Since 2009, the Air Quality Model Evaluation International Initiative (AQMEII; Rao et al., 2011) has focused on evaluating regional-scale air quality models used for research and regulatory applications. The goal of AQMEII is to conduct coordinated research projects and model inter-comparisons to advance model evaluation practices and inform model development. This initiative is promoted by the European Commission Joint Research Center, the U.S. Environmental Protection Agency (EPA), and Environment and Climate Change Canada and involves the regional-scale air quality research communities active in both North America and Europe.

AQMEII has been executed in phases that each focused on a critical aspect of modelling systems. The phases were conducted as multi-model comparisons that were analysed through the organization of common modelling activities and supported by gathering specific monitoring data needed to evaluate model performance. Each of the phases required developing innovative evaluation and data reconciliation techniques to provide scientific insight across disparate modelling systems. AQMEII phase 1 provided the first detailed annual ensemble comparison of air quality model predictions for North America and Europe (Galmarini et al., 2012). AQMEII phase 2 examined the impacts of feedbacks between air quality and weather on forecasting skill and identified the key sources of uncertainty in feedback model forecasts (Galmarini et al., 2015). AQMEII phase 3, in collaboration with the Task Force on Hemispheric Transport of Air Pollution (TF HTAP) (http://www.htap.org, last access: 7 October 2021), studied the effects of intercontinental transport on regional air quality predictions (Galmarini et al., 2017). Details and findings of the past three phases of AQMEII can be found in journal special issues dedicated to these activities (Galmarini et al., 2012, 2015, 2017). The AQMEII initiative is based on the four pillars of model evaluation described by Dennis et al. (2010): operational, diagnostic, dynamic, and probabilistic evaluation, which will be partly described hereinafter.

This fourth phase of AQMEII (AQMEII4), detailed in this special issue and introduced by a pair of technical notes, focuses on the processes of wet and especially dry deposition, including the parameterized approaches used within current air quality models, and how these approaches and the details of their implementation influence model predictions and performance across multiple modelling systems. Deposition is critical to the life cycle of a pollutant, as it regulates the rate of pollutant removal from the atmosphere and determines the net flux of that pollutant to the earth's surface. This latter point is particularly important when the pollutants have a known deleterious effect on ecosystems (e.g. the deposition of acidifying compounds to aquatic ecosystems, or the dry deposition of ozone on vegetation). By affecting the pollution remaining in the atmosphere, deposition estimates also modulate predictions of ambient pollutant concentrations that affect human health through inhalation exposure.

Deposition has only been peripherally investigated in past phases of AQMEII. The operational evaluation of air quality models, in which modelled concentrations are directly compared to monitoring network observations, quantifies the extent to which an air quality model meets expected performance. However, operational evaluation does not provide the process-level understanding of the extent to which the performance results from correct representation of model physical and chemical processes. In this context, dry and wet deposition are key processes within air quality models because they represent removal, which can affect the concentrations of key atmospheric species. Several past AQMEII publications were dedicated specifically to wet and dry deposition (Vivanco et al., 2018; Hogrefe et al., 2020; Solazzo et al., 2018). However, only wet deposition fluxes could be evaluated against observational data in these papers. The causes of differences in model predictions for dry deposition were not determined. Some of the studies performed within AQMEII also addressed dynamic evaluation (i.e. the performance of a model in capturing changes in concentrations or deposition fluxes when subjected to variations in meteorology or emissions). The effects of these variations on deposition were therefore investigated, but without analysis at the process level on the extent to which the details of deposition algorithms influenced model performance.

