Journal cover Journal topic
Atmospheric Chemistry and Physics An interactive open-access journal of the European Geosciences Union
Journal topic
Atmos. Chem. Phys., 19, 7347–7376, 2019
https://doi.org/10.5194/acp-19-7347-2019
Atmos. Chem. Phys., 19, 7347–7376, 2019
https://doi.org/10.5194/acp-19-7347-2019

Research article 04 Jun 2019

Research article | 04 Jun 2019

# Modelling CO2 weather – why horizontal resolution matters

Modelling CO2 weather – why horizontal resolution matters
Anna Agustí-Panareda1, Michail Diamantakis1, Sébastien Massart1, Frédéric Chevallier2, Joaquín Muñoz-Sabater1, Jérôme Barré1, Roger Curcoll3, Richard Engelen1, Bavo Langerock4, Rachel M. Law5, Zoë Loh5, Josep Anton Morguí3, Mark Parrington1, Vincent-Henri Peuch1, Michel Ramonet2, Coleen Roehl6, Alex T. Vermeulen7, Thorsten Warneke8, and Debra Wunch9 Anna Agustí-Panareda et al.
• 1European Centre for Medium-Range Weather Forecasts, Shinfield Park, Reading, RG2 9AX, UK
• 2Laboratoire des Sciences du Climat et de l'Environnement, LSCE/IPSL, CEA-CNRS-UVSQ, Université Paris-Saclay, 91198 Gif-sur-Yvette, France
• 3Environmental Science and Technology Institute, Universitat Autònoma de Barcelona, ICTA-UAB, Bellaterra, Spain
• 4Royal Belgian Institute for Space Aeronomy (BIRA-IASB), Brussels, Belgium
• 5CSIRO Oceans and Atmosphere, PMB 1, Aspendale, Victoria 3195, Australia
• 6California Institute of Technology, Pasadena, California, USA
• 7ICOS ERIC Carbon Portal, Sölvegatan 12, 22362 Lund, Sweden
• 8University of Bremen, Institute of Environmental Physics, Otto-Hahn-Allee 1, 28359 Bremen, Germany
• 9University of Toronto, Department of Physics, Toronto, Ontario, Canada

Correspondence: Anna Agustí-Panareda (anna.agusti-panareda@ecmwf.int)

Abstract

Climate change mitigation efforts require information on the current greenhouse gas atmospheric concentrations and their sources and sinks. Carbon dioxide (CO2) is the most abundant anthropogenic greenhouse gas. Its variability in the atmosphere is modulated by the synergy between weather and CO2 surface fluxes, often referred to as CO2 weather. It is interpreted with the help of global or regional numerical transport models, with horizontal resolutions ranging from a few hundreds of kilometres to a few kilometres. Changes in the model horizontal resolution affect not only atmospheric transport but also the representation of topography and surface CO2 fluxes. This paper assesses the impact of horizontal resolution on the simulated atmospheric CO2 variability with a numerical weather prediction model. The simulations are performed using the Copernicus Atmosphere Monitoring Service (CAMS) CO2 forecasting system at different resolutions from 9 to 80 km and are evaluated using in situ atmospheric surface measurements and atmospheric column-mean observations of CO2, as well as radiosonde and SYNOP observations of the winds.

The results indicate that both diurnal and day-to-day variability of atmospheric CO2 are generally better represented at high resolution, as shown by a reduction in the errors in simulated wind and CO2. Mountain stations display the largest improvements at high resolution as they directly benefit from the more realistic orography. In addition, the CO2 spatial gradients are generally improved with increasing resolution for both stations near the surface and those observing the total column, as the overall inter-station error is also reduced in magnitude. However, close to emission hotspots, the high resolution can also lead to a deterioration of the simulation skill, highlighting uncertainties in the high-resolution fluxes that are more diffuse at lower resolutions.

We conclude that increasing horizontal resolution matters for modelling CO2 weather because it has the potential to bring together improvements in the surface representation of both winds and CO2 fluxes, as well as an expected reduction in numerical errors of transport. Modelling applications like atmospheric inversion systems to estimate surface fluxes will only be able to benefit fully from upgrades in horizontal resolution if the topography, winds and prior flux distribution are also upgraded accordingly. It is clear from the results that an additional increase in resolution might reduce errors even further. However, the horizontal resolution sensitivity tests indicate that the change in the CO2 and wind modelling error with resolution is not linear, making it difficult to quantify the improvement beyond the tested resolutions.

Finally, we show that the high-resolution simulations are useful for the assessment of the small-scale variability of CO2 which cannot be represented in coarser-resolution models. These representativeness errors need to be considered when assimilating in situ data and high-resolution satellite data such as Greenhouse gases Observing Satellite (GOSAT), Orbiting Carbon Observatory-2 (OCO-2), the Chinese Carbon Dioxide Observation Satellite Mission (TanSat) and future missions such as the Geostationary Carbon Observatory (GeoCarb) and the Sentinel satellite constellation for CO2. For these reasons, the high-resolution CO2 simulations provided by the CAMS in real time can be useful to estimate such small-scale variability in real time, as well as providing boundary conditions for regional modelling studies and supporting field experiments.

1 Introduction

Over synoptic weather timescales of hours to days and spatial scales less than 1000 km, the assumption that atmospheric CO2 is well-mixed into a homogeneous background does not hold, as shown by the observed variability at baseline in situ stations (e.g. Halter and Harris1983). CO2 weather is defined here as the atmospheric CO2 variability at timescales and spatial scales of weather systems as depicted in Fig. 1. It reflects a complex combination of anthropogenic and natural CO2 fluxes near the Earth's surface and transport by weather systems in the atmosphere . This synergy of CO2 fluxes and weather results in intricate atmospheric CO2 patterns of positive and negative anomalies, collocated with weather variations on top of the well-mixed CO2 background that varies slowly on timescales of weeks to years .

Figure 1Snapshots of column-averaged CO2 (XCO2) (ppm) above (in reds) and below (in greens) the global mean on 15 January (a) and 15 July (b) at 12:00 UTC from the Copernicus Atmosphere Monitoring Service global CO2 forecast at high horizontal resolution (∼9km).

Modelling the synoptic-scale transport that modulates the CO2 weather is crucial for interpreting the variability of surface CO2 concentrations from in situ observations and column-averaged CO2 from satellite and ground-based observations and for forecasting CO2 from 1 to 10 d ahead in order to examine the predictive skill of the models. Tracer transport models use the numerical schemes and meteorological information of numerical weather prediction (NWP) to simulate the tracer variability in the atmosphere. Increasing the horizontal resolution associated with the grid spacing of tracer transport models has the benefit of reducing the numerical errors in tracer simulations, leading to convergence of the transport solution from different transport schemes . NWP models for weather forecasting have been doubling the global horizontal resolution approximately every 8 years (Wedi2014) in order to improve the forecast skill. But until now, global tracer transport models have generally used lower resolution than NWP models, as chemical transport models including chemistry and/or long window data assimilation cannot afford such computational expense.

Observations of atmospheric CO2 are used in data assimilation systems based on tracer transport models to produce optimal estimates of atmospheric CO2 concentrations (e.g. Massart et al.2016) or model parameters and CO2 fluxes in atmospheric inversion systems . If tracer transport models cannot represent the synoptic variability accurately, then the resulting errors when comparing the tracer from the model with observations will prevent these observations from being used effectively in the data assimilation systems (e.g. Brooks et al.2012). The model–observation mismatch caused by differences in the resolution of the tracer transport model – including both the resolution of the meteorological fields and the resolution of the fluxes on the model grid – and the resolution of the observation footprint is also known as representativeness error. Failure to properly account for representativeness errors in data assimilation will lead to errors in the optimised parameters, whether atmospheric concentrations, model parameters or surface fluxes.

Several studies have investigated the spatial representativeness errors of CO2 by analysing the CO2 distribution within model grid cells, based on nested high-resolution simulations on limited domains over Europe, North America and South America for certain months or by statistical parameterisation of CO2 covariances based on lower-resolution simulations . The importance of high resolution over complex terrain has also been demonstrated on regional scales, e.g. in Europe and in North America using very high resolution simulations (down to 1 km). However, other studies with coarser global tracer transport models have compared CO2 simulations with a range of resolutions from a few degrees down to 0.5 without finding significant improvements with respect to observations .

The full impact of horizontal resolution on the simulated tracer variability depends on the resolution of transport and emissions/biogenic fluxes (e.g. Vogel et al.2013) as well as the resolution of the topography and the winds (e.g. Sekiyama et al.2015) in the model. In this study the full sensitivity of CO2 synoptic variability to the model horizonal resolution (including all the aspects mentioned above) is investigated by quantifying the change in model error with horizontal resolution at observing stations. Three main questions are addressed:

1. What is the sensitivity of the modelled atmospheric CO2 variability at diurnal and synoptic timescales to horizonal resolution?

2. How is horizontal resolution affecting the medium-range (1–10 d) forecast error growth of atmospheric CO2?

3. What are the typical CO2 representativeness errors in models with horizontal resolutions of 1× 1, currently considered as high resolution in tracer transport models, and where and when are these representativeness errors largest?

The model simulations use the operational Copernicus Atmosphere Monitoring Service (CAMS) global CO2 forecasting system which is based on the Integrated Forecasting System (IFS) model of the European Centre for Medium-Range Weather Forecasts (ECMWF). They are performed over a range of resolutions currently used operationally in NWP from 9 to 80 km. A detailed description of the simulations, observations and tools used to assess the importance of horizontal resolution for simulating atmospheric CO2 variability related to weather is presented in Sect. 2. Section 3 shows the impact of horizontal resolution on the error of simulated horizontal winds (Sect. 3.1) and atmospheric CO2 (Sect. 3.2 and 3.3). The results of the sensitivity to horizontal resolution are explained in the context of the small-scale variability in Sect. 3.4. The diagnostics of small-scale variability provide an estimate of the expected representativeness errors for CO2 simulations with coarser horizontal resolutions. Finally, an example of an urban site is shown in Sect. 3.5, where the impact of horizontal resolution is positive in January and negative in July. The implications of the results for CO2 forecasting and atmospheric inversion systems are discussed in Sect. 4, with a summary of the main findings on why and where horizontal resolution matters.

2 Methodology

## 2.1 Observations

Continuous in situ observations near the surface and column-averaged observations from the Total Carbon Column Observing Network (TCCON) provide the reference for atmospheric CO2 variability. Figure 2 shows the spatial distribution of the CO2 observing stations used in this study. Hourly near-surface CO2 observations are provided by 51 in situ stations operated by various organisations throughout the period of the simulations: data from 44 stations are taken from the cooperative GLOBALVIEWplus (ObsPack 2015) data set, and additional data have been obtained from three additional stations from CSIRO in Australia and Antarctica and four stations from the ClimaDat network over the Iberian Peninsula. The cooperative GLOBALVIEWplus (ObsPack 2015) data set is coordinated by NOAA, and it comprises data collected by various institutions and laboratories including AEMET, AGH, CSIRO, ECCC, ECN, EMPA,FMI, HMS, LSCE, NCAR, NOAA, JMA, NIWA, SAWS, TU, UBA-SCHAU, UEA, UHEI-IUP and UR (see Tables A1 and A2 for full list of stations with their organisations and associated references). No selection criteria are applied to the stations from the GLOBALVIEWplus , CSIRO and ClimaDat data sets, other than availability of hourly data for January and July 2014.

Figure 2Map of in situ (blue squares) and TCCON (red triangles) stations. Detailed information on each station is provided in Tables A1 and A3.

Most stations are on the World Meteorological Organization (WMO) CO2 scale, although the inter-calibration of standard gases is not critical for this study because the focus is on the relative difference between the high- and low-resolution simulations to quantify the sensitivity of modelled CO2 to horizontal resolution in the model. The distribution of the stations is not homogeneous over the globe. However, there is a wide variety of locations that sample synoptic variability on various types of terrain including many coastal, mountain, continental and oceanic sites over different continents on both hemispheres. Wind observations from around 400 radiosondes stations and all the operational 10 m SYNOP stations around the globe are used to evaluate the sensitivity of wind errors to the model horizontal resolution at different atmospheric levels in the troposphere.

Total column observations from 18 TCCON Fourier transform spectrometers (FTSs) available in January and July 2014 – shown as red triangles in Fig. 2 – are also used to evaluate the variability of the column-averaged dry-air mole fraction of CO2 – hereafter referred to as XCO2 – (Table A3). These TCCON observations are retrieved from direct solar near-infrared spectra (http://www.tccon.caltech.edu, last access: 30 May 2019), and they provide a ground reference to the GOSAT , OCO-2 and TanSat satellite observations (e.g. Inoue et al.2016; Wunch et al.2017). Total column averages are less sensitive to the uncertainties associated with vertical mixing than the CO2 abundances near the surface. However, the temporal coverage of TCCON observations is limited to clear-sky and sunny conditions, which means there are generally more gaps in the TCCON data than in near-surface in situ data.

## 2.2 Global atmospheric CO2 model

The model used in this study is the Integrated Forecasting System (IFS), the same model used in NWP at ECMWF and in the CAMS atmospheric composition analysis and forecasting system to issue 5 d CO2 and CH4 forecasts (https://atmosphere.copernicus.eu/charts/cams/carbon-dioxide-forecasts, last access: 30 May 2019), as well as reactive gases and aerosol forecasts relevant for air quality . The IFS model version is CY43r1, the operational weather forecast model at ECMWF from 22 November 2016 to 10 July 2017. A full evaluation of this model cycle can be found in . The high horizontal resolution is based on a cubic octahedral reduced Gaussian – called hereafter octahedral – grid . The implementation of the octahedral grid has allowed a substantial increase in the grid point resolution from 16 km to approximately 9 km, without having to increase the spectral resolution of the model . The 9kmEXP simulation comprises up to 904 million model grid points, 137 levels and a time step of 7.5 min.

The tracer transport is modelled by three different numerical schemes to represent (i) the resolved advection of CO2 by the winds, and the sub-grid-scale (ii) convection and (iii) turbulent mixing processes that need to be parameterised. The tracer advection is computed using a semi-implicit semi-Lagrangian scheme which is an unconditionally stable method for the integration of the transport equations and for the fast terms associated with gravity waves. Semi-Lagrangian advection schemes have small dispersion and phase speed errors despite using long time steps . In practice, these properties mean that the time step is limited only by the local truncation error and not by numerical stability bounds. The semi-Lagrangian advection scheme in the IFS is not mass-conserving. Thus, a mass fixer is required to ensure mass conservation at every time step . The turbulent mixing scheme is described in , and . The convection scheme is based on (see Bechtold et al.2008, 2014, for further details). Full documentation of the IFS can be found at https://www.ecmwf.int/en/forecasts/documentation-and-support/changes-ecmwf-model/ifs-documentation (last access: 30 May 2019).

