The EMEP MSC-W model (hereafter EMEP model) is an Eulerian chemical transport model that can be used on scales ranging from local (e.g., Vieno et al., 2016) to global (e.g., Schwede et al., 2018). Here it is run with 20 vertical layers and a horizontal resolution of 0.1 × 0.1 degrees driven by ECMWF IFS meteorology (https://www.ecmwf.int/en/research/modelling-and-prediction). It includes about 170 reactions between approximately 130 gas- and particulate-phase species and tracers (EmChem16 scheme, see Simpson et al., 2017; Simpson et al., 2020), and uses the EQSAM module (Metzger et al., 2002) to describe equilibria between the inorganic aerosols. To calculate dry deposition, the model includes a stomatal conductance algorithm, applied to all pollutants where stomatal control is important (e.g. ozone, N and S oxides and ammonia). Non-stomatal uptake is also included, and for NH₄ ⁺ and SO₄ ⁻ this is calculated as a function of temperature, humidity, and the molar ratio between S dioxide and ammonia. The model version used here (rv4.17a) is described in Simpson et al. (2012, 2017, and references therein). Evaluation of the model against EMEP data is performed every year as part of the annual EMEP MSC-W reports (https://www.emep.int/mscw/mscw_publications.html), as well as in several publications (e.g. Simpson et al., 2006; Bian et al., 2017; Theobald et al., 2019).

For this study, the EMEP model was run on a monthly basis for the period 2010–2014, The simulations used here are part of the time-series simulations discussed in Tsyro and Mortier (2018). For this study, monthly results from these model runs were extracted for the period 2010–2014. Anthropogenic emissions for these runs were taken from a harmonized data set used for trend analysis, as documented in Tista et al. (2018).

For this comparison exercise, wet and dry depositions of sulfur and nitrogen calculated by the EMEP model were retrieved for each deposition measurement site of both ICP Forests and EMEP/CCC networks. Modeled total deposition is obtained by summing up modeled wet and dry deposition. The ICP Forests measurement sites have been classified into broadleaf and coniferous plots depending on the main tree species. The corresponding land-use specific dry deposition rates from the EMEP for “deciduous” and “coniferous” forests were therefore used for this study.

Annual depositions of NO₃ ⁻, NH₄ ⁺ , and SO₄ ⁼ were obtained by multiplying the volume weighted average concentrations by the annual amount of precipitation. As the deposition of marine aerosol may represent an important contribution to the total deposition of SO₄ ⁼, a sea-salt correction was applied on the annual deposition values, subtracting the marine contribution from S deposition in order to obtain an estimate of non-marine oxidized S deposition (SO₄*). The correction factor was obtained as fractions of the chloride or sodium deposition, using the formulas reported in the manual of the ICP Modelling and Mapping (CLRTAP, 2004). To limit the bias due to local sources of either chloride or sodium, the smaller correction was applied.

The contribution of marine aerosol to SO₄ ⁼ deposition, which may be relevant in coastal areas, is not included in the EMEP model, and the results are directly comparable with sea-salt corrected SO₄ ⁼ deposition provided by the monitoring networks.

The comparison between modeled and measured deposition and between measurement networks was performed both graphically and statistically, using regression analysis and testing the significance of the differences between the bias of the model using a Student’s t-test for independent samples for each combination of substance and sampler. Bonferroni correction (Dunn, 1961) was applied to account for repeating the same test for several combinations.

For each variable and sampling method, modeled deposition was plotted against measured deposition, considering an average deposition value for each site, without distinguishing between sites running for the whole study period (5 years) or for less. In this latter case, we averaged only those years of modeled data that are also available in the measured data on a plot-by-plot basis.

Bias was calculated as the difference between mean modeled and measured values. Percent bias was obtained by dividing bias by the measured value, and multiplying by 100.

Measured deposition data obtained in the same location (i.e., at a distance lower than 10 km) using different samplers were compared with each other. Also in this case, data were averaged using only the years in which all the different samplers were active, and the data obtained were validated.

To stabilize data variance, measured and modeled deposition values were log-transformed before performing regression analysis and the Student’s t-test was applied on percent bias. One NH₄ ⁺ value reported in Iceland as zero because all data were below the detection limits was replaced with the value 0.01 before log transformation.

