GSL presents at the 2023 AGU Fall Meeting!

Are you going to the American Geophysical Union Fall Meeting? Here is a list of GSL presentations and posters to consider!

Dave Turner GSL "A containerized thermodynamic profile retrieval system for ground-based remote sensors"

The choice of algorithm to derive thermodynamic profiles matters. The TROPoe algorithm can derive these profiles from a range of different remote sensors, and derives uncertainty and other useful information so that the resulting data can be properly used. TROPoe can be used to help create more homogeneous datasets from a range of very different instruments.

Julia Simonson GSL/CIRES "Modeling Land-Atmosphere Interactions During the Morning Transition at the ARM Southern Great Plains Site"
Land-surface properties play important roles in the diurnal evolution of the atmospheric boundary layer (ABL). Changes in soil moisture and vegetation alter surface water and energy fluxes, which can then have an outsized impact on the ABL during transition periods, as there are no dominant forcings and few terms are negligible. Here we investigate land-atmosphere interactions and ABL evolution during the morning transition using observations from the DOE Atmospheric Radiation Measurement (ARM) Southern Great Plain site and simulations from the Common Community Physics Package (CCPP) single column model (SCM)(CCPP-SCM). The mixing diagram framework is applied to the observational data and model output to quantify the relative contributions of surface fluxes, advection, radiation, and entrainment to the evolution of ABL heat and moisture. For cases with clear skies and low advection, CCPP-SCM replicates the general trend in potential temperature and specific humidity evolution during the morning transition shown in observations. However, the contribution of surface fluxes and entrainment differ substantially, with greater differences in the modeled trend in moisture. Soil moisture in particular plays a significant role in modeled low-level moisture biases.

Dave Turner GSL "Evaluating the evolution of the convective boundary layer using mixing diagrams: Comparing models with observations"
We demonstrate the new use of an older analysis framework to evaluate the evolution of the lower part of the atmosphere during daytime using advanced ground-based remote sensing data. In particular, we apply this framework to evaluate two different numerical weather prediction models, which have different spatial resolutions, using data from a site in north-central Oklahoma. During this time period, the land surface conditions varied considerably from a wet spring to a dry summer condition, which provides a good test of the modeling systems.

Christina Kumler GSL/CIRES "Using Random Forests for Hourly Prediction of Wildfire Intensity with Inputs from Weather Forecast and Combined Satellites Fire Radiative Power Observations"
This project uses a new satellite product, which combines multiple satellite detects of fire power, and combines it with atmospheric variables from a numerical weather model using machine learning. The machine learning structure is tree-based and we use 3 years of data to test and train the computer to learn to model the fire power each hour into the near future. After the model is created, different tools are used to understand why the model behaved the way it did.

Jordan Schnell GSL/CIRES "A near-real time verification system for air quality forecast models"A near-real time verification system for air quality forecast models"

Multiple organizations and agencies produce daily air quality forecasts over the US and world ranging from simple tracer models to models that include full gas- and aerosol-phase chemistry and aerosol-physics interactions. The predictions are utilized by local, state, and federal agencies to provide guidance to the stakeholder population. Some air pollutants are predicted in multiple models, and it is not generally clear to the end-user which model should be used for guidance. Indeed, one model may outperform another during one event, but not in another. Moreover, as new models undergo development to replace existing operational models, they need to be frequently evaluated against observations to ensure they are providing accurate predictions. Very few air quality forecasts have a dedicated and publicly-accessible tool to evaluate predictions in a timely manner, such that stakeholders can quickly access the fidelity of the models against observations. Moreover, no publicly accessible tool allows comparison against multiple air quality forecast models for the US. Here we we use the joint NOAA/NCAR MELODIES-MONET analysis tool (https://melodies-monet.readthedocs.io/en/latest/) to produce daily evaluations for a suite of models against EPA AirNOW, NASA AERONET, and NOAA ISD surface stations. On a given analysis day, up to 12 individual forecasts (with up to four initializations) are compared. We include time series, Taylor, box, and bias plots and statistics aggregated over CONUS and EPA Regions. We have additionally built an interactive site-level analysis interface, capable of displaying 1000’s of plots over locations in North America. Here we will demonstrate the capabilities of the system by highlighting model abilities during notable events (e.g., wildfires) as well as discuss future additions and enhancements.