The CO2 surface fluxes from the ocean, biomass burning and anthropogenic emissions are prescribed using inventories or climatologies, while the biogenic fluxes over land are modelled online (see Table 1). The anthropogenic CO2 emissions come from the EDGAR v4.2 FT2010 inventory for 2012 (last year with gridded emissions). They are extrapolated in time to the year of the simulation with country trends provided by the EDGAR database (http://edgar.jrc.ec.europa.eu, last access: 30 May 2019). The biogenic CO2 emissions from land vegetation are modelled with the A-gs photosynthesis scheme and an empirical model to simulate the ecosystem respiration fluxes which are integrated in the CHTESSEL land surface model of the IFS . The fluxes have been evaluated with FLUXNET data and compared to different models (e.g. CASA and ORCHIDEE) showing a comparable performance on synoptic to seasonal scales . An online bias correction scheme is applied to the modelled gross primary production (GPP) and ecosystem respiration (Reco) fluxes to correct for biases in the net ecosystem exchange (NEE) budget on a timescale of 10 d compared to a climatology of optimised fluxes . Figure S1 in the Supplement shows the monthly mean NEE for the highest and lowest resolutions.

Table 1Datasets and models of CO2 fluxes used in simulations listed in Table 2.

The atmospheric tracer transport and CO2 biogenic fluxes are two of the largest contributors to the synoptic variability of atmospheric CO2 globally . Thus, the modelling of these two components online in the IFS allows us to investigate the full impact of the resolution coming from the winds and the tracer transport, as well as the fluxes.

## 2.3 Global atmospheric CO2 simulations

A set of global simulations are performed at several resolutions from 9 to 80 km (Table 2) to investigate the impact of horizontal resolution on the modelled CO2 variability at diurnal and synoptic scales. These are the resolutions that are currently used operationally in global meteorological reanalysis – e.g. ERA-Interim at 80 km – widely used in tracer transport models and the typically higher resolutions of operational weather forecasts models. For instance, the deterministic weather forecast at ECMWF currently runs at 9 km resolution, and it was the global forecasting system with the highest resolution in the world when it was introduced on 8 March 2016 .

Table 2List of simulations with different resolutions given by different model grids. All simulations use 137 vertical model levels. All the experiments have been performed in January and July 2014 using the same CO2 surface fluxes (see Table 1).

The octahedral grid is used for all simulations, except for the lowest-resolution simulation at 80 km which uses a reduced linear Gaussian grid as in the ERA-Interim and CAMS Reanalysis . The time steps are also dependent on the horizontal resolution and range from 7.5 to 45 min. As described in Sect. 2.2, the semi-implicit semi-Lagrangian method used in the IFS is free from stability restrictions. Thus, the model uses the longest possible time step that provides the most accurate result for each spatial resolution. This is selected through experimentation and validation, but a rule of thumb is that as the horizontal resolution increases, the time step decreases to keep the mean Courant–Friedrichs–Lewy (CFL) number constant. This typically leads to much longer time steps than Eulerian models for which their time step is restricted by the typical CFL stability limit (i.e. the maximum CFL number being less than 1).

All the simulation experiments are conducted for a winter and a summer month, in January and July 2014, as we expect that winter and summer periods will show markedly different variability patterns in CO2. Figure 3 shows the configuration of the simulations. A 10 d forecast is performed at 00:00 UTC each day of the month. The meteorological initial conditions of each forecast come from the ECMWF operational NWP analysis , whereas the atmospheric CO2 tracer is initialised with the previous 1 d forecast, which means CO2 is essentially free-running, as in . The first initial conditions for CO2 on 1 January and 1 July 2014 are extracted from the CAMS CO2 analysis . NWP analysis of meteorological fields is one of the main elements determining the quality of the tracer transport . Keeping the meteorological fields close to the analysis by having a sequence of 1 d forecasts ensures the tracer transport is as realistic as possible. Therefore, the sequence of 1 d forecasts is used as the standard (cyclic forecast) configuration for the simulations at different resolutions.

Figure 3Schematic of CO2 simulations with cyclic forecast configuration with 10 d forecasts initialised every day from 1st of the month to the 10th day of the following month. Initial conditions are depicted by arrows (see legend), and the period of evaluation in which several forecast lead times can be compared is delimited by the red dashed line. The standard CO2 simulations are composed by a series of 1 d forecasts as shown by the green rectangles.

The extension to the 10 d forecasts allows us to assess the impact of errors in the meteorological fields – which grow during the forecast – on the CO2 simulations. There are 10 realisations of CO2 for each day, one for each forecast lead time (Fig. 3). Each forecast lead time is evaluated separately in order to estimate the error growth during the forecast. For consistency in the evaluation of the different forecast lead times, the periods from 10 January to 10 February and 10 July to 10 August are used in the validation diagnostics.

The simulations also include an additional CO2 tracer which is only transported (i.e. does not respond to CO2 surface fluxes) during the forecast. We refer to this tracer as NFX. This tracer is still initialised with the standard CO2 at the beginning of each forecast. The difference between the NFX CO2 and the standard CO2 tracers can provide insight into the sensitivity to local flux at different horizontal resolutions. Similarly, the change in the error of the simulation with resolution for both the standard and the NFX tracers can be used as an indicator of transport versus local flux influence in the assessment of the impact of horizontal resolution.

## 2.4 Diagnostics for model evaluation

The focus of this paper is on assessing the skill of the model in simulating CO2 weather with short-term variability over a period of a month. For this purpose, the root mean square error

$\begin{array}{}\text{(1)}& \mathrm{RMSE}=\sqrt{\frac{\mathrm{1}}{N}\sum _{i=\mathrm{1}}^{N}\left({m}_{i}-{o}_{i}{\right)}^{\mathrm{2}}},\end{array}$

the systematic error or bias

$\begin{array}{}\text{(2)}& \mathit{\mu }=\frac{\mathrm{1}}{N}\sum _{i=\mathrm{1}}^{N}\left({m}_{i}-{o}_{i}\right)\end{array}$

and the random error

$\begin{array}{}\text{(3)}& \mathrm{STDE}=\sqrt{\frac{\mathrm{1}}{N}\sum _{i=\mathrm{1}}^{N}\left({m}_{i}-{o}_{i}-\mathit{\mu }{\right)}^{\mathrm{2}}}\end{array}$

of the modelled CO2 dry molar fraction (m) are computed with respect to N hourly observations (o) at each observing site. The standard deviation of the site error – also known as inter-station error – is used as an indicator of the spatial variability of the error e (e.g. RMSE, μ) between the M observing sites:

$\begin{array}{}\text{(4)}& {\mathit{\sigma }}_{e}=\sqrt{\frac{\mathrm{1}}{M-\mathrm{1}}\sum _{s=\mathrm{1}}^{M}\left({e}_{s}-\stackrel{\mathrm{‾}}{e}{\right)}^{\mathrm{2}}},\end{array}$

where $\stackrel{\mathrm{‾}}{e}$ is the mean error of all sites. It reflects the skill of the model in representing spatial gradients between the sites. The Pearson’s correlation coefficient is also used to assess the skill of the model in simulating the diurnal and synoptic variability at the sites.

The model is sampled in the horizontal by taking the nearest grid point to the station over land. This approach is widely used in model evaluation as it allows assessment of the model directly at grid point scale. At coastal locations, coarse-resolution models can find a better fit to observations by sampling the nearest ocean grid point as land grid points tend to overestimate the diurnal cycle . For this reason, the sampling protocol for observations in the atmospheric inversion system moves some stations offshore . However, coastal sites can be influenced by both ocean and land, which means that they will have contrasting periods sampling baseline air associated with low variability and periods with land and local influences associated with high CO2 variability . In this study we have chosen to consistently sample the nearest land point over land because we are interested in assessing the capabilities of the model to represent both baseline and local influences. The temporal sampling is performed with a linear interpolation from the 3-hourly archived model fields to the observation time as in .

At the surface stations, the model is also interpolated to the altitude of the sampling height above ground level (a.g.l.). This ensures the same model levels are used for the different horizontal resolutions. The model has hybrid coordinates that follow the terrain close to the surface. Selecting the model level at the station height above mean sea level (a.m.s.l.) would imply the use of different model levels for different resolutions when the orographic height varies between the horizontal resolutions. It would therefore lead to comparisons of CO2 in the planetary boundary layer and free troposphere at mountain sites where the low-resolution model underestimates the orographic height. tested both approaches at several mountain sites. They found that the sampling at a.m.s.l. greatly underestimates the amplitude of the diurnal cycle, as the sensitivity to local fluxes is reduced at higher levels above the ground. Since most low-resolution models used in atmospheric inversions tend to use the model sampling a.m.s.l. at mountain sites (e.g. Wang et al.2018), a comparison of the two approaches (a.g.l. and a.m.s.l.) is provided in the Supplement. At the TCCON stations, the model profile is processed with the TCCON averaging kernel and prior, as described in Appendix A of .

Atmospheric CO2 variability is subject to local- or small-scale influences (<100km) associated with complex topography, coastal boundaries, local fluxes and mesoscale atmospheric flow (Lin2007). Most models used in carbon cycle studies are unable to represent such local variability. The resulting representativeness errors reflect the sub-grid-scale variability associated with the coarse resolution of the models (e.g. Tolk et al.2008). At high resolution it is possible to estimate part of this sub-grid-scale variability of coarser models. In order to do that, the 9kmEXP simulation is interpolated onto a 0.1× 0.1 regular lat–long. grid, and subsequently it is sampled for each time zone (computed hourly along longitude) at 13:00 ± 0.5 h local time. This temporal sampling at 13:00 is consistent with the GOSAT (http://www.gosat.nies.go.jp/en/, last access: 30 May 2019) and OCO-2 (http://oco.jpl.nasa.gov/mission, last access: 30 May 2019) overpass time. Thus, it provides a more relevant estimate of the potential representativeness error for lower-resolution inversion systems, which use daytime surface in situ data and satellite data (e.g. Chevallier et al.2010, 2014). The representativeness error is estimated by computing the standard deviation of the CO2 dry molar fraction at 0.1 resolution within the coarser grid boxes of 1× 1 over the whole globe:

$\begin{array}{}\text{(5)}& {\mathit{\sigma }}_{{\mathrm{CO}}_{\mathrm{2}}}=\sqrt{\frac{\mathrm{1}}{n-\mathrm{1}}\sum _{j=\mathrm{1}}^{n}\left({m}_{j}-\stackrel{\mathrm{‾}}{m}{\right)}^{\mathrm{2}}},\end{array}$

where $\stackrel{\mathrm{‾}}{m}=\frac{\mathrm{1}}{n}\sum _{j=\mathrm{1}}^{n}{m}_{j}$; n is the number of 0.1 resolution grid cells within the coarser grid cell of 1× 1; m is the CO2 dry molar fraction at 0.1 resolution; and $\stackrel{\mathrm{‾}}{m}$ is the average within the coarser grid cell.

3  Results

## 3.1 Impact of horizontal resolution on winds

The accuracy of the winds is a crucial aspect of the CO2 transport quality, as winds drive the advection of CO2 across the resolved gradients in the model. In this section we investigate the benefit of increasing the resolution from 80 to 9 km on the RMSE of the zonal and meridional components of the wind. We investigate the changes in the global wind error with model resolution based on 12-hourly radiosonde observations which measure the horizontal wind components throughout the troposphere. Figure 4 shows there is a consistent and significant RMSE reduction of the vector wind for the 1 d forecast with resolution. The impact of resolution – quantified here by the difference in RMSE between the 80kmEXP and 9kmEXP simulations – is largest near the surface at 850 and 1000 hPa, with a RMSE reduction ranging between 0.2 and 0.6 m s−1. This is equivalent to a reduction in RMSE of around 15 % near the surface. In the middle and upper troposphere (500 and 200 hPa) there is a consistent but smaller RMSE reduction, ranging between 0.1 and 0.2 m s−1.

Figure 4Mean RMSE of vector wind (m s−1) at different model resolutions in (a) January and in (b) July for around 400 radiosonde stations over the globe. Different colours represent different pressure levels (see legend). All the model simulations are based on the standard 1 d forecast configuration shown in Fig. 3. Note that the number of data at the 1000 hPa level might be slightly smaller than at the other pressure levels, as the observations at 1000 hPa are not available when the surface pressure is lower than 1000 hPa.

The RMSE reduction extends throughout the 10 d forecast for the two components of the wind between 1000 and 850 hPa with values around 0.4 m s−1, and it is consistent in both the Northern and Southern Hemisphere and the tropics (not shown). The results are also in agreement with the RMSE with respect to 10 m wind speed from SYNOP observations, with a mean RMSE reduction over the global domain of 0.34 m s−1. The reduction of the mean error is smaller than the RMSE (<0.2m s−1) throughout the troposphere, which means the largest component of the wind error is random.

## 3.2 Impact of horizontal resolution on CO2 diurnal and synoptic variability

The sensitivity of the model skill at hourly and daily timescales to the horizontal resolution of the model is assessed with the error of the CO2 simulations with respect to hourly mean observations. The change in the RMSE with horizontal resolution based on the surface CO2 and XCO2 observations (see Sect. 2.1) is shown in Figs. 5 to 7.

Figure 5Mean RMSE of near-surface CO2 (ppm) (a, b) and Pearson's correlation coefficient R (c, d) at different model resolutions in January (a, c) and July (b, d) for all 51 stations (see Table A1). The standard deviation of the plotted variable from each station is shown by the numbers below the horizontal resolution for each temporal resolution (hourly, daily mean, daily min and daily max). All the model simulations are based on the 1 d forecast. Note that different scales are used in each panel.

Figure 6Mean RMSE of near-surface CO2 (ppm) at different model resolutions in January (a, c) and July (b, d) for (a, b) 37 lowland stations (below 1000 m a.m.s.l.) and (c, d) 12 mountain stations (1000 m a.m.s.l., excluding bao and spo, as listed in Table A1). The standard deviation of the plotted variable from each station is shown by the numbers below the horizontal resolution for each temporal resolution (hourly, daily mean, daily min and daily max). All the model simulations are based on the 1 d forecast. Note that different scales are used in each panel.

