Stoichiometry of hydrological C, N, and P losses across

Preparing to load PDF file. please wait...

0 of 0
100%
Stoichiometry of hydrological C, N, and P losses across

Transcript Of Stoichiometry of hydrological C, N, and P losses across

Click Here
for
Full Article

GLOBAL BIOGEOCHEMICAL CYCLES, VOL. 22, GB1026, doi:10.1029/2007GB003005, 2008

Stoichiometry of hydrological C, N, and P losses across climate and
geology: An environmental matrix approach across New Zealand
primary forests
M. E. McGroddy,1,2 W. T. Baisden,3,4 and L. O. Hedin1
Received 2 May 2007; revised 22 September 2007; accepted 7 December 2007; published 21 March 2008.
[1] Hydrologic losses can play a key role in regulating ecosystem nutrient balances, particularly in regions where baseline nutrient cycles are not augmented by industrial deposition. We used first-order streams to integrate hydrologic losses at the watershed scale across unpolluted old-growth forests in New Zealand. We employed a matrix approach to resolve how stream water concentrations of dissolved organic carbon (DOC), organic and inorganic nitrogen (DON and DIN), and organic and inorganic phosphorus (DOP and DIP) varied as a function of landscape differences in climate and geology. We found stream water total dissolved nitrogen (TDN) to be dominated by organic forms (medians for DON, 81.3%, nitrate-N, 12.6%, and ammonium-N, 3.9%). The median stream water DOC:TDN:TDP molar ratio of 1050:21:1 favored C slightly over N and P when compared to typical temperate forest foliage ratios. Using the full set of variables in a multiple regression approach explained approximately half of the variability in DON, DOC, and TDP concentrations. Building on this approach we combined a simplified set of variables with a simple water balance model in a regression designed to predict DON export at larger spatial scales. Incorporating the effects of climate and geologic variables on nutrient exports will greatly aid the development of integrated Earth-climate biogeochemical models which are able to take into account multiple element dynamics and complex natural landscapes.
Citation: McGroddy, M. E., W. T. Baisden, and L. O. Hedin (2008), Stoichiometry of hydrological C, N, and P losses across climate and geology: An environmental matrix approach across New Zealand primary forests, Global Biogeochem. Cycles, 22, GB1026, doi:10.1029/2007GB003005.

1. Introduction
[2] In terrestrial ecosystems, losses of nutrients via hydrological vectors (i.e., via ground and stream water) are of fundamental importance for determining ecosystem functional properties such as nutrient balances [Likens and Bormann, 1995] and nutrient limitation [Hedin et al., 2003]. In addition to the amount of nutrients lost, whether they occur as dissolved inorganic (e.g., NO3À or PO43À) or dissolved organic (e.g., dissolved organic nitrogen or phosphorus; DON and DOP, respectively) forms have important implications for our understanding of the factors that control nutrient budgets and cycles within ecosystems. For example, elevated losses of the inorganic forms of a nutrient can be taken as evidence of sufficient supply of that nutrient relative
1Department of Ecology and Evolutionary Biology, Princeton University, Princeton, New Jersey, USA.
2Now at Department of Environmental Sciences, University of Virginia, Charlottesville, Virginia, USA.
3Landcare Research, Palmerston North, New Zealand. 4Now at Environmental Isotope Section, National Isotope Centre, GNS Science, Lower Hutt, New Zealand.
Copyright 2008 by the American Geophysical Union. 0886-6236/08/2007GB003005$12.00

to biological demands at the ecosystem scale [A˚ gren and Bosatta, 1988; Aber et al., 1989; Hedin et al., 2003]. In turn, high losses of organic forms of a nutrient can constitute a plant-unavailable path of nutrient loss, which over time acts to constrain the buildup of nutrients in internal ecosystem pools [Hedin et al., 1995; Vitousek et al., 1998; Hedin et al., 2003; Neff et al., 2003; Baisden and Amundson, 2003].
[3] Hedin et al. [1995] pointed out that we know rather little about how and why losses of major nutrients vary across forested ecosystems that have little to no history of modern human impact. And yet, such information can offer insights about preindustrial conditions of nutrient cycling – a kind of natural baseline against which modern ecosystem changes and trajectories of change can be better understood [Hedin et al., 1995; Lewis, 2002; Smith et al., 2003].
[4] It is, perhaps, reasonable to expect that differences in nutrient losses across forested watersheds are shaped by the broad ‘‘state factors’’ proposed by Jenny [1941] as ultimate determinants of ecosystem properties: parent material, climate, topography, time of ecosystem development, and potential biota. The actual biota found in any given location would be a subset of the potential biota, to a large extent shaped by local variations in these state factors within and across landscapes. This ‘‘state factor approach’’ has been an important aspect of understanding geographically broad

