Deriving turbulence characteristics from the COSMO numerical weather prediction model for dispersion applications

Improving the Turbulence Coupling between High Resolution Numerical Weather Prediction Models and Lagrangian Particle Dispersion Models

In this paper, COSMO numerical weather prediction model is used to simulate land-sea breeze circulation in the north coast of Oman and to analyze the variability of the PBL depth. Typical summer day 1st August 2009 and winter day 1st January 2009 were simulated and verified by surface observations. The COSMO model was initialized using the German global model which runs as 40km resolution and 40 vertical layers. The simulations show differences in the onset, strength, and inland penetration of the sea breeze. The high sea-land thermal contrast in the summer season induces earlier onset and progression of the sea breeze with larger horizontal and vertical extents. The summer and winter inland penetration is simulated around 150 and 65km, respectively. The PBL depth was defined from the verticals profiles of the turbulence kinetic energy (TKE). The simulated PBL height varied at 1,600m in summer and 700m in the winter case. These results are in general in good agreement with the simulations of the National Centers for Environmental Prediction Global Data Assimilation System (NCEP-GDAS) model. The dispersive ability of the atmosphere was tested through the vertical mixing coefficient (VMC) generated by NCEP-GDAS simulations. The VMC during the summer simulated case is approximately four times larger than the respective one during the winter case. Consequently, during the advection of winter sea breeze, a shallow dome less than 100m is shaped and forms an obstacle to the air-pollutant dispersion.

<p>Stable water isotopes in marine boundary layer water vapour are strongly influenced by the strength of air-sea moisture fluxes and are thus tracers of air-sea interaction. Air-sea moisture fluxes in the extratropics are modulated by large-scale air advection, for instance the advection of warm and moist air masses in the warm sector of extratropical cyclones. A distinct isotopic composition of water vapour in the latter environment has been observed in near-surface water vapour over the Southern Ocean during the 2016/17 Antarctic Circumnavigation coordinated by the Swiss Polar Institute. Most prominently, the second-order isotope variable d-excess shows negative values in the cyclones’ warm sector. Here, we present three single-process air parcel models, which simulate the evolution of d-excess and specific humidity in an air parcel induced by dew deposition, decreasing ocean evaporation or upstream cloud formation, respectively. The air-parcel models are combined with simulations with the isotope-enabled numerical weather prediction model COSMO<sub>iso</sub> (i) to validate the air parcel models, (ii) to study the extent of non-linear interactions between the different processes, and (iii) to quantify the relevance of the three processes for stable water isotopes in the warm sector of the investigated extratropical cyclone. This analysis reveals that dew deposition and decreasing ocean evaporation lead to the strongest d-excess decrease in near-surface water vapour in the warm sector. Furthermore, COSMO<sub>iso</sub> air parcel trajectories show that the persistent low d-excess observed in the warm sector of extratropical cyclones is not a result of material conservation of low d-excess. Instead the latter feature is sustained by the continuous production of low d-excess values in new air parcels entering the warm sector. We show that with the mechanistic approach of using single-process air parcel models we are able to simulate the evolution of d-excess during the air parcel’s transport. This improves our understanding of the effect of air-sea interaction and boundary layer cloud formation on the stable water isotope variability of marine boundary layer water vapour.</p>

Abstract. A rigorous methodology for the evaluation of integration schemes for Lagrangian particle dispersion models (LPDMs) is presented. A series of one-dimensional test problems are introduced, for which the Fokker–Planck equation is solved numerically using a finite-difference discretisation in physical space and a Hermite function expansion in velocity space. Numerical convergence errors in the Fokker–Planck equation solutions are shown to be much less than the statistical error associated with a practical-sized ensemble (N = 106) of LPDM solutions; hence, the former can be used to validate the latter. The test problems are then used to evaluate commonly used LPDM integration schemes. The results allow for optimal time-step selection for each scheme, given a required level of accuracy. The following recommendations are made for use in operational models. First, if computational constraints require the use of moderate to long time steps, it is more accurate to solve the random displacement model approximation to the LPDM rather than use existing schemes designed for long time steps. Second, useful gains in numerical accuracy can be obtained, at moderate additional computational cost, by using the relatively simple “small-noise” scheme of Honeycutt.

