Ecosystems

, Volume 14, Issue 5, pp 848–863 | Cite as

Hot Spots of Inorganic Nitrogen Availability in an Alpine-Subalpine Ecosystem, Colorado Front Range

Article

Abstract

Inorganic nitrogen (N) availability hot spots have been documented in many ecosystems, but major uncertainties remain about their prevalence, timing, and causes. Using a novel mathematical definition of hot spots, spatially explicit measurements of KCl-extractable inorganic N, 2-week soil incubations in the field, ion-exchange resins deployed for 1 year, and a set of associated biotic and abiotic variables, we investigated inorganic N availability hot spots within a 0.89 km2 alpine-subalpine ecosystem in the Colorado Front Range. Measurements of KCl-extractable NH4+ and NO3 taken on multiple dates showed that hot spots of N availability were present in some but not all parts of the study site and that hot spot location varied over the course of the season. Ion-exchange resins showed that over a 1-year period hot spots were important contributors to resin-available N at the landscape level, with 14% of resin locations accounting for 58% of total resin-extractable inorganic N. The KCl-extractable and resin-available inorganic N measurements showed that although spatial variation in the timing of hot spots (that is, hot moments) spreads the influence of short-term hot spots across the landscape to some extent, spatial variation in inorganic N availability is still important when integrated over 1 year. Resin-available N was poorly correlated with the biotic and abiotic variables that we measured, though we did observe that hot spots of resin-available N were twice as common below tree and shrub canopies than in herbaceous areas. Beyond this relationship with canopy structure, neither KCl-extractable nor resin-available inorganic N hot spots were closely related to plant species identity. Instead, the most effective predictor of KCl-extractable NH4+ was the size of the soil organic matter (SOM) N pool, with nearly all hot spots appearing in soils that had greater than 1.4% SOM N.

Key words

nutrient availability Lorenz curve spatially explicit inequality Niwot Ridge LTER random forest model disproportion 

Introduction

Nitrogen (N) availability is a critical biogeochemical variable in soils because of the role of N as a limiting nutrient (Vitousek and Howarth 1991). Understanding N availability in soils is also important due to increasing rates of anthropogenic N deposition that are reducing N limitation in many ecosystems (Fenn and others 2003; Vitousek and others 1997). One of the challenges of understanding N availability in soils is that inorganic N pools and the fluxes that control these pools are often highly variable in space and time. Spatially explicit studies have shown high variation in measurements of KCl-extractable inorganic N (Gallardo and others 2006; Mellert and others 2008; Stenger and others 1998) and in rates of net N mineralization, net nitrification, and denitrification (Fraterrigo and others 2005; Gallardo and Covelo 2005; Robertson and others 1988; Robertson and others 1997). High temporal variation in inorganic N pool sizes is often observed as well (Aber and others 1993; Weintraub and Schimel 2005). These measurements of inorganic N are often lognormally distributed, with many low values accompanied by a few disproportionately high values, also known as biogeochemical hot spots and hot moments (McClain and others 2003). Although often observed, major uncertainties remain about the prevalence, timing, and causes of inorganic N hot spots.

The presence of both high spatial and high temporal variability raises questions about the relationship between spatial and temporal variation in inorganic N availability. In this study, we explore the hypothesis that observations of high spatial variation in short-term inorganic N availability (for example, extractable pools) are due to spatial variation in the timing of hot moments. If this hypothesis is correct, we would expect spatial patterns in N availability to change substantially over time as hot moments occur in different parts of the landscape, potentially spreading their influence by covering many locations over time. Conversely, if this hypothesis is not correct, we might expect more consistent spatial patterns in N availability over time, potentially magnifying the extremity of hot spots in locations where hot moments repeatedly occur. Repeated measurements of landscape variation in inorganic N pool sizes over time can help to test this hypothesis; however, the few such data sets available have not been analyzed with this hypothesis in mind and present conflicting results. For example, in a German winter wheat field, spatial patterns of KCl-extractable inorganic N levels were not consistent over time (Stenger and others 1998) whereas in a fallow field in Belgium and a restored wetland in Georgia, USA, modest correlations with landscape features were found, suggesting more consistency in spatial patterns over time (Ettema and others 1998; Goovaerts and Chiang 1993). Another approach is the use of ion-exchange resins or short-term soil incubations to make measurements of inorganic N availability that are integrated over time (Eno 1960; Sibbeson 1977). Resins and soil incubations are known to produce lognormally distributed data (for example, Johnson and others 2010; Robertson and others 1988), suggesting that spatial variation in the timing of hot moments cannot solely account for high levels of spatial variation.