Recent studies of dry deposition of ozone have been fuelled by the need to quantify impacts on global to regional water and carbon cycles (Lombardozzi et al., 2015; Oliver et al., 2018), vegetation damage including crop yields (McGrath et al., 2015; Emberson et al., 2018; Schiferl and Heald, 2018; Hong et al., 2020), and ozone air pollution (Andersson and Engardt, 2010; Silva and Heald, 2018; Baublitz et al., 2020). In particular, reduced stomatal dry deposition of ozone during droughts may contribute to high-ozone-pollution episodes (Vautard et al., 2005; Solberg et al., 2008; Emberson et al., 2013; Huang et al., 2016; Anav et al., 2018; Lin et al., 2020). Dry deposition of ozone occurring through nonstomatal deposition pathways, on average 45 % of the total (Clifton et al., 2020a), has also been shown to be more variable and more important than predicted by current chemical transport models, with implications for background and extreme ozone pollution (Clifton et al., 2017, 2020b). Previous intercomparisons at the global scale suggest large differences in simulated ozone deposition velocities with implications for the simulated tropospheric ozone budgets and the models' ability to quantitatively capture the drivers of recent trends and interannual variability in observed ozone pollution (Hardacre et al., 2015; Wong et al., 2019). However, process-oriented evaluation in regional to global models is missing, in large part because key process-oriented diagnostics have not been archived and different land use–land cover (LULC) inputs across models have inhibited the systematic elucidation of processes driving the noted differences (Hardacre et al., 2015; Clifton et al., 2020a). One way in which discrepancies between observed and modelled deposition has been addressed is through model–measurement fusion approaches (Schwede and Lear, 2014; Makar et al., 2018; Robichaud et al., 2020a, b). Such approaches could benefit from an improved characterization of process-level uncertainty in modelled dry deposition.

Despite the great advancements in regional-scale air quality modelling, the primary schemes used for dry and wet deposition in today's models originated in the 1980s and 1990s. Moreover, while the role of deposition as a persistent sink has been known for a long time (e.g. Chang et al., 1987; Irving and Smith, 1991; Borrell and Borrell, 2000), its relative importance in regulating trace species budgets has become more prominent in recent years as the magnitude of the anthropogenic emission source term has generally decreased. The evaluation studies performed within AQMEII (e.g. Solazzo et al., 2017; Hogrefe et al., 2018) and other recent work reaffirmed that deposition is a process of paramount importance within an air quality model (e.g. Knote et al., 2015; Huang et al., 2016; Beddows et al., 2017; Matichuk et al., 2017; Campbell et al., 2019; Sharma et al., 2020) with consequences of primary relevance in a number of sectors (human health, agriculture, forestry, hydrology, soil management, ecosystem management). Thus, there is renewed focus on better characterization of this term and its magnitude.

All the above points were the motivation to make use of the AQMEII community and evaluation infrastructure to construct an AQMEII phase dedicated to deposition. This phase was designed to compare deposition predictions from multiple regional models by isolating specific deposition pathways across multiple modelling systems and across multiple LULC classification systems using common diagnostic tools. Analysing dry deposition of gaseous species, especially ozone and nitrogen species, is a particular focus, as is quantifying the range of model predictions for acidifying wet and dry deposition. A process-level diagnostic intercomparison of particle dry deposition is not conducted here due to the complexity added by model-to-model differences in the representation of aerosols (size and composition) themselves. We also note that some previous work (e.g. Makar et al., 2018) suggests that the impact of particle deposition on total nitrogen and sulfur deposition is relatively small, although particle deposition is the main source of base cations transferred from the atmosphere to ecosystems. However, more recent work (Saylor et al., 2019; Emerson et al., 2020) suggests that particle dry deposition algorithms used in current modelling systems are highly uncertain, suggesting a need for performing further process-level diagnostic intercomparisons.

AQMEII4 has the following research goals.

Quantify the performance and variability of dry and wet deposition fields simulated by multiple state-of-the-science regional air quality models.

Most model dry deposition schemes are derived from Wesely (1989). However, their implementation in regional and global models has considerable variation (a comparison with global models may be found in Hardacre et al., 2015). Specifically, most schemes follow the parameterization structure used by Wesely (1989) but may differ in the details of their representation of individual parameters and processes. This is discussed in more detail in Sect. 3.