GB1026

1 of 14

GB1026

MCGRODDY ET AL.: HYDROLOGIC NUTRIENT LOSSES IN NZ FORESTS

GB1026

variations in nutrient accumulation, availability and cycling within terrestrial ecosystems [Jenny, 1930; Harradine and Jenny, 1958; Syers et al., 1970; Walker and Syers, 1976; Crews et al., 1995; Vitousek, 2004; Hedin et al., 2003].
[5] While other studies have considered DOC and NO3À export from terrestrial ecosystems [Clark et al., 2000; Smith et al., 2003; Detenbeck et al., 2003], most of these concern larger streams that drain areas subject to human influences from land use change and/or atmospheric deposition. Even in studies designed to measure ‘‘background’’ concentrations, the effects of atmospheric deposition need to be taken into account [Clark et al.,. 2000; Smith et al., 2003]. In fact, recent analyses of nitrogen export from rivers throughout northern temperate regions have found that human sources dominate, with substantial effects on patterns of nitrogen export [Boyer et al., 2002].
[6] To better understand nutrient cycles in unimpacted temperate forests, Hedin and colleagues studied patterns of nitrogen loss from old-growth forests throughout unpolluted areas of temperate South America [Hedin et al., 1995; Perakis and Hedin, 2001]. They found a geographically consistent pattern of loss that differed dramatically from polluted, old-growth forests in the northern hemisphere: exceedingly low losses of plant-available N forms (NO3À and NH4+) coupled with high concentrations and losses of DON [Hedin et al., 1995; Perakis and Hedin, 2001]. This striking loss pattern is consistent with the idea that these forests are naturally N-poor (thus low losses of plantavailable forms), and that the relatively high rates of DON export contribute to and maintain the scarcity of N availability at the ecosystem scale [Hedin et al., 1995; Perakis and Hedin, 2001; Hedin et al., 2003].
[7] What is less clear, however, is whether these patterns are diagnostic for unpolluted temperate forests in other regions, and across broad variations in Jenny’s state factors. Here we take advantage of the low relative impact of human activities that can be found across old-growth forests in different locations across the North and South Islands of New Zealand. In addition, we take advantage of existing broad variations in Jenny’s state factors across the New Zealand landscape. Building on previous studies of inorganic and organic losses of nitrogen from South American watersheds by Hedin and colleagues, we add observations on losses of dissolved organic carbon (DOC) and dissolved inorganic (DIP) and organic (DOP) phosphorus. To integrate carbon, nitrogen and phosphorus we take a stoichiometric approach, in which we examine variations in ratios of these different elements. The largely unpolluted nature of these forests, and the existence of dramatic differences in climate, geology and vegetation across New Zealand forests, offers a unique opportunity to test and expand earlier findings from unpolluted South American forests.
[8] We are interested in how climate and geology and their interactions shape variations in stoichiometry and forms of hydrologic C, N and P losses. To address this question we used both direct measures of climate and geologic variables as well as current vegetation cover as an integrated index of both factors. We adopted the geographically extensive sampling approach of Perakis and Hedin [2002], which differs from traditional intensive

studies of a single watershed. The geographically extensive approach is designed to include natural variability due to differences in environmental factors across watersheds (e.g., climate, slope, flow paths, et cetera) as well as temporal variations within watersheds. The approach assumes that within-watershed variations by and large remain small when compared to between-watershed trends that are governed by differences in state factors. While the approach captures more residual variation than intensive, single watershed studies, it is precisely this variation that we wish to examine for evidence of macroscopic structure across state factors. Sampling the headwaters of first-order streams also minimizes variability in nutrient forms associated with in-stream processing and connectivity between streams and riparian wetlands.
[9] Of central concern is the cycling and losses of N since New Zealand, like many temperate forests is thought to be primarily limited by N [Vitousek and Howarth, 1991]. If, in addition, New Zealand forests display high DON losses as found in South American forests, our study may offer the opportunity to determine how differences in DON losses across forests depend on climatic, vegetation and physical factors such as rainfall, plant community composition and geological substrate. Phosphorus provides an interesting contrast to N dynamics. In contrast to N, ecosystem P pools are highly dependent on the parent material characteristics with respect to mineral composition and resistance to weathering. For example, Dillon and Kirchner [1975] noted greater P export from watersheds draining sedimentary rocks when compared igneous rocks. More recent efforts have begun to consider parent material, soil properties, watershed slope, and vegetation as determinants of watershed P export [e.g., van der Perk et al., 2007; Heathwaite et al., 2003; Cooper et al., 1992]. Most these studies, however, focus on agriculture, pastures, and other managed landscapes, in which erosion of soil particles commonly control hydrological P export. None of these studies offer the kind of empirical comparison across broad geological and climatic factors that we present here.
[10] Regulation of P export, beyond biological demand, may be largely due to the reactivity of the phosphate ion in soils. Under basic conditions phosphate rapidly binds to calcium to form insoluble calcium phosphates while under acidic conditions phosphate binds insolubly to iron and aluminum oxides and hydroxides on clay particle surfaces [McDowell and Wood, 1984; Sposito, 1989]. Organic P compounds are considered to be much more mobile and likely to leach from terrestrial ecosystems because of their reduced chemical reactivity. The amount of P exported in dissolved versus particulate forms is spatially and temporally highly variable, varying with soil texture, [Djodjic et al., 1999], amount and type of water flow in the soil [Chapman et al., 1997; Djodjic et al., 1999] and disturbance [Schelde et al., 2006]. In stream water, particulate constituents represent both P exported in particulate form from soil as well as interactions of dissolved forms of P with the streambed and stream organic matter load. Previous work in undisturbed forest ecosystems has shown that dissolved forms dominate stream TN, TP and TC pools [Kortelainen et al., 2006], despite localized hotspots of particulate C, N and P export within the landscape [Scott et al., 2006].

2 of 14

GB1026

MCGRODDY ET AL.: HYDROLOGIC NUTRIENT LOSSES IN NZ FORESTS

GB1026

Figure 1. Map of soil parent materials across New Zealand used with permission from Land Environments of New Zealand (LENZ). Sampling locations for this study are indicated on the inset.