Lagrangian particle dispersion models (LPDMs) have been used extensively to calculate source-receptor relationships (“footprints”) for use in applications such as greenhouse gas (GHG) flux inversions. Because a single model simulation is required for each data point, LPDMs do not scale well to applications with large data sets such as flux inversions using satellite observations. Here, we develop a proof-of-concept machine learning emulator for LPDM footprints over a ~350 km by 230 km region around an observation point, and test it for a range of in situ measurement sites from around the world. As opposed to previous approaches to footprint approximation, it does not require the interpolation or smoothing of footprints produced by the LPDM. Instead, the footprint is emulated entirely from meteorological inputs. This is achieved by independently emulating the footprint magnitude at each grid cell in the domain using gradient-boosted regression trees (GBRTs) with a selection of meteorological variables as inputs. The emulator is trained based on footprints from the UK Met Office Numerical Atmospheric dispersion Modelling Environment (NAME) for 2014 and 2015, and the emulated footprints are evaluated against hourly NAME output from 2016 and 2020. When compared to CH4 concentration time series generated by NAME, we show that our emulator achieves a mean R-squared score of 0.69 across all sites investigated between 2016 and 2020. The emulator can predict a footprint in around 10 ms, compared to around 10 minutes for the 3D simulator. This simple and interpretable proof-of-concept emulator demonstrates the potential of machine learning for LPDM emulation.

The parameterization for transilient turbulence coefficients suggested by Stull and Driedonks (SD. 1987) is tested against the large-eddy simulations (LES) of Ebert et al. (ESS, 1989) for the special case of an idealized convective boundary layer. The SD parameterization is based on a nonlocal approximation to the turbulence kinetic energy (TKE) equation, and requires turbulent exchange (i.e., the matrix of transilient mixing coefficients is assumed to be symmetric) and dominance of the smaller eddies (i.e., elements closer to the main diagonal of the matrix are greater). Measurements from the LES model, however, show that the transilient matrix is asymmetric in convective situations, with larger eddies dominating. Mean-state conditions such as the deep convective mixing and mixed-layer growth are satisfactorily described by the parameterization, but the surface layer is too deep and the entrainment zone thickness is poorly defined. Turbulence properties such as skewed vertical velocity distributions are not possible within the constraint of a symmetric matrix, and partial convective overturning is also not possible because of the restriction that small eddies dominate. Future improved parametcrizations might continue to be based on the TKE equation, but should allow transilient matrix asymmetry.

Flux inversion is the process by which sources and sinks of a gas are identified from observations of gas mole fraction. The inversion often involves running a Lagrangian particle dispersion model (LPDM) to generate simulations of the gas movement over a domain of interest. The LPDM must be run backward in time for every gas measurement, and this can be computationally prohibitive. To address this problem, here we develop a novel spatio‐temporal emulator for LPDM sensitivities that is built using a convolutional variational autoencoder (CVAE, a two‐piece neural network capable of condensing and reconstructing images). With the encoder segment of the CVAE, we obtain approximate (variational) posterior distributions over latent variables in a low‐dimensional space. We then use a spatio‐temporal Gaussian process emulator on the low‐dimensional space to emulate new variables at prediction locations and time points. Emulated variables are then passed through the decoder segment of the CVAE to yield emulated sensitivities. We show that our CVAE‐based emulator outperforms the more traditional emulator built using empirical orthogonal functions and that it can be used with different LPDMs. We conclude that our emulation‐based approach can be used to reliably reduce the computing time needed to generate LPDM outputs for use in high‐resolution flux inversions.

Impact of data assimilation and aerosol radiation interaction on Lagrangian particle dispersion modelling

Haline stratification in the Rhine-Meuse freshwater plume: a three-dimensional model sensitivity analysis

The growing volume of methane measurements from space provides new opportunities for evaluating and improving countries' self-reported emissions. Surface emissions can be estimated from atmospheric observations using inverse modelling systems, which often rely on Lagrangian Particle Dispersion Models (LPDMs) to simulate how methane is transported through the atmosphere. Ensembles of particles are transported backwards in time from the measurement point, to define source-receptor relationships (“footprints”), which reflect the sensitivity of a measurement to all potential upwind sources within the domain. However, LPDM-based techniques are computationally costly, struggling to scale to the size of modern satellite datasets and limiting the amount of data that can be used for emissions inference. Previously, we presented a machine learning-driven LPDM emulator that can approximate satellite footprints using only meteorology and topography, and demonstrated its use over the South American continent, achieving speed-ups of over three orders of magnitude compared to the LPDM. We integrated the emulator into an emissions inference pipeline to estimate Brazil’s methane emissions from GOSAT observations in 2016 and 2018, and found that the emulator-based estimates were consistent with those obtained using the more expensive physics-based LPDM. Here, we show preliminary results of applying the emulator to other regions with high natural methane emissions, like North Africa and India. We compare the emulator’s performance across the selected time periods and geographical domains as well as the estimated emissions. Furthermore, we discuss solutions to improve performance and reduce the training data needed, like transfer learning across domains.