Figure 7Mean RMSE of XCO2 (ppm) and Pearson's correlation coefficient R with respect to observations from 18 TCCON stations (see Table A3) at different model resolutions in (a) January and (b) July. The standard deviation of the plotted variable from each station is shown by the numbers below the horizontal resolution for each temporal resolution (hourly, daily mean, daily min and daily max). All the model simulations are based on the 1 d forecast. Note that different scales are used in each panel.

At the surface there is an overall substantial reduction of RMSE between 80 and 9 km (i.e. between 1.8 and 3.5 ppm for hourly data) which is clearly not linear (Fig. 5a and b). The RMSE difference between the 80kmEXP and 40kmEXP simulations or the 40kmEXP and 25kmEXP simulations is not as large as the difference between the higher-resolution simulations (e.g. the 25kmEXP and 16kmEXP or the 16kmEXP and 9kmEXP simulations). This is particularly pronounced for the daily maximum CO2 occurring usually at night-time, which is generally controlled by local fluxes and small-scale transport of tracers, and therefore it is more sensitive to resolution. The daily maximum values are generally much better captured at 9 km resolution compared to 80 km, with a reduction in the RMSE of around 2.5 ppm in January and 6 ppm in July. Indeed, there are large differences between the RMSE of the daily maximum and minimum CO2 values. As expected, daily minimum values that emerge during daytime have a smaller RMSE. This is because during daytime the minimum CO2 values are influenced by the larger-scale fluxes and tracer transport which are less sensitive to high resolution. The reduction in RMSE of the daily minimum CO2 is therefore smaller than for the daily maximum, but it is still considerable, with an RMSE decrease of around 0.75 ppm from 80 to 9 km resolutions in both January and July. These differences reflect the ability of the model to represent the diurnal cycle. The 9kmEXP simulation clearly shows a general improvement in the CO2 diurnal cycle near the surface, with smaller differences in the RMSE of the two daily extremes. The largest RMSE reduction comes from mountain sites (over 1000 m a.m.s.l.), ranging between 6 and 10 ppm for hourly CO2 (Fig. 6a and b), compared to the lowland sites, which can see improvements between 0.5 and 2 ppm for hourly CO2 RMSE near the surface (Fig. 6c and d).

In general there is also a notable reduction in the spread of the RMSE at the different sites with resolution, as shown by the σ RMSE values below the panels in Figs. 5 and 6. This implies that the spatial gradients between stations are better represented at higher resolutions. The global mean correlation coefficient also increases with resolution from 0.47 to 0.56 in January and 0.51 to 0.59 in July for the hourly CO2, with consistently higher correlations for the daily mean, minimum and maximum CO2 at higher resolution.

As expected, the sensitivity to the strategy of sampling the model level at observing stations is generally small over lowlands but large over mountains, particularly at low resolution (Fig. S2). At mountain sites, the model level at the real station height above mean sea level is predominantly in the free troposphere, and therefore it has a small sensitivity to the local fluxes and flow, whereas taking the model level with respect to the model ground generally exhibits larger errors associated with local influences in the boundary layer. The difference between the two sampling strategies in the RMSE and correlation coefficients becomes smaller at high resolution (Fig. S3). This reflects an improvement in the capability of the model to represent the flow and fluxes around complex topography at higher horizontal resolution.

The XCO2 RMSE at the TCCON sites during daytime also displays a general decrease with resolution (Fig. 7), with differences of the order of 0.1 ppm from 80 to 9 km resolutions and increases in the correlation coefficients (r) of up to 0.05. In boreal summer, the XCO2 daily minimum has the largest/smallest RMSE/r because it reflects the uncertainty associated with modelled photosynthesis and negative XCO2 anomalies, whereas in boreal winter, the XCO2 daily maximum has the largest RMSE because ecosystem respiration associated with positive XCO2 anomalies is the dominant process at most TCCON sites. It is likely that the larger footprint of XCO2 at most TCCON stations – associated with its sensitivity to large-scale flux patterns – is causing most TCCON sites to be less sensitive to horizontal resolution. However, there is a large variation in RMSE between sites (see σ RMSE in Figs. 7 and S9), which is reduced at high resolution. In particular, the TCCON site at Pasadena (California, USA), located near the anthropogenic emission hotspot of the megacity of Los Angeles, stands out (Fig. S9). The improvement associated with high resolution at Pasadena is indeed remarkable in January (i.e. approximately 2 ppm RMSE reduction). A more detailed study for Pasadena is provided in Sect. 3.5.

The change of RMSE with resolution is partly associated with the improvement in the transport and also the representation of the local fluxes at higher resolutions. Figure 8 shows that when the fluxes are switched off during the 1 d forecast, there is still an improvement with resolution at most sites, but the magnitude of the error reduction is smaller (see symbols to the right of the dashed line). This is very clear for a large number of mountain sites and TCCON sites affected by anthropogenic emissions such as Pasadena (USA) in January and Saga (Japan) in July. However, there are also some sites and months in which the impact of resolution is better without fluxes than with fluxes (e.g. Pasadena in July). This would indicate that in this case the errors in the fluxes are the main cause of the deterioration in RMSE with resolution.

Figure 8Sensitivity of horizontal resolution impact to surface CO2 fluxes at (a, b) surface stations and (c, d) TCCON stations in January and July, as shown by the difference in RMSE between different CO2 tracers, i.e. the standard tracer (RMSE9 kmRMSE80 km) and the CO2 tracer with just transport (RMSENFX9kmRMSENFX80km) in the y axis and x axis respectively. The symbols that are close to the dashed line correspond to stations that have a small sensitivity to local fluxes, while at the stations associated with symbols that are located above/below the dashed line there is a negative/positive contribution of the local fluxes at high resolution. The further from the dashed line, the larger the contribution of the local fluxes. The stations located along the y axis are mainly impacted by local fluxes. The surface stations in (a, b) are depicted with different symbols depending on whether they are classified as mountain, continental, coastal or remote (see Table A1), while TCCON stations that are strongly influenced by fluxes are labelled with station names. Note that different scales are used in each panel.

The overall global error statistics of the 9kmEXP and 80kmEXP simulations including the systematic (or bias) error and the standard (or random) error are shown in Table 3. The reduction in RMSE at 9 km is associated with a decrease in the magnitude of the CO2 biases on average of 1.5 to 2 ppm near the surface and up to 0.2 ppm for XCO2 and a general reduction in the CO2 random error of 1 to 1.5 ppm near the surface and 0.1 ppm for XCO2 (Figs. S4 and S5). The biases depend largely on the bias of the CO2 initial conditions, as well as the biases of the fluxes and tracer transport. What is important in this sensitivity study is that the standard deviation of the bias at each station – i.e. the inter-station bias – is reduced at 9 km with respect to 80 km, as shown by the shaded area in Figs. S4 and S5. The largest decrease in the inter-station bias between 80kmEXP and 9kmEXP simulations occurs in January, when it is almost halved near the surface. The errors at the individual observing stations are listed in Tables S1, S2, S3 and S4.

Table 3Surface CO2 and XCO2 mean statistics for bias, STDE and RMSE of all stations and the standard deviation of inter-station statistics (in brackets and bold font) from the highest- and lowest-resolution simulations, i.e. 9kmEXP and 80kmEXP respectively. All the SFC CO2 stations used in January and July are listed in Tables S1 and S2; the TCCON stations used in the XCO2 statistics are listed in Tables S3 and S4.

## 3.3 Impact of horizontal resolution on CO2 forecast error growth

In 10 d the global mean RMSE of CO2 forecast at the in situ surface stations grows by around 1.4 ppm in January and around 1 ppm in July (Fig. 9). It is worth noting that this error growth is smaller in magnitude than the impact of increasing horizontal resolution from 80 to 9 km. Namely, the 10 d forecast at 9 km is better than the 1 d forecast at 80 km near the surface. At the TCCON sites the XCO2 RMSE grows on average between 0.2 and 0.5 ppm in 10 d (Fig. 10). The forecast RMSE growth for near-surface CO2 and XCO2 does not appear to be linear, with a slow growth until day 4 and a faster increase from day 5 onwards. The CO2 RMSE growth at 80 km is slightly faster than at 9 km. In summary, the gain in skill from horizontal resolution is maintained throughout the 10 d forecast. Thus, the results suggests that the horizontal resolution has a small but positive impact on the short- and medium-range forecast skill for CO2 and XCO2.

Figure 9Mean RMSE of near-surface CO2 (ppm) at different forecast lead times for the 9kmEXP (red) and 80kmEXP (blue) in (a) January and (b) July. The errors are computed with respect to hourly continuous in situ surface measurements from 51 stations (see Table A1). The error standard deviation between the different stations is shown with the shaded area: red for 9kmEXP, blue for 80kmEXP and grey for overlap. Note that different scales are used in each panel.

Figure 10Same as Fig. 9 but for XCO2 (ppm). The errors are computed with respect to hourly TCCON observations from 18 TCCON stations (see Table A3).

As expected, the RMSE in July is largest because of the high uncertainty associated with the modelled biogenic fluxes at synoptic scales which influence the variability at continental sites . There is also a larger uncertainty in the meteorology driving the tracer transport during summer compared to winter . The fact that the forecast RMSE for day 1 is larger than for day 2 in July is associated with a sporadic overestimation of daily maximum CO2 peaks at sites influenced by strong local fluxes. There are several potential causes of the overestimation (e.g. biogenic fluxes responding to rapid adjustments in meteorology after analysis re-initialisation at 00:00 UTC or issues with the tracer transport associated with the short spin-up period), but these are beyond the scope of this study.

The near-surface CO2 RMSE increase during the forecast appears to come mostly from an increase in random error in January and from both mean and random error in July (Fig. S4), whereas for XCO2, both mean and random errors contribute equally to the forecast RMSE growth in January and July (Fig. S5). This is probably linked to the distribution of the stations, as most in situ stations are located in the Northern Hemisphere, whereas TCCON stations are more equally distributed in both hemispheres, and thus, the mean error at all stations does not show differences between summer and winter conditions.

## 3.4 Impact of horizontal resolution on CO2 small-scale variability

The sensitivity of the RMSE to resolution is generally associated with regions that are affected by small-scale variability that cannot be properly represented by typical global tracer transport models . Figure 11 shows that the mean small-scale variability, given by the standard deviation within 1× 1 grid box, can be as large as 10 ppm near emission hotspots at the surface during daytime. The representation of the CO2 small-scale variability at the surface in the 9kmEXP compared to the 80kmEXP is also illustrated in Fig. S6. Larger values than 10 ppm can be found over most land areas at night-time (Fig. 12). These values are likely to be underestimated, since we expect horizontal gradients to become steeper as the resolution increases, the point sources associated with anthropogenic activities become stronger at the grid cell scale and part of the sub-grid-scale flow is resolved.

Figure 11Monthly mean surface CO2 small-scale variability (σ) within 1× 1 grid boxes (ppm) at 13:00 local time in (a) January and (b) July. Grey shading denotes σ<1.0.

Figure 12Monthly mean surface CO2 small-scale variability (σ) within 1× 1 grid boxes (ppm) at 01:00 local time in (a) January and (b) July. Grey shading denotes σ<1.0.

Coastal sites and mountain sites have a typical sub-grid-scale variability of around 5 ppm during daytime. This variability varies from January to July, depending on meteorological conditions (e.g. stagnant or windy conditions) and the magnitude and sign of fluxes (e.g. biogenic activity shifting northwards in Northern Hemisphere summer). Over land, the patterns of sub-grid-scale variability of surface and total column are consistent (Figs. 11 and 13), as both are subject to surface heterogeneity in terms of topography and fluxes. However, there is a difference in magnitude because the variability of the total column average is much smaller than the variability at the surface.

Figure 13Monthly mean XCO2 small-scale variability (σ) within 1× 1 grid boxes (ppm) at 13:00 local time in (a) January and (b) July. Grey shading denotes σ<0.1.

XCO2 has a maximum standard deviation of 1 ppm near surface flux hotspots and typically less than 0.5 ppm in most regions (Fig. 13), which is consistent with other estimates from regional studies (see also Fig. S7 for a visual illustration). The differences in the small-scale XCO2 variability between day and night appear to be small. Interestingly, the small-scale variability of XCO2 is much larger in summer than in winter (both in the Northern and Southern Hemisphere). During the growing season, negative CO2 anomalies associated with plant photosynthesis and positive anomalies associated with ecosystem respiration and anthropogenic emissions combine to create steeper gradients throughout the troposphere – as illustrated in Fig. 1b – that contribute to the enhanced sub-grid-scale variability in summer compared to winter. Over the ocean, the small-scale variability of XCO2 ranges between 0.1 and 0.3 ppm, with lower values in the winter and higher values in the summer. In the Northern Hemisphere summer, the values over the ocean and over the land are comparable, whereas near the surface, the mean sub-grid-scale variability is an order of magnitude smaller over the ocean than over land. This is because over land the surface fluxes dominate the gradients resulting in the steepest gradients being near the surface, while over the ocean, the transport associated with the weather systems creates steep CO2 gradients in the free troposphere. Therefore, column-averaged CO2 is much more likely to be influenced by sub-grid-scale variability associated with weather systems than by surface CO2 fluxes over the ocean.

## 3.5 Example of horizontal resolution impact at an urban site

Although the winds, the topography and the spatial heterogeneity of the fluxes are generally better represented at high horizontal resolution, there can still be a deterioration in the RMSE scores at sites where the local influence is strong and the emissions/biogenic fluxes have large errors in the model. In this section we present an example of such a case at the Caltech TCCON site in Pasadena (California, USA; see Table A3) with XCO2 under clear-sky and daylight conditions. The variability of the simulated XCO2 exhibits a substantial improvement with high resolution in winter and an equally considerable deterioration in summer (Fig. 15). Thus, it illustrates some of the challenges associated with urban regions.

Figure 14Monthly mean XCO2 small-scale variability (σ) within 1× 1 grid boxes (ppm) at 01:00 local time in (a) January and (b) July. Grey shading denotes σ<0.1.

Figure 15Hourly XCO2 (ppm) at TCCON site of Pasadena (CA, USA) in (a) January and (b) July from 80kmEXP (blue) and 9kmEXP (red) simulations. Hourly observations are shown by black circles. Triangles represent the model XCO2 after smoothing with TCCON averaging kernel and prior. The bias (μ), standard error (STDE) and Pearson's correlation coefficient (R) from 80kmEXP (blue) and 9kmEXP (red) are shown at the top of each panel.

