Analysis of CO₂ Concentration and Fluxes of Lisbon Portugal Using Regional CO₂ Assimilation Method Based on WRF-Chem

Abstract

Cities house more than half of the world’s population and are responsible for more than 70% of the world anthropogenic CO₂ emissions. Therefore, quantifications of emissions from major cities, which are only less than a hundred intense emitting spots across the globe, should allow us to monitor changes in global fossil fuel CO₂ emissions in an independent, objective way. The study adopted a high-spatiotemporal-resolution regional assimilation method using satellite observation data and atmospheric transport model WRF-Chem/DART to assimilate CO₂ concentration and fluxes in Lisbon, a major city in Portugal. It is based on Zhang’s assimilation method, combined OCO-2 XCO₂ retrieval data, ODIAC 1 km anthropogenic CO₂ emissions and Ensemble Adjustment Kalman Filter Assimilation. By employing three two-way nested domains in WRF-Chem, we refined the spatial resolution of the CO₂ concentrations and fluxes over Lisbon to 3 km. The spatiotemporal distribution characteristics and main driving factors of CO₂ concentrations and fluxes in Lisbon and its surrounding cities and countries were analyzed in March 2020, during the period affected by COVID-19 pandemic. The results showed that the monthly average CO₂ and XCO₂ concentrations in Lisbon were 420.66 ppm and 413.88 ppm, respectively, and the total flux was 0.50 Tg CO₂. From a wider perspective, the findings provide a scientific foundation for urban carbon emission management and policy-making.

1. Introduction

With the intensification of global climate change, as the main area for human activities, cities are the main carriers of energy consumption and greenhouse gas emissions. They have become a key issue in climate governance in terms of their CO₂ concentration and CO₂ emission characteristics. The global average concentration of atmospheric CO₂ reached 413.2 ± 0.2 ppm in 2020 [1]. The rapid increase in atmospheric CO₂ concentration caused by human activities is currently the main cause of global climate change [2]. According to the International Energy Agency [3], cities as the centers of human socio-economic activities gather 50% of the world’s population and contribute about 70% of the world’s fossil energy-related CO₂ emissions. Europe as a highly urbanized region has large city clusters such as Paris, Berlin, and Milan with CO₂ concentrations consistently exceeding 400 ppm, far beyond preindustrial levels of 280 ppm. This phenomenon is closely related to dense transportation networks, building energy consumption, and industrial activities. Peter’s [4] research showed that for the year 2000, GHG emissions in Europe were about 23% of the global total, 50.8% of which were used for energy conversion, followed by transportation (14.9%), residential (12.1%), industry (11.2%), agricultural (7.8%), and water (3.2%). At the same time, CO₂ is the main Greenhouse Gas (GHG) leading to global warming [5].

As a highly urbanized region, Europe exhibits high heterogeneity in its CO₂ concentration emission sources (transportation, construction, industry, etc.). Research on urban CO₂ concentration and emissions started earlier, forming a multi-scale and interdisciplinary research system. In terms of monitoring technology, Europe relies on large-scale scientific research infrastructure such as the Integrated Carbon Observing System (ICOS) [6] and the Total Carbon Column Observing Network (TCCON) [7] to establish the stereoscopic monitoring network covering urban stations, transportation, and industrial areas. By combining satellite remote sensing data from Orbiting Carbon Observatory-2 (OCO-2) and OCO-3 with ground-based sensor measurements, real-time dynamic tracking of CO₂ concentrations has been achieved.