Samples with a Cook (1977) distance larger than 4/n (where n is the number of data pairs) in the log-log regression were considered as outliers and excluded from the data before statistic treatment, but these outliers are present in the plots, identified with a different symbol in the maps and discussed in the text.

Most calculations were performed in a spreadsheet, Cook’s distance was obtained using the base package in the R statistical environment (R core team, 2020).

The mean annual wet, bulk open field and throughfall deposition of SO₄* measured in Europe span over more than one order of magnitude (Table 1), with average values across time and plots of 2.5, 3.0 and 4.2 kg S ha⁻¹ y⁻¹, respectively. Considering bulk open field deposition, the average values measured in the ICP Forests network are higher than those measured in the EMEP network, on average by 12%, at least in part because most ICP Forests samplers are located in central Europe, in areas with high S emission (Figure 1). After outlier rejection and Bonferroni correction, the Student’s t-test on log-transformed deposition values found significant (p < 0.05) differences between the networks for wet deposition, but not for bulk open field deposition. In the case of wet deposition, the results are similar, but the number of wet samplers in the ICP Forests network is too low to make the comparison meaningful.

In only four ICP Forests plots, wet, bulk open field and throughfall collectors were all running in the same period, and four ICP Forests plots with bulk samplers and 6 with throughfall samplers were located at a distance smaller than 10 km from an EMEP wet sampler. In these sites wet SO₄* deposition showed lower values than both bulk open field and throughfall deposition, which include, beside wet deposition, the amount of dry deposition collected by the constantly open bulk sampler in dry periods and by tree canopies, respectively (Figure 2A).

In 75% of the forest plots where both were measured, SO₄* throughfall deposition was higher than open field bulk deposition, as expected because the former also contains dry deposition collected by forest canopy and washed off during rainfall. However, in the remaining sites, throughfall deposition was slightly lower than open field bulk deposition, most probably because of the low levels of dry deposition coupled with analytical uncertainty. It is also possible that there might be a small direct uptake of S by foliage that becomes apparent at low levels of dry deposition (Staelens et al., 2008).

Sites with the highest SO₄* wet deposition values (Figure 3A) are located in eastern Europe (in Croatia, Slovakia, and Hungary) and in Italy, at the margin of the strongly industrialized Po plain. A high SO₄* deposition value was measured in Montenegro, in a site relatively close (150–250 km) to three large point sources of S emission, namely the Kostolac, Nikola Tesla, and Tuzla coal power plants (Fioletov et al., 2016).

The geographical distribution of high values of bulk deposition of SO₄* (Figure 4A) extends more to eastern Europe than in the case of wet deposition, also including Poland, Czechia, and Slovenia. This different pattern may be partially due to the different distribution of wet and bulk sampling sites: for example there are no wet sampler in some countries where bulk deposition is high, such as Poland, Belgium, or Romania. A smaller area with high S deposition level includes Belgium and some sites in Germany, Denmark, and Norway. A high S bulk deposition value was measured in Serbia, not far from the site where high wet deposition was measured.

The mean annual throughfall deposition of SO₄* showed a larger range than those of wet and open field bulk deposition (Table 1), and the geographical distribution of throughfall deposition (Figure 5A) is similar to the distribution of wet-only and open field bulk deposition, but high levels are also found in Greece and the United Kingdom.

Modeled SO₄∗ wet deposition (Figures 3B, 4B) compared well (r² = 0.61) with measured wet deposition (Figure 2B). After removing two outliers (one in Montenegro and one in Spain), the percent bias of the model is not significantly different from zero, and no significant difference was found among the percent bias for the different networks and sampling periods (Table 2). Apart from one case, the differences between modeled and measured S wet deposition are lower than 2 kg S ha⁻¹ y⁻¹ (Figure 3C). The differences are generally lower than 40%, but higher percentages were found in Spain, in one site in France and one in Montenegro (Figure 3D).

In spite of the high correlation (r² = 0.73 after the rejection of 13 outliers), measured SO₄* bulk open field deposition was generally higher than modeled wet deposition (Figure 2C), and the percent bias (Figure 2D) was significantly different from zero (p < 0.05 after Bonferroni correction) in sites with monthly samples (Table 2), as expected considering that constantly open bulk samplers collect some amount of dry deposition in dry periods.