In addition, dry deposition algorithms require, as a key input, information on LULC and vegetation. It is therefore important to determine how the deposition modules themselves work, both as stand-alone physical descriptions and within a regional air quality model. AQMEII4 has been organized as two parallel activities to address the research goals outlined above. AQMEII4 Activity 1 (introduced in this technical note) focuses on the detailed diagnostic comparison of predictions of air quality model deposition fields, along with evaluation of model concentration and wet deposition flux performance at routine monitoring stations in North America (NA) and Europe (EU). Activity 2 (introduced in separate special issue publications) evaluates only the dry deposition schemes used in air quality models, and other models used for impact assessments, such as zero-dimensional single-point models, driven by observed meteorology, biophysics, and ecosystem characteristics, at specific sites across the Northern Hemisphere where ozone flux measurements have been collected continuously over at least a year, with many datasets spanning 3 years or more. AQMEII4 will provide the most comprehensive analyses yet performed on dry deposition schemes, since the schemes will be tested both within and independently from the air quality model, under controlled conditions, and when subjected to variable meteorological and surface characteristic conditions. The single-point modelling component allows a very detailed analysis of how ozone dry deposition is modelled; recent work comparing five deposition algorithms at a single site (Wu et al., 2018) here has been extended to multiple sites with additional deposition algorithms and takes advantage of a new collection of ozone flux measurements at sites around the Northern Hemisphere and new process-oriented diagnostics.

This technical note is designed to summarize all relevant information that constitutes the set-up and organization of AQMEII4 Activity 1. Its intent is to provide both the readers and authors of this special issue with a common reference for the description of the AQMEII4 aims, scientific protocols, and analysis approaches; the model reporting framework; the model input data and monitoring data used for model evaluation; and the descriptions of the model deposition algorithms themselves. By serving as common point of reference for the individual studies undertaken through AQMEII4 Activity 1, the technical note reduces the need for repetition of background material by individual study papers, which allows these papers to focus on specific analyses and the presentation of the results. It also allows the reader to access all relevant background material in a single location rather than spread out over several papers. Because of this design, this technical note should not be viewed as a stand-alone scientific paper as it does not contain any results but rather as laying the groundwork for subsequent scientific papers contributed by Activity 1 modelling groups to the AQMEII4 special issue.

Activity 1 like the previous phases of AQMEII includes the evaluation of regional air quality model simulation on the NA, EU, or both domains for at least a 1-year period. Prior to describing the requested output that pertains strictly to dry deposition, we briefly summarize the modelling periods and domains, common inputs, and standard concentration, meteorology, and wet deposition outputs for Activity 1 in this section.

Analysis of dry deposition is the focus of AQMEII4. In particular, AQMEII4 intends to go beyond an operational evaluation of ambient concentrations and comparison of total deposition across models because this approach does not provide enough information to determine the causes of different deposition totals among regional models. The novelty of AQMEII4 is that we request additional and very detailed diagnostic-evaluation outputs related to dry depositional from all of the models. With these very detailed outputs, we can compare the important elements of the model machinery and understand model differences.

Many regional models use the Wesely (1989) dry deposition scheme, but several variants have been developed and implemented with different levels of sophistication. Dry deposition schemes are mostly resistance frameworks – by framework, we mean the structure of the scheme with respect to how processes relate to one another – and all of the regional models in AQMEII4 use resistance frameworks for dry deposition. Resistance frameworks are based on the representation of series and parallel resistors in electrical circuits. Differences in resistance frameworks across regional models imply that comparing a given process among the regional models is not straightforward. Thus, diagnostic variables that account for differences in resistance frameworks need to be reported. Below, we present the strategy devised to reduce any dry deposition scheme to the essential set of comparable variables regardless of the differences in the frameworks of the schemes that generated them.