[11] We here ask three questions concerning nutrient losses from these New Zealand forests: (1) How variable is the stoichiometry of dissolved C:N:P losses across a matrix of unpolluted ‘‘baseline’’ forests? (2) Are variations in N and P losses related to one or more key state factors? (3) How predictable are nitrogen losses (DON in particular) across the NZ landscape?
[12] These questions regarding the organization of nutrient losses can be of tremendous value for the development of integrated Earth-climate biogeochemical models across complex natural landscapes, in which the ‘‘baseline’’ biogeochemical system must be represented to understand human-induced changes. An important litmus test for such models is their ability to predict patterns of nutrient loss across forests and across biomes.
2. Methods
2.1. Study Sites
[13] We sampled first-order streams in 97 forested watersheds on both the North and South Islands of New Zealand

extending from 39° 120 S to 45° 290 S (Figure 1). Across sampling sites, mean annual temperature (MAT) ranges from 6.4 to 13.9°C and mean annual rainfall (MAR) from 1170 to 6909 mm. To capture maximum seasonal variation we collected water in both winter and summer periods (September 2003, March 2004, and August 2004) with 42% of all watersheds sampled in both seasons. Sampling decisions were not based on weather or flow conditions.
[14] Our goal was to sample small streams draining intact native forests with no evidence of human disturbance within recent decades, across a wide range of climate and soil parent materials. We considered most sites pristine, but could not rule out limited human disturbance. Cold montane and warm lowland sites might be especially sensitive as they are either inherently slower to recover from disturbances (cold montane) or located more closely to human populations (warm lowland sites). We viewed disturbances including fires, windstorms and extreme erosion or deposition events, or limited human disturbance from pre-European settlement, as influences consistent with natural sources of variability we did not wish to exclude.

3 of 14

GB1026

MCGRODDY ET AL.: HYDROLOGIC NUTRIENT LOSSES IN NZ FORESTS

GB1026

[15] We targeted our sampling of watersheds to capture the range of common parent materials in New Zealand, but also sought to maximize the range of climatic variability available within the most widespread parent materials (Figure 1). We used soil parent data mapped in the New Zealand Land Resources Inventory (LRI) and grouped for the Land Environments of New Zealand (LENZ) classification system [Leathwick et al., 2003]. We used an ArcGIS (Redlands, California, United States) coverage based on original 1:250,000 mapping (LRI version 1) overlain by improved mapping at 1:50,000 for some regions (LRI version 2). Where possible, parent materials were validated against recent 1:250,000 geologic maps [Rattenbury et al.,1998; Begg and Johnston, 2000; Turnbull, 2000; Nathan et al., 2002; Turnbull and Allibone, 2003; Edbrooke, 2005], geologic field studies [Wandres et al., 1998] and field inspection of our watersheds. In cases where older LRI mapping disagreed with recent geologic mapping or field data, parent materials were reclassified to the most appropriate units within the LRI/LENZ system.
[16] We either avoided sampling, or discarded sites on the basis of subsequent GIS investigation using criteria that could not be evaluated in the field. We excluded sites for which the vegetation cover was characterized by LRI mapping from the 1970s as early successional vegetation (<20– 30 years old). We did not exclude sites in which some large trees were selectively removed 50– 150 years ago, as such exclusion would have limited our analysis of native forests in productive environments where protected forest remnants are rare. We eliminated another seven sites because estimates of N deposition indicated substantial influence from adjacent animal agriculture, in excess of 2 kg N haÀ1 aÀ1 based on the deposition model given by Parfitt et al. [2006]. Geographic coordinates, forest type, parent material and climate data are provided for each site in Table S11.
2.2. Field and Laboratory Analyses
[17] We collected duplicate 60 ml stream water samples from each stream. Samples were immediately filtered through prerinsed Gelman 0.45 micron glass fiber filters and stored in well-rinsed HDLP bottles. Samples were kept cold (<10°C) and subsequently frozen until analysis. During March and August 2004 we measured stream temperature and pH using a handheld pH/conductivity probe (Fisher Accumet 1003 with an Ag/AgCl reference probe). Streamflow was measured with a Pygmy meter in larger streams, and by direct measurement of water volume collected during a time interval in small streams.
[18] We measured ammonium-N (NH4+-N) colorimetrically on an Astoria-Pacific automated spectrophotometer (Astoria-Pacific Incorporated, Clackamas, Oregon, United States) using an alkaline phenol and hypochlorite color reaction amplified by sodium nitroferricyanide. Nitrate-N (NO3À-N) was determined using a Dionex Ion Chromatograph (AS-4 column, Dionex Corporation, Sunnyvale, California, United States). We measured DIP and TDP colorimetrically using the Astoria-Pacific automated spec-
1Auxiliary materials are available in the HTML. doi:10.1029/ 2007GB003005.

