Ecological Applications, 16(4), 2006, pp. 1590–1607
Ó 2006 by the Ecological Society of America
EMPIRICAL MODELING OF ATMOSPHERIC DEPOSITION
IN MOUNTAINOUS LANDSCAPES
KATHLEEN C. WEATHERS,1,4 SAMUEL M. SIMKIN,1,2 GARY M. LOVETT,1
AND
STEVEN E. LINDBERG3
1
3
Institute of Ecosystem Studies, Millbrook, New York 12545 USA
2
Cornell University, Ithaca, NewYork 14853 USA
University of Tennessee, Knoxville, and Oak Ridge National Laboratory, Oak Ridge, Tennessee 37831 USA
Abstract. Atmospheric deposition has long been recognized as an important source of
pollutants and nutrients to ecosystems. The need for reliable, spatially explicit estimates of
total atmospheric deposition (wet þ dry þ cloud) is central, not only to air pollution effects
researchers, but also for calculation of input–output budgets, and to decision makers faced
with the challenge of assessing the efficacy of policy initiatives related to deposition. Although
atmospheric deposition continues to represent a critical environmental and scientific issue,
current estimates of total deposition have large uncertainties, particularly across heterogeneous landscapes such as montane regions.
We developed an empirical modeling approach that predicts total deposition as a function
of landscape features. We measured indices of total deposition to the landscapes of Acadia
(121 km2) and Great Smoky Mountains (2074 km2) National Parks (USA). Using ;300–400
point measurements and corresponding landscape variables at each park, we constructed a
statistical (general linear) model relating the deposition index to landscape variables measured
in the field. The deposition indices ranged over an order of magnitude, and in response to
vegetation type and elevation, which together explained ;40% of the variation in deposition.
Then, using the independent landscape variables available in GIS data layers, we created a
GIS-relevant statistical nitrogen (N) and sulfur (S) deposition model (LandMod). We applied
this model to create park-wide maps of total deposition that were scaled to wet and dry
deposition data from the closest national network monitoring stations. The resultant
deposition maps showed high spatial heterogeneity and a four- to sixfold variation in ‘‘hot
spots’’ and ‘‘cold spots’’ of N and S deposition ranging from 3 to 31 kg Nha1yr1 and from
5 to 42 kg Sha1yr1 across these park landscapes. Area-weighted deposition was found to be
up to 70% greater than NADP plus CASTNET monitoring-station estimates together. Modelvalidation results suggest that the model slightly overestimates deposition for deciduous and
coniferous forests at low elevation and underestimates deposition for high-elevation
coniferous forests. The spatially explicit deposition estimates derived from LandMod are an
improvement over what is currently available. Future research should test LandMod in other
mountainous environments and refine it to account for (currently) unexplained variation in
deposition.
Key words: Acadia National Park (USA); atmospheric deposition; elevation; GIS; Great Smoky
Mountain National Park (USA); hotspots; landscape features; model; nitrogen; sulfur; throughfall;
vegetation type.
INTRODUCTION
Atmospheric deposition is an important source of
nutrients and pollutants to ecosystems. For many
terrestrial ecosystems it is the primary source of nitrogen
and sulfur (Likens and Bormann 1995, Schlesinger
1997). Accurate measures of total atmospheric deposition are critically important for many reasons, including
the calculation of input–output budgets, assessment of
biological, ecological, and watershed responses to
pollutant and nutrient loading, and for linking air
emissions with actual deposition to landscapes (WeathManuscript received 27 April 2005; revised 27 October 2005;
accepted 1 November 2005. Corresponding editor: J. S. Baron.
4
E-mail: [email protected]
ers and Lovett 1998, Driscoll et al. 2001, Holland et al.
2005). For example, ecosystem theory and mechanisms
of nutrient cycling have been proposed based on budget
imbalances: the difference between watershed outputs
and inputs has been used in support of biogeochemical
‘‘theory’’ that suggests that old-growth, temperate
forests are less retentive of nitrogen than vigorously
growing or disturbed watersheds (Vitousek and Reiners
1975, Hedin et al. 1995, Goodale et al. 2000, Aber et al.
2003). In addition, for much of Europe, deposition
estimates are critical for setting emission-control policies
and legislation (Hettelingh et al. 1995a, b).
Substances are delivered from the atmosphere to
Earth’s surface in three forms: wet deposition (rain and
snow), dry deposition (gases and particles), and cloud or
1590
August 2006
ATMOSPHERIC DEPOSITION: EMPIRICAL MODEL
fog (occult) deposition (Lovett 1994). We know much
about the contribution of wet deposition to nutrient and
pollutant budgets, but relatively little about dry and
cloud deposition (Weathers et al. 2000).
Wet deposition
As a result of intensive, site-specific, long-term
monitoring efforts (Likens and Bormann 1995, Kelly
et al. 2002) as well as the existence of the national
monitoring networks, both wet-chemistry and deposition data are readily available from more than 250
monitoring stations in the United States (National
Atmospheric Deposition Program [NADP; information
available online]5 and AIRMoN [information available
online] ).6 Wet deposition is the product of precipitation
amount and chemistry, so point estimates can also be
extended with some accuracy to weather stations where
only precipitation amount is monitored. Thus temporal
trends (since the late 1970s) and spatial distribution of
wet deposition are well documented in the United States
(Lynch et al. 1995, 2000, Driscoll et al. 2001, Grimm and
Lynch 2004, Holland et al. 2005) and in many parts of
Canada and Europe (Hedin et al. 1994, Summers 1995,
van Leeuwen et al. 1996, Holland et al. 2005).
Dry deposition
Estimates of dry deposition are far less certain than
wet deposition, both because of the paucity of airchemistry monitoring stations and the inadequacy of the
models used for estimating dry deposition. Air chemistry
is measured routinely at independent research sites
(Kelly et al. 2002) as well as nationally through the
Clean Air Status and Trends Network (CASTNET;
information available online)7 and AIRMoN (see footnote 6) monitoring programs. There are relatively few
air-chemistry monitoring stations in the United States
(CASTNET [see footnote 7] and AIRMoN [see footnote
6]), with most of these located in the eastern half of the
country. Most estimates of dry-particulate and gas
deposition are based on measured air concentrations,
which then are used in dry-deposition models (Meyers et
al. 1998, Finkelstein et al. 2000) that consider the
multiple resistances, canopy characteristics, and meteorological variables that influence deposition (see Lovett
1994). Current dry-deposition models are limited by
simplifying assumptions, which include homogeneous
canopies and flat terrain (Weathers et al. 1995, 2000).
Dry-deposition model estimates are reasonably robust in
grasslands and croplands with flat terrain, but perform
more poorly in forests, and are largely untested in
uneven terrain (Baumgardner et al. 2002, Lavery et al.
2003). It is therefore difficult to extrapolate estimates of
wet plus dry deposition beyond the immediate vicinity of
dry-deposition monitoring stations.
5
6
7
hhttp://nadp.sws.uiuc.edui
hhttp://nadp.sws.uiuc.edu/airmoni
hhttp://www.epa.gov/castneti
1591
Cloud or fog deposition
The third form of deposition, cloud or fog, represents
an important water and chemical flux primarily in
montane and coastal regions. However, cloud chemistry
measurements and monitoring programs are few. Cloud
chemistry data for the United States have been collected
from fewer than 20 mountains and even fewer Atlantic
and Pacific coastal areas (Weathers et al. 1986, 1988,
Kimball et al. 1988, Vong et al. 1991, Mohnen and Vong
1993, Anderson et al. 1999, Baumgardner et al. 2003).
Cloud-deposition models suffer from the same limitations as dry-deposition models in that they rely on cloud
chemistry data from single stations, and on measured or
modeled meteorologic and canopy variables, and do not
account for complex topography or heterogeneous
vegetation (Weathers et al. 2000).
Total deposition
Total deposition (wet þ dry þ cloud) estimates for
complex terrain are either inadequate or lacking,
primarily as a result of the vagaries of measuring dry,
cloud, and/or snow deposition in complex terrain. A
sulfur- and nitrogen-deposition map was created in the
early 1990s for a limited area of the northeastern United
States (Ollinger et al. 1993). It has been extensively used
by the research community to determine site-specific
deposition estimates, even though it does not have
sufficient resolution to accurately discriminate between
locations that are 10s of kilometers apart. This map does
not include cloud deposition, and its dry-deposition
estimates do not take into account elevational effects or
variations in land cover, reducing its accuracy in
complex terrain. In recent years, other important
deposition maps have been published for the United
States (Holland et al. 2005) and limited areas in the
western United States (Fenn et al. 2003, Nanus et al.
2003). Despite these efforts, no empirical model exists
for total deposition to heterogeneous terrain.
Objectives
We developed an empirical modeling approach to
explore controls on deposition across heterogeneous
landscapes. The specific goals of this work were: (1) to
develop a generalizable method whereby both total
deposition to heterogeneous terrain and its spatial
heterogeneity across a landscape could be more accurately measured; (2) to determine which readily measured field- and GIS-derived independent variables
control deposition rate in an empirically based statistical
model; and (3) to develop landscape-scale maps of
nitrogen (N) and sulfur (S) deposition (LandMod) from
dry- and wet-deposition monitoring data (‘‘reference
deposition’’) and a separate empirically based statistical
model using only independent variables that were
available as geographic information system (GIS) data
layers. Finally, (4) we examined the behavior of our
model by performing validation analyses.
1592
Ecological Applications
Vol. 16, No. 4
KATHLEEN C. WEATHERS ET AL.
FIG. 1. Flow chart showing the steps involved in this research to create an empirical model of atmospheric deposition
(LandMod) for Acadia (ACAD) and Great Smoky Mountain National (GRSM) parks. Abbreviations: IV, independent variables;
f, function [thus, SF ¼ f(IV) means scaling factors are a function of independent variables].
