The air-quality model used in our study is the WRF/Chem [22] [23] . It was set up identical to [11] : Cloud processes on the resolvable and subgrid-scale were calculated by the WRF-Single-Moment 5-class cloud-microphysics scheme [24] and a further-developed version of [25] ’s ensemble convective scheme, respectively. The Goddard two-stream multi-band scheme [26] and Rapid Radiative Transfer Model [27] served to determine shortwave and long-wave radiation including cloud and aerosol-radiation feedbacks [28] . Surface and atmospheric boundary layer physics were treated in accord with [29] . A modified version of the so-called NOAH land-surface model [30] calculated the exchange of heat and matter at the surface-atmosphere interface, as well as the snow, soil-temperature and soil-moisture and frozen ground conditions. Gas chemical pro- cesses were calculated by the Regional Acid Deposition Model version 2 chemical mechanism [31] and used inline calculated photolysis rates [32] . Aerosol dynamics, physics, and chemistry were considered by the Modal Aerosol Dynamics Model for Europe [33] and Secondary Organic Aerosol Model [34] .

All simulations were driven by the 1˚ × 1˚, 6 h-resolution National Centers for Environmental Prediction global final analyses (FNL) data [35] as lateral boundary conditions for the meteorological fields. The FNL snow, soil-temperature and moisture conditions, as well as sea-surface temperatures (SST) served to initialize the meteorological and surface fields. All simulations ran in “forecast mode” with re-initialization of the meteorology every five days. Note that from a modeling standpoint, the simulations could have been run using reanalysis data instead of analysis data, and in nudging mode rather than forecast mode. We chose the forecast mode because managers have experience with the uncertainty associated with weather forecasts. Weather forecasts base on numerical weather prediction (NWP) that uses analysis data for initialization to produce a forecast. Thus, performing the study in forecast mode provides a framework of “known performance” for stakeholders.

In all four WRF/Chem simulations, the same idealized profiles of clean air background concentrations served as initial conditions of the chemical fields at the start of the simulation and provided the lateral boundary conditions over the May 15 to September 15, 2008 “tourist season” [11] .

The model domain encompassed the atmosphere over Southeast Alaska with 28 layers from the surface to 100 hPa, and 120 × 120 grid-points of 7 km horizontal increment centered at 58.5N, 135.5W. To permit adjustment of the meteorological and chemical fields and minimize errors from lateral boundary effects, we discarded five grid-points on each lateral boundary from the results (Figure 1).

Anthropogenic emissions, except from cruise ships were derived from the 0.5˚ × 0.5˚ Reanalysis of the Tropospheric Chemical Composition (RETRO) data. For cruise ships, we calculated activity-based emissions using Automated Information System (AIS) voyage data (ship position, cruise speed, operation mode) and the individual ships’ characteristics (engine power, size, fuel type, maximum cruise speed, etc.) [11] . We refer to this emissions inventory as REF. The emissions inventory ALL combined the RETRO and REF inventories (Figure 1). WRF/Chem calculated the biogenic emissions (e.g. NO_x, various VOCs) inline depending on vegetation and soil type, soil temperature and moisture, and atmospheric conditions following [36] [37] in all simulations.

We performed four WRF/Chem simulations for the tourist “season”. For all simulations, model setup, boundary conditions, and initialization were identical except for the choice of the emissions inventory.

Our CLN simulation included only biogenic emissions and, due to the absence of anthropogenic emissions, represented a “clean” atmosphere [38] . Our reference (REF) simulation included the biogenic emissions plus the activity-based cruise-ship emissions. Evaluation of this simulation with respect to various aspects can be found in [9] [11] . The RETRO simulation included biogenic emissions and other anthropogenic emissions including commercial shipping, but no cruise-ship emissions. Finally, the ALL simulation represented the atmospheric composition of the 2008 tourist season as it included activity-based cruise ship and other anthropogenic emissions as well as biogenic emissions.

