The study focuses on Catalonia, a region in northeast Spain with an area of approximately 32 108 km2. Its geographic diversity, illustrated in Fig. , which includes mountains over 3000 m high, plains, and coasts, plays a significant role in the region's air quality .

Catalonia presents a variety of landscapes and climates that influence the dispersion and concentration of atmospheric pollutants. Key geographic features include the Pyrenees, the Central Catalan System, and the Lleida Plain. Climatically, most of Catalonia experiences a Mediterranean climate, while the Pyrenees have a mountainous climate characterized by colder and wetter conditions . The Pyrenees influence air currents and can act as a barrier to pollutant dispersion. The Central Catalan System, separated by the coastal plain, presents a relief that affects air circulation and pollutant dispersion, creating local microclimates. Catalonia also has a significant urban center, Barcelona, whose surrounding areas are known as the Barcelona Metropolitan Area (BMA), a region of considerable demographic and economic impact. In the interior, Lleida's extensive agricultural activity contributes to thermal inversion conditions. Along the coast, sea breezes can help disperse pollutants in coastal areas, but they can also transport them inland.

Other factors such as hydrography and anthropogenic influences also play a significant role in air quality. Major rivers such as the Ebro, Ter, Llobregat, and Besós are key features of the region. These river valleys frequently experience thermal inversion phenomena during winter, trapping pollutants close to the ground and affecting nearby populations.

The ground data for this study are obtained from the Xarxa de Vigilància i Previsió de la Contaminació Atmosfèrica (XVPCA) network of Catalonia, which provides near-real-time (NRTI) air quality data across the study area. The XVPCA network operates monitoring stations that measure various pollutants, including NO2, PM10, PM2.5, O3, carbon monoxide (CO), and sulfur dioxide (SO2). These stations are strategically distributed across different geographic zones and typologies, providing hourly data and comprehensive coverage of the region's air quality .

The XVPCA network follows the European standard EN 14211:2012 for the measurement of NO2 concentrations, which defines the chemiluminescence method for determining NO2 and NOx levels . This standardized approach ensures typical uncertainties of approximately ±10% under standard operating conditions .

The S5p satellite is part of the European Union's Copernicus Earth observation program, and it is the first Copernicus mission specifically dedicated to atmospheric monitoring. The S5p data, freely accessible through the Copernicus platform, are essential for global air quality monitoring .

S5p provides daily and global data on various atmospheric pollutants, including O3, CO, NO2, SO2, methane (CH4), formaldehyde (HCHO), and aerosols (AOD) . It uses the TROPOMI instrument to collect these data, covering a width of 2600 km, allowing the entire planet to be mapped in a 24 h period . Of the different data provided by S5p on NO2, in this study, the band containing the information on the tropospheric column of NO2 has been specifically used. This field provides the total amount of NO2 present in the troposphere column, which is the most relevant atmospheric region for air quality monitoring, as it reflects the direct impact of NO2 at ground level (see Fig. ).

Datasets downloaded directly from the Copernicus platform (level 2 data) have a native spatial resolution of approximately 3.5km×5.5km . Based on mission requirements and validation reports, the random precision (1σ) of the tropospheric NO2 column is 0.7×1015 moleccm-2, while the systematic uncertainty (bias) ranges from 25 % to 50 % . In our study, however, NO2 values were extracted from the Sentinel-5P level 3 dataset using the Google Earth Engine (GEE) platform. These level 3 data are provided on a regular 0.01°×0.01° grid (approximately 1.1km×0.85km over Spain), generated by applying the harpconvert tool to the level 2 product after filtering out pixels with qa_value <0.75 . This process remaps the data spatially but does not recalculate or propagate the precision band, so the official level 2 uncertainties are retained.

It is important to note that these data can occasionally show negative values – an observation that has been documented in previous studies . These negative values in tropospheric NO2 columns can arise due to uncertainties inherent to the data retrieval process, which involves separating the tropospheric component column from the total NO2 column by subtracting the estimated stratospheric contribution using differential optical absorption spectroscopy (DOAS) techniques . Since this separation process can be subject to errors, especially under very low NO2 concentrations, negative values are generally considered to be retrieval artifacts or instrumental noise rather than physically meaningful concentrations . In particular, these cases are more likely to appear in regions with minimal NO2 concentrations or complex atmospheric conditions, being within the typical instrument uncertainty range.