Pasadena is located 14 km north-east of the megacity of Los Angeles (LA) with a large local anthropogenic emission influence . The XCO2 variability in the model is also mainly explained by the local anthropogenic emissions (Figs. S10 and S11) producing very large CO2 enhancements in the planetary boundary layer (PBL) (Fig. S12) and therefore in XCO2. The CO2 budget of the anthropogenic emissions used at 9 and 80 km is the same. However, the instantaneous values of the emissions per square metre are much higher at 9 than at 80 km, representing some of the steep gradients and heterogeneous distribution of fossil fuel emissions within the LA basin, with higher emissions in downtown LA and lower emissions in Pasadena (e.g. Feng et al.2016). At 80 km, Pasadena and downtown LA are in the same model grid box, which means this gradient cannot be represented. In addition to the influence of anthropogenic emissions, the seasonal variation of the winds is very pronounced in Pasadena, with a large contrast in the origin or air masses between winter and summer .

In winter, air masses originate from various directions: from the prevailing westerly and southerly winds, bringing and accumulating polluted air from the LA megacity, to northerly and easterly flow, characterised by cleaner air with lower CO2 values from the surrounding desert and mountains . Persistent low wind conditions lead to a large accumulation of CO2 in the LA basin as it remains trapped by the mountains. These episodes result in large enhancements in XCO2 and high CO2 anomalies over periods of a few days (e.g. 26 to 30 January in Fig. 15a). In those stagnant conditions, the 9kmEXP simulation is in much closer agreement with the observed XCO2 peaks than the 80kmEXP simulation, which overestimates the XCO2 anomalies. This is because at 80 km resolution there is an effectively uniform emission for the whole LA basin. Note that the CO2 and XCO2 small-scale variability around LA appears to be larger in winter than in summer (Figs. 11 and 14). Without preserving the sharp gradient in emissions between Pasadena and downtown LA, the CO2 accumulation is overestimated in Pasadena.

The atmospheric circulation in summer is mainly controlled by the sea–mountain breeze . Daytime advection of anthropogenic CO2-rich air from the city of LA results in XCO2 peaking in the afternoon before it is vented over the mountains . The overestimation in the summer XCO2 peaks at 9 km likely reflects an overestimation of the emissions in downtown LA. The enhancement of CO2 from anthropogenic emissions is larger at 9 than at 80 km (Fig. S11). This suggests an overestimation of the hotspot emissions over the LA basin in the temporally extrapolated EDGAR inventory, which is smoothed and less noticeable at lower resolution. There are many reasons why the anthropogenic emissions used in the model can be overestimated, including the temporal extrapolation based on country-scale scaling factors and the use of annual constant emissions in EDGAR v4.2 FT2010 instead of seasonally varying emissions .

Differences in the sampling location (centre of grid is 3 and 34 km from station location at 9 and 80 km respectively) and orography (15 m below and 46 m below the station height at 9 and 80 km respectively), as well as differences in flow and local biogenic fluxes can also play a role in explaining the differences between the simulations at 80 and 9 km resolutions. The results are consistent with previous studies by and . They found that uncertainties in the fluxes and their high-resolution representation in the LA basin are as important as the atmospheric tracer transport in the representation of the CO2 enhancement and its variability in Pasadena.

This example at Pasadena highlights the importance of horizontal resolution in representing local gradients of CO2 fluxes in order to reduce the atmospheric CO2 representativeness error. It emphasises that the impact of increasing horizontal resolution is not only to reduce the error of atmospheric CO2 simulations but to enhance the sensitivity of the modelled atmospheric CO2 variability to the CO2 fluxes in urban regions characterised by emission hotspots. Therefore horizontal resolution is crucial for atmospheric inversion systems that aim to estimate anthropogenic emissions.

4 Discussion and conclusions

This paper addresses the importance of horizontal resolution in the representation of CO2 variability at diurnal and synoptic scales, referred to here as CO2 weather. The CO2 simulations performed with the ECMWF IFS model allow the combined impact of horizontal resolution associated with (i) the online modelling of the winds, (ii) the numerical tracer transport model and (iii) the spatial–temporal distribution of CO2 fluxes over land to be quantified. The assessment is done by comparing the model errors at various horizontal resolutions with respect to a wide range of observations with hourly resolution and distributed around the globe. The horizontal resolution of the model ranges from 9 km – as in current operational high-resolution weather and CO2 forecasts at ECMWF – to 80 km, which corresponds to the ERA-Interim Reanalysis resolution, widely used by many offline tracer transport models. The conclusions to the three main questions addressed in the paper are summarised below.

1. What is the sensitivity of the modelled atmospheric CO2 variability at diurnal and synoptic timescales to horizonal resolution?

The high horizontal resolution of 9 km leads to a general improvement in the simulated variability of hourly near-surface and column-averaged atmospheric CO2 compared to the resolution of 80 km. This is shown by a reduction in the mean RMSE of around 1.8 ppm in winter and 3.5 ppm in summer (equivalent to 33 % error reduction) and 0.1 ppm (i.e. around 10 % error reduction) at in situ and TCCON sites respectively, which is associated with a reduction of both the mean and random errors in the model. The inter-station variability is also generally improved in the 9kmEXP simulation for near-surface and column-averaged CO2 in January and July, with the standard deviation of station biases reduced up to 50 % compared to the 80kmEXP simulation in January for near-surface CO2.

Column-averaged CO2 is not as sensitive to horizontal resolution as near-surface CO2 because it has a larger footprint or area of flux influence, except for sites like Pasadena which are close to CO2 emission hotspots. Similarly, minimum daily values of atmospheric CO2 are less sensitive to the horizontal resolution than maximum daily values because their footprint tends to be larger in size.

This study also shows that the RMSE reduction with horizontal resolution is not linear. This implies that results from sensitivity studies exploring the impact of resolution based on coarse simulations which show small sensitivity to horizontal resolution cannot be extrapolated to higher horizontal resolutions. These results are consistent with the findings of the study based on a wider range of model resolutions from ∼100 down to 1 km and observations at three mountain sites. The reduction in model error associated with the increase of horizontal resolution to 9 km emanates from four different well-known and connected aspects, as listed below.

• a.

Better accuracy of the horizontal winds is integral to the reduction of error. The strength of the winds determines the observed CO2 variability – i.e. the detected CO2 enhancement – close to emission hotspots like in urban regions . Therefore, the error in the wind will affect the value of the enhanced CO2 as much as the error in the fluxes. In this context, for example, a wind speed error reduction of 0.5 m s−1 – as shown in Sect. 3.1 – across a gradient of 10 ppm per degree – typical of urban areas as shown in Sect. 3.4 – throughout a 6 h period can result in a CO2 error reduction of around 1 ppm. Uncertainty in the winds has been shown to be one of the largest contributors to the uncertainty in the estimated fluxes over urban areas (e.g. Hedelius et al.2018).

• b

An overall reduction of the numerical error associated with lower spatial and temporal truncation errors leads to a reduction in tracer advection errors .

• c.

There is a general improvement in the horizontal and vertical sampling at the station locations in the model associated with a more realistic representation of orography and coastal boundaries.

• d.

A more realistic representation of CO2 flux distribution at the surface is key. High resolution gives an increased capability to represent small-scale sharp gradients associated with complex topographical boundaries at coastal and mountainous terrain sites, as well as the presence of strong local surface fluxes of CO2 such as anthropogenic emission hotspots.

2. How is the horizontal resolution affecting the forecast error growth of atmospheric CO2?

The horizontal resolution has a consistent positive impact on the error reduction at all forecast lead times, from day 1 to day 10, implying a long-lived improvement in the prediction skill. The RMSE growth is small from days 1 to 4, namely less than 0.5 ppm near the surface CO2 and less than 0.05 ppm for XCO2. Over the 10 d there is an increase in RMSE of 1 to 1.5 ppm at the surface and 0.1 to 0.5 ppm for the total column. This error growth is not linear. For example, in July the error of the 1 d forecast is worse than the 2 d forecast, with a slower error increase during the 2 to 4 d forecast and a generally faster error increase from day 5 to day 10 in the forecast. This incoherent change in the error evolution at the beginning of the forecast is likely linked to the strong influence of the biogenic surface fluxes, which respond very fast to changes in temperature, moisture and radiation forcing in the model. Inconsistencies between the initial conditions from the analysis and the model forecast can cause spin-up adjustments which may lead to a degradation of the 1 d forecast.

Generally, the improvement of forecast skill with increased horizontal resolution is most pronounced in January, when at 9 km resolution the skill of the 10 d forecast is better or equal to the accuracy of the 1 d forecast at 80 km both near the surface and for the column average CO2. It is likely that the skill of the 10 d forecast to represent variability of CO2 during summer conditions is hampered by the growing errors in the surface biogenic fluxes during the forecast, as they can be an important contributor to synoptic variability in the summer .

3. Where and when are the typical representativeness errors associated with unresolved small-scale variability largest?

During daytime, the CO2 small-scale variability of the 9 km resolution forecast ranges from 1 to 10 ppm at the surface and is an order of magnitude smaller (0.1 to 1 ppm) for the total column average. It points to the areas associated with small-scale gradients where horizontal resolution matters: coastal boundaries and mountain regions have typical values of 5 ppm per degree, and CO2 flux hotspots have the highest variability of up to 10 ppm per degree. During night-time, the small-scale variability tends to be larger than 10 ppm over most areas near the surface, whereas that of column-averaged CO2 shows small differences between day and night.

The high horizontal resolution gives us an insight into the areas with high sensitivity to uncertainty associated with both local tracer transport and fluxes. It is in these areas where improvements in the tracer transport and increased understanding of the heterogeneity and complexity of the surface will be crucial in future model developments. Since these areas are close to emission hotspots, it is clear that in order to monitor CO2 emissions, particularly from cities and power stations such as in the new Carbon Human Emission project (http://www.che-project.eu, last access: 30 May 2019), it is paramount to invest in high horizontal resolution models.

Interesting differences are found between surface and column-averaged variability. Near the surface the variability is most pronounced close to emission hotspots and complex terrain. For column-averaged CO2 the sub-grid-scale variability is also substantial over the ocean downstream from emissions. This emphasises the importance of the transport influence on XCO2 variability. Small-scale variability is also found to be more pronounced in summer than in winter, as biogenic CO2 fluxes of opposite sign in summer enhance the CO2 gradients in the atmosphere.

In summary, this paper has shown that model simulations using the CAMS CO2 forecasting system at 9 km resolution can provide a more accurate representation of tracer transport and the local influences of surface fluxes than at lower resolutions ranging from 80 to 16 km, resulting in an overall better representation of the atmospheric CO2 variability at diurnal and synoptic timescales. However, at higher horizontal resolution there is also higher sensitivity of atmospheric CO2 to CO2 flux errors, as emissions and biogenic flux hotspots are not diffused over large areas like in lower-resolution models. Thus, higher-resolution models also risk deterioration in the forecast RMSE, e.g. near emission hotspots associated with larger errors. With the enhancement of the model uncertainty at high resolution, the prospect of further increasing the horizontal resolution needs to be carefully balanced with improvements in the most uncertain model processes.

The impact of horizontal resolution on the accuracy of the winds highlights that errors in the wind need to be considered as an important source of uncertainty both in the atmospheric CO2 analysis and forecast and in the inversion systems . The findings in this study also suggest that increasing horizontal resolution up to kilometric scales in atmospheric data assimilation and inversion systems would allow the use of more in situ and high-resolution satellite observations close to strong sources and sinks and over complex terrain. found that a minimum horizontal resolution of 4 km is required to simulate a realistic diurnal cycle of CO2 at mountain sites.

Currently, the precision of XCO2 from satellite observations is around 1.0 to 1.5 ppm for ACOS-GOSAT data and OCO-2 data . However, if tracer transport models cannot represent their variability accurately in space and time, all the efforts to reduce the errors from the satellite retrievals of CO2 will not be fruitful in their attempt to reduce the uncertainty in the estimation of surface fluxes. This is because relatively small differences in atmospheric mixing ratios are associated with significant differences in surface fluxes . The benefits of high resolution in inversion systems will also need to be balanced with the costs of running a model at such high resolution.

Finally, the CAMS high-resolution forecast running currently at 9 km resolution can provide benchmarks for other simulations using coarser grids or offline meteorology . Both CAMS analysis and high-resolution forecasts are freely available to users (https://atmosphere.copernicus.eu, last access: 30 May 2019). Potential applications include the estimation of representativeness errors and data selection screening of observations from satellites and in situ stations in data assimilation systems, spatial collocation of XCO2 from satellite and TCCON data for validation purposes (e.g. Guerlet et al.2013) or as boundary conditions for high-resolution simulations and/or inversions at regional scales.

Data availability
Data availability.

The data are accessible by contacting the corresponding author (anna.agusti-panareda@ecmwf.int).

Appendix A

Table A1Continuous in situ stations (surface and tower) used to evaluate synoptic variability. NA denotes references that are not available. The full names for the abbreviations of the network organisations are provided in Table A2.

Table A2Organisations associated with observing stations.

Table A3TCCON stations

Supplement
Supplement.

Author contributions
Author contributions.

The simulations were performed by AAP. The coding of the mass fixer required for the high-resolution transport in the IFS was done by MD. The concept and ideas to design the high-resolution simulations were devised by FC, AAP, MD, SM and JMS in discussion with RE and VHP. RML, ZL, JAM and RC provided additional observations at crucial sites and guidance on the evaluation of the simulations. CR and DW provided data and input on the interpretation of the model evaluation at the TCCON site of Pasadena. The CO2 validation tools have been developed by SM and AAP. The paper was prepared by AAP with input and feedback from MD, SM, FC, JMS, JB, RE, BL, RML, ZL, JAM, MP, VHP, MR, CR, ATV, TW and DW.

Competing interests
Competing interests.

The authors declare that they have no conflict of interest.

Acknowledgements
Acknowledgements.