Traditional emission inventories, such as Emissions Database for Global Atmospheric Research (EDGAR) [8], are often based on macro statistics and static assumptions, making it difficult to capture the complex spatiotemporal dynamics within cities, and it is often delayed by several years. In recent years, the atmospheric assimilation inversion algorithms have provided a new path for refined inversion of urban CO₂ emissions by integrating multi-source observation data (ground monitoring stations, satellite remote sensing, mobile sensors) with atmospheric transport models such as Global Chemistry Transport Model 5 (TM5) [9], Weather Research and Forecasting model coupled with Vegetation Photosynthesis and Respiration (WRF-VPRM) [10,11], Global Modeling and Assimilation Office model coupled with Chemistry (GEOS-Chem), and Weather Research and Forecasting model coupled with Chemistry (WRF-Chem) [12,13]. This has been used successfully to estimate regional fluxes at finer scales [14,15,16,17,18] and compared to existing agricultural inventories [19]. Lin Wu et al. [20] assessed the performance of atmospheric CO₂ inversion for the monitoring of total and sectoral fossil fuel emissions in the Paris metropolitan area by the Bayesian inversion method, based on observing system simulation experiments (OSSEs). Lauvaux et al. [21] used a state-of-the-art mesoscale model to characterize daily CO₂ emissions from Davos, Switzerland. Lowry et al. [22] used a mass-balance approach to quantify London, England, CH₄ emissions and their trend over several years. Using amultiple box model, Strong et al. [23] estimated the emissions from Salt Lake City, Utah and the contributions from biogenic and anthropogenic sources. However, urban assimilation inversion still faces such challenges of the heterogeneity of the surface properties [24] and the simulation of fine-scale structures such as plumes from point sources [25]. The accuracy of the atmospheric transport models can lead to large uncertainties unless the full complexity of the local dynamics is correctly simulated [26].

COVID-19 (Coronavirus 2019) in 2020 led to the stagnation of global economic activities, and the global CO₂ emissions fell sharply in the short term, but the atmospheric CO₂ concentration continued to rise, becoming a “natural experiment” to study the relationship between anthropogenic emissions and atmospheric CO₂ concentration. In 2020, global CO₂ emissions decreased by about 7% year-on-year, but atmospheric CO₂ concentrations continued to increase at a rate of 2.5 ppm per year [27]. And this phenomenon was also observed from space by the OCO-2 satellite [28]. This contradictory phenomenon highlights the long-term and complex nature of climate action, triggering in-depth discussions in the scientific community on the effectiveness of emission reduction and carbon cycle response mechanisms.

As a major country in southwestern Europe, Portugal’s spatiotemporal distribution characteristics and driving factors of urban CO₂ concentration have not been fully studied. As the capital of Portugal and one of the largest cities on the Iberian Peninsula, Lisbon’s urbanization process and energy consumption patterns have a significant impact on regional air quality and climate change.

This study aims to use Mizzi’s and Zhang’s assimilation system [29,30,31,32] combined with OCO-2 retrieval data and three two-way nested domains in WRF-Chem to analyze the CO₂ concentrations and fluxes of Lisbon, Portugal in March 2020. The paper is organized as follows. Section 2 describes research materials and methods. Results and discussion are given in Section 3, followed by conclusion in Section 4.

2. Materials and Methods

2.1. ODIAC Emission Dataset

We use the Open-Source Data Inventory for Anthropogenic CO₂ (ODIAC) as prior fluxes in our experiment. The ODIAC dataset was created using a hybrid approach that disaggregates national emissions estimates produced by Carbon Dioxide Information Analysis Center (CDIAC), but it distributes them using nightlight data and the location of Local Point Sources (LPSs) inventoried in Carbon Monitoring and Action (CARMA). ODIAC is published at the same resolution as the nightlight data, approximately 1 km, or 0.008333^∘ [33].

Using the version ODIAC 2023 which is the latest version with anthropogenic CO₂ emissions in prior fluxes, the spatial distribution of monthly anthropogenic CO₂ emissions in March 2020 in the assimilation experiment area is shown in Figure 1.

2.2. OCO-2 XCO₂ Retrieval Dataset

We used the Orbiting Carbon Observatory-2 (OCO-2) V11.1r as satellite observation data for our experiment, which is NASA’s first dedicated CO₂ monitoring satellite, launched on 2 July 2014. OCO-2 satellite routinely makes around 1 million observations each day, and more than 10% of these observations are sufficiently cloud-free to retrieve XCO₂ precisely [34,35,36]. The XCO₂ retrieved from OCO-2 V11.1r are consistently lower than Total Carbon Column Observing Network (TCCON) (about 1 ppm).