Being in control of the air quality in Glacier Bay would require that in the bay, air quality only or at least to a high percentage depends on the emissions occurring inside the bay. This means that advection of pollutants from other sources including cruise ships outside of Glacier Bay would have to be negligibly small relative to the pollutant concentrations from sources occurring inside the bay. Furthermore, the concentrations inside

the bay should not have been modified by chemical reactions with pollutants from local non-controllable and/or distant sources.

To test the hypothesis of No interaction of pollutants from different local and/or distant sources we applied the principle of superposition [20] [21] . Applied to our task it means that the sum of the concentrations of the simulations considering only one type of emission sources should equal the concentrations of the simulation that considered all emission sources (ALL).

Since both REF and RETRO considered background concentrations and biogenic emissions, we subtracted the concentrations obtained by the CLN simulation from both REF and RETRO to account for the contribution of natural emissions and background concentrations to the total atmospheric concentrations only once. After some algebra, we obtain, which we compared with the concentration of C_i,j,k,t,m(ALL)

Here, i, j, k are the indices of a model grid-cell, m is the chemical species or particulate matter, and t is time.

If there were no interactions of pollutants from different local and/or distant emission sources, the left hand side of Equation (1) would not differ significantly, at the 95% confidence level in a two-tailed t-test [39] from the concentration C_i,j,k,t,m(ALL) on the right hand side. Any difference between the left and right hand side of Equation (1) can be expressed as a fraction, α, stemming from interactions of pollutants from the three groups of sources (biogenic emissions and background, cruise-ship emissions, anthropogenic emissions other than cruise-ship emissions).

Due to atmospheric transport, the atmospheric composition can contain pollutants emitted at different places. For example, assume a wind of 10 m∙s⁻¹ blows from Juneau to Glacier Bay that are about 150 km apart. Pollutants emitted in Juneau by anthropogenic sources including cruise ships would reach Glacier Bay about 4.2 hours later. During transport, chemical reactions and gas-to-particle conversion of the primary pollutants and particles emitted at Juneau produce secondary pollutants and particles. Furthermore, chemical reactions, removal processes, and emissions from sources along the way, as well as mixing processes modify the air composition. Once the aged air reaches Glacier Bay, reactions between these pollutants from various different distant sources and those from the local cruise-ship emissions may occur. Following [20] [21] the fraction α_i,j,k,m,t to the concentration of ALL due to interaction of species from different emission sources can be determined as

The emission-source contribution ratios (ESCR) [19] of the various emission sources, i.e. natural/biogenic, cruise ships, the collective anthropogenic emissions except cruise ships, and all anthropogenic emissions to the ALL concentrations reads

Here is the fractional contribution from natural/biogenic emissions and the background concentrations to the overall concentration C_i,j,k,t,m(ALL) for i, j, k, t, and m. This contribution refers to the natural background chemistry for which it is always positive.

Furthermore, is the emission source contribution ratio to the ALL concentrations due to cruise ships. The emission source contribution ratio

represents the non- cruise ship related anthropogenic emissions. Multiplication of α and the ESCR_x (x = CLN, REF, RETRO) terms by 100% provides the contribution of interaction and the respective emission sources to the concentration in ALL, C_i,j,k,t,m(ALL) in percent.

According to Equation (2), interaction of pollutants from different local and distant sources, and non-linear chemical reactions can enhance or diminish ALL concentrations as compared to the concentrations obtained by adding the concentrations in response to the individual emissions inventories.

Equation (3) permits negative ESCR values, when the concentration in CLN exceeds that of REF or RETRO, respectively. Negative ESCR_x (x = REF, RETRO) values result from interaction. They mean that the emissions from the respective source permitted chemical reactions and processes that reduced the ALL concentration of the respective species. In other words, if the emitted species would not have reacted, the ALL concentration would be higher than it is by the absolute of the percentage. Adding reactive species m to a (clean) air sample reduces the species it reacts with and increases the concentrations of the reaction products. The products may further react when they are not an end product of the reaction chain. When the species m and its initial reactant(s) are not reproduced in the reaction chain, their concentrations decrease [6] .