Many biotic and abiotic factors are potentially important in creating hot spots of soil inorganic N (Townsend and others 2008). The majority of NH4+ in soils is created by enzymatic decomposition of plant and microbially derived N-containing compounds by microbial enzymes (Schimel and Bennett 2004). In some conditions, this NH4+ can then be used as an energy source by nitrifying bacteria, creating NO3. Given the importance of microbially mediated reactions in producing soil inorganic N, abiotic factors that are important to microbes such as temperature, moisture, and pH likely play an important role in hot spot creation. Studies in tropical ecosystems also suggest that topography may be a useful predictor of nutrient availability (John and others 2007; Vitousek and others 2003). However, it is likely that guilds of fast-growing microbes that could create hot spots of inorganic N are dependent on labile carbon from plant rhizodeposition or litter inputs (Moorhead and Sinsabaugh 2006) and some plant species may have mycorrhizal associations that facilitate hot spot creation. Measurements of net N mineralization below particular plant species can show large interspecific variation (Steltzer and Bowman 1998), and variation in litter chemistry arising from the mixture of chemically different plants can affect rates of microbial activity and N cycling (Meier and Bowman 2008). These potential mechanisms and empirical data suggest the hypothesis that, in addition to abiotic conditions, plant species identity may be a useful predictor of soil inorganic N availability and hot spot formation.

One obstacle to investigating biogeochemical hot spots is that we lack a suitable method with which to quantitatively identify hot spots and compare their influence among data sets. There are several mathematical approaches used to identify statistical outliers, including quantile-based approaches (for example, Pennock and others 1992) and statistical distribution-based approaches (for example, Grubbs 1950; van der Loo 2010). However, hot spots are not equivalent to statistical outliers because statistical outliers are large relative to other measured values whereas hot spots must be disproportionately large on an absolute scale. Identifying hot spots is important because they can affect our ability to measure ecosystem processes at larger scales: if hot spots account for a large percentage of total biogeochemical activity, we can focus measurement and modeling efforts primarily on the hot spots; if, on the other hand, few hot spots are found, random sampling across the landscape or through time will effectively capture the variation (Groffman and others 2009; McClain and others 2003; Strayer 2005). Quantifying the magnitude of disproportion for each hot spot is also important because hot spots that are more disproportionate or less disproportionate relative to other values may be caused by different phenomena. In this study, we describe a mathematical technique for identifying hot spots within a set of measured values and quantifying the magnitude of disproportion for each hot spot.

Using this mathematical definition of hot spots in conjunction with other statistical tools, we investigate hot spots of inorganic N availability in a 0.89 km2 site at the alpine-subalpine ecotone in the Front Range of the Rocky Mountains. Nitrogen availability has received considerable attention at this site due to the importance of N limitation to ecosystem processes (Bowman and others 1993) and the effects of increasing anthropogenic N deposition (Burns 2003). The complex terrain also creates high levels of heterogeneity in biotic and abiotic conditions that are ideal for investigating the potential relationships between these conditions and inorganic N availability. To test the hypothesis that spatial variation in the timing of hot moments underlies the high levels of spatial variation in inorganic N availability, we made spatially explicit measurements of the KCl-extractable inorganic N pool across the study site on three dates. We also made measurements of inorganic N availability integrated over the course of 2 weeks using short-term in situ soil incubations and over the course of 1 year using ion-exchange resins. We compare the prevalence and importance of hot spots among these different dates and between KCl-extractable N and resin-available N. To test the hypothesis that plant species identity is a useful predictor of soil inorganic N availability, we compare the effectiveness of plant species identity and composition in predicting inorganic N values with the effectiveness of abiotic and soil chemical variables using random forest models, a multivariate machine learning-based regression technique (Breiman 2001).