This research was generated using Copernicus Atmosphere Monitoring Service (2018) information. Anna Agustí-Panareda has been partly funded by the CHE project. The CHE project has received funding from the European Union's Horizon 2020 Research and Innovation programme under grant agreement no. 776186. Frédéric Chevallier received funding from the Copernicus Atmosphere Monitoring Service, implemented by the European Centre for Medium-Range Weather Forecasts (ECMWF) on behalf of the European Commission. Thanks are given to all the station principal investigators from the numerous individual stations and networks like NOAA, ICOS, AEMET, AGH, CSIRO, ECCC, ECN, EMPA, FMI, HMS, LSCE, NCAR, JMA, MPI-BGC, NIWA, SAWS, TU, UBA-SCHAU, UEA, UHEI-IUP, UR, UBremen-IUP, BIRA-IASB, Caltech, KIT, NASA, JAXA, NIES and UOW that contributed their CO2 observations to the cooperative GLOBALVIEWplus (ObsPack 2015) product and TCCON which are fundamental for the evaluation of the model simulations. The ClimaDat Network has received funding from the “la Caixa” Foundation, under agreement 2010-002624. We are grateful to many colleagues at ECMWF for their support and fruitful discussions, particularly to Gabor Radnoti, Thomas Haiden and Martin Janouseck for their technical support in the evaluation of the winds; Miha Razinger for his technical support in the production of Fig. 1; Johannes Flemming and Zak Kipling for their support in the implementation of the additional CO2 tracers in the IFS model; Sylvie Malardel for her support and discussions on the general aspects of atmospheric tracer modelling in the IFS; and Gianpaolo Balsamo, Souhail Boussetta, Zak Kipling and Johannes Flemming for their technical support in the implementation of a bug fix in the CTESSEL model of biogenic emissions. Many thanks are given to Paul Wennberg (Caltech) for his advice on the use of the TCCON data; Martin Krol (Wageningen University) for his suggestions on the evaluation of the daily maximum CO2; and Britton Stephens (NCAR) for his comments on the mountain site evaluation which helped improve the description of the vertical sampling strategy and emphasise the importance of high resolution at mountain sites.

Financial support
Financial support.

This research has been supported by the Copernicus Atmospheric Monitoring Service and by the European Commission (CHE project, grant no. 776186).

Review statement
Review statement.

This paper was edited by Christoph Gerbig and reviewed by two anonymous referees.

References

Agustí-Panareda, A., Massart, S., Chevallier, F., Boussetta, S., Balsamo, G., Beljaars, A., Ciais, P., Deutscher, N. M., Engelen, R., Jones, L., Kivi, R., Paris, J.-D., Peuch, V.-H., Sherlock, V., Vermeulen, A. T., Wennberg, P. O., and Wunch, D.: Forecasting global atmospheric CO2, Atmos. Chem. Phys., 14, 11959–11983, https://doi.org/10.5194/acp-14-11959-2014, 2014. a, b, c, d, e, f, g

Agustí-Panareda, A., Massart, S., Chevallier, F., Balsamo, G., Boussetta, S., Dutra, E., and Beljaars, A.: A biogenic CO2 flux adjustment scheme for the mitigation of large-scale biases in global atmospheric CO2 analyses and forecasts, Atmos. Chem. Phys., 16, 10399–10418, https://doi.org/10.5194/acp-16-10399-2016, 2016. a, b

Agustí-Panareda, A., Diamantakis, M., Bayona, V., Klappenbach, F., and Butz, A.: Improving the inter-hemispheric gradient of total column atmospheric CO2 and CH4 in simulations with the ECMWF semi-Lagrangian atmospheric global model, Geosci. Model Dev., 10, 1–18, https://doi.org/10.5194/gmd-10-1-2017, 2017. a

Ahmadov, R., Gerbig, C., Kretschmer, R., Körner, S., Rödenbeck, C., Bousquet, P., and Ramonet, M.: Comparing high resolution WRF-VPRM simulations and two global CO2 transport models with coastal tower measurements of CO2, Biogeosciences, 6, 807–817, https://doi.org/10.5194/bg-6-807-2009, 2009. a

Alkhaled, A. A., Michalak, A. M., and Kawa, S. R.: Using CO2 spatial variability to quantify representation errors of satellite CO2 retrievals, Geophys. Res. Lett., 35, L16813, https://doi.org/10.1029/2008GL034528, 2008. a

Andrews, A. E., Kofler, J. D., Trudeau, M. E., Williams, J. C., Neff, D. H., Masarie, K. A., Chao, D. Y., Kitzis, D. R., Novelli, P. C., Zhao, C. L., Dlugokencky, E. J., Lang, P. M., Crotwell, M. J., Fischer, M. L., Parker, M. J., Lee, J. T., Baumann, D. D., Desai, A. R., Stanier, C. O., De Wekker, S. F. J., Wolfe, D. E., Munger, J. W., and Tans, P. P.: CO2, CO, and CH4 measurements from tall towers in the NOAA Earth System Research Laboratory's Global Greenhouse Gas Reference Network: instrumentation, uncertainty analysis, and recommendations for future high-accuracy greenhouse gas monitoring efforts, Atmos. Meas. Tech., 7, 647–687, https://doi.org/10.5194/amt-7-647-2014, 2014. a, b, c, d, e, f, g

Bechtold, P., Köhler, M., Jung, T., Doblas-Reyes, F., Leutbecher, M., Rodwell, M., Vitart, F., and Balsamo, G.: Advances in simulating atmospheric variability with the ECMWF model: From synoptic to decadal time-scales, Q. J. Roy. Meteor. Soc., 134, 1337–1351, 2008. a

Bechtold, P., Semane, N., Lopez, P., Chaboureau, J.-P., Beljaars, A., and Bormann, N.: Representing equilibrium and nonequilibrium convection in large-scale models, J. Atmos. Sci., 71, 734–753, 2014. a

Belikov, D. A., Maksyutov, S., Ganshin, A., Zhuravlev, R., Deutscher, N. M., Wunch, D., Feist, D. G., Morino, I., Parker, R. J., Strong, K., Yoshida, Y., Bril, A., Oshchepkov, S., Boesch, H., Dubey, M. K., Griffith, D., Hewson, W., Kivi, R., Mendonca, J., Notholt, J., Schneider, M., Sussmann, R., Velazco, V. A., and Aoki, S.: Study of the footprints of short-term variation in XCO2 observed by TCCON sites using NIES and FLEXPART atmospheric transport models, Atmos. Chem. Phys., 17, 143–157, https://doi.org/10.5194/acp-17-143-2017, 2017. a

Balzarolo, M., Boussetta, S., Balsamo, G., Beljaars, A., Maignan, F., Calvet, J.-C., Lafont, S., Barbu, A., Poulter, B., Chevallier, F., Szczypta, C., and Papale, D.: Evaluating the potential of large-scale simulations to predict carbon fluxes of terrestrial ecosystems over a European Eddy Covariance network, Biogeosciences, 11, 2661–2678, https://doi.org/10.5194/bg-11-2661-2014, 2014. a

Beljaars, A. and Viterbo, P.: The role of the boundary layer in a numerical weather prediction model, in Clear and cloudy boundary layers, Royal Netherlands Academy of Arts and Sciences, North Holland Publishers, Amsterdam, 1998. a

Blumenstock, T., Hase, F., Schneider, M., García, O. E., and Sepúlveda, E.: TCCON data from Izana, Tenerife, Spain, Release GGG2014R1, TCCON data archive, hosted by CaltechDATA, California Institute of Technology, Pasadena, CA, USA, https://doi.org/10.14291/tccon.ggg2014.izana01.R1, 2014. a

Boussetta, S., Balsamo, G., Beljaars, A., Agustí-Panareda, A., Calvet, J.-C., Jacobs, C., van den Hurk, B., Viterbo, P., Lafont, S., Dutra, E., Jarlan, L., Balzarolo, M., Papale, D., and van der Werf, G.: Natural carbon dioxide exchanges in the ECMWF Integrated Forecasting System: Implementation and offline validation, J. Geophys. Res.-Atmos., 118, 1–24, https://doi.org/10.1002/jgrd.50488, 2013. a, b

Brooks, B.-G. J., Desai, A. R., Stephens, B. B., Bowling, D. R., Burns, S. P., Watt, A. S., Heck, S. L., and Sweeney, C.: Assessing filtering of mountaintop CO2 mole fractions for application to inverse models of biosphere-atmosphere carbon exchange, Atmos. Chem. Phys., 12, 2099–2115, https://doi.org/10.5194/acp-12-2099-2012, 2012. a

Brunke, E.-G., Labuschagne, C., Parker, B., Scheel, H. E., and Whittlestone, S.: Baseline air mass selection at Cape Point, South Africa: Application of 222Rn and other filter criteria to CO2, Atmos. Environ., 38, 5693–5702, 2004. a

Chevallier, F., Ciais, P., Conway, T. J., Aalto, T., Anderson, B. E., Bousquet, P., Brunke, E. G., Ciattaglia, L., Esaki, Y., Fröhlich, M., Gomez, A., Gomez-Pelaez, A.J., Haszpra, L., Krummel, P. B., Langenfelds, R. L., Leuenberger, M., Machida, T., Maignan, F., Matsueda, H., Morguí, J. A., Mukai, H., Nakazawa, T., Peylin, P., Ramonet, M., Rivier, L., Sawa, Y., Schmidt, M., Steele, L. P., Vay, S. A., Vermeulen, A. T., Wofsy, S., and Worthy, D.: CO2 surface fluxes at grid point scale estimated from a global 21 year reanalysis of atmospheric measurements, J. Geophys. Res., 115, D21307, https://doi.org/10.1029/2010JD013887, 2010. a, b, c

Chevallier, F., Palmer, P. I., Feng, L., Boesch, H., O'Dell, C. W., and Bousquet, P.: Toward robust and consistent regional CO2 fluxestimates from in situ and spaceborne measurements of atmospheric CO2, Geophys. Res. Lett., 41, 1065–1070, https://doi.org/10.1002/2013GL058772, 2014. a

Conway, T. J. and Thoning, K. W.: Short-term variations of atmospheric carbon dioxide at the South Pole, Antarct. J. US, 236–238, 1990. a

Corbin, K., Denning, A. S., Wang, J.-W., Lu, L., and Baker, I. T.: Possible representation errors in inversions of satellite CO2 retrievals, J. Geophys. Res., 113, D02301, https://doi.org/10.1029/2007JD008716, 2008. a, b, c

Corbin, K. D., Denning, A. S., and Gurney, K. R.: The space and time impacts on U.S. regional atmospheric CO2 concentrations from a high resolution fossil fuel CO2 emissions inventory, Tellus B, 62, 506–511, https://doi.org/10.1111/j.1600-0889.2010.00480.x, 2010. a

Crisp, D., Pollock, H. R., Rosenberg, R., Chapsky, L., Lee, R. A. M., Oyafuso, F. A., Frankenberg, C., O'Dell, C. W., Bruegge, C. J., Doran, G. B., Eldering, A., Fisher, B. M., Fu, D., Gunson, M. R., Mandrake, L., Osterman, G. B., Schwandner, F. M., Sun, K., Taylor, T. E., Wennberg, P. O., and Wunch, D.: The on-orbit performance of the Orbiting Carbon Observatory-2 (OCO-2) instrument and its radiometrically calibrated products, Atmos. Meas. Tech., 10, 59–81, https://doi.org/10.5194/amt-10-59-2017, 2017. a

Dee, D. P., Uppala, S. M., Simmons, A. J., Berrisford, P., Poli, P., Kobayashi, S., Andrae, U., Balmaseda, M. A., Balsamo, G., Bauer, P., Bechtold, P., Beljaars, A. C. M., van de Berg, L., Bidlot, J., Bormann, N., Delsol, C., Dragani, R., Fuentes, M., Geer, A. J., Haimberger, L., Healy, S. B., Hersbach, H., Hólm, E. V., Isaksen, L., Kallberg, P., Köhler, M., Matricardi, M., McNally, A. P., Monge-Sanz, B. M., Morcrette, J.-J., Park, B.-K., Peubey, C., de Rosnay, P., Tavolato, C., Thépaut, J.-N., and Vitart, F.: The ERA-Interim re-analysis: Configuration and performance of the data assimilation system, Q. J. Roy. Meteor. Soc., 137, 553-–597, https://doi.org/10.1002/qj.828, 2011. a

De Mazière, M., Sha, M. K., Desmet, F., Hermans, C., Scolas, F., Kumps, N., Metzger, J.-M., Duflot, V., and Cammas, J.-P.: TCCON data from Reunion Island (La Reunion), France, Release GGG2014R0, TCCON data archive, hosted by CaltechDATA, California Institute of Technology, Pasadena, CA, USA, https://doi.org/10.14291/tccon.ggg2014.reunion01.R0/1149288, 2014. a

Deutscher, N., Notholt, J., Messerschmidt, J., Weinzierl, C., Warneke, T., Petri, C., Grupe, P., and Katrynski, K.: TCCON data from Bialystok, Poland, Release GGG2014R1, TCCON data archive, hosted by CaltechDATA, California Institute of Technology, Pasadena, CA, USA, https://doi.org/10.14291/tccon.ggg2014.bialystok01.R1/1183984, 2014. a

Diamantakis, M. and Agustí-Panareda, A.: A positive definite tracer mass fixer for high resolution weather and atmospheric composition forecasts, ECMWF Technical Memorandum, No. 819, available at: https://www.ecmwf.int/en/elibrary/17914-positive-definite-tracer-mass-fixer-high-resolution-weather-and-atmospheric (last access: 30 May 2019), 2017. a

Diamantakis, M. and Magnusson, L.: Sensitivity of the ECMWF model to semi-Lagrangian departure point iterations, Mon. Weather Rev., 144, 3233–3250, https://doi.org/10.1175/MWR-D-15-0432.1, 2016. a

Feist, D. G., Arnold, S. G., John, N., Geibel, M. C., and Christoph, M.: TCCON data from Ascension Island (SH), Release GGG2014R0, TCCON data archive, hosted by CaltechDATA, California Institute of Technology, Pasadena, CA, USA, https://doi.org/10.14291/tccon.ggg2014.ascension01.R0/1149285, 2014. a

Feng, S., Lauvaux, T., Newman, S., Rao, P., Ahmadov, R., Deng, A., Díaz-Isaac, L. I., Duren, R. M., Fischer, M. L., Gerbig, C., Gurney, K. R., Huang, J., Jeong, S., Li, Z., Miller, C. E., O'Keeffe, D., Patarasuk, R., Sander, S. P., Song, Y., Wong, K. W., and Yung, Y. L.: Los Angeles megacity: a high-resolution land–atmosphere modelling system for urban CO2 emissions, Atmos. Chem. Phys., 16, 9019–9045, https://doi.org/10.5194/acp-16-9019-2016, 2016. a, b

