This upgrade encompasses several significant scientific advancements. Detailed documentation of the CAMS Cy48R1 upgrade for aerosols and reactive gases is available in . More information regarding the aerosol/chemistry changes and the meteorological changes is provided in and , respectively. In the following subsections the model, assimilation and emissions updates are summarised.

The ozone concentrations simulated by the CAMS e-suite and o-suite were evaluated using several surface measurements provided by WMO-GAW, NOAA Earth System Research Laboratories (ESRL), IASOA, the European Environment Agency air quality database (EEA-Airbase), the China National Environmental Monitoring Centre (CNEMC) and the US AirNow services; profile measurements from IAGOS and ozonesondes; and satellite total column observations from the Infrared Atmospheric Sounding Interferometer (IASI).

The comparison with surface observations over Europe (EEA-Airbase), Fig. , shows that MNMB of the e-suite is higher than that of the o-suite by up to 15 %, especially over central and northern Europe, changing some small underestimates into small overestimates. The e-suite temporal correlation improved slightly compared to the o-suite but not everywhere. The bias in the e-suite control run is slightly lower and improved compared to the o-suite control run, in agreement with comparisons with WMO-GAW and ESRL Global Monitoring Laboratory (ESRL-GML) observations.

The surface ozone validation using observations from the China National Environmental Monitoring Centre, Fig. , shows that both the e-suite and the e-suite control runs reduce the negative bias observed for the o-suite and o-suite control runs over the northeastern region extending from Shanghai to Beijing. A slight improvement is also found over the megacity of Guangzhou in the south. This indicates less titration of ozone due to improved NO2 in the e-suite and e-suite control run – see also Sect.  below – which is likely linked to the anthropogenic emission update. The performance of the e-suite remained almost similar to that of o-suite in central and western China. The correlation on average is about 0.75 and is similar for all experiments.

The evaluation of surface ozone using the AirNow ground-based stations over North America, Fig. , shows that the overestimation is slightly higher in the e-suite than in the o-suite (MNMB from +13 % to +22 %), while the temporal correlation (R) and FGE improved, especially for the control run. The higher ozone bias may be related to a slight underestimation of NO2 in the e-suite compared to the o-suite (Fig. S3). The differences between the assimilation and control runs demonstrate the impact of the assimilation.

Surface ozone from the e-suite and o-suite was also compared to WMO-GAW, ESRL-GML and Arctic IASOA surface station observations (Fig. S4). During October 2022 to February 2023, there was no significant change in the surface ozone bias between the CAMS e-suite and o-suite, with relative biases of less than -40 % for most of the stations. An exception is the South Pole Observatory (SPO) station over the Antarctica, where the e-suite shows a higher negative bias (-50 %) than in the o-suite (-40 %).

The e-suite and o-suite have been evaluated with IAGOS aircraft measurements of tropospheric ozone. The time series of the daily profiles at Frankfurt Airport show that the e-suite and the o-suite perform similarly in general (Fig. ). However, in the free troposphere (between 850 and 350 hPa, approximately between 1.5 and 8.5 km) the e-suite control run shows a reduced bias compared to the other runs from both the e-suite and the o-suite from the beginning of the evaluation period until April. For May–June, the e-suite control run develops a negative bias and the results of the other e-suite and o-suite runs are similar. This is more clearly depicted in the time series of the monthly scores by vertical layers at Frankfurt, which shows a better performance (MNMB and FGE) in the e-suite control run in the free troposphere in January–April and the degradation for May and June. The two controls runs show a seasonal difference, which is compensated for by the assimilation. At most airports worldwide, the bias in the lower troposphere (pressure > 850 hPa) is slightly larger for the e-suite than for the o-suite, in particular over airports located in western Africa and eastern Asia (not shown). In the free troposphere, the bias is about the same between the e-suite and the o-suite for most visited airports for the analysis or the 1 d forecast (Fig. S5). For the control run we observe larger differences, and the o-control shows mainly positive biases and the e-control mainly negative biases compared to IAGOS; see also Fig. . These differences are reduced and results are improved by the data assimilation in the free troposphere.