Methods

Study Site

The study site is a 0.89 km2 area within the Niwot Ridge Long Term Ecological Research (LTER) site in the Front Range of the Rocky Mountains, Colorado, USA (Figure 1; 40.0490°, −105.5730°). The Niwot Ridge landscape includes both an elevation gradient (3,290–3,510 m a.s.l.) from the subalpine forest to the alpine tundra and a topographic/snow gradient of snow depth and snow cover duration that varies with landscape slope (0°–22°) and aspect in relation to the prevailing westerly winds (Seastedt and others 2004). The study site also contains features such as krummholz trees, sparsely vegetated talus slopes, and small seasonal wetlands. The vegetative canopy structure is 4% willow and other shrubs, 24% tree cover, and 72% open areas dominated by low-growing herbs or rock talus.
Figure 1

Aerial photo of the study site, which is on Niwot Ridge in the Front Range of the Rocky Mountains, Colorado, USA (40.0490°, −105.5730°). The predominant wind direction is shown by the arrow in the upper left. The star indicates the location of Niwot Ridge within Colorado. The contour interval is 10 m.

KCl-Extractable Inorganic N

Measurements of KCl-extractable NH4+ and NO3 were made three times. The first time was the period of June 28 to July 25, 2007 (205 samples). The other two times were single field days: July 9, 2008 (98 samples) and July 30, 2008 (104 samples). Sampled locations correspond with the geographic centroids of composite soil samples collected in a previous study (Darrouzet-Nardi 2010). Those composite samples were haphazardly chosen in such a way as to ensure complete coverage across the site. The samples on July 30, 2008 cover the entire study site whereas the samples on July 9, 2008 exclude the still snow-covered center of the study site. Samples on the two dates in 2008 are a subset of the sample locations from July 2007. Soil core samples (3 cm diameter, 10 cm depth) were either sieved through a 2 mm mesh (July 2007 and July 9, 2008) or homogenized in a bucket (July 30, 2008). Hand mixing was used on July 30, 2008 to facilitate simultaneous field collection of samples, though roots and large rocks were excluded from extracted subsamples as they would be in sieving. For all samples, an approximately 20 g subsample was extracted with 40 ml of 2 M KCl and a second approximately 10 g subsample was weighed and placed into a 100°C drying oven for calculation of gravimetric moisture content. July 2007 samples were transported to the lab on ice before being sieved and extracted the same day and shaken on a rotary shaker for 1 h. For the samples collected on July 9, 2008 and July 30, 2008, extractions were made in the field directly after soil core collection using pre-weighed specimen cups containing KCl, which were then stored at 4°C overnight before filtering. The field extractions in 2008 were used in 2008 to facilitate simultaneous field collections by reducing the time between soil excavation and extraction. All samples were filtered through pre-KCl-wetted filter paper and then frozen. Frozen samples, including blanks to verify that no contamination occurred, were later thawed and analyzed for ammonium and nitrate on an OI Analytical Flow Solution IV with a detection limit of 0.04 μEQ/l. For the samples on July 30, 2009, an additional 10 g subsample was analyzed for total carbon and nitrogen content using a Shimadzu elemental analyzer.