In general, reactions requiring low energy are favored over those needing comparatively higher energy at same temperature. Reaction rates differ among reactive species. Furthermore, reactions may be limited by low concentrations of their reactants. For instance, in a closed air parcel of NO-NO₂-O₃ under quasi-steady state conditions, the ozone concentration remains constant despite of the occurring reactions. Adding NO_x to this air parcel decreases the O₃ concentration. Adding VOC instead introduces competing reactions. As long as the VOC/NO_x ratio of the parcel is low, its air is VOC-sen- sitive. It is NO_x sensitive for low NO_x and high VOC concentrations. In this case, O₃ concentration increases with increasing NO_x and little response of VOC.

In the clean atmosphere, background chemistry also occurs [6] . Vegetation emits various types of VOCs [36] [40] that may differ strongly in their reaction rates [41] . Emission rates depend on plant-available water in the root zone, air temperature and moisture, and photosynthetic active radiation, among other things [37] [42] . Soil bacteria emit NO_x as part of their metabolism that depends on soil temperature and moisture [36] [37] [40] [42] . Note that WRF/Chem considers these processes [22] [23] .

Chemical processes conserve the total mass of atoms, but not molecule species. Anthropogenic emissions may add species that also have natural sources and/or species that do not occur in the clean atmosphere [6] . Consequently, the competition for reactants changes and interaction of species from natural and anthropogenic sources and/or different anthropogenic sources may occur. In the case of cruise-ship emissions, for instance, both faster and slower reacting VOCs join the natural VOCs. As illustrated in the above example of the reaction system NO_x-O₃-VOC, anthropogenic emissions may shift the chemical regime leading to reduced concentrations even of emitted species. The chemical regime also alters the concentrations of secondary pollutants, i.e. those pollutants that form by reaction of the primary, i.e. emitted pollutants. Finally, and with respect to managing park resources, it is worth mentioning that anthropogenic emissions also change the fractional composition of particles. However, particles of different composition have different impacts on visibility [12] .

We examined the mean ESCRs from the different sources and pollutant interactions to the atmospheric composition of ALL to assess the degree to which NPS managers have the possibility to manage air quality in Glacier Bay. We used comparison to the mean concentrations in Southeast Alaska with the assumption that differences between the mean composition of ALL in Southeast Alaska vs. Glacier Bay provide evidence that local emissions within the bay’s air shed govern its air quality. The value of ESCR_REF represents the degree to which cruise-ship emissions in Glacier Bay determine the concentration of the respective species inside the bay. The difference 1-ESCR_REF is the limit for management. The lower this limit is for an emitted species (1-ESCR_REF ESCR_REF), the greater are the possibilities of the NPS for managing the pollutant by emission- control measures.

Anthropogenic emissions contributed all coarse particulate matter, i.e. PM₁₀ with diameters exceeding 2.5 μm, but being less than or equal to 10 μm in diameter in ALL, RETRO, and REF. In CLN, all PM₁₀ was PM_2.5. Comparison of the REF and RETRO seasonal mean profiles of SO₂ over the Glacier Bay water proper (Figure 2) reveals that transport of pollutants from other sources than cruise ships contributed to the ALL concentrations in the maritime ABL, and up to 700 hPa (~3 km above sea level). The SO₂ peak around 1 km (900 hPa) was related to frequent inversions. The second peak at

about 1.5 km above water level (850 hPa) relates to the background profile (compare to CLN).

The height of highest daily mean SO₂ concentrations in the ABL differed with time due to local weather conditions (not shown; cf. [9] ). On days with low inversion heights, highest concentrations in Glacier Bay occurred below 1 km. When bay-wide, highest daily mean SO₂ concentrations existed at about 850 hPa at the top of the ABL, SO₂ concentrations were low close to the water. This finding points to large-scale forcing with vertical mixing and/or convection on those days.

Comparison of bay-wide seasonal means of CLN, REF, RETRO, and ALL NO_x concentrations (Figure 2) supports the above finding that pollutants from emissions outside the bay affected the ALL concentrations below 700 hPa. Seasonal bay-wide mean O₃ profiles barely differed indicating that they mainly represent O₃ background concentrations.