This paper describes the coupling between a mesoscale numerical weather prediction model, the Weather Research and Forecasting (WRF) model, and a Lagrangian Particle Dispersion Model, the Stochastic Time-Inverted Lagrangian Transport (STILT) model. The primary motivation for developing this coupled model has been to reduce transport errors in continental-scale top–down estimates of terrestrial greenhouse gas fluxes. Examples of the model’s application are shown here for backward trajectory computations originating at CO2 measurement sites in North America. Owing to its unique features, including meteorological realism and large support base, good mass conservation properties, and a realistic treatment of convection within STILT, the WRF–STILT model offers an attractive tool for a wide range of applications, including inverse flux estimates, flight planning, satellite validation, emergency response and source attribution, air quality, and planetary exploration.

Simulation of diffusion phenomena under unstable conditions using a Lagrangian particle dispersion model

Global anthropogenic CO2 sources are dominated by power plants and large industrial facilities. Quantifying the emissions of these point sources is therefore one of the main goals of the planned constellation of anthropogenic CO2 monitoring satellites (CO2M) of the European Copernicus program. Atmospheric transport models may be used to study the capabilities of such satellites through observing system simulation experiments and to quantify emissions in an inverse modelling framework. How realistically the CO2 plumes of power plants can be simulated and how strongly the results may depend on model type and resolution, however, is not well known due to a lack of observations available for benchmarking. Here, we use the unique data set of aircraft in-situ and remote sensing observations collected during the CoMet measurement campaign down-wind of the coal-fired power plants at Bełchatów in Poland and Jänschwalde in Germany in 2018 to evaluate the simulations of six different atmospheric transport models. The models include three Large-Eddy-Simulation (LES) models, two mesoscale numerical weather prediction (NWP) models, and one Lagrangian particle dispersion model (LPDM) and cover a wide range of model resolutions from 200 m to 2 km horizontal grid spacing. At the time of the aircraft measurements between late morning and early afternoon, the simulated plumes were slightly (at Jänschwalde) to highly (at Bełchatów) turbulent, consistent with the observations, and extended over the whole depth of the atmospheric boundary layer (ABL, up to 1800 m a.s.l. in the case of Bełchatów). The stochastic nature of turbulent plumes puts fundamental limitations to a point-by-point comparison between simulations and observations. Therefore, the evaluation focused on statistical properties such as plume amplitude and width as a function of distance from the source. LES and NWP models showed similar performance and sometimes remarkable agreement with the observations when operated at comparable resolution. A resolution of 1 km or better, however, appears to be necessary to realistically capture turbulent plume structures. At coarser resolution, the plumes disperse too quickly especially in the near field (0–8 km from the source) and turbulent structures are increasingly smoothed out. Total vertical columns are easier to simulate accurately than the vertical distribution of CO2, since the latter is critically affected by profiles of vertical stability, especially near the top of the ABL. Cross-sectional flux and integrated mass enhancement methods applied to synthetic CO2M data generated from the model simulations with a random noise of 0.5 ppm–1.0 ppm suggest that emissions from a power plant like Bełchatów can be estimated with an accuracy of about 20 % from single overpasses. Estimates of the effective wind speed are a critical input for these methods. Wind speeds in the middle of the ABL appear to be a good approximation for plumes in a well-mixed ABL as encountered during CoMet.