In the upper troposphere (evaluations based on flight level data with a potential vorticity below 2), ozone is overestimated and the results from the e-suite and the o-suite are very similar for all runs (Fig. S6), except for the o-suite control run, which presents slightly larger biases than the other runs over the northern Atlantic and North America.

Tropospheric ozone profiles have also been compared to ozonesonde observations; see Fig. . The Northern Hemisphere profile difference can be considered statistically robust (p value for tropospheric columns is below 0.01), and little difference is seen between the o-suite and e-suite in the troposphere. The e-suite control run shows a large negative bias in the upper troposphere and lower stratosphere (below 100 hPa), while almost no bias is found in the lower troposphere, in agreement with the IAGOS comparisons. For the other regions, the number of profiles is much lower, and the differences between the o-suite and e-suite are mixed: in the tropics and Antarctica, the bias in the lower troposphere is smaller for the e-suite, while the opposite is found in the Arctic and Southern Hemisphere mid-latitudes. The spread in the differences is similar for both runs. Note that the ozonesonde results in the troposphere, including the change in bias around March–April, are consistent with the IAGOS plots. Note that there is no significant change in the number of sondes around this time.

The global maps of monthly mean total column ozone for October 2022 are compared with satellite observations from IASI; see Fig. , top row. Overall, the performance is somewhat improved in the e-suite, with slightly better distributions of the ozone columns compared to the o-suite. A bias feature over the tropical Pacific Ocean in October 2022 has disappeared in the e-suite. This feature may be attributed to the issue with the background covariances in Cy47R3, which was solved on 15 December. The positive bias at low latitudes in October is smaller in the e-suite than in the o-suite. For December 2022, the difference between the e-suite and the o-suite is less prominent. The positive bias above the Pacific Ocean is improved in the e-suite and remains mostly below 5 %. There is hardly a difference between the e-suite and o-suite for June 2023.

The simulated carbon monoxide of the CAMS e-suite and o-suite was evaluated using surface observations from the WMO-GAW, EEA-Airbase, CNEMC and AirNow networks; vertical profiles from IAGOS aircraft and NDACC FTIR measurements; and satellite total column retrievals from MOPITT and IASI.

Over China, the comparison with surface CO observations indicates an overall reduction in CO concentrations in the e-suite, which leads to a reduced positive bias and better performance in the megacities over northeastern China and particularly over Shanghai, Hangzhou and Beijing (Fig. ). The negative bias and FGE increased in the e-suite in the rest of China, including central China. Overall correlation range between 0.3 and 0.8 for most stations, with a mean of 0.53 for the e-suite. The e-suite shows similar correlations to the o-suite. In the densely populated northeastern part of China, correlations exceed 0.6 over most stations. The reduction in CO can be linked to the emission update in Cy48R1.

Compared to surface CO observations from five WMO-GAW stations (Hohenpeißenberg, Jungfraujoch, Sonnblick, Zugspitze and Monte Cimone, located in Europe), there was no significant change in the bias between CAMS e-suite and o-suite runs, which does not exceed ±10 % in general (Fig. S7). The correlation for the e-suite has slightly improved compared to the o-suite over Europe and the Cabo Verde station in the tropics.

The comparison of the CO mixing ratios to EEA-Airbase observations over Europe shows that the e-suite performs better than the o-suite in terms of bias and correlation over most stations (Fig. S8). In all cases note that e-control scored better than o-control as well; hence the detected improvements at the surface mainly originate from the model changes rather than assimilation changes.