We start with a description of the Wesely (1989) resistance framework, one of the earliest literature examples of a resistance framework for dry deposition and arguably the most popular dry deposition scheme, and follow with both generic and specific examples of other resistance frameworks as a guide to the AQMEII4 output protocol. The components of the deposition velocity are process-based resistances (units are s cm-1) that impede the transfer of mass to a variety of surfaces. Resistances are added in series for processes operating on the same depositional pathway, and in parallel when multiple surfaces for dry deposition exist. In the original Wesely (1989) scheme, four deposition pathways were used: soil, “lower canopy and exposed surfaces”, leaf cuticles, and plant stomata. Gases are first impeded by an aerodynamic resistance to deposition (ra), second impeded by a quasi-laminar sublayer resistance (rb), and third impeded by a bulk surface resistance term (rc) composed of a parallel summation of the resistances associated with each pathway. The three impedances to deposition are added into a total resistance, the inverse of which is the deposition velocity of the gas (units = cm s-1): 1vd=(ra+rb+rc)-1. The bulk surface resistance (rc) in Wesely (1989) follows 2rc=rs+rm-1+rlu-1+rdc+rcl-1+rac+rgs-1-1. The component resistances used in rc are defined in Fig. 2, which is a schematic of the Wesely (1989) resistance framework.

Work subsequent to Wesely (1989) also uses the resistance approach, but sometimes with considerable variation in the resistance framework, the number of surfaces to which dry deposition occurs, and/or the processes represented by individual resistances. Several motivating factors likely led to the development of a diversity of resistance frameworks. In the intervening years subsequent to Wesely's introduction of the resistance framework concept, new measurement capabilities (for higher-time-resolution information, for greater chemical speciation, higher-precision measurements) allowed the original algorithms to be tested and modified. Developments in plant physiology understanding have also resulted in improved stomatal resistance parameterizations. Examples include the observation-based introduction of bidirectional fluxes for ammonia gas and improved understanding of the role of CO2 fluxes in the deposition of other gases. Also, some divergence in approaches is likely due to algorithm developments having been made in the context of specific regional models – each of which encompasses a diverse range of process representation algorithms, vertical resolutions, horizontal resolutions, etc. An algorithm which provided good performance relative to surface concentration observations within the context of one regional model thus may not have resulted in as good of performance in another model, further spurring model-specific development. These factors have resulted in the variety of approaches for gas-phase deposition in current regional models and provide the part of the motivation for this first attempt at cross-comparing the results of the models' deposition algorithms in detail – to show and explain the causes for these differences.

Schematics of resistance frameworks as two generic examples are shown in Fig. 3. In these examples, the Wesely (1989) deposition pathway for “lower canopy buoyancy and exposed surfaces” deposition is not included. The example of Fig. 3a also lacks a quasi-laminar sublayer resistance rb applied across all surface types. Instead, surface-specific quasi-laminar sublayer resistances are used: rsoil2 for soil and rleaf1 for leaves. The examples in Fig. 3 demonstrate two ways in which the resistance framework has been adapted from Wesely (1989). In general, the diversity in resistance frameworks across models complicates model intercomparison of individual resistances.

When there are differences in resistance frameworks across models, the deposition pathways may be compared across models using a construct we will refer to here as effective conductance (Paulot et al., 2018; Clifton et al., 2020b). While generally a conductance is simply the inverse of a resistance, an effective conductance is the contribution of a given depositional pathway to the deposition velocity, expressed in the same units as the deposition velocity. The sum of the effective conductances for all deposition pathways is the deposition velocity. The effective conductances of the soil (ESOIL), lower canopy (ELCAN), cuticle (ECUT), and stomata (ESTOM) branches specifically for Wesely (1989) are given by

Note that the depositing gases in each pathway are influenced by ra and rb prior to encountering the different resistances that make up rc, and this is why vd, which includes the influence of ra and rb, is scaled by the fraction of the inverse of rc occurring through a given pathway. Some models include surface-specific quasi-laminar sublayer resistances; when this is the case, these terms appear in the pathway-specific fractions of the total uptake terms.