trophotometer and an acidified ammonium molybdate color reaction. Total dissolved P was analyzed as PO4-P after high-temperature persulfate digestion [Cabrera and Beare, 1993]. We measured TDN and DOC on a Shimadzu Total Carbon analyzer with a Total Nitrogen unit (Shimadzu Corporation, Kyoto, Japan). Dissolved organic N and P were then calculated as the difference between the concentrations of TDN or TDP and the inorganic forms (PO4-P for P and NH4-N + NO3-N for N).
2.3. Spatial Data and GIS
[19] We obtained spatial data from a variety of sources and matched it against our watersheds using ArcGIS, on the basis of sample locations recorded with handheld GPS receivers. Forest composition data was acquired from EcoSat classifications combining recent Landsat ETM+ imagery with LRI vegetation mapping [Dymond and Shepherd, 2004]. Elevation and slope were obtained from the Landcare Research digital elevation model. Average climate surfaces for a number of variables were obtained either directly from the underlying data used to construct the LENZ ecoregion classification [Leathwick et al., 2002a] or layers created using similar methods [Leathwick et al., 2002b] Topographic, climate, and vegetation data were available as 20, 25, and 15 m rasters, respectively, and were queried over watershed areas identified using algorithms in GRASS GIS 5.2 (ITC, Trento, Italy), or where the watersheds could not be identified from the DEM, by averaging the data within a 50 m radius centered 50 m upslope from the sampling site.
[20] Variables describing soil parent material were obtained directly from the underlying data used to create the ecoregions classification for New Zealand, which compiled average soil properties (exchangeable Ca, drainage class, acid P, induration, and particle size for all soil parent material classes recorded in the LRI [Leathwick et al., 2002a]). We classified these parent materials into the ten broad groups for statistical analysis. We created two additional spatially based categorical variables for interpreting our data. The first recognized a binary distinction rather than a continuum of soils with the value 1 identifying poorly and very poorly drained soils, and 0 signifying imperfectly, well and excessively drained soils. We estimated this property manually in the field for our sites, and reclassified the LENZ drainage layer to obtain a raster map of NZ. Second, we noted that the LENZ did not capture differences between soils with andic properties and other soils [Perakis and Hedin, 2007]. For soils formed on andesitic or basaltic parent material we defined categorical variable ‘‘andesitic/mafic’’ as 1, while soils derived from sandstones with a partial contribution of andesitic or basaltic sources were given the value 0.5 and for all other soils the value 0.
2.4. Seasonal and Hydrological Variability
[21] While our approach incorporates within-watershed variations (as discussed above in section 2.1) it assumes such variability to be relatively low if strong patterns appear across state factors. We explicitly tested whether stream chemistry within watersheds was stable relative to trends

4 of 14

GB1026

MCGRODDY ET AL.: HYDROLOGIC NUTRIENT LOSSES IN NZ FORESTS

GB1026

Table 1. Dissolved Nutrient Concentrations in First-Order
Streams, Under Mature Forest Cover in Chile and Argentina and New Zealanda

Nutrient

New Zealandb

Chile/Argentinac

NO3-N, mg LÀ1 NH4-N, mg LÀ1 N organic, mg LÀ1
NO3-N, % NH4-N, % N organic, % PO4-P, mg LÀ1 P organic, mg LÀ1
PO4-P, % P organic, % DOC, mg LÀ1

7.4 (1.8 – 22.5) 2.5 (1.6 – 3.5) 47.5 (34.6 – 74.7) 12 (2 – 34) 4 (2 – 6) 81 (57 – 93) 3.0 (1.1 – 10.8) 2.7 (1.1 – 4.2) 61 (22 – 97) 44 (12 – 79) 3.3 (1.6 – 5.7)

1.9 (0.02 – 7.1) 4.9 (0.5 – 11.0) (8 – 135) 5 (0.1 – 18) 15 (3 – 36) 80 (61 – 97) — — — — —

aData are reported as medians with the 25 and 75 quartile values reported
in parentheses. bData from this study. cData from Perakis and Hedin [2002].

trations and ratios as the dependent factors. As described above, variables were log transformed where necessary to meet the assumptions of parametric statistical analyses. The above statistical analyses were performed using SYSTAT vers.10.2 (SSI, San Jose, California, United States).
[25] Regressions were performed in JMP (SAS Institute, Cary, North Carolina, United States) using backward stepwise regression with a significance threshold of p > 0.001 to exclude variables. A map of DON export was produced using raster algebra in ArcGIS using the relationships described in Table 5, according to the following regression equation

log10 ðDON ExportÞ ¼ b þ Si¼1:4 aiPi

ð1Þ

where b is the regression intercept, and ai and Pi are regression coefficients and predictor variables. DON export was calculated as an annual flux in kg N haÀ1 aÀ1 as follows

between watersheds, by revisiting roughly half (47 of 97) of our watersheds in the summer following the winter sampling. This comparison revealed strong similarity within watersheds, across seasons. Summer versus winter correlations were in all cases highly significant (p < 0.001) across watersheds: TDP (Pearson’s r = 0.98), TDN (0.80), PO43À (0.98); NO3À (0.76); DOC (0.74); DON (0.72). We found no significant correlation for NH4+-N, however, likely because levels were uniformly near the detection limit. Further analysis indicated that seasonal variations in concentrations were minor relative to the much larger trends that we observed across watersheds and state factors. Specifically, type II regression slopes between winter versus summer values were close to unity for TDP (0.91 ± 0.04), TDN (0.83 ± 0.14), DOC (0.90 ± 0.04), and DON (0.99 ± 0.11). For PO43À, the slope (1.47 ± 0.07) indicated somewhat higher concentrations in winter than summer.
[22] In addition, during our summer campaign we manually estimated water flow rates for individual watersheds. We found no correlation between water chemistry and flow rate (for all constituents, r < 0.3 and p > 0.16). We conclude that differences in flow rate could not explain the observed differences between watersheds during this period. It is, however, difficult to extrapolate this finding to periods with different flow regimes. In our predictive model of N fluxes we have therefore included an error analysis that considers the potential bias due to correlations between concentration and flow (see section 3).
2.5. Data Analysis
[23] We used ANOVAs to examine relationships between absolute and relative (i.e., ratios among nutrients) nutrient concentrations and the independent factors parent material and climate as well as forest type which was understood to depend on and integrate both climate and geologic conditions. In all analyses data were log (base 10) transformed to meet assumptions of the ANOVA.
[24] The state factor ANOVA approach was complemented by a multiple regression analysis where climatic, geologic and vegetative site variables were included in a backward stepwise linear regression with nutrient concen-

