. It is found that along the path of cold air transport, particularly near the coldest surface center, there exists a large enhancement of O3, deep tropopause folding, significant downward transport of stratospheric dry air, and a warm center above the tropopause. The upper warm center can be observed directly using the brightness temperature (BT) of CrIS stratospheric sounding channels. While similar large-scale patterns of temperature (T), relative humidity (RH), and ozone (O3) are captured from CrIS, MERRA-2, and ERA-5 products, it is found that, in the regions impacted by CAO, MERRA-2 has a thicker dry layer under the tropopause (with the difference of RH up to ~10%) and the total column ozone (TCO) from ERA-5 has a relatively large positive bias of 2.8 ± 2.8% compared to that measured by Ozone Mapping and Profiler Suite (OMPS). This study provides some observational evidence from CrIS that confirm the impact of the stratosphere on CAO through downward transport and demonstrates the value of the SFOV retrieval products for CAO dynamic transport study and model evaluation.
The medium-range forecasting of global weather plays a pivotal role in decision-making processes across various societal and economic sectors. Recent years have witnessed a rapid evolution in machine learning (ML) models applications in weather prediction, demonstrating notably superior performance compared to traditional numerical weather prediction (NWP) models. These cutting-edge models leverage diverse ML architectures, such as Graph Neural Networks (GNNs), Convolutional Neural Networks (CNNs), Fourier Neural Operators (FNOs), and Transformers. Notably, Google DeepMind has pioneered a novel ML-based approach known as GraphCast, enabling direct training from reanalysis data and facilitating global predictions for numerous weather variables in less than a minute. Impressively, GraphCast forecasts show improved accuracy in predicting severe weather events, including phenomena like tropical cyclones, atmospheric rivers, and extreme heat. However, the efficiency of GraphCast relies on high-quality historical weather data for training, typically sourced from ECMWF's ERA5 reanalysis. Concurrently, the National Centers for Environmental Prediction (NCEP) has initiated efforts in collaboration with the research community, focusing on developing Machine Learning Weather Prediction (MLWP). This study assesses the pre-trained GraphCast model, leveraging Global Data Assimilation System (GDAS) data from the operational GFSv16 model as initial states. Additionally, we explore the potential use of GDAS data as an alternative training source for GraphCast. Notably, GDAS data is available at a 0.25-degree latitude-longitude resolution and with a temporal resolution of 6 hours. Our investigation involves a comparative analysis of GraphCast's performance when initiated on GDAS data versus ERA5 and HRES data. Alongside this comparative analysis, we investigate the advantages and limitations of utilizing GDAS data for GraphCast while proposing potential approaches for enhancing future iterations of this study.