Net N Mineralization and Net Nitrification: 2-Week Incubations

In July 2007, we measured net N mineralization and net nitrification using the buried bag technique (Eno 1960) at the 205 locations described above. At each location, two soil cores (3 cm diameter, 10 cm deep) were collected 1–3 weeks after snowmelt. Both cores were placed in polyethylene bags. One of these bags was kept on ice in a cooler, and then extracted the same day for mineral nitrogen as described above. The second bag was left unsealed and placed in the extraction hole in the field where it was left for 14 days. It was then removed from the field and kept on ice until it was extracted. Mineralization was calculated as initial minus final NH4+ plus NO3. Nitrification was calculated as initial minus final NO3.

Resin-Available Inorganic N

We measured soil inorganic N availability across the study site over the period of 1 year (October 2006 to October 2007) using ion-exchange resin bags (Sibbeson 1977). Resin bags were made from 2.54 cm diameter, 2.54 cm long polyethylene tubes filled with mixed-bed ion-exchange resins (Baker) and wrapped in nylon stockings. The resin bags were buried along a network of transects within the watershed. The transects were placed so as to encompass as much of the heterogeneity within the study site as possible. Resin bags were placed in herbaceous areas, under trees, or under shrubs, with tree and shrub samples placed within several meters of nearby herbaceous samples. At each of 364 locations, two resin bags were buried 3 cm under the surface. A small hole was dug, and the resins were embedded in the sides of the hole to leave intact soil above them. The hole was then refilled. This depth was chosen to maximize the exposure of resins to N cycling in the rooting zone. The resin bags were collected 1 year later, at which point they were extracted and analyzed for NH4+ and NO3 using the procedure described above for soil samples. The two resins from each hole were measured separately and averaged. Analytical blanks without resin were also run through the same extraction and measurement procedure, and the blank values were subtracted from the sample values.

Mathematical Definition of Hot Spots

In this study, we use a mathematical definition of hot spots (Appendix 1 in Supplementary material) to identify hot spots in KCl-extractable and resin-available inorganic N. This mathematical definition identifies values within a statistical population that are not only outliers, but are also disproportionately large. For a set of sample values \( x_{i}^{n} :\,i = 1,\, \ldots ,\,n, \), we can calculate a hot spot cutoff, Ch, above which values can be considered hot spots:
$$ C_{\text{h}} = ({\text{med}}\left( {\frac{{x_{i}^{n} }}{{x_{\text{rrms}} }}} \right) + F^{ - 1} (0.99))x_{\text{rrms}} $$
(1)
where med is the median of \( x_{i}^{n} ,\,x_{\text{rrms}} \) is a robust version of the root mean square used to scale the data, F is a cumulative distribution function for the t-distribution, and 0.99 is a subjective parameter indicating the quantile of F beyond which we wish to identify hot spots. Intuitively, the hot spot cutoff can be thought of as a multiple of a combination of the median and the dispersion (in this case quantified using the median absolute deviation) of the data instead of a multiple of the dispersion alone as would be used in the definition of an outlier. With F−1(0.99), the multiple that is used is based on the 99th percentile of a normal distribution centered at zero, allowing the identification of values that exceed what would be expected at this percentile. To quantify how disproportionately large a particular measured value is—that is, the degree to which it represents an extreme value—we use this equation in which the distance of a value from the median is expressed in relation to the distance between the median and the hot spot cutoff, causing all hot spots to have disproportion greater than 1:
$$ {\text{Disproportion}} = \frac{{x_{i}^{n} - {\text{med}}(x_{i}^{n} )}}{{C_{\text{h}} - {\text{med}}(x_{i}^{n} )}} $$
(2)