The differences between modeled wet and measured bulk SO₄* wet deposition were in general lower than 2 kg S ha⁻¹ y⁻¹, but higher differences were found in France, Czechia, Lithuania and particularly in the southern Alps (Italy and Slovenia) a region with strong orographic precipitation and close to large industrial areas (Figure 4C). In spite of the small absolute differences, percent differences were marked: in most of central Europe, modeled deposition was lower than measured deposition by more than 40% of the measured value (Figure 4D).

The comparison between measured throughfall (Figure 5A) and modeled total (Figure 5B) SO₄* deposition showed higher scatter, and 15 outliers were detected. After outlier rejection, a good correlation between log-transformed modeled and measured deposition (r² = 0.68) was found, and the percent bias was low (17%) but significantly different from zero. Large differences were found in central Europe, from England to Poland, and in Greece and Romania (Figure 5C), while high percent differences were also found in region with low deposition values, such as in Spain and in Scandinavia (Figure 5D).

In the case of NO₃ ⁻, the range of the mean annual wet, bulk open field and throughfall depositions is large (from 0.2 to 15 kg N ha⁻¹ y⁻¹, Table 1), with average values of 2.5, 3.1 and 4.7 kg N ha⁻¹ y⁻¹. As for SO₄*, the average values measured in the ICP Forests network are higher than those measured in the EMEP network, on average by 20% for wet deposition and 44% for bulk open field deposition. The difference between the mean log-transformed values for the two networks was significant only for bulk open field deposition (p < 0.001 after Bonferroni correction).

In sites where different samplers were exposed, NO₃ ⁻ wet deposition results were lower than from both bulk open field and throughfall deposition, presumably because of the amount of dry deposition collected by the constantly open bulk sampler in dry periods and by the tree canopy, respectively (Figure 6A). As expected, in 71% of the forest sites where both were measured, NO₃ ⁻ throughfall deposition was higher than open field bulk deposition, probably because of dry deposition collected by forest canopy.

However, in most sites where open field deposition was lower than 2.5 kg N ha⁻¹ y⁻¹ and in a large share of sites where it was lower than 3.5 kg N ha⁻¹ y⁻¹, NO₃ ⁻ throughfall deposition was lower than open field bulk deposition, reflecting the presence of processes decreasing the throughfall deposition of NO₃ ⁻ (“canopy effects”), such as foliar uptake of N compounds (e.g., Garten and Hanson, 1990) and the consequences of metabolism of microorganisms in the phyllosphere (e.g. Guerrieri et al., 2015).

The area with high NO₃ ⁻ wet deposition values (Figure 7A) is smaller than in the case of SO₄*, including Italy, part of France, Switzerland, and Germany and some sites in Sweden and Lithuania. The highest average NO₃ ⁻ deposition was measured in Montenegro, in the same site relatively close to large point sources where the maximum SO₄* deposition was measured, and in Italy, in the southern slope of central Alps, in a site receiving a high amount of orographic precipitation and located close to the Po plain, where most of Italian industry and agriculture are located.

The geographical distribution of high values of bulk deposition of NO₃ ⁻ (Figure 8A) is larger, including Slovenia, Belgium, Czechia, Poland, southern Norway and southern Sweden, and extends westwards to sites in southern France and Spain. High values of throughfall NO₃ ⁻ deposition (Figure 9A) were found in the same area and in Switzerland, Austria, Wales, Denmark and in central Sweden.

As in the case of SO₄*, modeled NO₃ ⁻ wet deposition (Figures 7B, 8B) compare well (r² = 0.72) with measured wet deposition (Figure 6B). However, after removing three outliers (in Montenegro, Italy and Spain), the percent bias of the model still was significantly different from zero (Table 2). In fact, for the higher values of NO₃ ⁻ deposition, modeled wet deposition was higher than measured wet deposition, suggesting a small loss of N from the samplers. or chemical transformation within the sample, or a small model bias relating to high concentrations and/or precipitation amounts.

After the rejection of 19 outliers, measured NO₃ ⁻ bulk open field deposition and modeled NO₃ ⁻ wet deposition showed a high correlation (r² = 0.81) and the percent bias was not significantly different from zero.