METHODS
Study areas
General approach
We conducted our research in Acadia National Park
(ACAD) and in Great Smoky Mountains National Park
(GRSM). Atmospheric deposition of pollutants has
been identified as a primary threat to natural resources
in both of these parks by the National Park Service, and
both parks have wet and dry atmospheric deposition
monitoring stations. ACAD is located on the north
Atlantic coast of the United States in Maine (see Plate
1). The ACAD study area of interest was the entire
Mount Desert Island (278 km2 ). GRSM is located in the
southern Appalachian Mountains of the United States,
in Tennessee and North Carolina. GRSM has a total
area of 2074 km2, and is ;450 km from the nearest
coast.
We measured indices of total deposition in the field
and converted those measures to unitless ‘‘scaling
factors’’ by relating them to total deposition from a
low-elevation, flat area. Using the scaling factors, we
developed two statistical models, one that used independent variables measured in the field (e.g., elevation,
canopy type, diameter at breast height, canopy cover),
‘‘StatMod,’’ and one that used a subset of these variables
that were available as data layers in a GIS (e.g., aspect,
elevation, vegetation type). We used the latter, ‘‘LandMod,’’ plus monitoring data, in a GIS to create maps of
deposition for landscapes. StatMod identified which
independent variables, out of all available, control
atmospheric deposition in mountainous landscapes,
while LandMod (a combination GIS–statistical model)
allowed scaling deposition estimates over an entire large
area. Throughout the Methods section we refer to Fig. 1,
which outlines the complete procedure, from collecting
data in the field to creating deposition maps; this is our
process for scaling up from point measurements to the
landscape.
GIS data acquisition
Vegetation and elevation (DEM [digital elevation
model]) data layers were acquired for both ACAD and
GRSM. For each park we used a USGS 1:24 000 DEM
with 30 3 30 m pixels. Vegetation for GRSM was
derived from LandSat Thematic Mapper imagery
collected in September 1984 (MacKenzie 1993). Vegetation data for ACAD, derived from 1:15 840 color
August 2006
ATMOSPHERIC DEPOSITION: EMPIRICAL MODEL
1593
TABLE 1. Summary of landscape features for study area and sampling points at (a) Acadia National Park and (b) Great Smoky
Mountains National Park, eastern United States.
Study area
Feature
2
Size (km )
(%)
Sampling points No.
(%)
a) Acadia National Park, ACAD
Study area
Feature
2
Size (km )
(%)
Sampling points No.
(%)
b) Great Smoky Mountains National Park, GRSM
Vegetation
Conifer
Mixedà
Deciduous
Nonforested
93
100
34
52
34
36
12
19
162
29
94
5
56
10
32
2
Vegetation
Conifer
Mixedà
Deciduous
Nonforested
Slope (8)
0–0.001
0.001–10
10–20
20–30
30–50
14
204
45
12
3
5
73
16
4
0.9
1
98
116
54
21
0.3
34
40
19
7
Elevation (m)§
0–100
100–200
200–300
300–400
400–600
230
32
11
4
1
83
11
4
2
0.3
95
106
54
30
4
33
37
19
10
1
Aspect (8)
NA (no slope)
0–90
90–180
180–270
270–360
14
70
58
75
62
5
25
21
27
22
9
47
107
61
66
3
16
37
21
23
279.56
50.64
1695.48
48.32
13.5
2.4
81.7
2.3
49
178
151
0
13.0
47.1
39.9
0
Slope (8)
0
1–10
10–20
20–30
30–65
22.43
199.90
592.32
910.02
349.32
1.1
9.6
28.6
43.9
16.8
3
37
80
130
128
0.8
9.8
21.2
34.4
33.9
Elevation (m)§
266–500
500-1000
1000–1500
1500–2027
87.93
931.08
873.58
181.40
4.2
44.9
42.1
8.7
10
155
79
134
2.6
41.0
20.9
35.4
Aspect (8)
NA (no slope)
0–90
90–180
180–270
270–360
18.79
486.55
511.64
520.17
536.84
0.9
23.5
24.7
25.1
25.9
2
76
69
123
108
0.5
20.1
18.2
32.5
28.6
There were 290 total sampling points in Acadia and 378 total sampling points in GRSM.
à Mixed ¼ conifer and deciduous.
§ Elevation data are meters above sea level.
infrared aerial photographs collected 27–28 May 1997,
were used (Lubinski et al. 2003). Vegetation data were
reclassified as conifer forest, deciduous forest, mixed
conifer–deciduous forest, or nonforest and summarized
in the ‘‘study area’’ sections of Table 1a (ACAD) and
Table 1b (GRSM). Elevation data (meters above sea
level) were used to calculate slope (degrees) and aspect
(degrees) data layers for each park, using a 3 3 3 pixel
moving window (Table 1). Elevation data were also
used to calculate a unitless index of topographic
exposure, or landscape position, that specifies whether
a location is on an exposed ridge (negative values), a flat
area (values at or near zero), or in a depression (positive
values).
Field methods for Acadia National Park
The deposition index (DI; Fig. 1) for ACAD was
sulfur in throughfall (TF). Throughfall S has been
shown to be a conservative tracer of wet, dry, and cloud
deposition for forests of the eastern United States (see
Discussion: Indices of deposition, below, and Lindberg
and Lovett [1992], Weathers et al. [1992, 1995], Lovett
[1994]). Two hundred eighty-five resin throughfall
collectors (Simkin et al. 2004) were installed on National
Park Service property along a set of trail loops that
could be accessed in 1–2 days to minimize the risk of
precipitation events interrupting the sample collection.
Each throughfall collector was located halfway between
the trunk and drip line of a tree. Five resin bulkcollector locations were also installed in the open: one at
413-m elevation on Cadillac Mountain (see Plate 1), one
at 405-m elevation on Sargent Mountain, one at 17-m
elevation within 200 m of the coast, and two at 153–157
m elevations in the same clearing as the permanent
monitoring stations. The resin throughfall (and resin
bulk collector) sampling season was from 7 June 2000 to
18 September 2000, subdivided into three sampling
periods of 4–6 weeks each.
At each collector vegetation and terrain attributes
were recorded, as well as position coordinates. Forest
type, tree species name, tree diameter at breast height,
tree height, qualitative tree-canopy density, and understory type were recorded for vegetation in the area of
canopy influence (assumed to be an inverted cone
extending upward from the collector at 45 degrees). A
compass and a clinometer were used to record the aspect
and slope, respectively, of the terrain immediately
downslope of the collector. A digital hemispherical
photograph was taken during midday hours under
cloudy conditions to quantify tree-canopy closure.
Position coordinates were recorded with a GPS unit,
with the resulting horizontal precision ranging from
0.148 to 2.343 m. Vegetation, elevation, slope, and
aspect attributes of collectors were also extracted from
GIS data layers using GPS coordinates, and are
summarized in the sampling points section of Table 1a.
1594
Ecological Applications
Vol. 16, No. 4
KATHLEEN C. WEATHERS ET AL.
Field Methods for Great Smoky Mountains
National Park
The deposition index (DI; Fig. 1) for GRSM was the
concentration of lead (Pb) in the soil organic horizon. It
was impractical to measure spatially representative TF
in an area as large as GRSM. Lead has been shown to
preserve the pattern of long-term total deposition (see
Discussion: Indices of deposition, below, and Johnson et
al. [1982], Weathers et al. [1995, 2000]). As a test of this
method, we previously measured Pb concentrations in
forest-floor samples collected at seven NADP sites along
a regional deposition gradient in the northeastern
United States (Lovett and Rueth 1999). There was a
significant positive, linear relationship for Pb (in milligrams per kilogram) with both total (wet þ dry) S (r2 ¼
0.95, n ¼ 91 samples) and total (wet þ dry) inorganic N
(r2 ¼ 0.72, n ¼ 91 samples) deposition, demonstrating
that Pb in the forest floor reflects patterns of deposition
across latitudinal, temperature, and 3–4-fold depositional gradients spanning from Pennsylvania to Maine.
We collected Pb samples from 378 forest-floor
locations in GRSM during 1999 and 2000. Sampling
was constrained to a ;800-km2 rectangular swath (22
km wide and 37 km long; oriented diagonally from
northwest to southeast; northwest corner 35.698 N and
83.658 W; southeast corner 35.548 N and 83.238 W) that
passes through the center of the park. Samples were
collected .100 m from primary roads and were 15–25 m
upslope of trails. Areas with evidence of fire or organic
horizon disturbance were avoided. Potential sampling
areas were also rejected if they did not have an intact
forest canopy (.12 m tall or .25-cm-diameter trees)
that would have collected Pb during most if not all of the
50–60 years preceding sample collection. An additional
six samples from a deciduous forest adjacent to the
NADP site (TN11) were collected for use as an average
Pb base deposition value (BD, Fig. 1).
At the area to be sampled, a centerpoint was
established and two subsamples of the organic horizon
were collected in each cardinal direction, for a minimum
of 8 subsamples. If the organic horizon was thin with
insufficient sample volume, additional subsamples were
collected (up to 16 subsamples) within the 5-m-radius
sampling area. The subsamples of the organic horizon
(up to ;9 cm deep) were collected using a soil corer with
a diameter of 5.77 cm. Subsamples were composited into
a single sample.
Vegetation and terrain attributes, as well as position
coordinates, were recorded at each sampling area.
Forest type, trees species names, qualitative tree canopy
cover, understory type, slope and aspect were all
measured in the field. The coordinates of the centerpoint
of the organic matter sampling area were recorded with
a GPS unit, with the resulting proprietary figure-ofmerit (FOM) accuracy values ranging from 4.6 to 37.5
m. Vegetation, elevation, slope, and aspect attributes of
the 378 GRSM organic-horizon sampling locations were
also extracted from GIS data layers using GPS
coordinates, and are summarized in the ‘‘sampling
points’’ section of Table 1b.
During the second year of the study, throughfall
samples were collected, as described above, from 32
locations and were used to validate our LandMod
model.