For any given model, effective conductances are an invaluable tool for determining the extent to which each pathway impacts dry deposition velocity, and which deposition pathways drive spatiotemporal variability in dry deposition velocity. Key for AQMEII4, the effective conductances allow a cross-comparison of the main deposition pathways across different resistance frameworks. The primary terms of comparison for dry deposition schemes in AQMEII4 are thus the effective conductances. In addition, given that many models' resistance frameworks follow Wesely (1989), we also request those individual resistance terms held in common by most models to allow exact comparisons of individual processes which may influence or control a given pathway. These resistances include

a term for the aerodynamic resistance, ra;

With regards to (6), the implementation of quasi-laminar sublayer resistance (rb in Wesely, 1989) tends to differ among models. Some models use the Wesely (1989) concept of a pathway-independent quasi-laminar sublayer resistance. Others use quasi-laminar sublayer resistances as pathway-dependent (e.g. Fig. 2a, where the rsoil2 and rleaf1 represent quasi-laminar sublayer resistances for soil and leaf pathways, respectively). The quasi-laminar sublayer resistance is thus reported in AQMEII4 for each pathway, with the models for which the term is independent of pathway reporting the same value for each pathway. Pathway-dependent quasi-laminar sublayer resistances are to be reported as “not present” only if the given pathway does not exist in the framework.

Note that models that include a single deposition pathway to soil that incorporates rdc are requested to report that pathway as “lower canopy” not “soil”. For example, the LOTOS-EUROS dry deposition scheme (Fig. B4) reports the effective conductance calculated for the soil pathway as ELCAN due to the presence of the in-canopy resistance term in this pathway. In contrast, the CMAQ-M3DRY and CMAQ-STAGE dry deposition schemes (Figs. B2 and B3) have two separate pathways for deposition to soil, one for vegetation-covered soil and one for bare soil. Due to the inclusion of the in-canopy convective resistance in the computations for vegetation-covered soil, the effective conductance for that pathway is reported as ELCAN, while the effective conductance for the bare soil pathway should be reported as ESOIL.

Specific resistance terms for the soil deposition pathway and the lower canopy pathway have not been requested because the resistance frameworks for these pathways vary considerably across models, and therefore specific resistance terms are not easily comparable. For example, Wesely (1989) used a single term for the soil resistance (Fig. 1) while other models may use two or three resistances related to dry deposition to soil only and added in series (Fig. 2).

In addition to the effective conductances, another set of diagnostic fields is calculated during post processing: the time-aggregated fractional mass (or charge equivalent) flux transferred to the surface via each of the four deposition pathways (hereinafter, effective flux). The effective flux is calculated on an hourly basis prior to conversion to AQMEII4 time-aggregated gridded and station data using ENFORM and is the product of the hourly effective conductances, dry deposition mass fluxes, and inverses of the deposition velocity. Effective conductances provide an estimate of the importance of each pathway towards the deposition velocity. However, since the flux depends on the deposition velocity and the near-surface air concentration, which both vary on hourly timescales, estimating the aggregate importance of each deposition pathway towards the flux requires calculating the effective flux before time aggregation.

Figure 4 provides an example of the different yet complementary information resulting from effective conductances and effective fluxes, showing hourly SO2 concentrations, effective conductances, and effective fluxes for a boreal forest impacted by a large industrial SO2 stack source and hourly NO2 concentrations, effective conductances, and effective fluxes for a location to the north-east of New York City. In both cases, high concentrations of the pollutant gas (Fig. 4a, d) occur at night, while the deposition velocity, due to the stomatal pathway (Fig. 4b, e), maximizes during the day. As a result of the low daytime concentrations, the effective fluxes for SO2 (Fig. 4c) show a relatively minor contribution of the stomatal pathway to the deposited mass despite the major contribution of the stomatal pathway to the daytime deposition velocity. As the result of high night and morning concentrations, the effective fluxes for NO2 (Fig. 4f) show separate day and night peaks of about equal magnitude, with the stomatal pathway dominating daytime values and roughly equivalent contributions from stomatal and soil pathways at night.