Flemming, J., Huijnen, V., Arteta, J., Bechtold, P., Beljaars, A., Blechschmidt, A.-M., Diamantakis, M., Engelen, R. J., Gaudel, A., Inness, A., Jones, L., Josse, B., Katragkou, E., Marecal, V., Peuch, V.-H., Richter, A., Schultz, M. G., Stein, O., and Tsikerdekis, A.: Tropospheric chemistry in the Integrated Forecasting System of ECMWF, Geosci. Model Dev., 8, 975–1003, https://doi.org/10.5194/gmd-8-975-2015, 2015. a

Francey, R. J., Steele, L. P., Spencer, D. A., Langenfelds, R. L., Law, R. M., Krummel, P. B., Fraser, P. J., Etheridge, D. M., Derek, N., Coram, S. A., Cooper, L. N., Allison, C. E., Porter, L., and Baly, S.: The CSIRO (Australia) measurement of greenhouse gases in the global atmosphere, report of the 11th WMO/IAEA Meeting of Experts on Carbon Dioxide Concentration and Related Tracer Measurement Techniques, Tokyo, Japan, September 2001, edited by: Toru, S. and Kazuto, S., World Meteorological Organization Global Atmosphere Watch, 97–111, 2003. a

Gaudry, A., Monfray, P., Polian, G., Bonsang, G., Ardouin, B., Jegou, A., and Lambert, G.: Non-seasonnal variations of atmospheric CO2 concentrations at Amsterdam Island, Tellus B, 43, 136–143, 1991. a

Geels, C., Doney, S. C., Dargaville, R., Brandt, J., and Christensen, J. H.: Investigating the sources of synoptic variability in atmospheric CO2 measurements over the Northern Hemisphere continents: a regional model study, Tellus B, 56, 35–50, https://doi.org/10.1111/j.1600-0889.2004.00084.x, 2004. a, b

Gomez-Pelaez, A. J. and Ramos, R.: Improvements in the Carbon Dioxide and Methane Continuous Measurement Programs at Izana Global GAW Station (Spain) during 2007–2009, in: GAW report (No. 194) of the 15th WMO/IAEA Meeting of Experts on Carbon Dioxide, Other Greenhouse Gases, and Related Tracer Measurement Techniques (Jena, Germany; 7–10 September 2009), edited by: Brand, W. A., World Meteorological Organization (TD No. 1553), 133–138, 2011. a

Griffith, D. W. T., Deutscher, N., Velazco, V. A., Wennberg, P. O., Yavin, Y., Keppel Aleks, G., Washenfelder, R., Toon, G. C., Blavier, J.-F., Murphy, C., Jones, N., Kettlewell, G., Connor, B., Macatangay, R., Roehl, C., Ryczek, M., Glowacki, J., Culgan, T., and Bryant, G.: TCCON data from Darwin, Australia, Release GGG2014R0, TCCON data archive, hosted by CaltechDATA, California Institute of Technology, Pasadena, CA, USA, https://doi.org/10.14291/tccon.ggg2014.darwin01.R0/1149290, 2014. a

Griffith, D. W. T., Velazco, V. A., Deutscher, N., Murphy, C., Jones, N., Wilson, S., Macatangay, R., Kettlewell, G., Buchholz, R. R., and Riggenbach, M.: TCCON data from Wollongong, Australia, Release GGG2014R0, TCCON data archive, hosted by CaltechDATA, California Institute of Technology, Pasadena, CA, USA, https://doi.org/10.14291/tccon.ggg2014.wollongong01.R0/1149291, 2014. a

Guerlet, S., Butz, A., Schepers, D., Basu, S., Hasekamp, O.P., Kuze, A., Yokota, T., Blavier, J.-F., Deutscher, N. M., Griffith,  D. W. T., Hase,  F., Kyro, E., Morino, I., Sherlock, V., Sussmann, R., Galli, A., and Aben, I.: Impact of aerosol and thin cirrus on retrieving and validating XCO2 from GOSAT shortwave infrared measurements, J. Geophys. Res.-Atmos., 118, 4887–4905, https://doi.org/10.1002/jgrd.50332, 2013. a

Gurney, K. R., Law, R. M., Denning, A. S., Rayner, P. J., Baker, D., Bousquet, P., Bruhwiler, L., Chen, Y.-H., Ciais, P., Fan, S., Fung, I. Y., Gloor, M., Heimann, M., Higuchi, K., John, J., Kowalczyk, E., Maki, T., Maksyutov, S., Peylin, P., Prather, M., Pak, B. C., Sarmiento, J., Taguchi, S., Takahashi, T., and Yuen, C.-W.: TransCom 3 inversion intercomparison: 1. Annual mean control results and sensitivity to transport and prior flux information, Tellus B, 55, 555–579, 2003. a

Haiden, T., Janousek, M., Bidlot, J.-R., Ferranti, L., Prates, F., Vitart, F., Bauer, P., and Richardson, D.: Evaluation of ECMWF forecasts, including 2016–2017 upgrades, ECMWF Technical Memorandum, No. 817, ECMWF, available at: https://www.ecmwf.int/en/elibrary/17913-evaluation-ecmwf-forecasts-includuing-2016-2017-upgrades (last access: 30 May 2019), 2017. a, b

Halter, B. and Harris, J. M.: On the variability of atmospheric carbon dioxide concentration at Barrow, Alaska, during winter, J. Geophys. Res., 88, 6858–6864, 1983. a

Halter, B. C., Harris, J. M., and Conway, T. J.: Component signals in the record of atmospheric carbon dioxide concentration at American Samoa, J. Geophys. Res., 93, 15914–15918, 1988. a

Hase, F., Blumenstock, T., Dohe, S., Groß, J., and Kiel, M.: TCCON data from Karlsruhe, Germany, Release GGG2014R1, TCCON data archive, hosted by CaltechDATA, California Institute of Technology, Pasadena, CA, USA, https://doi.org/10.14291/tccon.ggg2014.karlsruhe01.R1/1182416, 2014. a

Haszpra, L., Barcza, Z., Bakwin, P. S., Berger, B. W., Davis, K. J., and Weidinger, T.: Measuring system for the long-term monitoring of biosphere/atmosphere exchange of carbon dioxide, J. Geophys. Res., 106, 3057–3070, 2001. a

Hatakka, J., Aalto, T., Aaltonen, V., Aurela, M., Hakola, H., Komppula, M., Laurila, T., Lihavainen, H., Paatero, J., Salminen, K., and Viisanen, Y.: Overview of the atmospheric research activities and results at Pallas GAW station, Boreal Environ. Res., 8, 365–384, 2003. a

Hedelius, J. K., Feng, S., Roehl, C. M., Wunch, D., Hillyard, P. W., Podolske, J. R., Iraci, L. T., Patarasuk, R., Rao, P., O'Keeffe, D., Gurney, K. R., Lauvaux, T., and Wennberg, P. O.: Emissions and topographic effects on column CO2 (XCO2) variations, with a focus on the Southern California Megacity, J. Geophys. Res.-Atmos., 122, 7200–7215, https://doi.org/10.1002/2017JD026455, 2017. a, b

Hedelius, J. K., Liu, J., Oda, T., Maksyutov, S., Roehl, C. M., Iraci, L. T., Podolske, J. R., Hillyard, P. W., Liang, J., Gurney, K. R., Wunch, D., and Wennberg, P. O.: Southern California megacity CO2, CH4, and CO flux estimates using ground- and space-based remote sensing and a Lagrangian model, Atmos. Chem. Phys., 18, 16271–16291, https://doi.org/10.5194/acp-18-16271-2018, 2018. a, b

Holm, E., Forbes, R., Lang, S., Magnusson, L., and Malardel, S.: New model cycle brings higher resolution, ECMWF Newsletter, No. 147, available at: https://www.ecmwf.int/en/elibrary/16299-newsletter-no-147-spring-2016 (last access: 30 May 2019), 2016. a, b

Houweling, S., Aben, I., Breon, F.-M., Chevallier, F., Deutscher, N., Engelen, R., Gerbig, C., Griffith, D., Hungershoefer, K., Macatangay, R., Marshall, J., Notholt, J., Peters, W., and Serrar, S.: The importance of transport model uncertainties for the estimation of CO2 sources and sinks using satellite measurements, Atmos. Chem. Phys., 10, 9981–9992, https://doi.org/10.5194/acp-10-9981-2010, 2010. a

Inness, A., Ades, M., Agustí-Panareda, A., Barré, J., Benedictow, A., Blechschmidt, A.-M., Dominguez, J. J., Engelen, R., Eskes, H., Flemming, J., Huijnen, V., Jones, L., Kipling, Z., Massart, S., Parrington, M., Peuch, V.-H., Razinger, M., Remy, S., Schulz, M., and Suttie, M.: The CAMS reanalysis of atmospheric composition, Atmos. Chem. Phys., 19, 3515–3556, https://doi.org/10.5194/acp-19-3515-2019, 2019. a

Inoue, M., Morino, I., Uchino, O., Nakatsuru, T., Yoshida, Y., Yokota, T., Wunch, D., Wennberg, P. O., Roehl, C. M., Griffith, D. W. T., Velazco, V. A., Deutscher, N. M., Warneke, T., Notholt, J., Robinson, J., Sherlock, V., Hase, F., Blumenstock, T., Rettinger, M., Sussmann, R., Kyrö, E., Kivi, R., Shiomi, K., Kawakami, S., De Mazière, M., Arnold, S. G., Feist, D. G., Barrow, E. A., Barney, J., Dubey, M., Schneider, M., Iraci, L. T., Podolske, J. R., Hillyard, P. W., Machida, T., Sawa, Y., Tsuboi, K., Matsueda, H., Sweeney, C., Tans, P. P., Andrews, A. E., Biraud, S. C., Fukuyama, Y., Pittman, J. V., Kort, E. A., and Tanaka, T.: Bias corrections of GOSAT SWIR XCO2 and XCH4 with TCCON data and their evaluation using aircraft measurement data, Atmos. Meas. Tech., 9, 3491–3512, https://doi.org/10.5194/amt-9-3491-2016, 2016. a

Iraci, L. T., Podolske, J., Hillyard, P. W., Roehl, C., Wennberg, P. O., Blavier, J.-F., Allen, N., Wunch, D., Osterman, G. B., and Albertson, R., TCCON data from Edwards (US), Release GGG2014R1, TCCON data archive, hosted by CaltechDATA, California Institute of Technology, Pasadena, CA, USA, https://doi.org/10.14291/tccon.ggg2014.edwards01.R1/1255068, 2016. a

Kaiser, J. W., Heil, A., Andreae, M. O., Benedetti, A., Chubarova, N., Jones, L., Morcrette, J.-J., Razinger, M., Schultz, M. G., Suttie, M., and van der Werf, G. R.: Biomass burning emissions estimated with a global fire assimilation system based on observed fire radiative power, Biogeosciences, 9, 527–554, https://doi.org/10.5194/bg-9-527-2012, 2012. a

Keeling, C. D., Bacastow, R. B., Bainbridge, A. E., Ekdahl, C. A., Guenther, P. R., Waterman, L. S., and Chin, J. F.: Atmospheric carbon dioxide variations at Mauna Loa Observatory, Hawaii, Tellus, 28, 538–551, https://doi.org/10.1111/j.2153-3490.1976.tb00701.x, 1976. a

Keppel-Aleks, G., Wennberg, P. O., and Schneider, T.: Sources of variations in total column carbon dioxide, Atmos. Chem. Phys., 11, 3581–3593, https://doi.org/10.5194/acp-11-3581-2011, 2011. a

Keppel-Aleks, G., Wennberg, P. O., Washenfelder, R. A., Wunch, D., Schneider, T., Toon, G. C., Andres, R. J., Blavier, J.-F., Connor, B., Davis, K. J., Desai, A. R., Messerschmidt, J., Notholt, J., Roehl, C. M., Sherlock, V., Stephens, B. B., Vay, S. A., and Wofsy, S. C.: The imprint of surface fluxes and transport on variations in total column carbon dioxide, Biogeosciences, 9, 875–891, https://doi.org/10.5194/bg-9-875-2012, 2012. a

Koehler, M., Ahlgrimm, M., and Beljaars, A.: Unified treatment of dry convective and stratocumulus-topped boundary layers in the ecmwf model, Q. J. Roy. Meteor. Soc., 137, 43–57, 2011. a

Kuze, A., Suto, H., Nakajima, M., and Hamazaki, T.: Thermal and near infrared sensor for carbon observation Fourier-transform spectrometer on the Greenhouse Gases Observing Satellite for greenhouse gases monitoring, Appl. Optics, 48, 6716–6733, 2009. a

Law, R. M., Peters, W., Rödenbeck, C., Aulagnier, C., Baker, I., Bergmann, D. J., Bousquet, P., Brandt, J., Bruhwiler, L., Cameron-Smith, P. J., Christensen, J. H., Delage, F., Denning, A. S., Fan, S., Geels, C., Houweling, S., Imasu, R., Karstens, U., Kawa, S. R., Kleist, J., Krol, M. C., Lin, S.-J., Lokupitiya, R., Maki, T., Maksyutov, S., Niwa, Y., Onishi, R., Parazoo, N., Patra, P. K., Pieterse, G., Rivier, L., Satoh, M., Serrar, S., Taguchi, S., Takigawa, M., Vautard, R., Vermeulen, A. T., and Zhu, Z.: TransCom model simulations of hourly atmospheric CO2: Experimental overview and diurnal cycle results for 2002, Global Biogeochem. Cy., 22, GB3009, https://doi.org/10.1029/2007GB003050, 2008. a

Law, R. M., Steele, L. P., Drummel, P. B., and Zahorowski, W.: Synoptic variations in atmospheric CO2 at Cape Grim: a model intercomparison, Tellus B, 62, 810–820, 2010. a, b

Lin, J. C., Mallia, D. V., Wu, D., and Stephens, B. B.: How can mountaintop CO2 observations be used to constrain regional carbon fluxes?, Atmos. Chem. Phys., 17, 5561–5581, https://doi.org/10.5194/acp-17-5561-2017, 2017. a, b, c, d