The e-suite has been evaluated with IAGOS measurements of CO (Level 1 data) at different airports. The time series of the daily profiles (curtain plots) of MNMB at Frankfurt Airport are presented in Fig. . The e-suite shows a smaller negative bias than the o-suite at all altitudes for the analysis and the 1 d forecast. This is also shown in the time series of the monthly scores by layers at Frankfurt, with an analysis and forecast improvement for the e-suite in both the lower troposphere (LT) and the free troposphere (FT) (Fig. S9). However, CO is still underestimated by both the e-suite and the o-suite, with a larger bias in the LT than in the FT. According to the fractional gross error (FGE) monthly values, the e-suite improvement over the o-suite is about -0.05 for most of the months in the LT compared to about -0.02 in the FT, while for correlation results, no notable difference is found (Fig. S9). The control runs from the e-suite and o-suite show a different seasonal pattern of the bias with a notable increase in the bias starting in early spring for both models. Like for the assimilated runs, the bias from the control run is smaller for the e-suite than for the o-suite but only until March. This can also be seen in the monthly scores time series (Fig. S9). For the remaining months the performance is similar for both control runs in the lower troposphere and slightly better for the o-suite control run in the free troposphere. For all other airports worldwide, the bias (MNMB) in the low troposphere is very similar for the e-suite and the o-suite. However and like at Frankfurt, at most visited airports the absolute differences (FGE) are better in the free troposphere for assimilated runs of the e-suite compared with those of the o-suite (Fig. S10). Regarding the control run results, they are similar for both models. In the upper troposphere, at cruise altitude, CO is underestimated by both the e-suite and the o-suite runs. Like in the free troposphere, there is a clear improvement in the bias in the upper troposphere over all regions in the e-suite compared to the o-suite and in particular for the analysis (Fig. S11).

The results of the comparison with satellite CO column retrievals from the IASI and MOPITT instruments are shown in Fig. . Averaging kernels have been used in these comparisons. The IASI total column CO observations are well reproduced by the CAMS e-suite in terms of absolute amounts and spatial distribution. The e-suite relative bias for December 2022 with reference to IASI stays mostly within 20 %, with wide areas below 5 %, some negative bias above the southern Africa biomass burning area (up to 20 %), a positive bias around 60° S and a negative bias over Antarctica. The e-suite performs much better than the e-control, which shows negative biases for the major part of the globe up to 20 % regularly and peak amounts above South Africa exceeding 40 %. The o-suite mostly has a negative bias compared to IASI, up to 20 % regularly and up to 30 % over the Pacific. Rows four and five of Fig.  show the e-suite relative biases for March and June 2023, which support the previous findings also for the other seasons. The MOPITT comparison looks very different, as overall biases are smaller for the o-suite than for the e-suite. The e-suite shows widespread positive biases of up to 20 % which are not present in the o-suite. Only the negative biases of the o-suite of up to 20 % above northern Africa improved to remain mostly below 10 % in the e-suite. Similar observations are made for the additional months from different seasons. These results for CO total columns demonstrate changes in the bias correction implemented in the e-suite, which now uses IASI-C and TROPOMI as reference. Note that TROPOMI CO data were not assimilated in the e-suite until 28 April, so the impact of TROPOMI has not been explicitly evaluated here.

In 2023, Canada suffered from a very extreme fire season, which started as early as May, with intense fires in June, and continued during the summer. The e-suite shows pronounced enhancements in CO in June, linked to the large GFAS fire emissions over Canada inserted in the model. As shown in Fig. , the amount of CO produced is underestimated by up to 20 % compared to IASI, and the largest relative error in the global map is located over Canada in June. However, the e-suite compares well with MOPITT over Canada, demonstrating the differences between these two satellite products. Negative CO biases over Canada compared to IASI (but positive compared to MOPITT) are also found in July and August, as shown in the JJA-2023 validation report available in .

Over the Arctic Sea we observe a large region with negative biases in March compared to IASI. Note that IASI data have less sensitivity in spring as compared to summer above the Arctic due to lower thermal contrast, which may explain part of the differences. The MOPITT results seem to indicate a similar underestimate, so a negative bias in the analysis is likely.

Comparisons were also made against CO partial columns and profiles from the FTIR instruments, part of the surface remote sensing network, NDACC (Fig. ). The scores for 13 NDACC stations are presented in Tables S1 and S2 in the Supplement. For tropospheric CO columns, the bias for the e-suite is reduced compared to the o-suite for almost all sites. Although correlations are similar for the o-suite and e-suite, the ratio of the standard deviation in the tropospheric columns for the e-suite and the FTIR time series is higher compared to the o-suite. The bias at the tropical sites has switched sign and is now positive. For stratospheric CO columns, the bias for the e-suite is reduced significantly for the southern hemispheric sites and is now of the order of the measurement uncertainty. The Southern Hemisphere correlations are also higher for the e-suite (in particular for the Antarctic Arrival Heights site). For the Northern Hemisphere, the e-suite and o-suite perform similarly. The o-suite has a degraded performance during this period in the Southern Hemisphere (Wollongong and Lauder stations), where the tropospheric columns show an increased negative bias compared to the months before October 2022 and the stratospheric columns show a strong (+20 %) positive bias. A direct comparison between the o-suite and e-suite should therefore be interpreted with care. The figure shows that for the e-suite, a reduced positive bias remains in the stratosphere for this period. This can be considered an improvement compared to the o-suite performance prior to October 2022 (negative bias < -20 %) and past October 2022 (positive bias > +20 %).