The products are available in NRTI and offline (OFFL) modes. NRTI data are typically made available within 2–3 h of acquisition by the satellite , making them ideal for applications requiring almost instantaneous information, such as air quality monitoring, early pollution alerts, or environmental emergencies. On the other hand, OFFL data, typically available 1–2 d after acquisition , offer more comprehensive processing, with more accurate atmospheric and meteorological models, making them more suitable for detailed scientific analysis and long-term climate studies.

In addition to being available for direct download from the Sentinel-5P Pre-Operations Data Hub, the datasets are also available on Google Earth Engine (GEE), a platform that allows accessing and working with historical and NRTI atmospheric data . Since analysis of large volumes of NetCDF data can be computationally expensive and slow, accessing and processing through GEE provides a more efficient alternative. In GEE, S5p data can be integrated with other datasets, such as from the XVPCA network, facilitating broader and faster analysis.

The main objective of this study is to characterize the correlation between NO2 measurements obtained from the S5p satellite and those recorded by the XVPCA surface monitoring network in Catalonia. This analysis takes into account the geographical diversity of the study area, as well as the distinctive characteristics of the monitoring stations and seasonal variations. Previous studies have explored the integration of meteorological data in air quality analysis. However, incorporating such data at this stage could mask the direct relationship we aim to establish between satellite- and ground-based NO2 measurements. By focusing on this direct relationship, we aim to identify the most appropriate methods to estimate ground-level air quality using satellite data in Catalonia, leaving the integration of meteorological variables for future research.

The methodology is organized into several steps. First, data preprocessing is performed, including spatial alignment of satellite pixels with ground station locations and temporal synchronization of measurement times. Next, the analysis begins with a global approach and becomes more specific, focusing on zone-based, typology-based, and seasonal aspects. Finally, statistical methods, including correlation plots, Pearson correlation coefficients (r), regression analysis, and coefficient of variation (CV), are applied to evaluate the relationship between satellite- and ground-based data.

Before presenting the results we would like to note that in the presentation of the dispersion results and statistical calculations, the original data types have been retained: surface concentration (µgm-3) for ground-based measurements and column-integrated values (molm-2) for satellite observations. This approach was chosen because the conversion from column-integrated values to surface concentration involves complex calculations that require additional assumptions about the vertical distribution of NO2 within the atmospheric column. Factors such as temperature, pressure, solar radiation, and local meteorological conditions significantly influence this conversion. Including these variables would introduce an additional layer of uncertainty into the analysis, potentially altering the direct relationship between the datasets.

Hereafter we present two scatter diagrams comparing the NO2 concentrations measured by the S5p satellite and by the XVPCA ground-based network (Fig. ), along with a table containing the correlation coefficient r and regression parameters (Table ). The global analysis of the data considers all the values measured in Catalonia during the years 2022 and 2023 without applying any filters or divisions to the dataset.

In this study, the negative values of satellite measurements were kept in the scatter diagrams and statistical calculations because the main objective is to understand the direct correlation between NO2 measurements from the S5p satellite and those recorded by the XVPCA terrestrial network in Catalonia. Keeping these values in the analyses provided us with valuable information on that correlation in urban zones.

In the global results (Fig.  and Table ), the r was 0.65 in 2022 and 0.66 in 2023, indicating a moderate positive correlation between measurements made by the S5p satellite and surface stations. Table  shows that the β1 values remained close between the two years, with a value of 2.4×10-6 in 2022 and 2.2×10-6 in 2023, indicating that the sensitivity of satellite measurements to changes in terrestrial NO2 concentrations is constant.