Laboratory methods
The resin throughfall columns were extracted in 1.0
mol/L potassium iodide (KI) and analyzed for sulfate
(SO42) with a Dionex ion chromatograph as described
in Simkin et al. (2004). Three resin column ‘‘blank’’
samples were extracted along with the field resin samples
for each sample period. Raw SO42 concentrations were
converted to S fluxes (in kilograms per hectare per
sampling period) by taking into account the collection
area and the period of time that resin columns were
exposed in the field.
Forest-floor samples were dried, sieved, ground, and
digested after the methods of Weathers et al. (1995,
2000). A laboratory blank (empty crucible) and sample
of standard reference material (NIST [National Institute
of Standards and Technology] 2781, domestic sludge)
were included with each extraction batch. Digested
material was filtered through number 42 ashless Whatman paper and brought to volume with doubledeionized water, and then analyzed for Pb using an
inductively coupled plasma atomic-emission spectrometer. Raw laboratory-sample Pb concentrations (in
milligrams per liter) were converted to surface organichorizon concentrations (milligrams of Pb per kilogram).
For both methods, NIST-traceable knowns, standardsas-samples, and replicate samples were included in each
instrument run for quality-control purposes. The quality
assurance–quality control (QA/QC) results for both
methods were within 610% accuracy and precision.
Calculation of base deposition (BD)
and scaling factors (SF)
Base deposition (BD; Fig. 1) was the deposition-index
measurement at a low-elevation site during the time
period that the field sampling occurred. At ACAD the
BD was calculated by summing weekly NADP wet
deposition and CASTNET dry deposition of S (in
kilograms per hectare) from the ACAD monitoring
station for the period (6 June 2000 to 19 September
2000) that most closely matched the throughfall
sampling period. At GRSM the BD was calculated
from the mean Pb concentration (in milligrams per
kilogram) of six organic-horizon samples collected in a
forest adjacent to the NADP monitoring location.
Deposition indices (kilograms S per hectare in
throughfall for ACAD and milligrams Pb per kilogram
in organic horizon for GRSM) were then divided by
respective BDs to produce unitless deposition scaling
factors (SFs) (Fig. 1). SFs for the five nonforested
collectors at ACAD were calculated by dividing by
NADP wet deposition only. It was not possible to
August 2006
ATMOSPHERIC DEPOSITION: EMPIRICAL MODEL
calculate SFs for nonforested areas at GRSM since it
was not possible to find a range of nonforested
elevations from which we could collect undisturbed
organic-horizon samples.
Reference deposition
Reference deposition (RD) was annual wet þ dry
deposition measured at NADP and CASTNET monitoring stations (Fig. 1). These stations were ME98
(NADP wet) and ACA416 (CASTNET dry) for ACAD
(44.37398 N and 68.26068 W) and TN11 (NADP wet;
35.66458 N and 83.59038 W) and GRS420 (CASTNET
dry; 35.63318 N and 83.94228 W) for GRSM. We used
the data for calendar year 2000 as a representative year
and scaled up from point measurements to create
spatially explicit deposition maps for each of the parks,
described below (see Results: Statistical model results:
Modeled N and S deposition maps: LandMod ).
Statistical analyses
Our first goal was to create an empirically based
statistical model of measured deposition as a function
of field and GIS variables (StatMod). All statistical
analyses were conducted using the SAS statistical
package (SAS Institute 1999). A forward stepwise
multiple-regression model (n ¼ 285 sampling points
for ACAD, n ¼ 378 sampling points for GRSM) was
used with forested deposition scaling factor as the
dependent variable and several independent variables:
elevation, (elevation)2, slope, and topographic-exposure
landscape variables, as well as dummy variables for
categorical vegetation and aspect class variables. In
ACAD, additional independent variables were available
for the stepwise regression: tree diameter, tree height,
tree-canopy openness, and distance from coast variables, as well as dummy variables for the presence of
krummholz vegetation, or occurrence of fire in 1947. A
general linear model (GLM) was then constructed with
forested deposition scaling factor as a function of the
landscape variables obtained from the stepwise regression, as well as significant interaction terms that
increased model r2. Deposition scaling factors were
regressed against elevation for the five ACAD nonforested locations.
GIS–statistical model and map
Our second goal was to create an empirically based,
combination GIS–statistical model in which the independent variables were restricted to only those variables
currently available as GIS data layers. We then used this
GIS–statistical model to create a new GIS data layer of
deposition (LandMod, Fig. 1).
After removing independent variables (IVs) not
currently available as GIS data layers, such as tree
diameter, canopy openness, and presence of krummholz,
GLM parameter estimates were recalculated. Using map
algebra within GIS software, the GLM parameter
1595
estimates from the above statistical analyses were
applied to each forested pixel (having coordinates x
and y) of the GIS database to calculate each pixel’s
deposition SF (Fig. 1):
modeled SF at x; y ¼ ½b0 ðat x; yÞ þ ½b1 3ðZ1 at x; yÞ
þ ½b2 3 ðZ2 at x; yÞ
þ ½b3 3ðZ1 at x; yÞ 3ðZ2 at x; yÞ
þ þ ½bn 3ðZn at x; yÞ
ð1Þ
where Z1, Z2, . . . , Zn are independent variables, b0 is
the intercept, b1 is the landscape variable Z1 coefficient,
b2 is the landscape variable Z2 coefficient, and b3 is the
Z1 3 Z2 interaction coefficient.
Scaling factors for nonforested pixels in ACAD were
calculated using the same procedure, but with different
coefficients and landscape variable(s). Scaling factors for
nonforested pixels in GRSM were assigned ‘‘no data’’
values.
This map of predicted scaling factors was then
multiplied by reference deposition (RD) to produce
map(s) of estimated annual deposition:
modeled deposition ðtime T; ion IÞ at x; y
¼ RD ðtime T; ion IÞ 3 predicted SF at x; y:
ð2Þ
For the vegetation type, slope, and aspect variables,
where both field- and GIS-derived data were available,
the field-derived data were deemed more accurate than
GIS data and were therefore used in developing the
GIS–statistical model, but by necessity the GIS versions
of these variables were used to generate the deposition
maps. For instance, if the GIS vegetation data layer
showed a location to be under coniferous canopy, but
our field data showed that the location was under
deciduous canopy, we used the deciduous classification
in creating the GIS-statistical model.
Map validation
Sulfur deposition residuals were plotted to examine
the behavior of the LandMod model. At ACAD, a
random subset of 30 sample locations was withheld from
the full ACAD dataset and sulfur deposition residuals
(measured throughfall subtracted from LandMod predicted) were plotted for those locations. At GRSM,
throughfall collectors (109-day throughfall sampling
period from 13 June 2000 to 30 September 2000) were
colocated with 32 forest-floor Pb sampling locations and
the sulfur-deposition residuals (measured throughfall
subtracted from LandMod predicted using forest-floor
Pb index) were plotted.
RESULTS
Deposition indices
Measured throughfall S deposition at Acadia National Park (ACAD) ranged from 0.85 to 9.83 kg S/ha for
1596
Ecological Applications
Vol. 16, No. 4
KATHLEEN C. WEATHERS ET AL.
FIG. 2. Spatial distribution of measured deposition indices (DI) for (a) sulfur flux in Acadia National Park (ACAD) (white
boundary areas indicate seawater) and (b) forest-floor lead concentrations in Great Smoky Mountains National Park (GRSM).
Larger bubbles indicate a higher deposition index; scale range categories overlap to reflect the impossibility of discerning fine
differences in this type of map.
the period 7 June to 18 September 2000, with a mean of
2.54 kg S/ha. Measured deposition indices at the five
nonforested ACAD locations ranged from 1.05 to 1.56
kg S/ha (Fig. 2a). Measured Pb concentration in the
forest floor at Great Smoky Mountains National Park
(GRSM) forested locations ranged from 7.06 to 122.8
mg Pb/kg, with a mean of 33.7 mg/kg (Fig. 2b).
Base deposition
Calculated base deposition at the ACAD monitoring
station site for the period 6 June 2000 to 19 September
2000 was 1.14 kg S/ha (0.99 kg/ha of wet S þ 0.15 kg/ha
of dry S). The base deposition at the forest adjacent to
the Great Smoky Mountains NADP site was 16.61 mg
Pb/kg.
August 2006
ATMOSPHERIC DEPOSITION: EMPIRICAL MODEL
1597
GRSM, scaling factors increased with proximity to
nearest road (r2 ¼ 0.04, P , 0.0001), but this relationship
was not significant for the subset of locations less than 1
km from roads (r2 , 0.005, P , 0.63). In contrast, at
ACAD, scaling factors decreased with proximity to
nearest road (r2 ¼ 0.03, P , 0.01).
Mean deposition enhancement, as indicated by mean
scaling factors, at coniferous forest locations was higher
than at deciduous locations at both ACAD (P , 0.0001)
(Fig. 3a) and GRSM (P , 0.0001) (Fig. 4a). Scaling
factors were not correlated with slope at either ACAD
or GRSM. Scaling factors were also not correlated with
aspect classes (centered on four cardinal directions) at
ACAD but they were weakly correlated with the same
aspect classes at GRSM; north aspect had significantly
higher SFs than either west or east aspect in Tukeyadjusted multiple pairwise comparisons. At ACAD, SFs
increased with increased qualitative canopy cover (P ¼
0.004), but at GRSM the SFs decreased with increased
qualitative canopy cover (P , 0.0001). GRSM qualitative canopy-cover data were missing for 58 locations.
At ACAD only, the additional variables of distance to
coast, tree species, tree height, tree diameter at breast
height (dbh), presence of krummholz, quantitative
canopy openness, and understory class were evaluated.
FIG. 3. Relationship between deposition scaling factors
(unitless values that show relative deposition; see Fig. 1 and
Methods) and individual landscape variables for Acadia
National Park (ACAD). Landscape variables shown are: (a)
forest type (boxes enclose 25th–75th percentiles; all other
observations are marked by dots) and (b) elevation (n ¼ 285
data points, r2 ¼ 0.21).