ðDON ExportÞ ¼ fDONg  ðMAP À PET þ SmonthsDÞ ð2Þ
where MAP is mean annual precipitation, PET is Penman potential evapotranspiration, and SmonthsD is the sum of monthly soil water deficits [Leathwick et al., 2002a]. These terms collectively represent an annual water balance.
3. Results
3.1. Overall Nutrient Concentrations
[26] In Table 1 we summarize patterns of dissolved nutrients across all watersheds: a total of 97 sampled watersheds covering broad geographic (10° latitude) and climatic (7° MAT and 5400 mm rainfall) and geologic (10 substrate and geomorphic classifications) ranges. Our findings show that losses of nitrogen were dominated by organic forms, with DON (median 81.3%) contributing substantially more than NO3À (median 12.6%) or NH4+ (median 3.9%) to total N concentrations. This pattern of N loss is similar to unpolluted temperate forests in southern South America [Perakis and Hedin, 2002, 2007]. While organic N forms dominated in most streams, we found that inorganic forms (sum of NO3À and NH4+) dominated in 18 out of 98 watersheds. In these streams we measured concentration ratios of inorganic to organic N that ranged from 1.02 to 3.91 with a single outlier as high as 9.49.
[27] In contrast, phosphorus concentrations were dominated by inorganic over organic forms (DOP), but the pattern was weak with a median inorganic P of 67.2% and a range between lower and upper quartiles of 24 to 98%.
[28] Dissolved organic carbon (DOC) showed a ten-fold variation across watersheds, with lower and upper quartiles ranging from 0.5 to 11 mg C LÀ1. While concentrations of DON were closely correlated with these variations in DOC (Pearson’s r = 0.70, p < 0.001, N = 96), we found less correlation between DOP and DOC concentrations (Pearson’s r = 0.30, p < 0.01, N = 98). In addition, DOP was only weakly correlated with DON (r = 0.26, p = 0.01, N = 96).
3.2. Stoichiometry
[29] We summarize in Figure 2 variations in the stoichiometric ratios of total and dissolved organic C, N and P

5 of 14

GB1026

MCGRODDY ET AL.: HYDROLOGIC NUTRIENT LOSSES IN NZ FORESTS

GB1026

Figure 2. Distribution of nutrient ratios for (a) DOC:DON, (b) DOC:DOP, and (c) total N:total P and DON:DOP. All ratios are molar ratios. Solid arrows indicate mean C:N, C:P, and N:P ratios for foliage in temperate broadleaf forests, while dashed arrows indicate the same for litter in temperate broadleaf forests [from McGroddy et al., 2004].

across our study watersheds. Concentrations of inorganic N and/or P were often too low for the inorganic N:P ratio to be of practical interest. Molar ratios of DOC relative to DON or DOP displayed substantial variability, increasing two-fold and five-fold, respectively, between upper and lower quartiles (33 to 79 for DOC:DON and 444 to 2445 for DOC:DOP). The DON:DOP ratio was also quite variable, ranging from a low quartile of 11 to a high quartile value of 41.
[30] The overall median ratio of total dissolved C:N:P of 1050:21:1 favored C slightly over N and P when compared to median C:N:P ratios observed in plant foliage across temperate broadleaf forests worldwide (990:28:1, indicated by arrows in Figure 2; [McGroddy et al., 2004]). Differences were slighter and less in favor of C relative to N or P when compared to litter (1699:29:1). If we consider only dissolved organic forms of nutrients the median ratio of 2640:49:1 favored the export of C over N or P more strongly, with C:N ratios similar to litter ratios, but C:P ratios that were substantially greater than both foliage and litter. In addition the DON:DOP ratio favors export of N over P when compared to either plant N:P ratios or the total dissolved stoichiometry of our streams (c.f. open versus hatched bars in Figure 2c).
[31] We found the highest DOC:DON ratios in watersheds on the geological parent materials mudstone/sandstones (median of 179, N = 9) followed closely by ultrabasic sandstones (median of 134, N = 5). Watersheds on calcareous parent materials had the highest DOC:DOP ratios measured (median of 10,486, N = 2). In contrast, watersheds on basaltic parent materials consistently produced the lowest measurements of both DOC:DON and DOC:DOP (medians of 25 and 963, respectively, N = 3).
[32] Dissolved organic N:P ratios were highest in watersheds on calcareous parent materials (median of 97, N = 2) and lowest in watersheds on mudstone/sandstone parent materials (median of 20, N = 9). While watersheds on calcareous parent materials also had high total dissolved N:P ratios (median of 50, N = 2), so did watersheds on unconsolidated till (median of 55, N = 15). We found the lowest total dissolved N:P ratios in watersheds on either basaltic or andesitic parent materials (median of 12.6 and 12.8 and N = 16 and 3, respectively).
[33] In order to examine the role of climate on stream water stoichiometry we classified our sites using the system developed by Holdridge [1947] based on mean annual rainfall and mean annual temperature. Our sites were distributed among three life zone categories: cool, temperate wet forest, cool temperate rain forest, and warm temperate moist forest. We found the highest DOC:DON ratios in watersheds classified as cool temperate rain forest (median of 72, N = 32). These sites also had the highest DOC:DOP ratios (median of 5034, N = 32) with similar values found in the warm temperate moist forest sites (median of 4995, N = 2). Dissolved organic C:TDP ratios in the rain forest sites were lower than those measured in the other two categories (median 2111 compared to 3064 and 8810 in cool wet and warm moist forests, N = 32, 63, and 2, respectively). Dissolved organic N:P ratios were lowest in watersheds classified as cool temperate wet forest (median of 45, N =