Data Analysis

For each of our data sets, we calculated robust summary statistics: median and coefficient of variation. We calculated the coefficient of variation as median/MAD. The MAD (median absolute deviation) is a robust alternative to the standard deviation and is calculated as the median of the absolute value of the differences between each value and the median (Hoaglin and others 1983). Robust statistical techniques are preferable for the study of hot spots because they prevent the hot spots themselves from unduly influencing the parameter estimation for values that are not disproportionately high. We then used equation (1) to calculate hot spot cutoffs (Ch) for each data set. For the measurements of net N mineralization and net nitrification, we also calculated a reverse hot spot cutoff (for disproportionately negative values). Using these hot spot cutoffs, we identified the number of hot spots present in each data set. We quantified the importance of hot spots by determining the percentage of ecosystem-wide pools or fluxes that can be accounted for by hot spots. We also analyzed the levels of inequality using Lorenz curves and their associated summary statistic, the Gini index (Gastwirth 1972). The Lorenz curve plots the cumulative percentage of data points against the cumulative sum of the represented variable, ranked low to high. The Gini index compares the Lorenz curve to a 1:1 line and calculates the percent of the area in the graph between the 1:1 line and the Lorenz curve. The Gini index ranges from 0 to 1, having a value of 0 if all values are identical.

To examine the relationship between soil inorganic N and biotic and abiotic conditions, the soil KCl-extractable inorganic N, net N mineralization, net nitrification, and resin-available inorganic N were modeled using random forest models (Breiman 2001). Random forest models are a nonlinear regression tree-based approach designed to maximize the predictive capability of a set of variables while avoiding overfitting (Hastie and others 2001). To quantify model-explained variance, we examined the correlation coefficients, denoted r2, calculated as 1 − mse/var(y), where mse is the mean squared error of the model-predicted values and var(y) is the variance of the response variable. To evaluate the importance of individual variables to particular models, fitted models were run with each variable randomly permuted and the reductions in fit were evaluated (Breiman 2001). All random forest models were fit using the package randomForest in R 2.6.0 (Liaw and Wiener 2002; R Development Core Team 2007).

Several groups of biotic and abiotic variables, as well as combined models, were tested as predictors of the KCl-extractable and resin-available inorganic N that we measured (Appendix 2 in Supplementary material). For abiotic variables, we used topographic variables only (elevation, slope, and aspect), climatic variables only (soil temperature, soil moisture, and maximum snow depth), and a combination of these groups with pH as the entire abiotic suite. For biotic variables, we examined plant canopy type only (a discrete variable with the categories open, tree, and shrub), plant community (plant canopy type, % unvegetated ground, and two ordinated variables—based either on the Jaccard or Canberra indices), and total plant community (plant canopy type, % unvegetated ground, and the abundances of the 50 most common plant species excluding trees; these plant species account for >90% of the plant cover in our study site). Models of both biotic and abiotic factors together were a combination of the entire abiotic suite and the total plant community variables. Finally, we fit models using soil organic matter (SOM) %N and SOM C:N ratio as predictors both by themselves and as additional predictors in the models containing both biotic and abiotic factors. In cases where the original values of the predictor variables were not co-located with the response variable, interpolated versions of the predictors were used. For the models of resin-available inorganic N, data from Darrouzet-Nardi (2010) were used for SOM %N and SOM C:N ratios. For more details on the specifications of the random forest models, see Appendix 2 in Supplementary material.

Results

KCl-Extractable Inorganic N

Hot spots accounted for a substantial fraction of the total observed NO3 and NH4+ in soil KCl-extractable pools (Table 1; Figure 2). In the data set collected on the morning of July 30, 2008, 19 of 100 soil cores (19%) were identified as inorganic N hot spots and accounted for 64% of the total inorganic N observed (Figure 3). The other two measurement dates had similar contributions to total inorganic N pools by hot spots (Table 1). Of the two single-day measurements of KCl-extractable inorganic N (July 9, 2008 and July 30, 2008), there were higher and more variable levels of inorganic N on July 9 (Table 1). The generally higher values were associated with a higher hot spot cutoff (29 μg N g−1 soil) on July 9 compared to the other two measurement times (19 μg N g−1 soil on July 30, 2008 and 17 μg N g−1 soil in July 2007). Overall, most of the KCl-extractable inorganic N was in the form of NH4+ (for example, 88% on July 9th; Table 1). In contrast to the importance of hot spots for inorganic N pools, no hot spots were identified in SOM N pools on July 30, 2008, and levels of inequality were lower, with a Gini coefficient of 0.25 for SOM N versus greater than 0.5 for resin-available and KCl-extractable inorganic N (Table 1; Figure 3).
Table 1