The differences between modeled NO₃ ⁻ wet deposition and both measured NO₃ ⁻ wet and bulk deposition were generally lower than 2 kg N ha⁻¹ y⁻¹ (Figures 7C, 8C) but percent differences larger than 40% were common (Figures 7D, 8D).

As in the case of SO₄*, the comparison between measured throughfall (Figure 9A) and modeled total (Figure 9B) NO₃ ⁻ deposition showed a high scatter. In most of the sites, and in particular all sites where throughfall deposition was lower than 2.9 kg N ha⁻¹ y⁻¹, modeled total deposition resulted higher than measured throughfall deposition, suggesting that canopy effects (see below) may also have an effect on NO₃ ⁻. After the rejection of 16 outliers, a good correlation between log-transformed modeled and measured deposition (r² = 0.71) was found, but the percent bias was high (72%) and consistently different from zero, for both coniferous and broadleaved forests and for both fortnightly and monthly sampling (Table 2). Large differences were mainly found in sites in central Europe (Figures 9C,D).

Average NH₄ ⁺ wet, bulk open field and throughfall deposition (3.2, 4.0 and 5.1 kg N ha⁻¹ y⁻¹, Table 1) were larger than the respective values for NO₃ ⁻. As for SO₄* and NO₃ ⁻, the average values measured in the ICP Forests network were higher than those measured in the EMEP network, on average by 48% for wet deposition and 51% for bulk open field deposition. The difference between the mean log-transformed values for the two networks was significant only for bulk open field deposition (p < 0.05 after Bonferroni correction).

As in the case of SO₄* and NO₃ ⁻, in sites where different samplers were used simultaneosly, wet deposition results were lower than both bulk open field and throughfall deposition. Measured throughfall deposition was generally higher than bulk open field deposition, because of the amount of dry deposition collected by tree canopy (Figure 10A), but it was very close to bulk deposition in all sites where the former was lower than 2.1 kg N ha⁻¹ y⁻¹, because of canopy effects, which may completely compensate both surface deposition of ammonia and dry deposition of NH₄ ⁺ collected by the forest canopy and washed off during rainfall on these plots.

The area with high NH₄ ⁺ wet deposition values (Figure 11A) was larger than in the case of SO₄* and NO₃ ⁻, including Germany, Switzerland, northern Italy, and eastern France. As in the case of NO₃ ⁻, the highest average NH₄ ⁺ deposition values were measured in Montenegro and in Italy, in the same sites remarked above. However, the geographical distribution of high values of bulk and throughfall NH₄ ⁺ deposition (Figures 12A, 13A) was large, covering central Europe from France to Poland, and from southern Scandinavia and England to Croatia.

As in the case of SO₄* and NO₃ ⁻, modeled NH₄ ⁺ wet deposition (Figures 11B, 12B) compared well with measured wet (r² = 0.61, Figure 10B) and bulk open field (r² = 0.56, Figure 10C) deposition. However, after removing 4 and 17 outliers, respectively, the percent bias of the model still was significantly different from zero, with modeled values higher than measured values. Modeled values higher than measurements suggests small nitrogen loss from the sampler, or an overestimation of the model, possibly due to the fact that the model provides higher value for the spatial cell (within the grid) in which the measurement site is located because the cell includes locations with locally high NH₃ sources while measurements are performed in more remote areas.

Measured throughfall NH₄ ⁺ deposition (Figure 13A) was lower than modeled total NH₄ ⁺ deposition (Figure 13B) in most sites (Figure 10D), indicating again the presence of relevant canopy effects. However, when throughfall deposition was higher than 8 kg N ha⁻¹ y⁻¹, measured throughfall deposition was higher than modeled total deposition at a number of sites, suggesting that the canopy effects affecting NH₄ ⁺ deposition may be less efficient when deposition values are very high.

After the rejection of 20 outliers, and in spite of a good correlation between log-transformed modeled and measured deposition (r² = 0.67), the percent bias was high (77%) and consistently different from zero, for both coniferous and broadleaved forest and for both fortnightly and monthly sampling (Table 2). As a consequence, large differences between modeled total and measured throughfall NH₄ ⁺ deposition were found almost everywhere (Figures 13C,D).