Deposition scaling factors
The scaling factors (SFs) for ACAD ranged 12-fold,
from 0.76 to 8.76. The GRSM scaling factors ranged 17fold: 0.43–7.40. Thus, hotspots of deposition are
characterized by ;12-fold higher deposition than ‘‘cold’’
spots of deposition at ACAD and 17-fold higher at
GRSM. The spatial distribution of deposition SFs for
the two parks indicates that many––but not all––of the
high-elevation locations have high SF values (Fig. 2a, b).
Statistical Model Results
Deposition as a function of individual landscape
independent variables.—The SFs were positively correlated with elevation at both ACAD (r2 ¼ 0.21, P ,
0.0001, first-order fit, Fig. 3b) and GRSM (r2 ¼ 0.42, P
, 0.0001, second-order fit, Fig. 4b). Scaling factors were
also positively correlated with increased topographic
exposure at both ACAD (r2 ¼ 0.12, P , 0.0001) and
GRSM (r2 ¼ 0.14, P , 0.0001), although topographic
exposure significantly covaried with elevation (r2 ¼ 0.33
and 0.30 at ACAD and GRSM, respectively). At
FIG. 4. Relationship between scaling factors (unitless values
that show relative deposition) and individual landscape
variables for Great Smoky Mountains National Park (GRSM).
Landscape variables shown are: (a) forest type (boxes enclose
25th–75th percentiles; all other observations are marked by
dots) and (b) elevation (n ¼ 378 data points, r2 ¼ 0.42).
1598
KATHLEEN C. WEATHERS ET AL.
FIG. 5. Scaling factors (unitless values that show relative
deposition) as a function of elevation and vegetation type for
(a) Acadia National Park and (b) Great Smoky Mountain
National Park.
Actual S deposition decreased from as much as 13.2 kg
Sha1yr1 (SF of 3.3) at 90 m from the coast to ,6 kg
Sha1yr1 of S at ;300 m from the coast, based on a
transect of four collectors that extended from coast to
inland, demonstrating a classic edge effect (e.g., Weathers et al. 1995, 2001). At the scale of the whole-park
landscape, when all 285 data points were considered,
there was a positive slope, but no statistically significant
relationship between deposition and distance to the
coast. There were significant differences in SFs under
different tree species (P , 0.0001) as well: Picea spp. had
the highest mean SF (2.9) and Acer pennslyvanicum L.
had the lowest mean SF (1.3, data not shown).
Deciduous species had mean scaling factors ,2 and
coniferous species all had mean scaling factors .2, with
the exception of Thuja occidentalis L. and Tsuga
canadensis (L.) Carr., which had mean SFs of 1.7 and
1.5, respectively. Scaling factors had a positive correlation with tree height (P ¼ 0.04), no correlation with tree
dbh or presence of krummholz, and a weak negative
correlation with quantitative hemispherical-photo canopy openness (P ¼ 0.01). Scaling factors for locations
with an understory were lower than for locations
without an understory (P ¼ 0.005), and for locations
with an understory there were no significant differences
Ecological Applications
Vol. 16, No. 4
in SF between coniferous and deciduous understories
(P ¼ 0.25).
At GRSM only, location with respect to the central
topographic divide, as well as rhododendron understory
presence, were evaluated as potential drivers of deposition. Deposition was somewhat higher on the Tennessee side of the central divide than on the North Carolina
side (P , 0.01). Deposition decreased at GRSM with
the presence of rhododendron, an evergreen, broadleaved, understory plant (P , 0.0001; note: rhododendron data were missing for 53 locations), but the
presence of rhododendrons is also uncommon at
elevations .1800 m.
Deposition as a function of multiple, landscape
independent variables: StatMod.—Stepwise multiple regression and general linear model (GLM) procedures for
ACAD forests showed that elevation and forest type
(coniferous or deciduous class) together explained 32%
of the variation in deposition (P , 0.0001, Fig. 5a).
When elevation, forest type, and elevation 3 forest type
interaction terms were combined with tree dbh, the
statistical model explained a total of 37% of the
variation in deposition scaling factors (P , 0.0001,
Table 2a). For the five nonforested open locations at
ACAD there was a weak positive correlation of
deposition SF with elevation in meters (deposition ¼
1.00760 þ 0.00081(elevation), r2 ¼ 0.60, P , 0.12).
Stepwise multiple regression and GLM procedures for
GRSM forests showed that elevation, (elevation)2, and
forest type (coniferous or deciduous class) together
explained 46% of the variation in deposition (Fig. 5b).
When these variables were combined with a forest type 3
elevation interaction term and slope (DEM-derived), the
statistical model explained a total of 48% of the variation
in our deposition indices (P , 0.0001, Table 2c).
GIS–statistical model and scaling-factor maps.—After
randomly withdrawing 30 locations at ACAD for map
validation purposes (see Discussion: How good are the
maps?, below), stepwise regression and GLM procedures
were used to select among the variables that were only
represented in GIS data layers so that they could be used
to generate deposition maps (Fig. 1, Table 2b). Of the
variables selected for the statistical model in the previous
step, at ACAD, krummholz presence and tree dbh did
not have comparable GIS data, and this, combined with
lower sample size (n ¼ 255 data points), reduced slightly
the explanatory power of the GIS–statistical relationships (r2 ¼ 0.31, P , 0.0001) for the GIS–statistical
model.
When the scaling-factor parameter estimates from
Table 2b were combined with the appropriate GIS data
layers, the resulting deposition SF map for ACAD
forests had values ranging from 1.1 to 4.5, a fourfold
difference. Nonforested SFs based on the elevation
regression ranged from 1.0 to 1.4.
At GRSM, 32 colocated deposition index locations
were withdrawn for map validation purposes (Table 2d).
The explanatory power of the resultant equations was
August 2006
ATMOSPHERIC DEPOSITION: EMPIRICAL MODEL
1599
TABLE 2. Statistical model parameters of deposition scaling factor (unitless values that show relative deposition; see Fig. 1 and
Methods) as a function of field- and GIS-measured (StatMod) landscape variables for (a) Acadia National Park, ACAD (n ¼ 285
data points), and (c) Great Smoky Mountains National Park, GRSM (n ¼ 378 data points). The table also reports mapping
equation parameters of deposition scaling factor as a function of GIS-measurable landscape variables for (b) ACAD (n ¼ 255
data points) and (d) GRSM (n ¼ 346 data points).
Variable
Coefficient
P
Partial r2
Model r2
Acadia National Park, ACAD
a) Statistical model parameters
Intercept
Elevation (m)
Conifer presence
Tree dbh (cm)
Elevation 3 conifer pres.
0.98720
0.00265
0.03922
0.01521
0.00482
,0.0001
0.8676
0.0098
0.0003
0.212
0.105
0.019
0.031
0.212
0.317
0.336
0.367
b) Mapping equation parameters
Intercept
Elevation (m)
Conifer presence
Mixed forest presence
Distance to coast (m)
Elevation 3 conifer pres.
1.60345
0.00263
0.03248
0.48995
0.00015
0.00481
0.0193
0.9061
0.0380
0.0399
0.0008
0.171
0.080
0.018
0.006
0.032
0.171
0.251
0.269
0.275
0.307
Great Smoky Mountains National Park, GRSM
c) Statistical model parameters
Intercept
(Elevation)2
Elevation (m)
Conifer presence
Slope (from DEM)
Elevation 3 conifer pres.
3.32326
0.000001995
0.00427
0.15380
0.01106
0.000648
,0.0001
,0.0001
0.6113
0.0250
0.0070
0.322
0.096
0.044
0.007
0.008
0.322
0.418
0.416
0.468
0.479
d) Mapping equation parameters
Intercept
(Elevation)2
Elevation (m)
Conifer presence
Elevation 3 conifer pres.
3.69836
0.000002195
0.00464
0.26948
0.000748
,0.0001
,0.0001
0.3737
0.0030
0.331
0.125
0.044
0.014
0.331
0.455
0.499
0.513
somewhat increased with the lower sample size (n ¼ 348)
(r2 ¼ 0.51, P , 0.0001). When the SF parameter
estimates from Table 2d were combined with the
appropriate GIS data layers, the resulting deposition
SF map for GRSM had values ranging from 0.72 to
4.56, a sixfold variation.
Modeled S and N deposition maps: LandMod.—The
ACAD scaling-factor map was multiplied by ACAD
reference deposition of 3.00 kg Nha1yr1 (2.51 kg
Nha1yr1 of wet deposition from NADP þ 0.49 kg
Nha1yr1 of dry deposition from CASTNET) for all
of calendar year 2000 to produce a 2000 modeled
ACAD N-deposition map (Fig. 6). Nitrogen-deposition
values ranged from 3.0 to 14 kg Nha1yr1, with an
area-weighted mean of 5 kgha1yr1 of N for the entire
ACAD study area (Fig. 6). For sulfur, the ACAD
scaling-factor map was multiplied by ACAD reference
deposition of 5.55 kg Sha1yr1 (4.87 kg Sha1yr1 of
wet deposition þ 0.68 kg Sha1yr1 dry deposition) for
all of calendar year 2000 to produce a 2000 modeled
ACAD S-deposition map (Fig. 6). Sulfur-deposition
values ranged from 5.6 to 25 kg Sha1yr1, with an areaweighted mean of 9.5 kg Sha1yr1.