Averaged over the season and to about 152 m above the water surface, ALL SO₂ emissions and concentrations correlated significantly at the 95% confidence level over the international shipping lane (southwestern corner of the model domain), port cities, some waterways, and most of Canada (Figure 3). In areas without anthropogenic emissions, for instance the glacier-covered mountains, absence of emissions correlated with

low concentrations. Below the first 152 m or so, significant, but low NO emission-con- centration correlations occurred over the major highways in Canada. NO emission- concentration correlations were comparatively higher over the waterways in the Alexander Archipelago including Glacier Bay, locally along the coast where cruise-ships travel density was high (cf. [43] ), and reached correlation coefficients about 0.5 over the international shipping lane (compare Figure 1, Figure 3).

Across Southeast Alaska waters, ALL PM₁₀ concentrations visibly reflected the ship routes as ships are the only source for coarse particulate matter over water (Figure 4). High SO₂ concentrations occurred in Icy Strait and just up the west side near the entrance of Glacier Bay park (Figure 3). Here cruise-ship speeds were higher than inside Glacier Bay. Note that under certain synoptic conditions, pollution was transported from Icy Strait into Glacier Bay elevating pollutant concentrations [9] .

In ALL and REF, the cruise path thru the bay was clearly visible at the time of the voyage for all emitted species, and even notable on season average for NO_x and PM_2.5 (Figure 4). In Glacier Bay, seasonal mean REF and ALL SO₂, and NO_x concentrations were highest in the lower inlet downstream of where cruise ships berth for several hours for glacier viewing, while running their auxiliary engines on low load to generate energy for hoteling [11] . Note that at low loads and during maneuvering, combustion is often incomplete meaning high emissions [4] [44] [45] .

Highest concentrations did not exactly correspond to the areas adjacent to the tidewater glaciers where ships berth for several hours to allow passengers time to experience the scenery. The slightly off-set of the “expected location” for concentration maxima was due to the coarse grid resolution, discretization of the advection-diffusion equations, and lost AIS signals. The former two are well-known modeling artifacts [41] [46] [47] .

Comparison of CLN and RETRO SO₂ concentrations revealed marginal increases in Glacier Bay due to advection of SO₂ from other anthropogenic sources than cruise ships (not shown). Comparing SO₂ concentrations simulated by REF and ALL confirmed these findings (Figure 4). The same was true for the NO_x concentrations in Glacier Bay. We conclude that cruise-ship emissions governed the magnitude and distribution of SO₂ and NO_x concentrations in the bay most of the time.

Over the bay, increases in ALL and REF hourly NO_x and hence daily mean concentrations corresponded to cruise-ship visits (Figure 5). The highest NO_x concentrations occurred when two large cruise ships visited the bay on the same day during strong inversions [9] . Since peroxide-acetyl-nitrate (PAN) is a reservoir for NO_x, NO_x concentrations increased at the cost of PAN when temperature increased, and vice versa [38] .

VOC has both anthropogenic and natural sources [6] [36] [37] [41] . In all four simulations, biogenic emissions governed the VOC concentrations over land in Southeast Alaska. In Glacier Bay National Park, the total amount of biogenic emissions is low because glaciers and water cover large areas of the park. Comparison of RETRO and CLN VOC concentrations showed marginal increases in RETRO VOC over Glacier Bay. This finding indicates that advection of anthropogenic VOC only marginally modified the

air composition over the water body (therefore not shown). On the contrary, cruise- ship emissions dominated the VOC concentrations over the water proper in REF [11] and ALL, as expected. Nevertheless, comparison of REF and ALL VOC concentrations confirmed that advection of VOC from anthropogenic sources outside of the bay had marginal impact on the VOC concentrations in the bay.