As of 22 May 2023, wildfires had been raging across Canada's Alberta Province for 3 weeks. IAGOS aircraft profiles of CO over Montréal and Calgary picked up the pollution plumes from these events and have been compared with the CAMS results (Fig. S12). For most profiles studied during this event, the e-suite performed better than the o-suite, with values of the CO mixing ratios higher than those provided by the o-suite, although the forecast model still faces difficulties in capturing the high concentrations in the plumes, while the altitude is simulated with success.

The CAMS configurations were also compared with surface observations in China, Europe and North America.

Over China (Fig. ), the e-suite and e-suite control run significantly reduce the Cy47R3 positive offset in surface NO2 over most regions in eastern China, which also leads to a reduced bias in surface ozone as discussed in Sect. . The correlation of the e-suite and e-suite control run is similar (slightly reduced) to that of the o-suite and o-suite control run. The mean correlation coefficient averaged over all stations is 0.52 for the e-suite and 0.54 for the o-suite.

The comparison with surface NO2 observations in Europe obtained from EEA-Airbase shows that the e-suite performs better than the o-suite in terms of bias for a majority of the stations especially in central and northern Europe (Fig. S13). Interestingly, this coincides with the regions where a slightly higher bias was found for ozone in the e-suite, indicating that a chemical regime changes in the model. In terms of correlations, the e-suite and o-suite performed almost equally.

The evaluation of surface NO2 using the AirNow ground-based stations over North America shows that the underestimation in e-suite and e-suite control run is higher than in the o-suite and o-suite control run (from about -30 % to -40 %), while the temporal correlation (R) slightly improves (from 0.49 to 0.54) (Fig. S3).

The CAMS e-suite Cy48R1 tropospheric NO2 column data are compared to the TROPOMI scientific IUP, University of Bremen (IUP Bremen), tropospheric NO2 product; see Fig. . Three different months are selected for performance demonstration, including those periods with the highest and lowest solar elevation for the two hemispheres. Overall, the e-suite results correspond well to the observational data in terms of spatial and temporal variations and absolute amounts. However, there are also apparent differences between model and measurements. The largest differences are mainly observed above regions with major emissions. Locations strongly affected by anthropogenic pollution such as hotspots in China can show strong positive biases partly above 100 %. Relative to the o-suite, however, the e-suite performs better compared to TROPOMI especially above strongly polluted regions such as eastern China. Here, the positive bias in the o-suite is reduced in the e-suite as well as above other hotspot regions. This effect is most visible in the December maps (left column of Fig. ). This improvement is likely due to the updated emissions used in Cy48R1. Most areas with strong positive biases are smaller in their spatial extent and smaller in their bias values in the e-suite compared to the o-suite. In addition, there are a few confined regions where the e-suite shows positive biases not present in the o-suite comparison such as above the North Sea. This can mainly be explained by the lower anthropogenic emissions in CAMS-GLOB-ANT v5.3 as applied in the e-suite, compared to CAMS-GLOB-ANT v4.2 in the o-suite.

Some typical biomass burning areas such as southern Africa or South America show negative biases around 40 % compared to TROPOMI. Negative biases are seen also above parts of the USA, the northern Atlantic, northern Pacific and parts of Asia. In contrast, positive biases are found over boreal fires. No noteworthy degradation in performance is seen when moving from the previous o-suite to the e-suite results. This is also true concerning the background regions: both the e-suite and the o-suite perform well in representing background values close to zero. The negative biases for some oceanic background regions are slightly larger for the e-suite than for the o-suite. The negative biases above the oceans on the Northern Hemisphere show up in areas where the absolute tropospheric NO2 column is small and close to background levels. Slight differences in absolute amounts thus cause comparably large relative bias values. Overall, the performance of the e-suite is slightly better than that of the o-suite.