Summary of KCl-Extractable Inorganic N, Resin-Available Inorganic N, and Soil Organic Matter N Measurements Made Across the Study Site

N cycling hot spots

n

Median

CV (robust)

Hot spot cutoff (Ch)

Number of hot spots

% Hot spots

% Sum

Max hot

Gini

KCl-extractable inorganic N pool

 NH4+ (July 2007)

205

3.2

0.71

12

17

8

72

129

0.77

 NH4+ (July 9, 2008)

94

6.4

1.04

28

10

11

45

4.8

0.61

 NH4+ (July 30, 2008)

100

3.7

1

16

17

17

60

4.6

0.6

 NO3 (July 2007)

205

0.85

0.8

3.4

22

11

44

6.3

0.53

 NO3 (July 9, 2008)

94

0.37

1.08

1.6

10

11

48

3.7

0.6

 NO3 (July 30, 2008)

100

0.39

1.22

1.8

21

21

86

35

0.81

 Total Inorganic N (July 2007)

205

4.4

0.67

17

14

7

65

95

0.73

 Total Inorganic N (July 9, 2008)

94

6.7

1.01

29

10

11

44

4.6

0.59

 Total Inorganic N (July 30, 2008)

100

4.5

0.95

19

19

19

64

6.6

0.61

Net N mineralization and net nitrification (measurement period 2 weeks)

 Net N mineralization

205

0.25

NA

[−0.9, 1.4]

[2, 25]

[1, 12]

NA

73

NA

 Net nitrification

205

0.087

NA

[−0.39, 0.57]

[3, 25]

[2, 12]

NA

7.1

NA

Resin-available inorganic N (measurement period 1 year)

 NH4+

364

77

1

336

42

12

55

21.3

0.65

 NO3

364

130

1.2

600

62

17

69

13.5

0.7

 Total Inorganic N

364

226

1.1

1000

49

14

58

15.3

0.65

Soil organic matter pool

 Total %N

103

0.95

0.48

3.4

0

0

0

0.47

0.25

The robust CV was calculated as median/MAD. Calculations of the hot spot cutoff (Ch from equation (1)) show the cutoff above which values are considered hot spots. The number of hot spots, the percentage of values that were hot spots (% hot spots), and the percentage of the total sum of the data accounted for by hot spots (% sum) are also shown. For net N mineralization and net nitrification, statistics for both reverse hot spots and hot spots (in that order) are shown in the brackets. Disproportion values for individual data points were calculated using equation (2) and the maximum disproportion value within each data set is shown (column “max hot”). The Gini coefficient (Gini) is an estimate of inequality among measured values.

Figure 2

Maps and density plots of: KCl-extractable inorganic N availability in July 2007 and on July 30, 2008 and July 9, 2008; net N mineralization and net nitrification in July 2007; soil organic matter (SOM) N on July 30, 2008; and resin-available inorganic N in open, tree, and shrub canopy locations (separated due to proximity). Plotting symbol shape shows whether the sample was collected in open, tree-covered, or shrub-covered areas (legend in bottom left). Dotted lines on the density plots indicate median values within those distributions. Plotting symbol color in both the density plots and maps corresponds to the amount of N in the sample. Samples determined to be hot spots using the hot spot cutoff (Ch) are shown in shades of red with larger plotting symbols; values below the median are shown in shades of blue; and values between the median and hot spot cutoff are shown in shades of gray. For net N mineralization and net nitrification, shades of blue show reverse hot spots whereas all values between the reverse hot spot cutoff and the hot spot cutoff are shown in shades of gray. Several of the more extreme outliers also have their disproportion values listed above them in the density plot. Both untransformed and log-transformed values are shown for resin-available N (Color figure online).