Global anthropogenic CO2 sources are dominated by power plants and large industrial facilities. Quantifying the emissions of these point sources is therefore one of the main goals of the planned constellation of anthropogenic CO2 monitoring satellites (CO2M) of the European Copernicus program. Atmospheric transport models may be used to study the capabilities of such satellites through observing system simulation experiments and to quantify emissions in an inverse modelling framework. How realistically the CO2 plumes of power plants can be simulated and how strongly the results may depend on model type and resolution, however, is not well known due to a lack of observations available for benchmarking. Here, we use the unique data set of aircraft in-situ and remote sensing observations collected during the CoMet measurement campaign down-wind of the coal-fired power plants at Bełchatów in Poland and Jänschwalde in Germany in 2018 to evaluate the simulations of six different atmospheric transport models. The models include three Large-Eddy-Simulation (LES) models, two mesoscale numerical weather prediction (NWP) models, and one Lagrangian particle dispersion model (LPDM) and cover a wide range of model resolutions from 200 m to 2 km horizontal grid spacing. At the time of the aircraft measurements between late morning and early afternoon, the simulated plumes were slightly (at Jänschwalde) to highly (at Bełchatów) turbulent, consistent with the observations, and extended over the whole depth of the atmospheric boundary layer (ABL, up to 1800 m a.s.l. in the case of Bełchatów). The stochastic nature of turbulent plumes puts fundamental limitations to a point-by-point comparison between simulations and observations. Therefore, the evaluation focused on statistical properties such as plume amplitude and width as a function of distance from the source. LES and NWP models showed similar performance and sometimes remarkable agreement with the observations when operated at comparable resolution. A resolution of 1 km or better, however, appears to be necessary to realistically capture turbulent plume structures. At coarser resolution, the plumes disperse too quickly especially in the near field (0–8 km from the source) and turbulent structures are increasingly smoothed out. Total vertical columns are easier to simulate accurately than the vertical distribution of CO2, since the latter is critically affected by profiles of vertical stability, especially near the top of the ABL. Cross-sectional flux and integrated mass enhancement methods applied to synthetic CO2M data generated from the model simulations with a random noise of 0.5 ppm–1.0 ppm suggest that emissions from a power plant like Bełchatów can be estimated with an accuracy of about 20 % from single overpasses. Estimates of the effective wind speed are a critical input for these methods. Wind speeds in the middle of the ABL appear to be a good approximation for plumes in a well-mixed ABL as encountered during CoMet.

Abstract. Power plants and large industrial facilities contribute more than half of global anthropogenic CO2 emissions. Quantifying the emissions of these point sources is therefore one of the main goals of the planned constellation of anthropogenic CO2 monitoring satellites (CO2M) of the European Copernicus program. Atmospheric transport models may be used to study the capabilities of such satellites through observing system simulation experiments and to quantify emissions in an inverse modeling framework. How realistically the CO2 plumes of power plants can be simulated and how strongly the results may depend on model type and resolution, however, is not well known due to a lack of observations available for benchmarking. Here, we use the unique data set of aircraft in situ and remote sensing observations collected during the CoMet (Carbon Dioxide and Methane Mission) measurement campaign downwind of the coal-fired power plants at Bełchatów in Poland and Jänschwalde in Germany in 2018 to evaluate the simulations of six different atmospheric transport models. The models include three large-eddy simulation (LES) models, two mesoscale numerical weather prediction (NWP) models extended for atmospheric tracer transport, and one Lagrangian particle dispersion model (LPDM) and cover a wide range of model resolutions from 200 m to 2 km horizontal grid spacing. At the time of the aircraft measurements between late morning and early afternoon, the simulated plumes were slightly (at Jänschwalde) to highly (at Bełchatów) turbulent, consistent with the observations, and extended over the whole depth of the atmospheric boundary layer (ABL; up to 1800 m a.s.l. (above sea level) in the case of Bełchatów). The stochastic nature of turbulent plumes puts fundamental limitations on a point-by-point comparison between simulations and observations. Therefore, the evaluation focused on statistical properties such as plume amplitude and width as a function of distance from the source. LES and NWP models showed similar performance and sometimes remarkable agreement with the observations when operated at a comparable resolution. The Lagrangian model, which was the only model driven by winds observed from the aircraft, quite accurately captured the location of the plumes but generally underestimated their width. A resolution of 1 km or better appears to be necessary to realistically capture turbulent plume structures. At a coarser resolution, the plumes disperse too quickly, especially in the near-field range (0–8 km from the source), and turbulent structures are increasingly smoothed out. Total vertical columns are easier to simulate accurately than the vertical distribution of CO2, since the latter is critically affected by profiles of vertical stability, especially near the top of the ABL. Cross-sectional flux and integrated mass enhancement methods applied to synthetic CO2M data generated from the model simulations with a random noise of 0.5–1.0 ppm (parts per million) suggest that emissions from a power plant like Bełchatów can be estimated with an accuracy of about 20 % from single overpasses. Estimates of the effective wind speed are a critical input for these methods. Wind speeds in the middle of the ABL appear to be a good approximation for plumes in a well-mixed ABL, as encountered during CoMet.