Sean Youn Professional Research Experience Program - GSL/CIRES (PREP-GC) "Applying the Visible Energy Fraction (VEF) to Investigate Nighttime Variations in Fire Combustion Phase"
The severity and frequency of wildfires have increased as a direct result of climate change driven by human activity. The emissions and products of wildfire combustion have been shown to be greatly affected by the fire combustion phase (flaming or smoldering) at a given point in time. Many models currently do not account for this variability in the combustion phase over time, which contributes to challenges in forming accurate emissions estimates. The Visible Infrared Imaging Radiometer Suite (VIIRS) Day/Night Band (DNB) sensor aboard the NOAA-20 and Suomi National Polar-orbiting Partnership (Suomi-NPP) satellites provides nighttime visible radiant energy information for active fires. Taking the ratio of the derived visible light power (VLP) from the DNB observations to the total radiative power from the fire (FRP) yields the visible energy fraction (VEF), which can be used to characterize the dominant combustion phase at the fire pixel level. Previous studies have shown that the VEF is closely linearly correlated to the modified combustion efficiency (MCE). We perform retrospective analysis on historical wildfire cases (most notably the Williams Flats fire in 2019 and the Creek Fire in 2020) to determine whether MCE values derived from VIIRS observations improved emissions estimates in the Rapid Refresh Forecast System Smoke/Dust (RRFS-SD) model relative to observational measurements (such as from the FIREX-AQ campaign). The relationship between the Hourly Wildfire Potential (HWP) index, a product of the High-Resolution Rapid Refresh (HRRR) model forecasts for temperature, wind, humidity, and soil moisture, and our calculated MCE values was investigated. A strong correlation between these meteorological variables and the MCE were found, which could be used to improve parametrization within existing models to provide better emissions estimations and help to drive this parametrization into air quality forecast models.

Ravan Ahmadov GSL "Forecasting smoke and dust in NOAA’s next-generation storm-scale numerical weather prediction model"
NOAA Global Systems Laboratory, in collaboration with other US laboratories, has been developing the high-resolution weather prediction model, Rapid-Refresh Forecasting System (RRFS). This model, based on the Unified Forecasting System [https://ufscommunity.org/], covers the entire North and Central Americas at a remarkable 3km resolution. To accurately simulate 3D concentrations of smoke and dust aerosols, we implemented aerosol emissions, turbulent mixing, advection, fire plume rise, and dry/wet removal schemes in the RRFS model. One smoke and two dust (fine and coarse mode) aerosol species were introduced as prognostic variables in the model, with their radiative effects included as well.

The RRFS-Smoke-Dust (RRFS-SD) model is being tested as an experimental model at NOAA/EMC, and the model output is displayed on [https://rapidrefresh.noaa.gov/RRFS-SD/]. Extensive sensitivity simulations were conducted, with a focus on the extreme fire seasons in the western US (2020) and Canada (2023). The direct feedback of smoke demonstrated a substantial improvement in the RRFS model's forecast skill for these cases. Additionally, a surface PM2.5/PM10 data assimilation (DA) system was developed for the RRFS-SD model and tested for September 2020. The results of the aerosol DA system were evaluated extensively.

In this presentation, we will provide an overview of the RRFS-SD model and discuss the findings from various numerical experiments. The future plans of the RRFS-SD model's development and research will also be explored. Our research aims to enhance weather prediction and air quality forecasting by incorporating smoke and dust aerosols, and their radiative impacts into the RRFS model, ultimately improving air quality and visibility forecasting, and their impact on weather.

Jordan Schnell GSL/CIRES "Sensitivity of parameterized heat flux and fire combustion phase in High-Resolution Rapid-Refresh with full chemistry (HRRR-Chem) simulations of wildfires"
Many current operational wildfire and air quality forecast models (e.g, HRRR-Smoke) parameterize the variation in wildfire intensity (i.e., fire radiative power, FRP) as the forecast evolves as a climatologically representative diurnal cycle. Within the model, the diurnal cycle is also applied to precomputed wildfire emission rates. An additional parameterization describes the combustion phase of the wildfire, which is held constant for and during all wildfires in the domain, and is assumed constant when generating the initial wildfire emissions. Here, we use the HRRR-Chem predicted Hourly Wildfire Potential (HWP) Index, measurements of trace gas (NO2, CO, HCHO) ratios from TropOMI, and measurements of the visible light fraction (VLF) from VIIRS to parametrize the FRP and combustion phase of wildfires based on internal model-generated weather (i.e., HWP). This modifies the magnitudes of the emission rate including the partitioning of nitrogen species as well as the fraction of the emission allowed to undergo plume rise. We additionally perform simulations where we turn off the parameterized plumerise and allow for inputs of the heat and moisture flux of the wildfire to explicitly simulate the buoyancy of the wildfire plume. We primarily investigate the Williams Flats Fire that occurred during the August 2019 FIREX-AQ field study, for which we have intensive surface and airborne measurements of the wildfire plume and surrounding areas.