Also with reference to Fig. 4, it should be noted that the effective conductances and effective fluxes show the relative contributions of the pathway towards the total deposition or the total flux at any given time. It should also be noted that the net surface resistance appearing in the denominator of these terms may drive the time variation. For example, the soil effective conductance of Fig. 4e minimizes at 6:00 LT – however, the factors contributing to the soil pathway itself for the model used in this example (see Appendix Table B1) are relatively time-invariant (seasonally varying). The temporal variation is driven by hourly variation in the stomatal term and hence the relative importance of the soil conductance varies with time in Fig. 4e.

We also consider that dry deposition strongly depends on LULC type, and different models use unique LULC databases. We thus request LULC-specific variables along with the fractional areal coverage for each LULC type, which allows quantification of not only the impacts of different LULC-specific processes and parameters on dry deposition, but also the impacts of different LULC databases. “Generic” AQMEII4 LULC types were devised due to the use of a wide variety of LULC databases across air quality models, in terms of both the source of the data and the number of LULC types employed. The AQMEII4 LULC types listed in Table 2 are broad LULC types into which the model-specific LULC types could be aggregated, to allow intercomparison between models. Study participants aggregated their LULC-model-specific diagnostic outputs to the set of common AQMEII4 LULC types using the fractional representation of each native LULC type contributing to the AQMEII4 type within each grid cell. Generic AQMEII4 LULC types were constructed after analysis of the LULC schemes in the participating models. A suggested mapping between model and AQMEII4 LULC types was provided to participants, along with the instruction that the mapping actually employed should be reported. The grid cell fractions of both the native model LULC types and the resulting fractions of AQMEII4 LULC types were reported by participants. Note that there is a large variety in number and therefore types of LULC across models, and thus each of the generic types represents a rather broad range of LULCs.

We also note that the mapping of LULC types from the individual model land use classifications to the AQMEII4 land use classifications is an unavoidable source of uncertainty in the land-use specific diagnostics. The 15 AQMEII4 land use types themselves were based on a survey of land-use classifications used in 17 regional models. For example, while “Herbaceous” is available as an AQMEII4 land use category, its intent is for use for moors and heathlands, while AQMEII4 land use category “Wetlands” encompasses wetlands which are diversely described in individual model land use categories such as herbaceous, wooded, and permanent wetlands, as well as swamps and peatbogs. However, some categories were included most models (e.g. evergreen needleleaf forest, deciduous broadleaf forest, snow and ice, mixed forest, usually taken as a combination of needleleaf and deciduous forests), while others could easily be classified according to the broader landscape type of which they were a member (e.g. different types of tundra were recommended to be classified as the AQMEII4 Tundra classification). Both the AQMEII4 and “native model” land use types were reported by participants – with the aim of using both sets of information to determine the extent to which land use database variation may be a factor in estimating deposition velocities and to provide information on specific land use types used by specific models when these differences appear to be large.

For AQMEII4, the terms listed in Table 4 were reported for SO2, NO2, NO, HNO3, NH3, PAN, HNO4, N2O5, organic nitrates, O3, H2O2, and HCHO, both as a function of the 16 generic AQMEII4 LULC types (Table 3) and for the net grid-scale calculation for each grid cell and/or receptor. Models employing bidirectional flux algorithms for the dry deposition of atmospheric NH3 reported a different set of terms, given in Sect. 4.2.

Table 4 summarizes the diagnostic variables related to gaseous dry deposition reported by all participants, the variable names as described in the AQMEII4 technical specification documents (TSDs), and a description of each variable. Equations (2) through (6) and the related text describe the terms specifically for the resistance framework of Wesely (1989); additional examples for participating models' resistance frameworks are provided in the Appendix tables and figures.

The presence of surface wetness or snow is incorporated into the effective conductance, effective flux, and component resistances. In other words, separate component resistances or effective conductances and fluxes for snow-covered or wet surfaces were not reported. In order to compare the impacts of the different models' predictions regarding snow cover or wetness, additional diagnostic variables were requested to describe surface state (e.g. fractional snow cover and either the values of binary wet–dry conditions or fractions in surface wetness).