According to the “xco2_quality_flag”, which indicates the quality of OCO-2 XCO₂ retrievals, a value of 0 represents good retrieval, while a value of 1 indicates poor retrieval quality and is not recommended for use. XCO₂ retrievals from the nadir observation mode with good quality were selected from the OCO-2 Level 2 (L2) Lite data product V11.1r. These selected retrievals were first screened by a filter to reject the outlies that drift away from the mean value of the background more than three times of the standard deviation of the background. The preprocessing approach it employs is outlined as follows [31]:

$$\widetilde{OCO{2}_{XC{O}_2}}=\sum _{i=1}^{n}OCO{2}_{XC{O}_{2i}}{\sigma }_{OCO{2}_i}^{-2}/\sum _{i=1}^n{\sigma }_{OCO{2}_i}^{-2}$$

(1)

$$\widetilde{{\sigma }_{OCO2}}=1/\sqrt{N^{-1}\sum _{i=1}^n{\sigma }_{OCO{2}_i}^{-2}}$$

(2)

where $\widetilde{OCO{2}_{XC{O}_2}}$ and $\widetilde{{\sigma }_{OCO2}}$ denote the representative mean XCO₂ value and its uncertainty of a model grid cell, respectively, which was calculated and assimilated according to the strategy of Crowell et al. [34]. The OCO-2 10s mean XCO₂ retrieval results over the assimilation experiment area in research time were processed by this strategy, as shown in Figure 2.

Within this area, there are 91 OCO-2 XCO₂ observations. To convert vertically layered CO₂ concentrations simulated by WRF-Chem into XCO₂ with the same physical dimensions as those of OCO-2 observations for data assimilation purposes, we adopted the observation operator H, as defined in Equation (3), following the approach described by Connor et al. [37]:

$$XC{O}_2^m=XC{O}_2^a+\sum _j{h}_j{a}_j(C{O}_2^m-C{O}_2^a)$$

(3)

where $XC{O}_2^a$, $h_j$, $a_j$ and $C{O}_2^a$ are the prior XCO₂ value, the pressure weighting function, the column averaging kernel, and the prior CO₂ concentration profile used by OCO-2 XCO₂ retrieval pretreatment, respectively. $C{O}_2^m$ is the optimal CO₂ concentration profile interpolated to the pressure levels of OCO-2 XCO₂ retrievals from the WRF-Chem model. $XC{O}_2^m$ is the column-average CO₂ concentration in the observation space transformed from $C{O}_2^m$ by the OCO-2 XCO₂ retrieval observation operator H.

2.3. CO₂ Assimilation System

The study is based on the assimilation system followed by Mizzi et al. [29,30] and Zhang et al. [31], a regional CO₂ concentration assimilation system with OCO-2 XCO₂ retrievals extended by DART [32]. We construct a high-spatiotemporal-resolution regional CO₂ assimilation system. It adopts three two-way nesting domains, based on the WRF-Chem/DART transport model, with resolutions of 27 km, 9 km, and 3 km, respectively. Figure 3 shows the structure of our CO₂ assimilation system.

Within the regional CO₂ assimilation framework, the OCO-2 XCO₂ V11.1r retrievals are employed as the satellite-based observational data. The ODIAC emissions inventory and Carbontracker CT2022 fluxes [38] serve as the prior CO₂ fluxes, while the MERRA2 dataset provides the necessary meteorological information.

The regional CO₂ assimilation method follows the approach described in Zhang et al. [31], employing an extended ensemble adjustment Kalman filter (EAKF) to assimilate WRF-Chem simulations with OCO-2 observation and to inverse the CO₂ emissions.