Using the same method as for ACAD, the GRSM
scaling-factor map was multiplied by GRSM reference
deposition of 6.8 kg Nha1yr1 (2.9 kg Nha1yr1 wet
deposition þ 3.9 kg Nha1yr1 of dry deposition from
the NADP and CASTNET programs, respectively) for
calendar year 2000 to produce a 2000 modeled GRSM
N-deposition map (Fig. 7). Nitrogen-deposition values
ranged from 5 to 31 kg Nha1yr1, with an areaweighted mean of 10 kg Nha1yr1 for the entire park
area. For sulfur, the GRSM scaling-factor map was
multiplied by GRSM reference deposition of 9.1 kg
Sha1yr1 of S (5.8 kg Sha1yr1 of wet deposition þ
3.3 kg Sha1yr1 of dry deposition) for all of calendar
year 2000 to produce a 2000 modeled GRSM Sdeposition map. The resulting LandMod GRSM Sdeposition map shows values that ranged from 7 to 42
kg Sha1yr1, with an area-weighted mean of 14 kg
Sha1yr1 (Fig. 7).
Validation of LandMod deposition maps
Sulfur-deposition residuals (LandMod predicted minus throughfall measured) were plotted as a function of
elevation to examine model function (Fig. 8a and b).
Thirteen out of 30 validation locations (43%) at ACAD
and 13 out of 32 validation locations (41%) at GRSM
had mismatches in vegetation type between GIS and
field data and were converted to the field value before
calculating residuals. After making those changes, the
correlation between measured and predicted deposition
1600
Ecological Applications
Vol. 16, No. 4
KATHLEEN C. WEATHERS ET AL.
As a further validation of the ACAD model, we
compared the LandMod predicted deposition with
sulfur deposition from a frequently used regional model
of deposition (ClimCalc [available online];8 Ollinger et al.
1993), as well as with reference data from the NADP
and CASTNET monitoring stations (Fig. 9). Although
our validation data suggest that the LandMod underestimates high deposition at high elevations, this analysis
demonstrates that the existing estimates for those
locations are considerably lower than what our approach produces (Fig 9).
DISCUSSION
Indices of deposition
FIG. 6. Modeled atmospheric deposition of N and S for the
year 2000 to Mount Desert Island study area of Acadia
National Park (ACAD). Modeling used the empirically based
LandMod (see Methods: GIS–statistical model and map).
Deposition was calculated as deposition scaling factor (unitless)
3 year 2000 base deposition from 52 weeks of NADP wet and
CASTNET dry deposition (5.5 kgha1 yr1 of S and 3.0
kgha1yr1 of N). Intermediate breakpoints for N deposition
are arbitrary integer values. Intermediate breakpoints for S
deposition are therefore the N breakpoints 3 1.850. Scale range
categories overlap to reflect the impossibility of discerning fine
differences in this type of map.
was quite high for ACAD (r2 ¼ 0.64, P , 0.0001) and
GRSM (r2 ¼ 0.80, P , 0.0001), although it was not a 1:1
relationship. At both ACAD and GRSM, LandMod
slightly overestimated deposition at low elevations and
underestimated deposition at high elevations.
The residual plots show interesting and consistent
patterns between the two sites. The model behaves
differently at low-elevation, low-deposition sites than at
high-elevation, high-deposition sites, and for different
vegetation types. At ACAD, the model overpredicts at
low elevations, more so for coniferous and mixed
vegetation than for deciduous. The residuals were closer
to zero for deciduous vegetation at low elevations. At
high elevation, the model under-predicts for coniferous
vegetation. In the subset of data used for validation
there is little representation of deciduous vegetation at
high elevation. For GRSM, the model also over-predicts
at low elevation for deciduous vegetation (the primary
vegetation type at low elevation) and under predicts
rather significantly for high-elevation conifer forests,
relative to field measurements.
To estimate spatial patterns of atmospheric deposition, our surrogate, or index, measures of total
deposition were sulfate in throughfall and lead in forest
floor. In general, important criteria for the use of
surrogates for total deposition are that the surrogate
element or ion used (1) is relatively biologically inert, (2)
preserves the pattern of deposition, and (3) represents
total (rain, snow, dry, cloud/fog) deposition.
Throughfall is the movement of water from the forest
canopy to the forest floor. Throughfall S has been shown
to be a robust surrogate measure of total S deposition
(Lindberg and Lovett 1992). The strengths of throughfall (TF) are that it encompasses all forms of deposition
(wet, gaseous, particulate, fog), and does not involve
assumptions about canopy or terrain characteristics.
The major weakness of TF in our application is that it
should be sampled continuously over all seasons and for
many years to reflect more accurately the long-term,
average patterns of deposition. Throughfall measurements also can be labor intensive to collect, and
stemflow (SF, water that moves from the canopy down
the stems of plants [to the forest floor]) should be
included (although often it is a small percentage of total
flux) but is more difficult to sample. Furthermore, TF
and SF could be influenced by leaching or uptake of S in
the canopy, although this has been shown to be minor in
most studies (Lindberg and Garten 1988, Cape et al.
1992).
In this study, because of the difficulty of accessing
several hundred (;300) TF collectors in roadless areas,
it was possible to measure throughfall S only during the
summer growing season, when deciduous trees have
leaves. In creating annual deposition maps from these
data, it was necessary to assume that the landscape
patterns of deposition for the entire year are similar to
the patterns we measured during the leaf-on season. An
analysis of the four-year average (1999–2003) NADP
plus CASTNET depositon data by season and by
element (S and N) showed surprisingly little variation
in wet-to-dry ratios: N dry deposition varied between
3% and 6% of total deposition across seasons, and S dry
8
hwww.pnet.sr.unh.edu/climcalci
August 2006
ATMOSPHERIC DEPOSITION: EMPIRICAL MODEL
1601
FIG. 7. Modeled atmospheric deposition of N and S for the year 2000 to Great Smoky Mountain National Park (GRSM),
using the empirically based LandMod. Deposition was calculated as deposition scaling factor (unitless) 3 year 2000 base deposition
from 52 weeks of NADP wet and CASTNET dry deposition (9.1 kgha1yr1 of S and 6.8 kgha1yr1 of N). Intermediate
breakpoints for N deposition are arbitrary integer values. Intermediate breakpoints for S deposition are therefore the N
breakpoints 3 1.345. Scale range categories overlap to reflect the impossibility of discerning fine differences in this type of map.
deposition varied between 3% and 10%. Nonetheless, we
recognize that seasonal variations in leaf area, prevailing
wind direction (aspect effect), and fog frequency and
extent can influence deposition patterns, so we recommend that future research be directed at measuring TF þ
SF fluxes year-round. We also note that winter
conditions (i.e., deciduous leaf drop) can be readily
simulated in LandMod to account for a known change
in leaf area.
Lead in the forest floor has also been used to indicate
regional patterns of pollution (Johnson et al. 1982,
Graustein and Turekian 1989) as well as more-local
deposition patterns (Weathers et al. 1995, 2000). Lead,
or any other atmospherically deposited and biologically
inert substance that is used to indicate the deposition of
some other element (S, for example) should be tested for
its appropriateness. This is especially true for the
pollutant Pb, whose emissions have been reduced
significantly since the 1970s (Johnson et al. 1995). Once
Pb has been deposited to the forest floor it moves,
though usually quite slowly (decades to centuries), from
forest floor to mineral soil (e.g., Wang et al. 1995).
Recent reports have documented faster-than-expected
Pb movement from forest floor to mineral soil (Johnson
et al. 1995, Kaste et al. 2003, Watmough et al. 2004),
which suggests the need for some independent measure
of the relationship between this deposition index and
atmospheric deposition to support its use as a deposition
index. Although we previously demonstrated the utility
of Pb as an index of deposition (Lovett 1994, Weathers
et al. 1995, 2000), for this work we performed an
additional test by measuring Pb concentrations in forest
floor samples from forested sites adjacent to seven
NADP monitoring stations. As noted in Methods (see
Field methods for Great Smoky Mountains National
Park, above), these locations span a 3–4 fold depositional gradient, as well as a large latitudinal gradient.
The concentration of Pb in forest floor at these sites was
strongly correlated with total deposition of both S and
N (P , 0.001), supporting its use for detecting patterns
of deposition in this study.
One of the most significant values of Pb as a surrogate
measure of atmospheric deposition is that continuous
sampling is not necessary—Pb in the forest floor reflects
long-term patterns of deposition, integrating both
seasonal and interannual variations. It is possible to
collect many more samples over a much larger area if
samples need to be collected only once. The weakness is
that, although Pb is carried on submicron particles that
exhibit some characteristics of gas behavior, there are no
true gaseous forms of Pb; therefore, it may not be an
accurate index of pattern for substances in which the
pattern of gaseous deposition is significantly different
from that of particulate and cloud-water deposition.
Finally, Pb in forest floor is truly a surrogate measure of
deposition; here we were not interested in Pb deposition,
per se, but rather, our goal was to use Pb as an indicator
of patterns of S and N deposition.
Because it integrates deposition over time, and
because sampling is logistically more feasible, we had
intended to use only Pb as the index of deposition for
this study at both parks. However, because approximately one third of the area of ACAD burned in the
1940s, the pattern of Pb in the forest floor may have
been altered by fire. Therefore, we measured sulfate in
throughfall as an index of deposition at ACAD.
1602
KATHLEEN C. WEATHERS ET AL.
FIG. 8. Residuals of predicted vs. observed sulfur deposition plotted as a function of elevation for (a) Acadia (ACAD)
and (b) Great Smoky Mountains (GRSM) National Parks. GIS
and field vegetation data were matched for this analysis.
Deciduous, coniferous, and mixed (ACAD only) stands are
shown.
Elevation and vegetation as
landscape-independent variables
For both ACAD and GRSM, elevation and vegetation were the independent variables that explained most
of the variance in the statistical models (StatMod) of
deposition.
Elevation.—It is not surprising that elevation was
found to be a primary control on the landscape pattern
of deposition. Orographic effects drive precipitation
increases with elevation (Lovett and Kinsman 1990).