6 of 14

GB1026

MCGRODDY ET AL.: HYDROLOGIC NUTRIENT LOSSES IN NZ FORESTS

GB1026

Table 2. Results From ANOVA Analyses Using Parent Material, Mean Annual Rainfall, and Forest Type as Independent Variables and Dissolved Nutrient Concentrations and Ratios as Dependent Variablesa

Parent Material

F

P

R2

Mean Annual Rainfall

F

P

R2

Forest Type

F

P

DON

6.130

<0.001

0.42

0.087

11.123

<0.001

DIN

2.332

0.02

0.21

4.11

0.004

0.15

0.06

TDN

4.232

<0.001

0.33

0.740

10.555

<0.001

DOP

0.08

0.166

9.763

<0.001

DIP

0.23

12.334

<0.001

0.35

5.156

<0.001

TDP

0.42

12.848

<0.001

0.36

0.748

DOCC

5.969

<0.001

0.41

0.174

3.502

0.01

DOC:TDN

5.485

<0.001

0.39

0.342

0.077

DOC:TDP

5.160

<0.001

0.38

2.863

0.028

0.11

3.018

0.022

TDN:TDP

0.053

9.342

<0.001

0.29

5.302

0.001

DIN:DON

4.041

<0.001

0.32

4.365

0.003

0.16

5.312

0.001

DIP:DOP

0.216

8.343

<0.001

0.27

8.288

<0.001

aAll data were log transformed in order to meet the assumption of normality. Results reported in bold highlight relationships significant at P with a R2 ! 0.3. For parent material, df = 9; for mean annual rainfall, df = 4; and for forest type, df = 4, for all N = 97.

R2 0.33
0.32 0.30 0.18
0.13
0.12 0.19 0.19 0.27 0.001

63), though total N:P was lowest in the cool temperate rain forest sites (median of 23). Warm moist forests had the highest DON:DOP and TDN:TDP ratios (63 and 114, N = 2).
3.3. Isolating State Factor Relationships
[34] To further examine factors that influence nutrient concentrations across the study watersheds we structured our site selection as a matrix of environmental variability encompassing two of Jenny’s state factors: climate and parent material. We examined the effects of these two factors directly using the measured and calculated variables provided by the LENZ national data set [Leathwick et al., 2003]. In addition since all sites had the same potential vegetation we assume that the actual forest type at each site is the result of variation in other state factors [Jenny, 1958, 1980]. We thus used forest type as an independent variable in ANOVA analyses to determine the integrated effects of the independent state factors of interest. 3.3.1. Parent Material
[35] Our ANOVA analyses showed that stream water nutrient concentrations differed significantly as a function of parent materials across NZ watersheds (Table 2 and Figure 3). Considering all watersheds, we found significant differences in DOC and DON across parent material (r2 = 0.41, and 0.42, respectively, p 0.001 for both, Figure 3). Levels of DOC and DON ranged by factors of 12 and 2.5 between different geological substrates. The highest concentrations of both DOC and DON were found in streams draining unconsolidated (alluvial till) terrains and the lowest levels in watersheds underlain by basaltic, andesitic and calcareous parent materials (Figure 3). Levels of NH4+ were low across all substrates, while NO3À was particularly high in basaltic watersheds, where levels of both DON and DOC were depressed (Figure 3). In contrast, parent material was not a statistically significant predictor of P concentrations across watersheds (r2 < 0.2 for both DOP and PO43À). However, as seen for NO3À, watersheds underlain by basalt displayed consistently high PO43À concentrations.

3.3.2. Climate [36] We evaluated the potential influence of climate
variables in three different ways. First, we used ANOVA analyses to evaluate the individual and combined influence of a broad variety of climate factors including MAT, MAR, insolation, PET, GDD, leaching and Holdridge life zones. While most of these variables were measured directly, we calculated values for leaching and Holdridge life zones in order to better capture some of the climatic influences. Leaching was calculated as the difference between MAR and PET at each site giving an estimate of the water pool available to be drained from the soil. Second, we focused our analysis on MAR since it captured much of the variability observed in other climate factors. Third, we examined stream water nutrient concentrations across watersheds on selected parent materials (andesite, granite, greywhacke/shists, and mudstone/sandstone) for which our sampling approach allowed us to maximize the observed range in MAT and MAR.
[37] Neither DOC nor any component of dissolved N (DON, NO3À, or NH4+) concentration was significantly influenced by climate related variables (r2 < 0.15 and p ! 0.01 ANOVA). In contrast, MAR and the variables derived from MAR (leaching and Holdridge life zones) were all significant predictors of TDP (r2 = 0.32, 0.41, and 0.26, respectively, p 0.001 for all, ANOVA, Figure 4), and DIP concentrations (p 0.001 for all), though no significant patterns were found for DOP. The effect of rainfall on TDP could most strongly be seen at sites that receive <2000 mm MAR, where the median concentration (14.9 mg LÀ1) was almost three-fold greater than the concentrations measured at the wetter sites (5.3 mg LÀ1). There was no evidence of any interaction effect between climate and parent material for any of the nutrient concentrations considered (p > 0.05 for all ANOVAs).
[38] Using regression analyses to examine variation in nutrient concentrations associated with climate variables within a single parent material we again found significant relationships only for dissolved P concentrations. The relationships between climate variables and P concentrations were not significant on all parent materials. For