In Southeast Alaska, the complex terrain prohibits dairy and agricultural activities. Consequently, the major ammonia (NH₃) emission sources were anthropogenic. Highest NH₃ concentrations occurred in the maritime ABL over the shipping lanes and waterways as well as in the urban ABL of port cities due to ship emissions. Ammonia can neutralize nitric acid (HNO₃), which forms by NO_x oxidation, to produce ammonium nitrate aerosol (NH₄NO₃) [6] [48] . NH₃ can also neutralize sulfuric acid (H₂SO₄) by forming ammonium sulfate aerosol ((NH₄)₂SO₄). SO₂ oxidation forms H₂SO₄ under all atmospheric conditions in the presence of reactive radicals and water vapor [6] [41] [48] . Thus, ammonia served as a sink for oxidation products of SO₂ and NO_x and as a source for PM from precursor gases (section 3.3). These aerosols affect visibility [12] .

In all four simulations, hourly O₃ concentrations were typically less than 50 ppb, sometimes even below 40 ppb in the maritime ABL of Glacier Bay. Thus, over the entire season, 8-hour mean near-surface O₃ concentrations remained below the EPA NAAQS of 70 ppb. Bay-wide season mean O₃ profiles differed negligibly between REF and ALL or CLN and ALL (Figure 2). Daily average O₃ concentrations remained less than 40 ppb in the first 200 to 500 m above the water surface and below 50 ppb up to about 2 km height (~800 hPa) most time. According to the O₃ concentrations obtained by RETRO and CLN, anthropogenic emissions occurring outside of Glacier Bay only marginally affected O₃ concentrations inside the bay. Comparison of REF and ALL-simu- lated O₃ concentrations confirmed this finding.

Over the entire season, ambient air temperatures governed PAN concentrations. Similar to O₃, anthropogenic emissions other than from cruise ships had small impact on the PAN concentrations found in Glacier Bay (therefore not shown). This finding indicates that besides temperature, cruise-ship emissions in the bay affected the PAN concentrations in Glacier Bay (Figure 5). The latter confirms [11] , who reached this conclusion by comparing PAN concentrations of days with and without cruise-ship visits in Glacier Bay.

While cruise ships emit PM, PM also forms naturally from gas-to-particle conversion from precursor gases [6] [33] [41] [49] . As aforementioned, over Southeast Alaska, all PM was in the fine-particle mode with diameters less than 2.5 μm under undisturbed conditions (CLN). Consequently, the bay-wide season mean PM₁₀ profile of CLN differed significantly from those obtained by REF, RETRO, and ALL with respect to concentrations (Figure 2), size distribution, and composition. This fact also means that all PM₁₀ in REF, RETRO, and ALL stemmed from anthropogenic emissions, gas-to-parti- cle-conversion, and/or particle growth.

Over the season, in the first layers above surface, PM_2.5 emission-concentration relations behaved similar as found for SO₂ (Figure 3). Emission-concentration relations (correlation coefficients R up to 0.5) were significant over the international shipping lane, port cities, some water ways, and most of Canada. These findings indicate that residential sources, commercial ship traffic including cruise ships governed the near- surface PM concentrations in Southeast Alaska.

While locally in the bay under inversion conditions, PM_2.5 concentrations exceeded 35 μg・m⁻³ for several hours when one or two large cruise ships visited the bay on the same day, daily mean PM_2.5 concentrations remained below the 24-hour average of 35 μg・m⁻³ of the NAAQS [50] (therefore not shown).

Comparison of season bay-wide mean profiles (Figure 2) showed that sources other than cruise ships also contributed to the PM₁₀ concentrations below 700 hPa (~3 km). In the case of PM₁₀, both advection of primary and secondary particles as well as gas-to- particle conversion from locally emitted and advected precursor gases (SO₂, NH₃, NO_x) played a role. Under strong inversion conditions, concentrations of PM increased strongly due to formation of secondary aerosols [50] . Consequently, concentrations of precursor gases like SO₂ and NO_x increased less (Figure 5) than expected from the emissions and/or by Equation (1) and were more sensitive to emissions and availability of oxidants.