Dry deposition may also increase with elevation,
because the total exposed surface area for collection of
particles and gases is often greater (Stachurski and
Zimka 2000) and windspeeds are higher, both of which
can influence dry deposition (Lovett and Kinsman
1990). For inland mountain ranges, cloud-water deposition is greatest at high elevations (Lovett and Kinsman
1990, Weathers et al. 2000). Previous studies have shown
nonlinear increases with elevation, and have attributed
these step increases at mid-to-high elevation to cloud
deposition, which occurs only above a clearly defined
cloud base (Siccama 1974, Lovett et al. 1999, Weathers
Ecological Applications
Vol. 16, No. 4
et al. 2000). As a result of increased wet, dry, and cloudwater deposition we had anticipated that total ion
deposition would increase with elevation and that
elevation would therefore be an important independent
variable in our model. Since the elevation range at
ACAD was ,500 m, it is not surprising that the
deposition relationship is weaker than at GRSM, where
there was a 1500-m elevation range. Fog-deposition
relationships with elevation are likely to vary between
inland and coastal regions as well. This may further
explain the stronger relationship of deposition with
elevation at GRSM compared to ACAD. Casual
observations suggest that fog deposition at ACAD near
the coast may alternate between being highest at
mountaintops and at low elevations, the latter when
advective fogs move from ocean to land. Systematic data
are needed to either refute or substantiate this observation, but if in fact there is only an intermittent cloud
base of irregular elevation this could be a contributing
factor that helps explain why ACAD had only a firstorder elevation relationship with deposition while
GRSM had a second-order relationship between elevation and deposition.
Vegetation.—It is also not surprising that vegetation
was found to be a primary control of the landscape
pattern of deposition. In general, evergreen coniferous
vegetation and mixed coniferous–deciduous forests, with
their greater leaf-area index and year-round collection
surface, have much more efficient scavenging surfaces
for particles (including cloud droplets) and gases than
does most deciduous vegetation. A few studies have
demonstrated differences in total deposition among
these broad vegetation classes (Ivens 1990, Weathers et
al. 1992, 1995, Lovett et al. 1999, Weathers et al. 2000).
It is important to reiterate that the GRSM Pb index
incorporates both the high magnitude and long duration
of leaf-area collection surface for coniferous forests,
FIG. 9. Comparison of S deposition (103-day sampling
period) at 32 validation data points from Acadia National Park
(ACAD; see Results for description of validation points) using
the LandMod described in this paper, deposition estimates for
each of the points based on Ollinger et al. (1993), and the
NADP and CASTNET programs; these monitoring data have
only one value for all locations in the region. The 1 : 1 line is
also shown.
August 2006
ATMOSPHERIC DEPOSITION: EMPIRICAL MODEL
1603
PLATE 1. A view of Cadillac Mountain from Otter Point, Acadia National Park, Maine, USA. Photo credit: Amanda Elliot.
while we are using the flux of S in summer throughfall S
to model annual deposition at ACAD.
At ACAD, we were able to discern differences in
deposition among species as indicated by the relative
difference in scaling factors. These data confirm that
most coniferous canopies are better scavengers than
deciduous canopies—there is on the order of twofold
enhancement.
Attempts to correlate leaf area with deposition
produced inconclusive, and in some cases unexpected,
results. Qualitative identification of ACAD leaf area as
low, intermediate, or high was a slightly better predictor
of deposition than quantitative canopy openness from
hemispherical photos, presumably because the latter did
not take into account overlapping leaf-area surfaces in
the upper canopy. Both measures suggested that
increased leaf area was associated with increased
deposition; these correlations were weak but in the
expected, positive direction. By contrast, a qualitative
assessment of canopy cover at GRSM suggested that
increased leaf area was (weakly) associated with
decreased deposition. In addition, locations with an
understory had slightly lower deposition at both ACAD
and GRSM. These results are not readily explicable with
the data that we have; more appropriate quantitative
measures of vegetation structure relevant to deposition
of dry and fog deposition are needed.
Elevation–Vegetation interaction.—One of the benefits
of this study is that by sampling hundreds of locations it
was possible to simultaneously address elevation and
vegetation variables. The strong interaction between
vegetation type and elevation we observed reflects the
much stronger relationship between elevation and
deposition for coniferous and mixed forests compared
to deciduous forest. Higher wind speeds at high
elevation create the potential for higher dry deposition.
Increased clouds at high elevation (for locations with a
distinct cloud base) create the potential for higher fog/
cloud deposition. It could be argued that this potential
for high deposition at high elevation is fully captured
only where there are coniferous forests with high leaf
area (Weathers et al. 2000).
How good are the maps?.—There are several important factors that determine the accuracy of the
deposition maps produced here. Errors in sampling
and analyzing point measurements of deposition chemistry would decrease accuracy, but results from our
quality-control procedures (610% precision and accuracy) lead us to believe that this is not a major source of
error. When comparing predicted vs. observed data,
errors in GIS data (e.g., vegetation) are clearly an
important factor that decrease the accuracy of the
deposition maps, as indicated by the major improvements in the validation analyses when the erroneous GIS
data were corrected to match the observed field data. It
seems reasonable to assume that, as remotely sensed
data with a resolution of 1 m or better become available,
the number of mismatches between GIS and field
vegetation should decrease. The sources of error
described above must be taken into consideration.
However, we think that the greatest and most interesting
sources of uncertainty in the deposition maps arise from
as-yet unquantified or unidentified variables that cause
real variability in deposition—variables that are not part
of our StatMod or LandMod, but are evident in the
1604
Ecological Applications
Vol. 16, No. 4
KATHLEEN C. WEATHERS ET AL.
pattern displayed in our validation data (Fig. 8a, b,
NIST/SEMATECH 2005).
The statistical relationships between independent
variables and deposition indices were highly significant,
yet they explained less than half of the variance in the
data. What explains the rest of the variance in actual
deposition? It is likely that the independent variables
that explain the rest of the variance, at a broad scale, are
those that affect the capture of particles and gases by
vegetation, such as canopy roughness and orientation
and topographic exposure. Everything else held equal,
there is an optimum canopy leaf area and morphology
for maximum deposition. For example, Lovett and
Reiners (1989) modeled cloud deposition as a function
of canopy leaf-area index (LAI) and found highest
deposition at a mid-range LAI. Our data suggest a
similar enhancement for predominately coniferous
forest—those regions with up to 25% deciduous
canopies—compared with pure deciduous or coniferous
forest at ACAD (Fig. 3a); however, this was not true at
GRSM (Fig. 4a). Canopy and leaf architecture and
morphology are also likely to influence deposition. Our
limited data from ACAD showing large differences in
fluxes among coniferous trees suggests that there may be
factors in addition to LAI that control deposition. To
wit, the highest measured deposition in the ACAD
validation data set––the biggest outlier for predicted vs.
measured values—was a large-statured (7-m-tall) conifer
that was situated on an exposed ridge. This is just the
kind of tree-specific––or site-specific––depositional environment that no general model can readily replicate.
A major variable controlling small-scale patterns in
deposition is likely to be vegetation exposure, including
trees whose crowns are exposed above their neighbors
and trees growing in unsheltered locations such as ledges
and promontories. A tree with high exposure is likely to
get much higher deposition than another tree of
equivalent LAI and height that is sheltered by neighbors. This has been demonstrated in several studies of
forest edges (Weathers et al. 1992, 1995, 2001, Lindberg
and Owens 1993). While we were able to generate a GIS
index of topographic exposure using elevation models of
the ground surface, we believe the horizontal and spatial
resolution of those data are not yet good enough to
generate an index of the combination of canopy
architecture and exposure that is relevant to rates of
deposition. However, the development of high-resolution remotely sensed LIDAR (light detection and
ranging) data shows promise in indicating both the
elevation of tree canopies and the density of tree
canopies, which should help considerably in filling this
data gap. Another challenge is the scale at which
topographic exposure might matter, which is likely to
be less than 10s of meters rather than the 30 3 30-m pixel
resolution of most GIS data layers. If it were possible to
measure canopy architecture and exposure (or surrogates) easily, and better yet, if they were available as GIS
data layers, landscape models of deposition could
probably be made more accurate. Thus an important
next step for deposition research will be to develop a
way to quantify different canopy LAIs and architectures
along with their exposure, perhaps using remotely
sensed data (e.g., LIDAR), and examine their relationship to deposition.
Map validation
The residual data show that at ACAD, the LandMod
map most significantly underestimated deposition in
high-elevation regions; these points also represent the
highest measured deposition values. The three data
points that represent high-elevation conifer forests show
the biggest discrepancies. The GRSM validation plots
show much bigger discrepancies for high-elevation
regions, indicating that high-elevation data reduce the
homogeneity of variance for the LandMod model. That
the variance is high, especially at high elevation, is
unsurprising given the large scatter of the primary data
(Fig. 3b).
We cannot say definitively how much of an influence
our use of Pb in forest floor to create the LandMod had
on the validation comparison. However the consistency
between the two validation plots (Fig. 8a, b) suggests
that there are missing variables in LandMod rather than
an effect of using different deposition indices at ACAD
vs. GRSM. We might predict that LandMod would
underestimate tree-with-leaf-on patterns of S deposition
(i.e., the measured values) since the Pb index integrates
deposition over seasons and years. It is likely that actual
sulfur deposition for a short time during the season of
highest deposition (i.e., when we made the S flux
measurements for the validation data) would be higher
than predicted deposition. In addition, LandMod
estimates deposition to a 30 3 30 m pixel that is
characterized by an average elevation and vegetation
type. We measured deposition to a small area that
surrounded individual trees.
Clearly these models need refinement to be able to
accurately predict deposition to heterogeneous landscapes. The validation data sets can be used to suggest
next steps, such as the need to identify and quantify
controls on deposition to high-elevation conifer forests.
However, it is important to consider our results within
the perspective of the data that are currently available.