2.4. Research Area

In the study, we aimed to acquire the high-spatial-resolution CO₂ concentration centered on Lisbon, Portugal. To achieve this, we devised three two-way nested domains based on WRF-Chem/DART for the purpose of regional assimilation of CO₂ concentration. The boundary conditions of the regional CO₂ concentration assimilation model were interpolated from the CO₂ assimilation result of GEOS-Chem, a Global Environmental Multiscale Model, which has 0.5^∘ × 0.625^∘ spatial resolution and 6-hour resolution.

The geographical coverage of the experiment illustrated in Figure 4. We designated the four domains, D00, D01, D02, and D03. The D00 domain includes the whole world. The D01 domain encompasses western Europe, northern Africa, and a portion of the North Atlantic Ocean. The D02 domain covers Spain and Portugal. Meanwhile, the D03 domain, which is centered on Lisbon, includes part of Portugal and a section of the North Atlantic Ocean.

2.5. Experiment Conditions and Materials

The WRF-Chem model version 4.4 was used as the CO₂ transport model [39,40]. The research area is shown in Figure 4. The research period is March 2020. The settings of the WRF-Chem parameter configurations are shown in

Table 1. The WRF-Chem parameter configuration setting.

Options	Configurations
WRF_Core	ARW
Domain center	38.717° N, −9.133° W
Max_dom	3
Grid resolution	27 km, 9 km, 3 km
Vertical level( nz)	48
Nx_CR, Nx_FR, Nx_IR	90, 76, 64
Ny_CR, Ny_FR, Ny_IR	90, 76, 64
Interval seconds	21,600 s/6 h
Time steps	90 s
Start date	1 March 2020 00:00:00
End date	31 March 2020 18:00:00
Microphysics process	WSM 5-class simple ice scheme [43]
Cumulus parameterization	Kain–Fritsch scheme [44]
Longwave atmospheric radiation	RRTM scheme [45]
Shortwave atmospheric radiation	Dudhia scheme [46]
Planetary boundary layer scheme	MYNN 2.5 level TKE [47]
Surface layer scheme	MYNN [48]
Land surface scheme	Unified Noah Land surface model
Chemistry option	chem_opt = 16 (CO2 only)

[41,42].

The WRF-Chem model used the ds083.2 dataset (DOI: 10.5065/D6M043C6) and MERRA2 (The Modern-Era Retrospective Analysis for Research and Applications, Version 2) data as initial and boundary meteorological conditions. The ds083.2 dataset, provided by the National Centers for Environmental Prediction (NCEP) Final (FNL) Operational Model Global Tropospheric Analyses, offers a spatial resolution of $1^{\circ }\times 1^{\circ }$ and a temporal resolution of 6 h, with continuous data available since July 1999. The MERRA-2 dataset, developed by the National Aeronautics and Space Administration (NASA), features a spatial resolution of 0.5^∘ × 0.625^∘ and a temporal resolution of one hour. It has been available since 1980, providing a long-term and comprehensive record for atmospheric modeling and analysis.

The prior CO₂ fluxes were obtained by interpolating ODIAC emissions combined with CarbonTracker CT2022 fluxes, where the CarbonTracker CT2022 fluxes only include fire emissions, biogenic fluxes, and ocean fluxes. These CarbonTracker CT2022 emissions have a spatial resolution of 1^∘ × 1^∘ and a temporal resolution of 3 h. ODIAC emissions serve as the source of anthropogenic emissions for the prior CO₂ fluxes. They have a high spatial resolution of 1 km and a monthly temporal resolution. However, for the purpose of this study, these emissions needed to be preprocessed to achieve a 3-hour temporal resolution. Its weekly/diurnal emissions can be modeled by applying the TIMES [49] temporal scaling factors to the ODIAC monthly emission fields. The scaling factor is defined on a 0.25^∘ × 0.25^∘ scale. The gridded scaling data product can be downloaded from the website of the Oak Ridge National Laboratory (https://data.ess-dive.lbl.gov/view/doi%3A10.15485%2F1463822 (accessed on 29 May 2025)). To ensure compatibility with the spatial resolutions of each domain within the regional CO₂ assimilation system, it was necessary to resample the ODIAC 1 km data and CarbonTracker CT2022 fluxes. The data were resampled to resolutions of 3 km, 9 km, and 27 km, respectively, to align with the different domain requirements of the assimilation system.