Figure 3

Lorenz curves of KCl-extractable inorganic N, resin-available inorganic N, and SOM N measured within the study site. Hot spot cutoffs (Ch) are shown as black circles on the Lorenz curves. Three different measurement dates are shown for the KCl-extractable inorganic N. Of the two overlapping black curves for KCl-extractable inorganic N in 2008, the upper curve shows July 9, 2008. The horizontal axis is log-transformed. A dotted 1:1 line indicating perfect equality of samples is shown for reference.

In the July 2007 measurements, several extreme outliers were observed. One soil core contained 1,188 μg N g−1 soil and another contained 253 μg N g−1 soil, with disproportion values of 128 and 20 (that is, 128 times larger than the difference between the sample median and the hot spot cutoff, Ch). These two points alone accounted for 54% of the inorganic N measured during that period. For both of these extreme values, high NH4+ levels were responsible for the high values, though in the case of the most extreme hot spot, NO3 levels were high as well (15 μg N g−1 soil).

The three KCl-extractable inorganic N data sets showed some consistent hot spot locations, though more so for the two data sets collected in the same season (July 9, 2008 and July 30, 2008). Of 100 measurements taken in both July 2007 and July 30, 2008, 13 NO3 hot spots were identified in 2007 and 21 in 2008. Of these, six locations were hot spots on both dates. Of the same 100 points, six NH4+ hot spots were identified in 2007 and 17 were identified in 2008. Two locations were hot spots on both dates. There was little correlation between points taken at the same locations on those two dates (r2 values for log-transformed NO3 values (0.12) and NH4+ values (0.03)). However, a much stronger correlation was seen between values measured on July 9, 2008 and July 30, 2008 at 54 shared locations (r2 = 0.44 for log-transformed values; Figure 4).
Figure 4

A Soil organic matter (SOM) %N versus log-transformed (ln(x + 1)) KCl-extractable NH4+, B SOM C:N ratio versus log-transformed KCl-extractable NH4+, and C SOM %N versus log-transformed (ln(x + 0.1)) KCl-extractable NO3. All 104 soil core samples were collected on the morning of July 30, 2008. Canopy structure is shown by plotting symbol. Black lines are loess curves.

Resin-Available Inorganic N, October 2006 to October 2007

Values of resin-available inorganic N collected over a 1-year period were well approximated by a lognormal distribution (for log-transformed values, skew = −0.26, kurtosis = 0.17), and hot spots accounted for a substantial fraction of the total observed resin-extractable NO3 and NH4+. During October 2006 to October 2007, 49 of 364 resin bag locations (14%) were identified as hot spots and accounted for 58% of the total resin-available inorganic N during the season (Figure 3). The most extreme outlier observed was under a willow in a krummholz area in the western corner of the study site and had both high NO3 (6490 μg N-NH4+) and high NH4+ (5594 μg N-NH4+). Across the study site, 40 of 49 hot spots were located in tree or shrub-covered areas. In contrast to the KCl-extractable inorganic N, NO3 accounted for a greater amount (69%) of the inorganic N measured in the resins. NO3 values were also more variable across the study area (MAD of 154 vs. 80). However, at smaller scales, correlations between dual resins buried in the same hole were better for NO3 (r2 = 0.54) than NH4+ (r2 = 0.31). Levels of inequality, as measured by Gini coefficients, were similar for resin-available inorganic N and KCl-extractable inorganic N (0.59, 0.61, and 0.73 for inorganic N in the three KCl-extractable datasets, and 0.65 for resin-available inorganic N).