Lin, X., Ciais, P., Bousquet, P., Ramonet, M., Yin, Y., Balkanski, Y., Cozic, A., Delmotte, M., Evangeliou, N., Indira, N. K., Locatelli, R., Peng, S., Piao, S., Saunois, M., Swathi, P. S., Wang, R., Yver-Kwok, C., Tiwari, Y. K., and Zhou, L.: Simulating CH4 and CO2 over South and East Asia using the zoomed chemistry transport model LMDz-INCA, Atmos. Chem. Phys., 18, 9475–9497, https://doi.org/10.5194/acp-18-9475-2018, 2018. a

Lin, Y.-L.: Mesoscale Dynamics, Cambridge University Press, Cambridge, https://doi.org/10.1017/CBO9780511619649, 2007. a

Locatelli, R., Bousquet, P., Chevallier, F., Fortems-Cheney, A., Szopa, S., Saunois, M., Agusti-Panareda, A., Bergmann, D., Bian, H., Cameron-Smith, P., Chipperfield, M. P., Gloor, E., Houweling, S., Kawa, S. R., Krol, M., Patra, P. K., Prinn, R. G., Rigby, M., Saito, R., and Wilson, C.: Impact of transport model errors on the global and regional methane emissions estimated by inverse modelling, Atmos. Chem. Phys., 13, 9917–9937, https://doi.org/10.5194/acp-13-9917-2013, 2013. a

Loh, Z. M., Law, R. M., Ziehn, T., van der Schoot, M. V., Krummel, P. B., Steele, L. P., Etheridge, D. M., Spencer, D. A., Gregory, R. L., Langenfelds, R. L., Stavert, A. R., and Thornton, D. P.: The Australian Greenhouse Gas Observation Network: Current status and vision for the future. 10th International Carbon Dioxide Conference (ICDC10), 21–25 August 2017, Interlaken, Switzerland, available at: http://www.icdc10.unibe.ch/unibe/portal/fak_naturwis/micro_icdc10/content/e342182/e604227/e604229/files623284/Loh_Zoe.pdf (last access: 30 May 2019), 2017. a

Lopez, M., Schmidt, M., Ramonet, M., Bonne, J.-L., Colomb, A., Kazan, V., Laj, P., and Pichon, J.-M.: Three years of semicontinuous greenhouse gas measurements at the Puy de Dôme station (central France), Atmos. Meas. Tech., 8, 3941–3958, https://doi.org/10.5194/amt-8-3941-2015, 2015. a

Lu, R. and Turco, R. P.: Air pollutant transport in a coastal environment. Part I: Two dimensional simulations of sea-breeze and mountain effects, J. Atmos. Sci., 51, 2285–2308, 1994. a

Malardel, S., Wedi, N., Deconinck, W., Diamantakis, M., Kühnlein, C., Mozdzynski, G., Hamrud, M., and Smolarkiewicz, P.: A new grid for the IFS, ECMWF Newsletter No. 146, available at: https://www.ecmwf.int/sites/default/files/elibrary/2016/15041-newsletter-no-146-winter-201516.pdf (last access: 30 May 2019), 2016. a

Massart, S., Agustí-Panareda, A., Heymann, J., Buchwitz, M., Chevallier, F., Reuter, M., Hilker, M., Burrows, J. P., Deutscher, N. M., Feist, D. G., Hase, F., Sussmann, R., Desmet, F., Dubey, M. K., Griffith, D. W. T., Kivi, R., Petri, C., Schneider, M., and Velazco, V. A.: Ability of the 4-D-Var analysis of the GOSAT BESD XCO2 retrievals to characterize atmospheric CO2 at large and synoptic scales, Atmos. Chem. Phys., 16, 1653–1671, https://doi.org/10.5194/acp-16-1653-2016, 2016. a, b, c, d

Miller, C. E., Crisp, D., DeCola, P. L., Olsen, S. C., Randerson, J. T., Michalak, A. M., Alkhaled, A., Rayner, P., Jacob, D. J., Suntharalingam, P., Jones, D. B. A., Denning, A. S., Nicholls, M. E., Doney, S. C., Pawson, S., Boesch, H., Connor, B. J., Fung, I. Y., O'Brien, D., Salawitch, R. J., Sander, S. P., Sen, B., Tans, P., Toon, G. C., Wennberg, P. O., Wofsy, S. C., Yung, Y. L., and Law, R. M.: Precision requirements for space-based XCO2 data, J. Geophys. Res., 112, D10314, https://doi.org/10.1029/2006JD007659, 2007. a

Morcrette, J.-J., Boucher, O., Jones, L., Salmond, D., Bechtold, P., Beljaars, A., Benedetti, A., Bonet, A., Kaiser, J. W., Razinger, M., Schulz, M., Serrar, S., Simmons, A. J., Sofiev, M., Suttie, M., Tompkins, A. M., and Untch, A.: Aerosol analysis and forecast in the European Centre for Medium-Range Weather Forecasts Integrated Forecast System: forward modeling, J. Geophys. Res., 114, D06206, https://doi.org/10.1029/2008JD011235, 2009. a

Morguí, J. A., Àgueda, A., Batet, O., Curcoll, R., Ealo, M., Grossi, C., Occhipinti, P., Sanchez-Garcia, L., Arias, R., and Rodo, X.: ClimaDat: A long-term network to study at different scales climatic processes and interactions between climatic compartments, Geophys. Res. Abstr., EGU13-10265, EGU General Assembly 2013, Vienna, Austria, 2013. a, b, c, d, e

Morguí, J. A., Curcoll, R., and Agustí-Panareda, A.: Using the CLIMADAT atmospheric CO2 dataset to evaluate (CAMS) the performance of a transport model over the Iberian Peninsula, 10th International Carbon Dioxide Conference (ICDC10) Theme 1.2: The contemporary carbon cycle – Emerging approaches and novel observation techniques, 21–25 August 2017, Interlaken, Switzerland, available at: http://www.icdc10.unibe.ch/unibe/portal/fak_naturwis/micro_icdc10/content/e342182/e588294/e591299/ICDC10_Abstracts_14082017.pdf (last access: 30 May 2019), 2017. a

Morino, I., Yokozeki, Y., Matzuzaki, T., and Shishime, A.: TCCON data from Rikubetsu, Hokkaido, Japan, Release GGG2014R1, TCCON data archive, hosted by CaltechDATA, California Institute of Technology, Pasadena, CA, USA, https://doi.org/10.14291/tccon.ggg2014.rikubetsu01.R1/1242265, 2017a. a

Morino, I., Matsuzaki, T., and Shishime, A.: TCCON data from Tsukuba, Ibaraki, Japan, 125HR, Release GGG2014R1, TCCON data archive, hosted by CaltechDATA, California Institute of Technology, Pasadena, CA, USA, https://doi.org/10.14291/tccon.ggg2014.tsukuba02.R1/1241486, 2014b. a

Necki, J., Schmidt, M., Rozanski, K., Zimnoch, M., Korus, A., Lasa, J., Graul, R., and Levin, I.: Six-year record of atmospheric carbon dioxide and methane at a high-altitude mountain site in Poland, Tellus B, 55, 94–104, 2003. a

Newman, S., Jeong, S., Fischer, M. L., Xu, X., Haman, C. L., Lefer, B., Alvarez, S., Rappenglueck, B., Kort, E. A., Andrews, A. E., Peischl, J., Gurney, K. R., Miller, C. E., and Yung, Y. L.: Diurnal tracking of anthropogenic CO2 emissions in the Los Angeles basin megacity during spring 2010, Atmos. Chem. Phys., 13, 4359–4372, https://doi.org/10.5194/acp-13-4359-2013, 2013. a, b

Newman, S., Xu, X., Gurney, K. R., Hsu, Y. K., Li, K. F., Jiang, X., Keeling, R., Feng, S., O'Keefe, D., Patarasuk, R., Wong, K. W., Rao, P., Fischer, M. L., and Yung, Y. L.: Toward consistency between trends in bottom-up CO2 emissions and top-down atmospheric measurements in the Los Angeles megacity, Atmos. Chem. Phys., 16, 3843–3863, https://doi.org/10.5194/acp-16-3843-2016, 2016. a, b, c

Notholt, J., Petri, C., Warneke, T., Deutscher, N., Buschmann, M., Weinzierl, C., Macatangay, R., and Grupe,P.: TCCON data from Bremen, Germany, Release GGG2014R0, TCCON data archive, hosted by CaltechDATA, California Institute of Technology, Pasadena, CA, USA, https://doi.org/10.14291/tccon.ggg2014.bremen01.R0/1149275, 2014a. a

ObsPack: Cooperative Global Atmospheric Data Integration Project, Multi-laboratory compilation of atmospheric carbon dioxide data for the period 1968–2014, obspack_co2_1_GLOBALVIEWplus_v1.0_2015-07-30, NOAA Earth System Research Laboratory, Global Monitoring Division, https://doi.org/10.15138/G3RP42, 2015. a, b, c, d

O'Dell, C. W., Connor, B., Bösch, H., O'Brien, D., Frankenberg, C., Castano, R., Christi, M., Eldering, D., Fisher, B., Gunson, M., McDuffie, J., Miller, C. E., Natraj, V., Oyafuso, F., Polonsky, I., Smyth, M., Taylor, T., Toon, G. C., Wennberg, P. O., and Wunch, D.: The ACOS CO2 retrieval algorithm – Part 1: Description and validation against synthetic observations, Atmos. Meas. Tech., 5, 99–121, https://doi.org/10.5194/amt-5-99-2012, 2012. a

Olivier, J. and Janssens-Maenhout, G.: CO2 Emissions from Fuel Combustion – 2012 Edition, IEA CO2 report 2012, Part III, Greenhouse-Gas Emissions, ISBN 978-92-64-17475-7, available at: https://edgar.jrc.ec.europa.eu/overview.php?v=CO2ts1990-2014 (last access: 30 May 2019), 2015. a

Ott, L. E., Pawson, S., Collatz, G. J., Gregg, W. W., Menemenlis, D., Brix, H., Rousseaux, C. S., Bowman, K. W., Liu, J., Eldering, A., Gunson, M. R., and Kawa, S. R.: Assessing the magnitude of CO2 flux uncertainty in atmospheric CO2 records using products from NASA's Carbon Monitoring Flux Pilot Project, J. Geophys. Res.-Atmos., 120, 734–765, https://doi.org/10.1002/2014JD022411, 2015. a

Parazoo, N. C., Denning, A. S., Berry, J. A., Wolf, A., Randall, D. A., Kawa, S. R., Pauluis, S. R., and Doney, S. C.: Moist synoptic transport of CO2 along the mid-latitude storm track, Geophys. Res. Lett., 38, L09804, https://doi.org/10.1029/2011GL047238, 2011. a

Patra, P. K., Law, R. M., Peters, W., Rödenbeck, C., Takigawa, M., Aulagnier, C., Baker, I., Bergmann, D. J., Bousquet, P., Brandt, J., Bruhwiler, L., Cameron-Smith, P. J., Christensen, J. H., Delage, F., Denning, A. S., Fan, S., Geels, C., Houweling, S., Imasu, R., Karstens, U., Kawa, S. R., Kleist, J., Krol, M. C., Lin, S.-J., Lokupitiya, R., Maki, T., Maksyutov, S., Niwa, Y., Onishi, R., Parazoo, N., Pieterse, G., Rivier, L., Satoh, M., Serrar, S., Taguchi, S., Vautard, R., Vermeulen, A. T., and Zhu, Z.: Transcom model simulations of hourly atmospheric CO2: analysis of synoptic scale variations for the period 2002–2003, Global Biogeochem. Cy., 22, GB4013, https://doi.org/10.1029/2007GB003081, 2008. a, b, c, d

Peterson, J. T., Komhyr, W. D., Waterman, L. S., Gammon, R. H., Thoning, K. W., and Conway, T. J.: Atmospheric CO2 variations at Barrow, Alaska, 1973–1982, J. Atmos. Chem., 4, 491–510, 1986. a

Pillai, D., Gerbig, C., Marshall, J., Ahmadov, R., Kretschmer, R., Koch, T., and Karstens, U.: High resolution modeling of CO2 over Europe: implications for representation errors of satellite retrievals, Atmos. Chem. Phys., 10, 83–94, https://doi.org/10.5194/acp-10-83-2010, 2010. a

Pillai, D., Gerbig, C., Ahmadov, R., Rödenbeck, C., Kretschmer, R., Koch, T., Thompson, R., Neininger, B., and Lavrié, J. V.: High-resolution simulations of atmospheric CO2 over complex terrain – representing the Ochsenkopf mountain tall tower, Atmos. Chem. Phys., 11, 7445–7464, https://doi.org/10.5194/acp-11-7445-2011, 2011. a

Polavarapu, S. M., Neish, M., Tanguay, M., Girard, C., de Grandpré, J., Semeniuk, K., Gravel, S., Ren, S., Roche, S., Chan, D., and Strong, K.: Greenhouse gas simulations with a coupled meteorological and transport model: the predictability of CO2, Atmos. Chem. Phys., 16, 12005–12038, https://doi.org/10.5194/acp-16-12005-2016, 2016. a, b

Prather, M. J., Zhu, X., Strahan, S. E., Steenrod, S. D., and Rodriguez, S. M.: Quantifying errors in trace species transport modeling, P. Natl. Acad. Sci. USA, 105, 19617–19621, https://doi.org/10.1073/pnas.0806541106, 2008. a, b

Rabier, F., Järvinen, H., Klinker, E., Mahfouf, J.-F., and Simmons, A.: The ECMWF operational implementation of four-dimensional variational assimilation. part I: Experimental results with simplified physics, Q. J. Roy. Meteor. Soc., 126, 1143–1170, 2000. a

Ramonet, M., Ciais, P., Aalto, T., Aulagnier, C., Chevallier, F., Cipriano, D., Conway, T. J., Haszpra, L., Kazan, V., Meinhardt, F., Paris, J.-D., Schmidt, M., Simmonds,  P., Xueref-Remy, I., and Necki, J.: A recent build-up of atmospheric CO2 over Europe. Part 1: observed signals and possible explanations, Tellus B, 62, 1–13, https://doi.org/10.1111/j.1600-0889.2009.00442.x, 2010. a

Rayner, P. J., Scholze, M., Knorr, W., Kaminski, T., Giering, R., and Widmann, H.: Two decades of terrestrial carbon fluxes from a carbon cycle data assimilation system (CCDAS), Global Biogeochem. Cy., 19, GB2026, https://doi.org/10.1029/2004GB002254, 2005. a