3. Results and Discussion

3.1. Assimilation Experiment Results

The OCO-2 XCO₂ retrievals were assimilated using the regional CO₂ concentration assimilation system. The optimized CO₂ concentration and CO₂ fluxes for the three experimental domains (D01, D02, and D03) in March 2020 are illustrated in Figure 5. For the month of March 2020, the assimilated monthly mean CO₂ concentration for the D01, D02, and D03 domains was 417.24 ppm, 419.14 ppm, and 420.62 ppm, respectively. Meanwhile, the monthly mean XCO₂ concentrations for these domains were 413.87 ppm, 413.83 ppm, and 413.87 ppm, respectively. Moving from the D01 domain to the D03 domain, the area covered by each domain decreases progressively. Notably, the monthly mean CO₂ concentration shows a gradual increase across these domains. In contrast, the monthly mean XCO₂ concentrations remain largely unchanged, indicating a relatively stable pattern in terms of column-averaged CO₂ across the different-sized domains.

Regarding the CO₂ fluxes, the total flux of D01, D02, and D03 domains in March 2020 were −43.68 Tg CO₂, −15.36 Tg CO₂, and 0.53 Tg CO₂, respectively. As illustrated in Figure 5, CO₂ emissions in the D01 and D02 dimains were negative, whereas in domain D03 they were positive. From Figure 5, it is evident that in March 2020, CO₂ emissions in the southwestern regions of Spain, including areas like the Andalusia region, were relatively low, even lower than those in Portugal. This phenomenon is mainly due to the blockade measures triggered by the COVID-19 epidemic. These measures led to a sudden drop in energy demand and were also influenced by the regional differences in economic structure. Spain was the first country to impose a nationwide lockdown on 14 March 2020. This lockdown involved the closure of non-essential businesses, travel restrictions, and other similar measures. In contrast, Portugal, with its high proportion of renewable energy, introduced its lockdown policy later, on 18 March. During the lockdown period, the southwestern region of Spain, which includes severely affected areas such as Seville and Malaga, witnessed a significant decline in transportation and industrial activities. Social activities in these regions also virtually ground to a halt, further contributing to the reduction in CO₂ emissions.

As depicted in Figure 6, the monthly mean CO₂ concentration of Lisbon was 420.66 ppm. The CO₂ concentration in the eastern of Lisbon was higher compared to the western coastal area. The monthly mean XCO₂ concentration in Lisbon stood at 413.88 ppm. Overall, the trend of XCO₂ changes within the city was not significant. However, there was a distinct high-concentration point located in the northeast region of Lisbon, and emissions at this specific location were also quite substantial. The total CO₂ fluxes for Lisbon were 0.50 Tg CO₂, and the CO₂ fluxes in the southern of Lisbon were higher than those in the northern regions.

3.2. Comparison with TCCON

Within the D01 domain of our experiment, there are two Total Carbon Column Observing Network (TCCON) sites: Orléans (located at ${47.97}^{\circ }$ N, ${2.113}^{\circ }$ E) and Paris (located at ${48.846}^{\circ }$ N, ${2.356}^{\circ }$ E). Both of these sites are situated in France. We extracted the XCO₂ concentration data from these two sites within the D01 domain, subsequently comparing them with the corresponding TCCON data. The outcomes of this comparison are presented in Figure 7 and

Table 2. The assimilation XCO₂ results compared with TCCON sites with hourly average.