LandMod appears to be conservative: it slightly overpredicts for low-elevation, low-deposition environments,
but significantly underestimates input for high-elevation,
high-deposition environments. The deposition data that
are currently available for these landscapes––exclusive
of our LandMod––either show no spatial variability (for
example, GRSM has one NADP þ CASTNET value for
the whole park), or some spatial variability over the
region (i.e., the Ollinger et al. [1993] regional-deposition
model for the northeastern United States), but do not
capture spatial variability at the scale of one to hundreds
of kilometers. Thus, LandMod S deposition for ACAD,
compared with S predictions based on Ollinger et al.
August 2006
ATMOSPHERIC DEPOSITION: EMPIRICAL MODEL
(1993; a widely used regional deposition model (see also
footnote 8), and with the monitoring-station data alone,
captures the response of measured deposition to both
elevation and vegetation type that the other two
estimates cannot (Fig. 9). For GRSM, in addition to
the monitoring station wet þ dry deposition estimate,
cloud deposition was estimated at the peak of Clingman’s Dome for the time period over which we sampled
(G. M. Lovett, unpublished manuscript). For this highelevation region, the predicted cloud þ wet þ dry
deposition based on the monitoring-station data, 27 kg
S/ha for a four-month growing season, is more than
twice the S measured in the throughfall at Clingman’s
Dome. Our LandMod predicts less than half the
measured S in throughfall for this same location. Again,
it appears that our measured as well as our modeled
deposition is quite conservative (i.e., underestimates
total deposition) compared to existing data.
How transferable is this approach?
The empirical scaling factors and models we have
created are likely to be applicable to other chemical
species that are deposited from the atmosphere in
approximately the same proportions and by the same
mechanisms as N and S are deposited. In addition,
where strong and predictable relationships can be
established between chemical species x and S, the
models might be extended to other atmospherically
deposited nutrients and pollutants.
Our approach may also be applicable to other sites.
We think that elevation and vegetation type are likely to
be important driving variables for deposition in other
mountain ranges around the world, albeit the coefficients may differ from landscape to landscape depending
upon such factors as the dominant mode of deposition,
for example fog deposition, dry deposition, and whether
snow is a dominant proportion of wet deposition,
patchiness of the landscape, and its elevational span.
Both of these topics—application of the model to other
chemical species and to other sites—warrant further
research and comparison.
Updating maps
The statistical relationships that underpin the atmospheric-deposition maps developed here are not expected to change from year to year; the independent
variables that drive LandMod (vegetation and elevation)
are likely to be stable across the landscape. However,
major vegetation changes as a result of an anthropogenic disturbance such as logging, or natural disturbances such as fire, major storms or insect outbreaks would
affect the spatial pattern of deposition. These large-scale
changes could be accommodated by recalculating
scaling factors. On the other hand, reference deposition
data from the NADP and CASTNET monitoring
stations do change over time. Since reference deposition
data are incorporated in the last step of creating the
deposition map, it is straightforward to update the
1605
deposition maps to reflect this temporal change. If data
layers with finer resolution or other key data become
available, they too could be substituted for the old data
layers at the stage where scaling-factor maps are
generated (Fig. 1).
Who cares?
Our park-wide LandMod deposition maps are based
on spatially explicit inputs, and as such they can be used
to identify regions of high and of low deposition
(hotspots and coldspots). But, at the scale of the whole
park, does it matter? Do we need to have spatially
explicit estimates of deposition? One way to answer this
question is to compare our area-weighted total deposition with currently available data from the NADP and
CASTNET monitoring stations. Our models indicate
that total, area-weighted N deposition is much greater
than suggested by the local monitoring-station data
(e.g., Fig. 9). ACAD has 70% greater, and GRSM 50%
greater, N deposition when our LandMod is compared
to NADP þ CASTNET deposition estimates. These
results demonstrate that even the average total deposition across mountainous landscapes is substantially
underestimated by using data from single-point, lowelevation monitoring stations.
At places such as GRSM where a complete, spatially
explicit biologic inventory is being compiled, it is
possible to overlay deposition maps of hotspots with
known sensitive species distributions to identify populations that are likely to be at greatest risk of damage
from air pollution. Knowledge of deposition patterns
can also be used to good advantage in designing effects
research. For instance, identifying strong deposition
gradients in areas where other factors (e.g., soil type or
temperature) are relatively constant could facilitate
comparative studies of the effects of deposition on
plants and animals. In general, knowing spatial variability in deposition across a landscape can enhance tests
of ecological responses to a range of inputs.
CONCLUSION
Estimates of atmospheric deposition in complex
terrain are currently inadequate; they are either incomplete (spatially representative, but only for wet deposition, for example), or not spatially explicit. In this
research, we have developed an empirically based model
for scaling up to the landscape (tens to hundreds of
kilometers) from point measurements to get spatially
explicit estimates of atmospheric deposition. Our LandMod was created by applying a statistical model to GIS
data layers. National monitoring network data were
used for reference deposition; the deposition estimate
from which to scale up. The resultant maps show
several-fold variation in deposition across the landscape
of two national parks. Vegetation type and elevation are
the independent variables that most strongly controlled
deposition.
1606
KATHLEEN C. WEATHERS ET AL.
This modeling effort and our validation exercises and
explorations were revealing: they confirmed that there is
large variance in deposition, especially at high elevations
and that, with current tools, landscape-scale (tens to
hundreds of kilometers) models are unlikely to be able to
capture extreme deposition over areas of small spatial
extent. Future research should test LandMod in other
mountainous environments, expanding it to other
chemical species, and refining it to account for
(currently) unexplained variation in deposition. We have
offered suggestions for how this might be accomplished.
We predict that the latter effort will benefit significantly
as new remote sensing technology and resulting GIS data
layers become available and are field tested.
ACKNOWLEDGMENTS
This research was funded by the EPA-NPS PRIMENet
Program, the National Park Service, and through a grant to the
Institute of Ecosystem Studies (IES) by the A. W. Mellon
Foundation. We are tremendously grateful for the support and
assistance offered to us by personnel from both Acadia
(especially D. Manski, B. Breen, B. Gawley, and K. Anderson)
and Great Smoky Mountains (especially J. Renfro) National
Parks. We received important advice and assistance from the
IES Analytical Laboratory, and field and laboratory assistance
from K. Schwarz, L. Chambers, D. Lewis, and J. Beeler. We
thank F. Vermeylen for statistical advice, and H. A. Ewing, T.
Blett, K. Morris, B. Breen, T. Maniero, J. Renfro, and two
anonymous reviewers for helpful comments on drafts of the
manuscript. This is a contribution to the program of the
Institute of Ecosystem Studies.
LITERATURE CITED
Aber, J. D., C. L. Goodale, S. V. Ollinger, M. Smith, A. H.
Magill, M. E. Martin, R. A. Hallett, and J. L. Stoddard.
2003. Is nitrogen deposition altering the nitrogen status of
northeastern forests? BioScience 53:375–389.
Anderson, J. B., R. E. Baumgardner, V. A. Mohnen, and J. J.
Bowser. 1999. Cloud chemistry in the eastern United States, as
sampled from three high-elevation sites along the Appalachian Mountains. Atmospheric Environment 33:5105–5114.
Baumgardner, R. E., S. S. Isil, T. F. Lavery, C. M. Rogers, and
V. A. Mohnen. 2003. Estimates of cloud water deposition at
mountain acid deposition program sites in the Appalachian
Mountains. Journal of the Air & Waste Management
Association 53:291–308.
Baumgardner, R. E., T. F. Lavery, C. M. Rogers, and S. S. Isil.
2002. Estimates of the atmospheric deposition of sulfur and
nitrogen species: Clean Air Status and Trends Network,
1990–2000. Environmental Science and Technology 36:2614–
2629.
Cape, J. N., L. J. Sheppard, D. Fowler, A. F. Harrison, J. A.
Parkinson, P. Dao, and I. S. Paterson. 1992. Contribution of
canopy leaching to sulfate deposition in a Scots pine forest.
Environmental Pollution 75:229–236.
Driscoll, C. T., G. B. Lawrence, A. J. Bulger, T. J. Butler, C. S.
Cronan, C. Eager, K. F. Lambert, G. E. Likens, J. L.
Stoddard, and K. C. Weathers. 2001. Acidic deposition in the
northeastern United States: sources and inputs, ecosystem
effects, and management strategies. BioScience 51:180–198.
Fenn, M. E., R. Haeuber, K. Tonnessen, J. S. Baron, S.
Grossman-Clarke, D. Hope, D. A. Jaffe, S. Copeland, L.
Geiser, H. M. Rueth, and J. O. Sickman. 2003. Nitrogen
emissions, deposition and monitoring in the Western United
States. BioScience 53:391–403.
Finkelstein, P. L., T. G. Ellestad, J. F. Clarke, T. P. Meyers, D.
B. Schwede, E. O. Hebert, and J. A. Neal. 2000. Ozone and
Ecological Applications
Vol. 16, No. 4
sulfur dioxide dry deposition to forests: observations and
model evaluation. Journal of Geophysical Research—Atmospheres. 105:15365–15377.
Goodale, C. L., J. D. Aber, and W. H. McDowell. 2000. The
long-term effects of disturbance on organic and inorganic
nitrogen export in the White Mountains, New Hampshire.
Ecosystems 3:433–450.
Graustein, W. C., and K. K. Turekian. 1989. The effects of
forests and topography on the deposition of sub-micrometer
aerosols measured by lead-210 and cesium-137 in soils.
Agricultural and Forest Meteorology 47:199–220.
Grimm, J. W., and J. A. Lynch. 2004. Enhanced wet deposition
estimates using modeled precipitation inputs. Environmental
Monitoring and Assessment 90:243–268.
Hedin, L. O., J. J. Armesto, and A. H. Johnson. 1995. Patterns of
nutrient loss from unpolluted, old-growth temperate forests:
evaluation of biogeochemical theory. Ecology 76:493–509.
Hedin, L. O., L. Granat, G. E. Likens, T. A. Buishand, J. N.
Galloway, T. J. Butler, and H. Rodhe. 1994. Steep declines in
atmospheric base cations in regions of Europe and North
America. Nature 367:351–354.