In REF and ALL, cruise ships increased the concentrations of aerosol precursor gases in the bay (see Figure 1, Figure 4, Figure 5 for SO₂, NO_x). Without cruise-ship emissions in Glacier Bay (CLN, RETRO) the PM distribution was nearly homogeneous over the bay (not shown, for CLN see Figure 6 in [38] ). On the contrary, in REF and ALL, enhanced PM concentrations occurred along the cruise path and close to the glaciers on season mean (Figure 4).

In the first decameters above the surface, PM₁₀ concentrations were low over the ice fields of Glacier Bay National Park (Figure 4). This finding suggests that upslope transport of PM₁₀ was low on season average. Indeed, the air in the ABL over the bay was decoupled from the air aloft frequently by inversions [9] [50] .

In all four simulations, days with worst visibility conditions were due to meteorology. At the height of cruise ships, bay-wide daily means of haze indices were high when relative humidity exceeded 90% (Figure 6). Daily bay-wide mean haze indices were correlated with relative humidity (R = 0.736, 0.618, 0.590, and 0.625 for CLN, REF, RETRO, and ALL, respectively).

CLN, REF, RETRO, and ALL daily bay-wide mean haze indices were also correlated with wind speed (R = 0.462, 0.825, 0.864, and 0.826, respectively). The high correlation coefficients found for REF, RETRO and ALL supported that transport of pollutants from emissions outside of Glacier Bay affected visibility inside the bay (Figure 6).

Daily bay-wide mean REF and ALL haze indices as well as RETRO and ALL haze indices were highly correlated (R = 0.991 and 0.995, respectively), while CLN haze indices were notably lower correlated with REF, RETRO, and ALL haze indices (R = 0.750, 0.730, and 0.756, respectively). The correlation of REF, RETRO and ALL haze indices with those of CLN confirmed [11] that meteorology governed visibility conditions in the bay. Correlations between REF and ALL haze indices and RETRO and ALL haze indices differed marginally. In RETRO, all ship emissions stemmed from commercial shipping, i.e. no cruise-ship emissions occurred in Southeast Alaska and Glacier Bay. Based on the low differences in correlations, we infer that transport of pollutants from ship emissions outside the bay affected the RETRO haze indices inside the bay.

In the bay, locally absolute differences between REF and ALL hourly haze indices reached up to 19.6 dv. Thus, transport of pollutants from emissions outside the bay influenced hourly visibility at least locally. Seasonal mean maximum (minimum) haze indices differed 2.0 (~0) between REF and ALL. These findings suggest that anthropogenic emissions occurring outside of Glacier Bay had small influences on worst visibility conditions that were caused by meteorological conditions.

Comparison of CLN and ALL haze indices revealed the total anthropogenic impact on visibility in Glacier Bay. On season average, the combined anthropogenic emissions increased the haze index by about 2 dv over wide areas of the bay compared to the natural conditions (Figure 6). Increases of about 1 dv or more occurred in ALL also in areas without cruise-ships traffic. These findings support that transport from outside and pollutants from emissions within the bay affected air quality, and visibility in the bay. Along the coast, in Icy Strait and Cross Sound, ALL anthropogenic emissions increased haze indices more than 3 dv on season average as compared to the background conditions represented by CLN. At these locations, cruise ships were the major emission source, but contributed only marginally to reduced visibility caused by meteorological conditions in the marine ABL.

For all trace-gas species and PM, the ESCRs to the ALL concentrations differed in space and time (e.g. Figures 7-11). The ESCRs also differed among species partly due to different spatial occurrences of their emissions and/or emissions of their precursors (cf. Figure 1). Here we discuss contributions from the various emission sources to ALL concentrations by species. Discussion starts with the contributions to Southeast Alaska daily means at various heights followed by the seasonal mean contributions in Southeast Alaska to assess the degree to which air quality in Glacier Bay depends on emissions in the bay itself. Important indicators of potential for managing air quality inside

the bay would be that the contribution from cruise-ships is highest in the bay, and the composition of pollutants inside and outside of the bay would differ.