	Vs. Orléans		Vs. Paris
	TCCON	DA Experiment	TCCON	DA Experiment
Numer	1639	733	2313	733
Mean XCO2	414.78 ppm	414.47 ppm	413.85 ppm	415.01 ppm
$\Delta XC{O}_2$	0.31 ppm		1.16 ppm

.

In March 2020, the Orléans TCCON site recorded 1639 data points with an average XCO₂ concentration of 414.78 ppm. Meanwhile, the Paris TCCON site gathered 2313 data points during the same period, and its mean XCO₂ concentration was 413.85 ppm. We compared the assimilated XCO₂ results with the data from these two TCCON sites on an hourly average, respectively. The compare results are presented in the right-hand panel of Figure 7. In this panel, the x-axis represents hourly time intervals, while the y-axis indicates the XCO₂ value. The comparison reveals that the difference in ΔXCO₂ between Orléans site and the assimilation results is 0.31 ppm. On the other hand, the ΔXCO₂ between Paris site and the assimilation results is 1.16 ppm.

3.3. Comparison with ObsPack

We compared the assimilated CO₂ concentration in the D01 domain with Observation Package Data Products (ObsPack) [50]. ObsPack data are specifically designed to stimulate and support carbon cycle modeling studies. The comparison was carried out for the month of March 2020. Within the D01 domain, we obtained 13 data points from ObsPack. Based on the statistical analysis of these points, the monthly mean CO₂ concentration from ObsPack was found to be 420.81 ppm. In contrast, the monthly mean CO₂ concentration from our assimilation experiment was 418.55 ppm, which is 2.26 ppm lower than the ObsPack data. The Root Mean Square Error (RMSE) between the assimilation results and ObsPack data was calculated to be 4.06 ppm, and the correlation coefficient (CORR) was 0.68. These comparative results are presented in

Table 3. The assimilation CO₂ results compared with ObsPack.

	ObsPack	DA Experiment
Numer	13	13
Mean CO2	420.81 ppm	418.55 ppm
$\Delta C{O}_2$	2.26 ppm
MBE	2.25 ppm
MAE	2.59 ppm
RMSE	4.06 ppm
CORR	0.68

and Figure 8. Figure 9 illustrates the spatial distribution of the monthly mean CO₂ concentration for both the CO₂ assimilation results and the ObsPack data within D01 domain for March 2020.

4. Conclusions

The paper presents a comprehensive assessment of CO₂ concentration and fluxes in Lisbon, Portugal during March 2020, employing a high-spatio-temporal resolution regional CO₂ assimilation system based on WRF-Chem/DART with three nested domains. The assimilation results indicate that in March 2020, Lisbon recorded CO₂ and XCO₂ concentrations of 420.66 ppm and 413.88 ppm, respectively, along with CO₂ fluxes of 0.50 Tg CO₂. We compared the XCO₂ concentration with the data from two TCCON sites, Orléans and Paris, revealing a monthly mean ΔXCO₂ of 0.74 ppm, which is less than 1 ppm. Moreover, the discrepancy with the Orléans site was merely 0.31 ppm. However, when comparing the surface CO₂ concentration with Obspack data, the monthly mean ΔCO₂ was found to be 2.26 ppm, with a Root Mean Square Error (RMSE) of 4.06 ppm, indicating a relatively significant gap on CO₂ concentration.

An examination of the CO₂ flux distribution across the D01 domain clearly illustrates that, in March 2020, Spain’s social and economic activities stagnated due to the blockade of COVID-19. This led to a sharp decline in CO₂ emissions in southwestern Spain. This short-term phenomenon underscores the strong correlation between human activities and carbon emissions. Additionally, CO₂ flux distribution in Lisbon pinpointed a hotspot for CO₂ emissions at coordinates ($9^{\circ }$ W, ${39}^{\circ }$ E).

This study provides valuable information on the dynamics of urban CO₂ and highlights the impact of sudden socioeconomic changes such as COVID-19 on carbon emissions. The regional CO₂ assimilation method employed can be adapted for other megacities to support the development of low-carbon policies.