7 of 14

GB1026

MCGRODDY ET AL.: HYDROLOGIC NUTRIENT LOSSES IN NZ FORESTS

GB1026

example, in watersheds on andesitic parent material both MAT and GDD were strong predictors of TDP in stream water (r2 = 0.64 and 0.66, p 0.001 for both,), but these variables were not significantly related to TDP concentrations in streams located on granitic or mudstone/sandstone parent materials. 3.3.3. Forest Composition as an Integrative Factor
[39] Variations in dissolved N, P and C stream concentrations corresponded weakly to forest type (Table 1 and Figure 4). The strongest effects of forest type occurred for DON (r2 = 0.33, p 0.001), with high concentrations occurring in Podocarp-dominated forests. However this effect disappeared entirely when Podocarp forests were removed from this analysis. It is difficult to separate the effects of forest composition and soil drainage since the Podocarp watersheds in this study were located in a narrow region characterized by flat terrain and poorly drained soils. Dissolved inorganic P was noticeably higher in watersheds under mixed forest cover as compared to other forest types though the difference was only marginally significant. Since all of the watersheds on basaltic parent materials were under mixed forest cover again it is difficult to separate the effects of parent material and forest composition in this case.

Figure 3. Median stream water concentrations of (a) carbon, (b) nitrogen, and (c) phosphorus for each parent material group. For the nitrogen (N) graph, solid fill represents organic forms of N, no fill represents nitrate-N, and diagonal lines represent ammonium-N. For the phosphorus (P) graph, solid fill represents organic forms of P, and no fill represents phosphate-P. Total number of streams sampled in each group is noted above the bars in Figure 3a.

3.4. A Matrix Approach to State Factors
[40] To complement the single state factor approach we used our full matrix of external variables to examine state factor control over nutrient losses across large spatial scales. This approach was made possible by quantifying rather than categorizing the variables, on the basis of LENZ ecoregion classification indices [Leathwick et al., 2003].
[41] Climate and geologic (including slope) variables explained up to 58% of variations in DON and approximately half in DOC and DIP across watersheds (Table 3). Examined independently, geologic variables were most effective in explaining variations in DON concentrations (r2 = 0.48, p < 0.001). Similar to the findings from the single factor analyses, regressions using climate variables were most effective predictors of DIP concentrations (r2 = 0.40, p < 0.001). Climate variables were significant, if less strong, predictors in regression analyses of DIN, DON and DOC (r2 = 0.23, 0.33 and 0.37, p < 0.001) despite their lack of significance in ANOVA analyses.
[42] No one variable was consistently significant in all regressions, but some variables were found to be significant repeatedly. Number of growing degree day was significant in 73% of the regressions analyzed using the full variable set and over half the regressions using only climate variables. When limiting the dependent variables in the regression analysis to the geological variable set, watershed slope was the one variable that was most commonly found to be significant (64% of the analyses).
[43] It is interesting that in the case of DIN and DOP we did not find any strong predictors or strong combination of predictors using the multiple regression approach which, in both cases, explained less than a third of the variation measured (Table 3). This might be expected for DIN which varied ten-fold between the low and high quartiles, and is expected to be strongly influenced by plant nutrient demand. Dissolved organic P, in contrast, only varied four-fold

8 of 14

GB1026

MCGRODDY ET AL.: HYDROLOGIC NUTRIENT LOSSES IN NZ FORESTS

GB1026

Figure 4. Median stream water concentrations of carbon, nitrogen, and phosphorus for forest type (a, b, and c, respectively) and rainfall categories (d, e, and f). For the nitrogen (N) graph, black shading represents organic forms of N, no fill represents nitrate-N, and gray shading represents ammonium-N. For the phosphorus (P) graph, solid fill represents organic forms of P, and no fill represents phosphate-P. Total number of streams sampled in each group is noted above the bars in Figures 4a and 4d.

over the same range and is not immediately available for plant uptake suggesting that the similar patterns in these two cases are the result of different regulatory processes.
3.5. Predictive Model of N Losses
[44] For the different forms of N in particular, the results from our multiple regression approach above (section 3.4) offer predictive power not found in the single-category ANOVA approach. In the case of DON inclusion of categorical variables to represent soil drainage and the presence of andesitic or mafic parent materials greatly improved the stability of predictions, which otherwise varied depending on the order in which predictor variables were considered (Table 4). The resulting regression in which DON concentrations decrease with good soil drainage, increasing catchment slope, and andesitic/mafic soil

parent materials explained 42% of the overall variance. Each component variable explained a similar proportion of the total variance. The categorical variables representing drainage and andesitic/mafic parent material describe changes in DON concentrations by factors of approximately 3 and 2, respectively. A change in catchment slope from approximately flat to 30° also represents a change in DON by a factor of 2. The effect of drainage is consistent with the idea that inundation influences DOM export [AitkenheadPeterson et al., 2003] while the parent material effect has also been identified in South American watersheds [Perakis and Hedin, 2007]. While the regression explains approximately 40% of the observed variance in DON concentrations across watersheds, it also implies the existence of a significant unexplained source of variation that was not captured by our geological or climatic variables.