Gridded dry deposition diagnostic variables were archived as hourly values for the native LULC types and then converted to the generic AQMEII4 LULC types during post-processing. The ENFORM Fortran code provided to all participants was used to convert gridded fields from the hourly values to temporal aggregations of the hourly values. Hourly diagnostics were converted to “monthly median diurnal” values using ENFORM by taking the medians of all values for a given UTC hour in a given month, thus reducing 8760 hourly values for each year to 288 values (24 h × 12 months). The use of monthly median diurnal values is motivated by the need to reduce the amount of data to be transferred and analysed on a single server (despite this aggregation, each year of gridded model output requires up to 200 Gb of storage), while preserving the key aspects of diurnal and seasonal variations.

The use of a median rather than an arithmetic mean for AQMEII4 diagnostic time aggregation resulted from consideration of the manner in which different dry deposition algorithms deal with pathways that effectively shut down under certain conditions. For example, some algorithms employ an upper-limit resistance to represent conditions under which the pathway transmits little mass to the surface (e.g. nighttime stomatal resistances may be set to very large values). Others simply use code branching to prevent a pathway from contributing to rc (e.g. the entire stomatal pathway is removed from rc at night). Others employ different resistance frameworks for different conditions (e.g. to account for snow-covered surfaces). However, the AQMEII4 protocol requires participants to submit “missing values” as a specific code (-9) in order to allow filtering of valid from invalid data during time aggregation. An algorithm removing a pathway may thus have a different number of valid values from an algorithm employing a large resistance. Similarly, a seasonal transition where the resistance network changes depending on whether a surface is snow-covered becomes difficult to interpret in a time average, whereas valid time-median values allow for a more meaningful comparison.

For example, if only 20 % of the resistances at 14:00 LT in a given month and grid cell are snow covered, then the monthly median for 14:00 LT would represent values typical of snow-free conditions, for both models representing resistances under snow-covered conditions as missing and models representing them as large values. Thus, the monthly median comparison represents the most common conditions encountered during the month for both models. On the other hand, while the monthly average resistance for 14:00 LT represents snow-free conditions for the model that treats snow-covered hours as missing, the monthly average for the model that represents snow-covered conditions as a large value is not meaningful and complicates inter-model comparison.

Monthly median diurnal values capture both seasonal and diurnal variations in the archived fields and allow comparisons between algorithms shutting off a pathway by removing the pathway and algorithms shutting off a pathway with high resistance values. Note that the same data completeness criterion used for comparing simulated and observed ambient concentrations was employed here for the construction of the median values. Specifically, more than 75 % of the values within a month were required for a median to be constructed.

The fourth phase of the Air Quality Model International Initiative has been introduced. The focus of this phase is on wet and especially dry deposition. The necessity of tackling this subject in a diagnostic way prompted us to divide the initiative into two activities, one dedicated to the evaluation of the process as described by four-dimensional air quality regional-scale models and the second dealing specifically with evaluating ozone dry deposition calculated by “single-point model” versions of the dry deposition modules used in the regional-scale models with a collection of ozone flux measurements. Here, the organization of Activity 1 has been formally introduced, whereas Activity 2 will be described in separate AQMEII4 special issue publications. In addition to presenting the standard and common input data and the way in which standard output is expected, we also presented the way in which the very diverse representations of dry deposition in participating models have been reduced to a common representation that will facilitate model inter-comparison. The essence of the adopted methodology is the transformation of individual resistances into effective conductances and effective fluxes, which represent the importance of deposition pathways held in common across the models to the total deposition velocity and flux. Resistances held in common across different modelling frameworks were also reported, to allow comparisons at the sub-pathway level, where possible. Thus, regardless of the level of sophistication of the resistance framework, one can meaningfully inter-compare the results produced by different models.