However, using satellite measurements in their original unit (molm-2) does not allow a direct interpretation for the analysis of the regression line. For this reason, we present a conversion to equivalent units (µgm-3) to facilitate result interpretation. This conversion does not take into account the atmosphere's complexity and assumes that NO2 is uniformly mixed throughout the tropospheric column. Consequently, the regression lines have been recalculated, this time using values measured by S5p converted to µgm-3. The results of these lines reveal, in general, a low β1 in the analyzed data. For the global dataset, the β1 value of 0.0163 indicates that, for each increase of 1 µgm-3 in the NO2 concentration measured at the surface, the concentration measured by the S5p satellite only increases by 0.0163 µgm-3. Additionally, the β0 of 11.5 µgm-3 suggests that even when surface measurements are zero, the satellite still predicts a positive concentration. The R2 obtained is 0.43 for both years, indicating that 43 % of the variance in satellite measurements is explained by surface measurements.

A correlation analysis was performed for two groups based on zones: one based on urban, suburban, and rural zones and the other distinguishing between BMA and non-BMA zones. This approach allowed us to assess variations in correlation patterns across different zone types.

Hereafter, we present the correlation analysis based on different station typologies: traffic, background, and industrial stations. This analysis allows us to assess how NO2 concentrations align with satellite data, depending on the type of activity in the vicinity of the stations. The results for 2022 and 2023 are illustrated in Fig. , which displays scatter plots and regression lines for each typology. The corresponding statistical values are summarized in Table .

For traffic stations, the correlation values were moderate, at 0.63 in 2022 and 0.61 in 2023. Background stations showed slightly higher correlations, with values of 0.66 in 2022 and 0.68 in 2023. Similarly, industrial stations demonstrated correlations of 0.68 in 2022 and 0.66 in 2023.

Finally, we present in Table  a seasonal analysis that covers all recorded values from 2022 and 2023 and provides a detailed view of how seasonal factors – such as temperature and atmospheric conditions – affect both the r correlation and the coefficient of variation (CV) between satellite and surface measurements of NO2 throughout the year. Additionally, Tables  and summarize the regression analysis results, including β1, β0, and R2 for NO2 measurements across different seasons and typologies, highlighting the seasonal variability observed in the data.

The correlation coefficients show marked seasonal variability throughout the year. Overall, winter and autumn stand out with the highest correlations, reaching a value of 0.70 in both years. In contrast, during spring and summer, significant decreases are observed in all zones and typologies, with the global minimum being 0.55 in the summer of 2022 and 0.57 in 2023. This seasonal reduction is especially evident in urban and the BMA zones, where correlations dropped to 0.38 and 0.23, respectively, in the summer of 2022.

The CV supports these observations by providing a measure of the relative variability of the seasonal correlations. The urban and BMA zones showed the highest CVs, at 16.61 % and 31.03 %, respectively, indicating a high degree of fluctuation. By contrast, the non-BMA zone exhibited a much lower CV (7.46 %), reflecting more consistent correlations with reduced variability throughout the year.

Results by station typology indicate that background stations maintained the highest correlations throughout the year, peaking at 0.74 in winter 2022 and a CV of 10.02 %, suggesting stable seasonal performance. Industrial stations also demonstrated strong seasonal stability (CV 11.28 %), with correlations exceeding 0.65 during the colder months. In contrast, traffic stations exhibited the greatest seasonal variability, with correlations falling to 0.47 in summer 2022 but recovering to 0.68 in winter 2022. Nevertheless, their CV of 11.75 % highlights this intermediate level of variability.

The regression analysis, summarized in Tables  and , supports these findings. β1 and R2 were highest in winter and autumn, reflecting a stronger sensitivity of satellite measurements to ground-level NO2 during these seasons. Conversely, the summer months exhibited lower values for both. The β0 values are consistently positive, with higher values observed in urban zones during summer and autumn.

The 2022 scatter plot (Fig. ) shows a greater dispersion of the measurements when compared to the 2023 graph. This difference is partly attributed to improvements implemented in the processing and calibration of the S5p data, which have reduced the bias and increased the accuracy of the measurements . These refinements demonstrate the impact of continuous updates in satellite data processing on improving the reliability of satellite–ground correlations.

The global results indicate a moderate positive correlation between measurements made by the S5p satellite and surface stations (Table ). This result verifies that there is a general correspondence between both datasets, although it is not strong enough to be considered a good agreement between the data. Similar observations have been reported in previous studies. For instance, found a moderate positive correlation coefficient (r=0.68) between TROPOMI NO2 total column and ground-based measurements in Helsinki.