Haiqin Li GSL/CIRES "The implementation of Grell-Freitas (GF) convection and smoke/dust aerosol feedback in the 3km storm-scale Rapid Refresh Forecast System (RRFS)"
The Rapid Refresh Forecast System (RRFS) is the National Oceanic and Atmospheric Administration (NOAA) next generation convection-allowing, rapidly-updated ensemble prediction system with a 3 km grid covering North America. The 3 km horizontal resolution falls into the “gray zone” of cumulus convection parameterization. It is also not clearly proven whether high resolution and explicit representation of convection can generate better numerical weather prediction. We implemented the Grell-Freitas (GF) scale-aware convection scheme with some modifications on the scale-awareness into RRFS, and executed a retrospective run in July 2022 with data assimilation, and compared with the control run, which explicitly represents convection in 3 km high resolution. The initial results from the run with GF convection significantly improved the upper air forecasts of temperature, humidity and wind speed. The GF run significantly reduces the large radar reflectivity bias for 35 dBz and above, but has low bias for 15 dBz. The GF run also significantly reduces the high precipitation bias above 1 inch threshold. We are working to further improve it. We also developed the RRFS – Smoke and Dust model (RRFS-SD) with the Common Community Physics Package (CCPP), which is designed to facilitate a host-model agnostic implementation of physics parameterizations. Here we embedded the plume rise modules for wildfire, sea-salt, dust, and anthropogenic emission modules into the RRFS using CCPP as subroutines of physics. We examined the smoke direct feedback to the radiation in the RRFS-SD with 3 km horizontal resolution and 65 vertical layers for September, 2020 during which the western US experienced extreme wildfires. The aerosol direct feedback run significantly improves the forecast of aerosol optical depth, surface 2m air temperature, 10m wind speed, and radiation fluxes.

Li Zhang GSL/CIRES "Developments and Applications of NOAA’s UFS-Aerosols and UFS-Chem Models for Global Aerosol Forecasts"
There are two global chemical forecast systems under development and online coupled with the Unified Forecast System (UFS) at NOAA. 1) UFS-Aerosols: the second-generation of a global coupled aerosol system was collaboratively developed by NOAA and NASA since 2021, which is planned to be implemented into Global Ensemble Forecast System (GEFS) v13. UFS-Aerosols embeds NASA’s 2nd-generation GOCART model in a NUOPC infrastructure. 2) UFS-Chem: an innovative community model of chemistry online coupled with UFS, which is a wide collaboration between NOAA Oceanic and Atmospheric Research (OAR) laboratories and NCAR. It utilizes the Common Community Physics Package (CCPP) infrastructure to link the gas and aerosol chemistry modules to the rest of the model that enhance the research capabilities to use different gas and aerosol chemical mechanisms to couple different physics options. One of the aerosol components based on the current operational GEFS-Aerosols, has been implemented into UFS-Chem with some updates to wet deposition, dust and fire emission etc. Both include the direct and semi-direct radiative feedback from online aerosols prediction. They also have the capability to be fully coupled with ocean, sea ice, and wave components for S2S forecasts. The capabilities of UFS-Aerosols and UFS-Chem in medium-range and S2S predictions are evaluated and compared using observations from reanalysis data, ground-based measurements, and satellite data.

Guoqing Ge GSL/CIRES "Towards a Digital Twin of the current atmosphere: a 3D/4D real-time mesoscale analysis system at NOAA"
Digital Twin, which entails creating a virtual replica of a physical entity, is an emerging technology and has gained significant momentum across various scientific and engineering disciplines. Creating a digital twin of the current atmosphere will enhance weather analysis and forecasting capabilities. This abstract presents an overview of the ongoing development of a 3D/4D Real-Time Mesoscale Analysis system (RTMA) at NOAA. This RTMA system can serve as a foundation for establishing a digital twin of the current atmosphere in the United States and/or North America.

The atmosphere is observed by an array of methods, including satellites, surface stations, wind profilers, weather balloons (radiosondes), Doppler radars, backyard weather stations, and smartphone pressure sensors, among others. These diverse platforms provide millions of observations at high temporal and spatial resolutions. Some platforms observe the atmospheric states directly (such as temperature, humidity, precipitation, air pressure, and wind), while others observe indirectly, yielding measurements that are a complex function or an integral of several atmospheric states, such as satellite radiance, GNSS radio occultation, radar reflectivity factor, etc. With so many sources of observations at different time/space scales and different data qualities, it is a non-trivial task to integrate and fuse these vast amounts of atmospheric observations into a coherent and physically-consistent representation of the current atmosphere.