Remaud, M., Chevallier, F., Cozic, A., Lin, X., and Bousquet, P.: On the impact of recent developments of the LMDz atmospheric general circulation model on the simulation of CO2 transport, Geosci. Model Dev., 11, 4489–4513, https://doi.org/10.5194/gmd-11-4489-2018, 2018. a

Rozanski, K., Necki, J., Chmura, L., Sliwka, I., Zimnoch, M., Bielewski, J., Galkowski, M.,Bartyzel J., and Rosiek, J.: Anthropogenic changes of CO2, CH4, N2O, CFCl3, CF2Cl2, CCl2FCClF2, CHCl3, CH3CCl3, CCl4, SF6 and SF5CF3 mixing ratios in the atmosphere over southern Poland, Geol. Q., 58, 673–684, 2014. a

Sandu, I., Beljaars, A., Bechtold, P., Mauritsen, T., and Balsamo, G.: Why is it so difficult to represent stably stratified conditions in numerical weather prediction (NWP) models?, J. Adv. Model. Earth Sy., 5, 1–17, https://doi.org/10.1002/jame.20013, 2013. a

Schibig, M. F., Steinbacher, M., Buchmann, B., van der Laan-Luijkx, I. T., van der Laan, S., Ranjan, S., and Leuenberger, M. C.: Comparison of continuous in situ CO2 observations at Jungfraujoch using two different measurement techniques, Atmos. Meas. Tech., 8, 57–68, https://doi.org/10.5194/amt-8-57-2015, 2015. a

Schmidt, M., Graul, R., Sartorius, H., and Levin, L.: The Schauinsland CO2 record: 30 years of continental observations and their implications for the variability of the European CO2 budget, J. Geophys. Res.-Atmos., 108, 4619–4626, 2003. a

Sekiyama, T., Kunii, M., Kajino, M., and Shimbori, T.: Horizontal Resolution Dependence of Atmospheric Simulations of the Fukushima Nuclear Accident Using 15-km, 3-km, and 500-m Grid Models, J. Meteorol. Soc. Jpn. Ser. II, 93, 49–64, https://doi.org/10.2151/jmsj.2015-002, 2015. a

Sherlock, V., Connor, B., Robinson, J., Shiona, H., Smale, D., and Pollard, D.: TCCON data from Lauder, New Zealand, 120HR, Release GGG2014R0, TCCON data archive, hosted by CaltechDATA, California Institute of Technology, Pasadena, CA, USA, https://doi.org/10.14291/tccon.ggg2014.lauder01.R0/1149293, 2014. a

Shiomi, K., Kawakami, S., Ohyama, H., Arai, K., Okumura, H., Taura, C., Fukamachi, T., and Sakashita, M.: TCCON data from Saga, Japan, Release GGG2014R0, TCCON data archive, hosted by CaltechDATA, California Institute of Technology, Pasadena, CA, USA, https://doi.org/10.14291/tccon.ggg2014.saga01.R0/1149283, 2014. a

Staniforth, A. and Côté, J.: Semi-Lagrangian integration schemes for atmospheric models – a review, Mon. Weather Rev., 119, 2206–2223, https://doi.org/10.1175/1520-0493(1991)119<2206:SLISFA>2.0.CO;2, 1991. a

Stavert, A. R., Law, R. M., van der Schoot, M., Langenfelds, R. L., Spencer, D. A., Krummel, P. B., Chambers, S. D., Williams, A. G., Werczynski, S., Francey, R. J., and Howden, R. T.: The Macquarie Island (LoFlo2G) high-precision continuous atmospheric carbon dioxide record, Atmos. Meas. Tech., 12, 1103–1121, https://doi.org/10.5194/amt-12-1103-2019, 2019. a

Stephens, B. B., Miles, N. L., Richardson, S. J., Watt, A. S., and Davis, K. J.: Atmospheric CO2 monitoring with single-cell NDIR-based analyzers, Atmos. Meas. Tech., 4, 2737–2748, https://doi.org/10.5194/amt-4-2737-2011, 2011. a, b, c

Sussmann, R. and Rettinger, M.: TCCON data from Garmisch, Germany, Release GGG2014R0, TCCON data archive, hosted by CaltechDATA, California Institute of Technology, Pasadena, CA, USA, https://doi.org/10.14291/tccon.ggg2014.garmisch01.R0/1149299, 2014. a

Takahashi, T., Sutherland, S. C., Wanninkhof, R., Sweeney, C., Feely, R. A., Chipman, D. W., Hales, B., Friederich, G., Chavez, F., Sabine, C., Watson, A., Bakker, D. C. E., Schuster, U., Metzl, N., Yoshikawa-Inoue, H., Ishii, M., Midorikawa, T., Nojiri, Y., Kortzinger, A., Steinhoff, T., Hoppema, M., Olafsson, J., Arnarson, T. S., Tilbrook, B., Johannessen, T., Olsen, A., Bellerby, R., Wong, C. S., Delille, B., Bates, N. R., and de Baar, H. J. W.: Climatological mean and decadal changes in surface ocean pCO2, and net sea-air CO2 flux over the global oceans, Deep-Sea Res. Pt. II, 56, 554–577, 2009. a

Tang, W., Arellano, A. F., DiGangi, J. P., Choi, Y., Diskin, G. S., Agustí-Panareda, A., Parrington, M., Massart, S., Gaubert, B., Lee, Y., Kim, D., Jung, J., Hong, J., Hong, J.-W., Kanaya, Y., Lee, M., Stauffer, R. M., Thompson, A. M., Flynn, J. H., and Woo, J.-H.: Evaluating high-resolution forecasts of atmospheric CO and CO2 from a global prediction system during KORUS-AQ field campaign, Atmos. Chem. Phys., 18, 11007–11030, https://doi.org/10.5194/acp-18-11007-2018, 2018. a

Temperton, C., Hortal, M., and Simmons, A.: A two-time-level semi-Lagrangian global spectral model, Q. J. Roy. Meteor. Soc., 127, 111–126, 2001. a

Thoning, K. W., Tans, P. P., and Komhyr, W. D.: Atmospheric carbon dioxide at Mauna Loa Observatory, 2. Analysis of the NOAA/GMCC data, 1974–1985, J. Geophys. Res., 94, 8549–8565, 1989. a

Tiedtke, M.: A comprehensive mass flux scheme for cumulus parameterization in large-scale models, Mon. Weather Rev., 117, 1779–1800, 1989. a

Tolk, L. F., Meesters, A. G. C. A., Dolman, A. J., and Peters, W.: Modelling representation errors of atmospheric CO2 mixing ratios at a regional scale, Atmos. Chem. Phys., 8, 6587–6596, https://doi.org/10.5194/acp-8-6587-2008, 2008. a, b

Tsutsumi, Y., Matsueda, H., and Nishioka, S.: Consistency of the CO2 primary standards in JMA, 12th WMO/IAEA meeting of experts on carbon dioxide concentration and related tracers measurement techniques (Toronto, Canada, 15–18 September 2003), Global Atmosphere Watch Report No. 161 (WMO/TD-No.1275), 23–31, 2005. a, b, c

van der Molen, M. K. and Dolman, A. J.: Regional carbon fluxes and the effect of topography on the variability of atmospheric CO2, J. Geophys. Res., 112, D01104, https://doi.org/10.1029/2006JD007649, 2007. a, b

Verhulst, K. R., Karion, A., Kim, J., Salameh, P. K., Keeling, R. F., Newman, S., Miller, J., Sloop, C., Pongetti, T., Rao, P., Wong, C., Hopkins, F. M., Yadav, V., Weiss, R. F., Duren, R. M., and Miller, C. E.: Carbon dioxide and methane measurements from the Los Angeles Megacity Carbon Project – Part 1: calibration, urban enhancements, and uncertainty estimates, Atmos. Chem. Phys., 17, 8313–8341, https://doi.org/10.5194/acp-17-8313-2017, 2017. a

Vermeulen, A. T., Hensen, A., Popa, M. E., van den Bulk, W. C. M., and Jongejan, P. A. C.: Greenhouse gas observations from Cabauw Tall Tower (1992–2010), Atmos. Meas. Tech., 4, 617–644, https://doi.org/10.5194/amt-4-617-2011, 2011. a

Vogel, F. R., Thiruchittampalam, B., Theloke, J., Kretschmer, R., Gerbig, C., Hammer, S., and Levin, I.: Can we assess a fine-grained emission model using high-resolution atmospheric transport modeling and regional fossil fuel CO2 observations? Tellus B, 65, 18681, https://doi.org/10.3402/tellusb.v65i0.18681, 2013. a

Wang, J. S., Kawa, S. R., Collatz, G. J., Sasakawa, M., Gatti, L. V., Machida, T., Liu, Y., and Manyin, M. E.: A global synthesis inversion analysis of recent variability in CO2 fluxes using GOSAT and in situ observations, Atmos. Chem. Phys., 18, 11097–11124, https://doi.org/10.5194/acp-18-11097-2018, 2018. a

Warneke, T., Messerschmidt, J., Notholt, J., Weinzierl, C., Deutscher, N., Petri, C., Grupe, P., Vuillemin, C., Truong, F., Schmidt, M., Ramonet, M., and Parmentier, E.: TCCON data from Orleans, France, Release GGG2014R0, TCCON data archive, hosted by CaltechDATA, California Institute of Technology, Pasadena, CA, USA, https://doi.org/10.14291/tccon.ggg2014.orleans01.R0/1149276, 2014. a

Wedi, N. P.: Increasing horizontal resolution in numerical weather prediction and climate simulations: illusion or panacea?, Philos. T. R. Soc. A, 372, 20130289, https://doi.org/10.1098/rsta.2013.0289, 2014. a

Wennberg, P. O., Wunch, D., Roehl, C., Blavier, J.-F., Toon, G. C., Allen, N., Dowell, P., Teske, K., Martin, C., and Martin, J.: TCCON data from Lamont, Oklahoma, USA, Release GGG2014R1, TCCON data archive, hosted by CaltechDATA, California Institute of Technology, Pasadena, CA, USA, https://doi.org/10.14291/tccon.ggg2014.lamont01.R1/1255070, 2016. a

Wennberg, P. O., Roehl,  C., Wunch, D., Toon, G. C., Blavier, J.-F., Washenfelder, R., Keppel-Aleks, G., Allen, N., and Ayers, J.: TCCON data from Park Falls, Wisconsin, USA, Release GGG2014R1, TCCON data archive, hosted by CaltechDATA, California Institute of Technology, Pasadena, CA, USA, https://doi.org/10.14291/tccon.ggg2014.parkfalls01.R0/1149161, 2014a. a

Wennberg, P. O., Wunch, D., Roehl, C., Blavier, J.-F., Toon, G. C., and Allen, N.: TCCON data from California Institute of Technology, Pasadena, California, USA, Release GGG2014R1, TCCON data archive, hosted by CaltechDATA, California Institute of Technology, Pasadena, CA, USA, https://doi.org/10.14291/tccon.ggg2014.pasadena01.R1/1182415, 2014b. a

Wilson, P. A.: Insight into the Carbon Cycle from Continuous Measurements of Oxygen and Carbon Dioxide at Weybourne Atmospheric Observatory, UK, PhD thesis, University of East Anglia, Norwich, UK, 2013. a

Worthy, D. E., Higuchi, K., and Chan, D.: North American influence on atmospheric carbon dioxide data collected at Sable Island, Canada, Tellus B, 55, 105–114, 2003. a, b, c, d, e, f, g, h, i, j, k, l

Wunch, D., Wennberg, P. O., Toon, G. C., Keppel-Aleks, G., and Yavin, Y. G.: Emissions of greenhouse gases from a North American megacity, Geophys. Res. Lett., 36, L15810, https://doi.org/10.1029/2009GL039825, 2009. a

Wunch, D., Toon, G. C., Blavier, J.-F. L., Washenfelder, R. A., Notholt, J., Connor, B., Griffith, D. W. T., Sherlock, V., and Wennberg, P. O.: The total carbon column observing network, Philos. T. R. Soc. A, 369, 2087–2112, https://doi.org/10.1098/rsta.2010.0240, 2011. a

Wunch, D., Wennberg, P. O., Osterman, G., Fisher, B., Naylor, B., Roehl, C. M., O'Dell, C., Mandrake, L., Viatte, C., Kiel, M., Griffith, D. W. T., Deutscher, N. M., Velazco, V. A., Notholt, J., Warneke, T., Petri, C., De Maziere, M., Sha, M. K., Sussmann, R., Rettinger, M., Pollard, D., Robinson, J., Morino, I., Uchino, O., Hase, F., Blumenstock, T., Feist, D. G., Arnold, S. G., Strong, K., Mendonca, J., Kivi, R., Heikkinen, P., Iraci, L., Podolske, J., Hillyard, P. W., Kawakami, S., Dubey, M. K., Parker, H. A., Sepulveda, E., García, O. E., Te, Y., Jeseck, P., Gunson, M. R., Crisp, D., and Eldering, A.: Comparisons of the Orbiting Carbon Observatory-2 (OCO-2) XCO2 measurements with TCCON, Atmos. Meas. Tech., 10, 2209–2238, https://doi.org/10.5194/amt-10-2209-2017, 2017. a, b

Xueref-Remy, I., Dieudonné, E., Vuillemin, C., Lopez, M., Lac, C., Schmidt, M., Delmotte, M., Chevallier, F., Ravetta, F., Perrussel, O., Ciais, P., Bréon, F.-M., Broquet, G., Ramonet, M., Spain, T. G., and Ampe, C.: Diurnal, synoptic and seasonal variability of atmospheric CO2 in the Paris megacity area, Atmos. Chem. Phys., 18, 3335–3362, https://doi.org/10.5194/acp-18-3335-2018, 2018.  a

Yang, D. X., Liu, Y., Cai, Z. N., Chen, X., Yao, L., and Lu, D. R.: First global carbon dioxide maps produced from TanSat measurements, Adv. Atmos. Sci., 35, 621–623, https://doi.org/10.1007/s00376-018-7312-6, 2018. a

Yu, K., Keller, C. A., Jacob, D. J., Molod, A. M., Eastham, S. D., and Long, M. S.: Errors and improvements in the use of archived meteorological data for chemical transport modeling: an analysis using GEOS-Chem v11-01 driven by GEOS-5 meteorology, Geosci. Model Dev., 11, 305–319, https://doi.org/10.5194/gmd-11-305-2018, 2018. a