In contrast, the results obtained by in a study conducted in Madrid reported moderately higher correlation coefficients (r=0.78) compared to those presented in Table  of this study. A notable difference between the two studies is that, while the analysis conducted here observed higher correlations in rural zones than in urban zones, the study in Madrid identified the opposite trend. Finally, in the same study, the authors agreed that the geographic distribution of monitoring stations can significantly influence the correlation coefficients obtained. These differences in results can be directly attributed to the size of the study areas. While their research focused exclusively on the city of Madrid (8028 km2), the present analysis covered the entire region of Catalonia (32 108 km2). This larger area introduces variations in geography and area characteristics that influence the results.

Regression analysis, using S5p values converted to µgm-3, reveals a β1 of 0.0163 and a β0 of 11.5 µgm-3. The low β1 suggests that variations in surface measurements are not proportionally captured by the satellite. The positive β0 implies a bias in the data; even when surface measurements are 0, the satellite predicts a positive value. As with r, where the results indicate a general correspondence, the R2 (Table ) obtained indicates that 43 % of the variance in satellite measurements is explained by surface measurements, leaving 57 % of the variance attributable to other factors not captured by ground-based monitoring. These values are essential for studies that seek to estimate surface concentrations from satellite data. These findings are important for understanding how the correlation between satellite and surface data evolves in response to changes in measured air pollutant levels.

This section analyzes how the correlation between satellite measurements and surface data varies depending on the two groups defined by zones (urban, suburban, and rural) and (BMA and non-BMA). The discussion focuses on interpreting the patterns observed in the correlation coefficients and in the regression adjustments.

The typology-based analysis allows for the assessment of how NO2 concentrations align with satellite data depending on the type of activity in the vicinity of the stations. (Fig.  and Table ).

The traffic data, shown in panels (a) and (b) of Fig. , do not exhibit negative satellite values at the monitoring station locations – similar to what is observed in the urban and BMA datasets (Figs. a and b and a and b). This result highlights the importance of retaining negative data measured by the S5p satellite in the analyses. Although traffic stations display the lowest correlations among typologies, they are still higher than those observed in urban and BMA zones.

In contrast, industrial and background stations show slightly stronger correlations compared to traffic stations, which is consistent with the R2 results (Table ). These findings underline the utility of separating the analysis by station typology to better understand the factors influencing satellite–ground measurement alignment.

The seasonal analysis, summarized in Table , reveals significant variations in the r throughout the four quarters of the year. In general, the highest correlations are observed during the coldest quarters, winter and autumn, with the maximum values reached in the first quarter of both years. This similarity in the results in both years is consistent across most zones and typologies and also indicates the higher correlation between satellite and surface measurements of NO2 during the coldest months of the year. During the second quarter, corresponding to spring, a slight decrease in correlation is observed across all zones and typologies, although it is not significantly pronounced. Similar trends are observed for both zone-based and typology-based stations. In the third quarter, which includes the summer months, correlation coefficients reach their lowest values.

The CV provides additional insight into the stability and variability of these seasonal correlations. Urban and BMA zones exhibit the highest CV values (16.61 % and 31.03 %, respectively), indicating a moderate degree of fluctuation due to environmental and anthropogenic factors, such as increased photochemical reactions. In contrast, non-BMA zones show CVs of 7.46 %, highlighting more stable values across seasons.

Research shows that increased solar radiation during the summer months intensifies the NO2 photolysis process , reducing its concentration near the ground. This seasonal effect is particularly pronounced during summer due to higher solar intensity and longer daylight hours, which enhance photochemical reactions and accelerate the conversion of NO2 into other compounds. Consequently, the lower correlations observed during this period can be attributed to this intensified photolysis and the resultant variability in NO2 concentrations . These seasonal dynamics help to explain the variations in correlations between satellite data and surface measurements across the different quarters analyzed.

As in the zone-based analyses for the entire year, suburban and non-BMA zones have higher correlations than urban and BMA zones, indicating a greater impact of urban-specific factors. Typology-based analysis reveals that background and industrial stations maintain relatively high correlations throughout the year, with CVs of 10.02 % and 11.28 %, respectively. Industrial areas reached their highest correlation in the first quarter of 2023, while traffic stations exhibited slightly greater seasonal variability (CV 11.75 %), with the lowest correlation occurring in the summer months, particularly in 2022, when it dropped to 0.47 (Table ).