The 3D/4D RTMA system employs state-of-the-art data analysis algorithms, specifically the hybrid ensemble 3D/4D Variational methods based on Bayesian theory, which combines a prior estimate of the state along with observations weighted by their respective error covariances. It also incorporates comprehensive, adaptive, and rigorous quality control procedures, best software practices, advanced supercomputing capabilities, and cutting-edge ML/AI technology. As a result, the RTMA system offers low-latency, real-time, and highly accurate high-resolution (2.5 km or less) estimations of current atmospheric conditions that closely align with observations while maintaining physical and dynamic consistency across diverse regions, altitudes, and timeframes. The 3D/4D RTMA system produces thousands of direct and derived atmospheric fields, encompassing, but not limited to, near-surface sensible weather elements (temperature, dew points, winds, etc.), three-dimensional cloud and visibility products, severe weather information (radar reflectivity, precipitable water, PBL height, precipitation types, CAPE, etc.), air quality-related fields (smoke, dust, other aerosols, etc.), turbine-height winds for renewable energy, and hydrology products.

All these developments will facilitate establishing a digital twin of the current atmosphere in real-time and provide highly comprehensive insights for interested parties across much of the weather and climate society, including general forecasting, severe weather applications, fire weather responses, aviation, renewable energy, hydrology, and more. It will enable better-informed decision-making regarding weather-related activities and management. As research continues, the Digital Twin of the current atmosphere holds vast potential for improving society's resilience to the challenges posed by an ever-changing climate.

Lindsey Anderson PREP-GC "Using Hourly Wildfire Potential Index and Satellite Data to Estimate the Impact of Changing Combustion Conditions on the Composition of Wildfire Emissions"
Human-caused climate change has led to an increase in the number, size, and intensity of wildfires. Wildfire smoke impacts air quality and climate, so it is important to understand smoke composition and how it changes over time. The chemical composition of wildfire smoke in current air quality forecasts is based on the vegetation type and area burned. However, it has been shown that the chemical composition of wildfire smoke also depends on whether the fire is more actively flaming or smoldering. Allowing the chemical composition of wildfire smoke to change based on the type of combustion could improve air quality forecasting. Previously, we showed that a space-based remote sensing instrument called TROPOMI could be used to study changes in the chemical composition of wildfire emissions over time, as they progress from more flaming to smoldering combustion. In this study, we use the relationship between modeled Hourly Wildfire Potential and remote sensing observations to inform how the chemical composition of wildfire smoke changes in time. We implemented this relationship in an air quality model and tested its performance on the Williams Flats fire, by comparing the modeled chemical composition with aircraft observations of the wildfire plume.

Tatiana Smirnova GSL/CIRES "Strategies for improving surface predictions in the operational weather prediction models"
The main purpose of land surface schemes in the weather prediction models is to provide more accurate lower boundary conditions for the atmosphere, especially important for improved predictions of severe weather having critical impact on aviation operations, ground transportation, and other areas of human activities. The required input to most boundary layer schemes are sensible and latent heat fluxes computed within the land surface models. These fluxes to a large extent depend on accurate initialization of initial land surface state, including soil moisture, soil temperature, vegetation greenness and snow in the cold season. Several functionalities within a coupled land data assimilation system have been developed for the High-Resolution Rapid Refresh (HRRR) model, operational at the National Center for Environment prediction (NCEP). These functionalities have been transferred to the next generation UFS-based regional Rapid Refresh FV3 Standalone (RRFS, under development at present time). Other strategies for improved surface predictions include identifying model biases in 2-m temperature and moisture, revealing the sources of errors and providing methods to alleviate these biases. Errors in surface fluxes could be caused by biases in clouds, precipitation and incoming solar radiation computed in the atmospheric components of the forecasting system. Land-surface-related factors would most likely include soil moisture affecting the Bowen ratio. Another important factor is the vegetation fraction that affects splitting evapotranspiration between direct soil evaporation, canopy evaporation and transpiration. Use of real-time greenness fraction rather than climatology should provide more realistic information about the state of the vegetation and thus provide more realistic evapotranspiration flux. For cold seasons snow cover depth and fraction can play a critical role in 2-m temperature over snow-covered areas. Experiments performed within the RRFS model in terms of described strategies for improved surface predictions will be presented at the meeting.