Hettelingh, J. P., R. J. Downing, P. A. M. de Smet, editors.
1995a. Calculation and mapping of critical loads in Europe.
RIVM Report Number 259101004. Coordination Center for
Effects, Bilthoven, The Netherlands.
Hettelingh, J. P., M. Posch, P. A. M. DeSmet, and R. J.
Downing. 1995b. The use of critical loads in emission
reduction agreements in Europe. Water, Air, and Soil
Pollution 85:2381–2388.
Holland, E. A., B. H. Braswell, J. Selzman, and J-F. Lamarque.
2005. Nitrogen deposition onto the United States and
western Europe: synthesis of observations and models.
Ecological Applications 15:38–57.
Ivens, W. P. M. F. 1990. Atmospheric deposition onto forests:
an analysis of the deposition variability by means of
throughfall measurements. Dissertation. Nederlandse Geografische Studies 118. University of Utrecht, Utrecht, The
Netherlands.
Johnson, C. E., T. G. Siccama, C. T. Driscoll, G. E. Likens, and
R. E. Moeller. 1995. Changes in lead biogeochemistry in
response to decreasing atmospheric inputs. Ecological
Applications 5:813–822.
Johnson, A. H., T. G. Siccama, and A. J. Friedland. 1982.
Spatial and temporal patterns of lead accumulation in the
forest floor in the Northeastern United States. Journal of
Environmental Quality 11:577–580.
Kaste, J. M., A. J. Friedland, and S. Sturup. 2003. Using stable
and radioactive isotopes to trace atmospherically deposited
Pb in montane forest soils. Environmental Science and
Technology 37:3560–3567.
Kelly, V. R., G. M. Lovett, K.C. Weathers, and G. E. Likens.
2002. Trends in atmospheric concentration and deposition
compared to regional and local pollutant emissions at a rural
site in southeastern New York, USA. Atmospheric Environment 36:1569–1575.
Kimball, K. D., R. Jagels, G. Gordon, K. C. Weathers, and J.
Carlisle. 1988. Differences between New England coastal fog
and mountain cloud water chemistry. Water, Air and Soil
Pollution 39:383–393.
Lavery, T. F., C. M. Rogers, H. K. Howell, and M. C. Burnett.
2003. Clean Air Status and Trends Network (CASTNET)
2002 annual report. Prepared by MACTEC (EPA Contract
number 68-D-03-052) for U.S. Environmental Protection
Agency, Office of Atmospheric Programs, Washington, D.C.,
USA.
Likens, G. E., and F. H. Bormann. 1995. Biogeochemistry of a
forested ecosystem, Second edition. Springer-Verlag, New
York, New York, USA.
Lindberg, S. E., and C. T. Garten. 1988. Sources of sulfur in
forest canopy throughfall. Nature 336:148–151.
August 2006
ATMOSPHERIC DEPOSITION: EMPIRICAL MODEL
Lindberg, S. E., and G. M. Lovett. 1992. Deposition and forest
canopy interactions of airborne sulfur: results from the
Integrated Forest Study. Atmospheric Environment 26A:
1477–1492.
Lindberg, S. E., and J. G. Owens. 1993. Throughfall studies of
deposition to forest edges and gaps in montane ecosystems.
Biogeochemistry 19:173–194.
Lovett, G. M. 1994. Atmospheric deposition of nutrients and
pollutants in North America: an ecological perspective.
Ecological Applications 4:629–650.
Lovett, G. M., and J. D. Kinsman. 1990. Atmospheric
pollutant deposition to high-elevation ecosystems. Atmospheric Environment, Part A, General Topics 24:2767–2786.
Lovett, G. M., and W. A. Reiners. 1989. Factors controlling
throughfall chemistry in a balsam fir canopy—a modeling
approach. Biogeochemistry 8:239–264.
Lovett, G. M., and H. Rueth. 1999. Soil nitrogen transformations in beech and maple stands along a nitrogen
deposition gradient. Ecological Applications 9:1330–1344.
Lovett, G. M., A. W. Thompson, J. B. Anderson, and J. J.
Bowser. 1999. Elevational patterns of sulfur deposition at a
site in the Catskill Mountains, New York. Atmospheric
Environment 33:617–624.
Lubinski, S., K. Hop, and S. Gawler. 2003. Revised edition.
U.S. Geological Survey–National Park Service Vegetation
Mapping Program: Acadia National Park, Maine. Project
Report. hhttp://biology.usgs.gov/npsveg/acad/report.pdfi
Lynch, J. A., V. C. Bowersox, and J. W. Grimm. 2000. Changes
in sulfate deposition in eastern USA following implementation of Phase I of Title IV of the Clean Air Act Amendments
of 1990. Atmospheric Environment 34:1665–1680.
Lynch, J. A., J. W. Grimm, and V. C. Bowersox. 1995. Trends
in precipitation chemistry in the United States: a national
perspective, 1980–1992. Atmospheric Environment 29:1231–
1246.
MacKenzie, M. D. 1993. The vegetation of Great Smoky
Mountains National Park: past, present, and future. Dissertation. University of Tennessee, Knoxville, Tennessee,
USA.
Meyers, T. P., P. Finkelstein, J. Clarke, T. G. Ellestad, and P.
F. Sims. 1998. A multilayer model for inferring dry
deposition using standard meteorological measurements.
Journal of Geophysical Research-Atmospheres 103:22645–
22661.
Mohnen, V. A., and R. J. Vong. 1993. A climatology of cloud
chemistry for the eastern United States derived from the
mountain cloud chemistry project. Environmental Review 1:
38–54.
Nanus, L., D. H. Campbell, G. P. Ingersoll, D. W. Clow, and
M. A. Mast. 2003. Atmospheric deposition maps for the
Rocky Mountains. Atmospheric Environment 37:4881–
4892.
NIST/SEMATECH [National Institute of Standards and
Technology/Semiconducting Manufacturing Technology].
2005. e-Handbook of statistical methods. hhttp://www.itl.
nist.gov/div898/handbook/, 26 January 2005i
Ollinger, S. V., J. D. Aber, G. M. Lovett, S. E. Millham, R. G.
Lathrop, and J. M. Ellis. 1993. A spatial model of
atmospheric deposition for the northeastern United States.
Ecological Applications 3:459–472.
SAS Institute. 1999. SAS/STAT user’s guide, version 8. SAS
Institute, Cary, North Carolina, USA.
1607
Schlesinger, W. H. 1997. Biogeochemistry: an analysis of global
change. Second edition. Academic Press, San Diego,
California, USA.
Siccama, T. J. 1974. Vegetation, soil, and climate on the
Green Mountains of Vermont. Ecological Monographs 44:
325–349.
Simkin, S. M., D. N. Lewis, K. C. Weathers, G. M. Lovett, and
K. Schwarz. 2004. Determination of sulfate, nitrate, and
chloride in throughfall using ion-exchange resins. Water, Air,
and Soil Pollution 153:343–354.
Stachurski, A., and J. R. Zimka. 2000. Atmospheric input of
elements to forest ecosystems: a method of estimation using
artificial foliage placed above rain collectors. Environmental
Pollution 110:345–356.
Summers, P. W. 1995. Time trend of wet deposition acidifying
potential at five ecological monitoring sites in eastern Canada
1981–1993. Water, Air, and Soil Pollution 85:653–658.
van Leeuwen, E. P., G. P. J. Erisman, and J. W. Erisman. 1996.
Mapping wet deposition of acidifying components and base
cations over Europe using measurements. Atmospheric
Environment 30:2495–2511.
Vitousek, P. M., and W. A. Reiners. 1975. Ecosystem
succession and nutrient retention—hypothesis. BioScience
25:376–381.
Vong, R. J., J. T. Sigmon, and S. F. Mueller. 1991. Cloud water
deposition to Appalachian forests. Environmental Science
and Technology 25:1014–1021.
Wang, E. X., F. H. Bormann, and G. Benoit. 1995. Evidence of
complete retention of atmospheric lead in the soils of
northern hardwood forested ecosystems. Environmental
Science and Technology 29:735–739.
Watmough, S. A., T. C. Hutchinson, and P. J. Dillon. 2004.
Lead dynamics in the forest floor and mineral soil in southcentral Ontario. Biogeochemistry 71:43–68.
Weathers, K. C., M. L. Cadenasso, and S. T. A. Pickett. 2001.
Forest edges as nutrient and pollutant concentrators:
potential synergisms between fragmentation, forest canopies,
and the atmosphere. Conservation Biology 15:1506–1514.
Weathers, K. C., et al. 1988. Cloudwater chemistry from 10
sites in North America. Environmental Science and Technology 22:1018–1026.
Weathers, K. C., et al. 1986. A regional acidic cloud/fog water
event in the eastern United States. Nature 319:657–658.
Weathers, K. C., and G. M. Lovett. 1998. Acid deposition
research and ecosystem ecology: synergistic successes. Pages
195–219 in M. L. Pace and P. M. Groffman, editors.
Successes, limitations and frontiers in ecosystem science.
Springer-Verlag, New York, New York, USA.
Weathers, K. C., G. M. Lovett, and G. E. Likens. 1992. The
influence of a forest edge on cloud deposition. Pages 1415–
1423 in S. E. Schwartzand and W. G. N. Slinn, coordinators.
Precipitation Scavenging and Atmosphere-Surface Exchange.
Volume 3. The Summers volume: applications and appraisals. Hemisphere Publishing, Washington, D.C., USA.
Weathers, K. C., G. M. Lovett, and G. E. Likens. 1995. Cloud
deposition to a spruce forest edge. Atmospheric Environment
29:665–672.
Weathers, K. C., G. M. Lovett, G. E. Likens, and R. Lathrop.
2000. The effect of landscape features on deposition to
Hunter Mountain, Catskill Mountains, New York. Ecological Applications 10:528–540.