The comparison to SO2 surface observations from the China National Environmental Monitoring Centre shows that the e-suite–e-suite control run significantly reduces the positive bias observed for the o-suite–o-suite control run over most regions in eastern China, as indicated by the difference in FGE between the e-suite and o-suite (Fig. ). However, a high overestimation (> 50 %) is still observed in the e-suite for many of the stations, indicating a remaining possible overestimation of SO2 emissions in the updated CAMS_GLOB_ANT inventory. Note that the temporal correlation (R) is slightly reduced in the e-suite compared to the o-suite.

The comparison of the SO2 mixing ratio against EEA-Airbase observations over Europe shows that the e-suite–e-suite control run performs better than the o-suite–o-suite control run in terms of bias for most stations (Fig. S14). The correlations are almost equal.

The tropospheric formaldehyde (HCHO) column from the CAMS e-suite is compared to the TROPOMI scientific IUP Bremen tropospheric HCHO product . Monthly mean global maps and differences are shown in Fig. .

The e-suite overestimates the tropospheric HCHO columns over South America and northern Australia by more than 60 % in December and March, and there is also an overestimation over Indonesia. The allocation of the HCHO columns over northern and southern Africa in the two months is well reproduced by the model, but there are biases in certain time periods, such as the overestimation north of the Equator and an underestimation in the south for June 2023.

In comparison to the o-suite, the overestimation above South America and northern Australia is stronger in the e-suite, visible in the relative biases of the e-suite and the o-suite in rows four and five of Fig. . This could be the effect of the updated isoprene chemistry, which induced higher HCHO production. Above northern Africa, the HCHO values show a larger positive bias, while over southern Africa the negative bias is slightly less pronounced in the e-suite. Overall, the performance is similar but slightly degraded for the e-suite.

Stratospheric ozone from the CAMS e-suite Cy48R1 and o-suite have been compared against satellite limb profiles and ozonesondes, considering the analyses, the fifth-day forecasts and the o-suite control run; see Fig. . The full e-suite period is considered, from 1 October 2022 to 27 June 2023.

The main conclusions are as follows:

In the lower stratosphere, at pressures higher than 10 hPa, the e-suite agrees better with the observations than the o-suite, with lower biases, lower FGE and higher correlations for the 9-month period evaluated.

Stratospheric ozone has also been compared to lidar profile observations from NDACC for the full e-suite period of 9 months; see Fig. . Above 10 hPa, the e-suite mean profile deviates from the o-suite, showing a negative bias compared to the lidar observations that increases with altitude. Between 50 and 10 hPa, the e-suite overestimates the ozone concentration. The negative bias feature observed around 20 hPa in the o-suite at Mauna Loa is no longer present in the e-suite. The dispersion (standard deviation, SD) in the differences is slightly, but systematically, improved for the e-suite. In general, the conclusions for the lidar comparisons are very similar to those for the satellite comparisons.

Other species simulated in the stratosphere in the e-suite have been compared with satellite limb profiles considering 3-hourly first-day forecast. The new CAMS cycle 48R1 introduces full stratospheric chemistry, and meaningful comparisons with the o-suite are not possible. However, reasonable concentrations of these species are important because of their impact on ozone. The species considered are as follows: CCl4, CFC-11, CFC-12, HCFC-22, ClO, ClONO2, HCl, CH4, H2O, HNO3, N2O, N2O5, NO2, NOx (NO+NO2) and O3.

There is a relatively good agreement between the e-suite and observations for long-lived species (CCl4, CFC-11, CFC-12, CH4, HCFC-22 and N2O) and HCl between 10 and 200 hPa; see Fig. . The chemical H2O tracer in the CAMS e-suite shows a negative bias. Even though not perfect, CAMS e-suite NO2 (and NOx) agrees relatively well with observations, which was not the case in the previous o-suite cycles. Above the 10 hPa level, the concentrations are overestimated compared to the Atmospheric Chemistry Experiment Fourier Transform Spectrometer (ACE-FTS), and there is room for improvement. For ClO, the agreement is good with MLS and less good with ACE-FTS, but this dataset is supposed to be less reliable than MLS for this species.