Ryan Harp GSL/NCAR "Climatic Influences on Seasonal Onset Timing and Annual Burden of West Nile Virus"
West Nile virus (WNV) is the primary mosquito-borne disease of concern in the continental United States. Since introduction in the country in 1999, WNV has become endemic with over 28,000 reported severe neuroinvasive disease cases and over 2,600 reported deaths. Reliable WNV forecasts could allow for more effective deployment of targeted public health preparations, but current forecasting capabilities are insufficient. However, climate-informed forecasts-of-opportunity might exist given demonstrated links between WNV disease burden and meteorological factors at a variety of spatiotemporal scales, though these opportunities have yet to be realized. To assess potential climatic drivers of WNV burden in the United States, we combined high-resolution statistically-downscaled gridded meteorological data from the Parameter-elevation Regressions on Independent Slopes Model (PRISM) and gridMET products with WNV case data provided by the Centers for Disease Control and Prevention ArboNET national surveillance system. Disease caseload and climatic variables of interest (e.g., mean temperature, precipitation, Palmer Drought Severity Index) were aggregated onto a spatially uniform hexagonal grid and analyzed at a regional level to identify critical climate anomalies and thresholds that impact both the timing of seasonal cycle and total annual disease burden; we present preliminary results showing regional clustering of climatic influences on WNV burden. Future work aims to build upon this analysis by projecting identified empirical relationships onto short-term and subseasonal-to-seasonal weather forecasts to increase the lead time of WNV forecasts.

Andres Ramos PREP-GC "RRFS vs. HRRR: An ML-based Model Verification Approach for Severe Winds"
The Rapid Refresh Forecasting System (RRFS) is a prototype, convection-allowing model (CAM) that is a candidate model to replace the current operational CAM, the High-Resolution Rapid Refresh (HRRR). This model is part of NOAA’s effort to unify and simplify the modeling suite based on the Finite-Volume Cubed-Sphere Dynamical Core. In this research project we use the RRFS model outputs from the months of May and June of 2023 to evaluate model performance in predicting severe convective winds. We developed a model evaluation approach that uses Storm Prediction Center severe wind reports as a ground truth and compares them to ”surrogate storm reports”. Surrogate storm reports are an idea used by Sobash et al. (2019) that represent the model’s best guess of where wind will occur. Once we have both sets of reports, surrogate and actual, we then use machine learning to cluster both sets into separate neighborhoods. By measuring the geographical overlap between these neighborhoods we can then evaluate model performance. We tested this approach with various variables. Preliminary results show that the
RRFS variable that maximizes CSI (Critical Success Index) is maximum 2-5km updraft helicity. The variable that had the worst CSI is maximum downdraft. We also applied this model evaluation approach to the HRRR model. Preliminary results show that RRFS marginally outperforms HRRR when comparing severe wind forecasts. These results can be used to help guide future RRFS development.

Jagger Alexander PREP-GC Visualizing weather hazard vulnerability and impact information in an ensemble forecast decision-support system
DESI (Dynamic Ensemble-based Scenarios for Impact-Based Decision Support Services) is a web tool created by the NOAA Global Systems Laboratory to provide forecasters, policy-makers, and other stakeholders with ensemble-based weather forecast information for decision making. However, decisions must also be made based on human population characteristics, which shape the impact of natural hazards. The CDC Social Vulnerability Index as well as other population and infrastructure characteristics, including households without access to internet or vehicles, mobile homes, and households without english fluency are provided as maps at the census tract level, allowing for visualization alongside ensemble weather forecast variables. Additionally, the application has been enhanced with the ability to provide historical weather hazard impact information from recent severe weather events. Data from 2003 through 2022 for seven weather hazard types, including hurricanes, tornadoes, winter storms, wind gusts, hail, wildfires, and floods, are pulled from the NOAA Storm Events Database. The user can select any location in the CONUS region, and for a given radius around the site, information on past hazards and their impacts are aggregated and plotted. Histograms of damages, injuries, and deaths are provided, and for certain hazards, plots of hazard magnitude against damages are shown. Statistical inference is conducted between hazard magnitude, damages, injuries, deaths, as well as social vulnerability indicators. This new information allows a user to understand the range of damages and injuries caused by hazards, as well as geographic variability in vulnerability, for a region where a hazard is being forecasted. This new suite of application functionality enables decision-makers to better understand the forecast's impact on public health, facilitating more informed decisions which best protect public well-being.