The R2 results also vary seasonally (Tables  and ), similar to the r coefficient, with the highest percentages of explained variance recorded in the first and fourth quarters. This trend highlights the stronger relationship between satellite- and ground-based NO2 measurements during the winter and autumn months. The summer months exhibited lower values for β1 and R2, suggesting greater unexplained variance and reduced comparability between datasets. The higher β0 observed in urban zones and during summer and autumn indicates a systematic overestimation of NO2 concentrations by the satellite.

The consistently positive β0 values across all areas and seasons indicate that satellite measurements capture a broader regional signal, integrating contributions from atmospheric layers above ground level. Meanwhile, variability in CV, β1, and R2 values highlights the influence of local environmental conditions, seasonal changes, and the spatial distribution of NO2 sources on satellite–ground comparability.

The strongest correlations are observed at 13:00 UTC – coincident with the S5p overpass – highlighting the need for temporal alignment in any satellite-based estimation model. Correlation strength varies with geographic location and station typology: urban and BMA zones display lower r values and larger seasonal variability than suburban and non-BMA zones, reflecting their greater emission density and environmental complexity.

The diversity of conditions within a satellite pixel can influence the observed correlations. A single TROPOMI pixel may cover diverse land uses, such as urban, suburban, and rural zones, or encompass areas with both point and diffuse emissions. This heterogeneity can reduce the correlation between satellite-derived NO2 values and ground measurements, particularly in complex zones such as urban and BMA zones. In contrast, more homogeneous regions, such as suburban and non-BMA zones, tend to have higher correlations also related to less variability within the pixel. In terms of typological stations, background and industrial stations record higher correlations compared to traffic stations, where local emissions and geographic heterogeneity reduce the representativeness of satellite data.

During winter and autumn, stable atmospheric conditions prevail, with colder air trapped in the lower boundary layers, limiting vertical mixing. This atmospheric stability reduces photochemical reactions, allowing NO2 to accumulate near the surface and contributing more significantly to the observed tropospheric column . As a result, correlations between surface and satellite data are stronger, particularly in background stations and in suburban and non-BMA zones, where lower CVs indicate more consistent seasonal alignment. In contrast, during summer months, atmospheric instability, characterized by anticyclones and higher temperatures, enhances NO2 vertical mixing. This process dilutes NO2 throughout the boundary layer, increases photolysis rates, and reduces the representativeness of ground-level measurements in satellite observations. This seasonal instability introduces greater variability, as reflected in higher CVs for urban and traffic stations.

The results obtained allow us to address the research questions posed at the beginning of this study. Regarding how closely S5p satellite NO2 data correlate with surface measurements from the official measurement of air pollution network in Catalonia, the study found moderate correlations on a global scale (r≈0.65) and highlighted significant geographic variability. Suburban and non-BMA zones showed the strongest correlations, suggesting that satellite data more accurately capture NO2 concentrations in less dense areas. Urban and BMA zones showed lower correlations due to their complexity and the influence of localized emissions. Moreover, the analysis of the correlation between satellite data and station data depending on zone or station type revealed that background and industrial stations align better with satellite data (r≈0.66–0.68) compared to traffic stations (r≈0.61–0.63). This highlights the impact of local emissions and environmental complexity in reducing correlation at traffic monitoring sites.

Regarding the seasonal variation, the analysis showed that winter and autumn reach the highest correlations (up to r=0.70), while summer and spring present lower values. The reduced correlation during summer is attributed to stronger photolysis processes and atmospheric variability, which decrease the representativeness of satellite observations. These findings emphasize the seasonal dependence of satellite–ground agreement and suggest a specific focus on winter data for modeling.

Finally, regarding the benefits and limitations of using S5p data to complement surface-measured NO2 information and how they could be integrated into monitoring strategies and public policies in Catalonia, the results highlight that S5p data provide valuable spatial coverage, especially in suburban zones and outside the BMA zone. These data allow for the identification of pollution trends and regions for prioritizing interventions, as well as the better allocation of resources for installing new ground-based monitoring stations.