Stratospheric NO2 columns have been compared with TROPOMI retrievals; see Fig. . The observational NO2 stratospheric column is represented well by the e-suite in terms of absolute amounts, latitudinal variation and temporal changes. Over the largest part of the globe, column amounts agree to within 2×1015 molec cm-2, deviating by less than 10 %. The e-suite performs well in reproducing the general strong increase in stratospheric NO2 at high latitudes in the summer hemisphere.

While no changes to the UV code were implemented in Cy48R1, the surface UV index (UVI) forecasts are affected by the changes in the optical depth of trace gases (most notably stratospheric ozone), aerosols and thick clouds. The net effect of these factors on the hourly forecast performance of the new model cycle was estimated using ground-based UVI measurements as the reference data at 38 stations, located in Europe, Israel, Thailand, Australia, New Zealand and the Antarctic.

Figure  illustrates the statistical improvement of Cy48R1 compared with Cy47R3 when evaluated against the ground-based observations in terms of MNMB, FGE and R. MNMB and FGE included all available hourly UVI forecasts, while R included only UV forecasts close to local noon in order to emphasise the importance of atmospheric composition (instead of the solar zenith angle, SZA) and to highlight the time of day when UV radiation is typically most intense and hazardous. Overall, the statistics (mean MNMB = 5 %, mean FGE = 0.30 and mean R = 0.82) indicate that CAMS UV forecasts are of good quality in Cy48R1. The mean values of R and FGE indicate no significant changes in correlation and scatter between the old and the new model cycles; however the mean MNMB increases slightly from 2 % to 6 %. This finding is supported by comparing the zonal mean of daily maximum UVI between the two model cycles, which increased by +0 % to +4 % in Cy48R1 compared to Cy47R3, with the largest increase of +2 % to +4 % occurring between 50° S and 60° N.

Cy48R1 introduces a redistribution of aerosol optical depth per species (Fig. , first column). Starting from Cy48R1, the secondary organic aerosol (SOA) optical depth is now provided separately, whereas prior to this update, the SOA optical depth was included as part of the organic matter (OM) optical depth. For this report, the OM optical depth in the e-suite includes the SOA in order to ensure comparability with the OM optical depth in the o-suite. Compared to the o-suite, there is less AOD for the e-suite, particularly due to a reduction in sulfate optical depth (Fig. , second column). For the e-suite, there is less black carbon optical depth, particularly over central Africa. Over Southeast Asia and the Sahel, nitrate optical depth decreases by about 50 % in the e-suite compared to the o-suite. Contrarily, the sea salt and ammonium optical depth increases, while dust and organic matter changes are regionally dependent. The Ångström exponent (AE), a parameter which is indicative of the aerosol size distribution, is considerably lower globally (reduction by -0.12) in the e-suite, especially over the Middle East and Sahara, while it increased over South Africa, South America and Australia.

These changes are related to (1) the introduction of two new secondary organic aerosol tracers (anthropogenic and biogenic) along with their respective precursor gas tracers (2) modifications to dust emissions and removal simulation that increased the global dust mass burden by a factor of 2, (3) a review of aerosol optical properties for dust and brown carbon, and (4) improvements to secondary inorganic aerosol simulation. Note that the AOD and AE mostly increase due to assimilation, except for dust (Fig. g, h, j).

The evaluation of daily aerosol optical depth (AOD) and the Ångström exponent (AE) against the ground-based network of AERONET version 3 Level 1.5 stations shows that e-suite overestimates less compared to the o-suite for both parameters (Figs.  and ). The e-suite improves in terms of AOD MNMB globally (from +21 % in the o-suite to +7 % in the e-suite), especially over North America, Europe, East Asia and the Middle East. Over Southeast Asia and the Sahara, the MNMB performance remains almost unchanged. Notably for South Asia, a deterioration in performance (higher underestimation) is observed compared to the o-suite. The correlation (R) performance remains unchanged (approximately 0.70 in both the o-suite and the e-suite). Note that both control runs underestimate AOD (not shown).