9 of 14

GB1026

MCGRODDY ET AL.: HYDROLOGIC NUTRIENT LOSSES IN NZ FORESTS

GB1026

Table 3. Results From Multiple Regression Analyses With Dissolved Nutrient Concentrations and Ratios in First-Order New Zealand Streams as the Dependent Factora

DON

DOC

DOP

DIN

DIP DOC:TDN DOC:TDP TDN:TDP DOC:DON DOC:DOP DON:DOP

(97)

(96)

(97)

(97)

(97)

(96)

(96)

(97)

(96)

(96)

(97)

R2

0.58

0.49

0.26

0.30

0.54

F 14.403 9.966 6.610 5.041 12.412

P <0.001 <0.001 <0.001 <0.001 <0.001

All Factors

0.46

0.50

9.933

10.576

<0.001

<0.001

0.59 10.719 <0.001

0.46 9.919 <0.001

0.46 10.034 <0.001

0.25 5.504 <0.001

R2

0.48

0.23

0.13

F 15.911 15.311 14.862

P <0.001 <0.001 <0.001

0.06 6.534 0.012

0.14 8.604 <0.001

Geologic Ns

0.13 7.793 0.001

0.23 8.346 <0.001

0.10 11.348 0.001

0.17 7.309 <0.001

0.10 6.459 0.002

R2

0.33

0.37

0.26

0.23

0.40

F 10.431 10.329 5.736 5.116 13.887

P <0.001 <0.001 <0.001 <0.001 <0.001

Climate 0.39 9.688 <0.001

0.22 7.499 <0.001

0.25 5.676 <0.001

0.43 11.064 <0.001

0.37 9.125 <0.001

0.10 4.552 0.004

aAnalyses were done with log transformed data in order to meet the condition of a normal distribution of the population; N for each analysis is indicated in parenthesis in the column heading.

[45] In order to translate our concentration measures into annual watershed export fluxes of DON across New Zealand we multiplied our predicted DON concentrations by a simple water balance model. This model considered monthly precipitation, evapotranspiration and soil water storage for each site. We then included the most readily available variable describing hydrologic flux, MAR, in a regression designed to predict DON export. The resulting regression for DON export (Table 5) includes three predictive variables, catchment slope, presence of andesitic/mafic parent materials, and soil drainage. However, MAR accounted for the majority of explained variance, confirming other studies indicating that hydrologic flux is the primary determinant of DOM export [e.g., Worrall and Burt, 2007; Moore, 1989]. The regression estimate in Table 5 can be extrapolated on the basis of readily available mapping of soil drainage and parent material, and MAR, as well as a digital elevation model sampled for slope at a similar scale to our small catchments (Figures 5 and S1).
[46] The pattern in Figure 5 represents our best quantitative spatial understanding of pristine N losses, given our study design. While the model makes predictions across nearly 2 orders of magnitude, local uncertainty (based on the root mean squared error) is roughly within a factor of 2. Additional uncertainty and/or bias could result from our lack of information of in-stream correlations between concentrations and flow; such information across 97 remote watersheds is well beyond the practical scope of this study.

Table 4. Regression Predicting log10DON (mg N LÀ1)a

Predictor

Estimate Standard Error % Variance Prob > F

Andesitic/maficb

À0.298

Average slope, degrees À0.0100

Well drainedc

À0.54

0.066 0.0026 0.14

9.0

<.0001

9.4

0.0003

12.6

0.0002

aR2 = 0.42 with n = 97 and intercept 2.49 ± 0.13. RMSE = 0.25. bAndesitic/mafic denotes soils forming on andesitic or more mafic (e.g.,
basaltic) parent materials, indicating soils where andic soil properties may
exist. cWell drained = 1 for catchments where most soils are adequately
drained and 0 where significant areas of poorly drained soils exist.

While information on the relationship of DON and flow is rare for unpolluted undisturbed forests, data from Oregon [Vanderbilt et al., 2002] and from our own studies in Chile (L. O. Hedin et al., unpublished data, 1995) show that DON concentrations are poorly related to discharge as evidenced by Pearson correlation r of 0.3 or less. If we conservatively assume that DON increases with discharge by a factor of 2, we conclude that this effect will be minor relative to the overall variations across state factors in our model.
[47] We can also examine this potential effect using absorbance at 340 nm (g340) as a surrogate variable for DON. Absorbance at 340 nm and total organic N (TON) were assessed in the National River Water Quality Network sites as surrogates for national river TN (DON + particulate ON) flux [Scott et al., 2006]. For six monitored catchments featuring primarily native vegetation typical of our sampling sites in the Southern Alps region, average bias associated with g340 and TON was much lower, averaging 14% and 17%, respectively. These calculations suggest that bias associated with correlations between DON concentrations and flow is less than the factor of 2 we report for RMSE in Table 5 and Figure 5. Thus we conclude that our geographically diverse sampling approach largely reflects real and large (order-of-magnitude) differences across watersheds and state factors.
4. Discussion
[48] Nitrogen concentrations in New Zealand streams were strikingly similar to those found in Chile and Argentina [Hedin et al., 1995; Perakis and Hedin, 2002] with

Table 5. Regression Predicting log10{DON Export} (kg N haÀ1 aÀ1), Calculated as DON Â Runoff a

Predictor

Estimate Standard Error % Variance Prob > F

Andesitic/mafic Average slope, degrees Well drained MAR, mm

À0.24 À0.011 À0.63 0.00018

0.067 0.003 0.14 0.00002

5.0

0.0005

6.2

0.0001

7.3

<.0001

29.2

<.0001

aR2 = 0.65 with n = 97 and intercept 0.27 ± 0.15. RMSE = 0.25.

10 of 14
WatershedsLossesConcentrationsVariablesForms