The AE improves in terms of global bias (mean bias goes from +0.28 in the o-suite to +0.19 in the e-suite), with regional improvements in Europe and North America as well as East China (Fig. ). The e-suite updated dust emissions and deposition fluxes, resulting in a coarser aerosol size distribution (smaller AE) and dust mass burden over the deserts in the e-suite control run. Results specifically related to dust changes are discussed in Sect. . The correlation (R) of AE improves globally (from 0.34 in the o-suite to 0.44 in the e-suite) and remains unchanged in Europe and East Asia, with a small improvement in South Asia, the Sahara and the Middle East and a considerable improvement over South America and North America, though R still remains below 0.5 in the latter case .

Global daily near-real-time data of particulate matter, under 10 and 2.5 µm (PM10 and PM2.5, respectively) in diameter, from the surface observational networks AirNow (North America, U.S. Environmental Protection Agency, EPA), EEA-Airbase (Europe, European Environmental Agency, EEA) and CNEMC (China National Environmental Monitoring Centre), are used to evaluate e-suite and o-suite experiments for the period October 2022 to June 2023 (Fig. ). Note that all stations are considered for the AirNow and CNEMC networks, including urban stations; hence CAMS may not be able to fully capture the very high PM2.5 and PM10 measured with observations at a local urban scale due to the coarse spatial resolution of the model. In this section we focus on the evaluation results for PM2.5. The results for PM10 were found to be very similar to those of PM2.5. The PM10 results are presented on the AeroVal website and are discussed in .

Over North America and Europe, the e-suite improves PM2.5 in terms of MNMB (-3.2 % and 8.9 %, respectively) compared to the o-suite, which exhibited a small overestimation of about +9.5 % and +16.2 %. In contrast, over China MNMB of the e-suite (-19.9 %) exhibits a higher and negative bias when considering the performance over all available stations compared to the o-suite (+3.8 %). The spatial distribution of MNMB over China forms a clear dipole pattern. The eastern side, which encompass most of China's megacities, mostly overestimates PM2.5, while in the western part, which contains less populated, high-altitude regions, PM2.5 is underestimated. The PM2.5 of the e-suite outperforms that of the o-suite in the eastern part of the country, with MNMB lower than 25 % at most stations, while the opposite is the case for the western part, where the e-suite displays MNMB that is more negative than -50 % at most stations. The measured PM2.5 over China displays a peak in January, which is strongly underestimated by both the e-suite and the o-suite (Fig. ). The correlation for the e-suite and o-suite remains unchanged in Europe (about 0.47), slightly improves over China (from 0.55 to 0.57) and is better in North America (from 0.39 to 0.55). The reduction in sulfates (Fig. ), the reduction in the PM2.5 positive bias (Fig. ) and the reduced positive bias of SO2 over China (Fig. ) are linked to the reduced SO2 emissions in the e-suite.

AOD coarse (AODc) values of the e-suite and o-suite were evaluated against the AERONET spectral deconvolution algorithm (SDA) version 3 Level 1.5 daily data (Fig. ). Overall, the e-suite AODc performs better than the o-suite AODc in terms of FGE globally (from 1.09 to 0.77) and for all regions (except the Middle East and the Pacific–Australia–New Zealand). The same applies to MNMB and R (not shown). Since the AE is indicative of the aerosol size distribution, its mean absolute bias (MAB) reveals similar results.

Over and around arid areas, AODc is represented mainly by aerosol dust. The new dust emission increased dust optical depth (DOD) over the Middle East and increased/decreased the DOD over the northern/southern part of the Sahara (Fig. e). Over northern Africa, the increasing dust concentration in spring, which increases AODc and decreases AE, is represented slightly better by the e-suite compared to the o-suite. In contrast, in the Middle East the e-suite displays too high an AODc (too low an AE) compared to the observations. In particular, the simulated AODc and AE of the e-suite are better than those of the o-suite over AERONET stations that are located over the westward transport of Saharan dust in the North Atlantic (e.g. Cabo Verde), as well as over other regions that are affected less by dust (e.g. southern Europe) .