Net Mineralization and Net Nitrification

Within the buried bag incubations, 13 and 14% of sample points were identified as hot spots (or reverse hot spots) for net N mineralization and net N nitrification, respectively (Figure 2; Table 1). Many more hot spots were observed than reverse hot spots. The only observed reverse hot spots for net N mineralization were relatively extreme outliers (disproportion = 73, 6.4). These two reverse hot spots were driven by the extreme inorganic N hot spots that are described above (1,188 μg N g−1 soil and 253 μg N g−1 soil) and thus occurred at the same locations (Figure 2). There were also three net nitrification reverse hot spots, but these were not as extreme. Positive hot spots for both net N mineralization and net nitrification were more common. Net N mineralization data from the higher elevation dry meadows showed low to intermediate values except for two cores that were beneath patches of the N-fixing Trifolium dasyphyllum (Figure 2). The hot spots in the net nitrification data set occurred widely across the study site.

Relationship Between Inorganic N Availability and Biotic and Abiotic Conditions

In the 104 soil cores collected across the study site on July 30, 2008, the strongest predictor of log-transformed soil KCl-extractable NH4+ was the percentage of SOM N in the extracted soil core (r2 = 44% with SOM %N as the sole predictor in a random forest model; Figure 4A). Of the values above the hot spot cutoff (16.2 g NH4+ g−1 soil), which accounted for 60% of total NH4+ measured on the sampling date, 14 of 15 hot spots occurred in soil cores with greater than 1.2% N. Several other measured variables were able to provide further small improvements to the modeled fit of NH4+ values for a combined best fit r2 value of 55% (Table 2). The most effective additional predictor variables (determined by permutations of predictor variables) were slope, soil C:N ratio, and the percentage of unvegetated ground in the 1 m2 quadrats. A positive relationship was also observed between soil C:N ratio and NH4+ (Figure 4B). SOM %N was not measured on July 9, 2008, but presumably a similar relationship existed given the strong correlation between measurements on the two dates (Figure 5). A positive but weaker and more heteroscedastic relationship was observed between SOM %N and log-transformed NO3 values (Figure 4C).
Table 2

Random Forest Model r2 Values for Models of Soil Inorganic N from KCl Extractions and Resins

 

Topographic

Climatic

Canopy

Jaccard

Canberra

Abiotic

Plants

Abiotic and plants

SOM

Combined with SOM

KCl extractions

 log(NH4+) July 2007

0

0

0

0

0

0

0

1

 log(NO3) July 2007

0

0

0

1

8

0

8

0

 log(inorganic N) July 2007

0

0

0

0

0

0

1

0

 log(NH4+) July 9

0

0

1

18

24

9

29

24

 log(NO3) July 9

0

0

0

0

6

0

8

14

 log(inorganic N) July 9

0

0

1

15

23

6

31

29

 log(NH4+) July 30

0

8

4

23

26

30

24

33

47

55

 log(NO3) July 30

0

0

0

0

0

5

13

14

13

33

 log(inorganic N) July 30

0

6

4

22

26

22

26

32

50

57

Ion-exchange resins

 log(NH4+)

0

2

5

5

8

4

7

13

0

17

 log(NO3)

0

13

8

12

11

21

23

25

8.5

26

 log(inorganic N)

0

5

8

9

10

13

17

20

1

23

Net mineralization and net nitrification

 log(net mineralization)

0

0

0

1

2

0

15

16

14

 log(net nitrification)

0

0

0

9

2

0

11

11

5

Each row shows one response (modeled) variable and each column shows a set of predictor variables that were used. Predictor variables used are as follows: topographic—slope, aspect, elevation; climatic—soil temperature, soil moisture, snow depth; canopy—plant canopy structure categories (open, tree, or shrub); Jaccard—canopy structure, % unvegetated, and two ordinated axes based on the Jaccard difference matrix; Canberra—same as Jaccard but with Canberra difference matrix; abiotic—topographic variables, climatic variables, and pH; plants—canopy structure, % unvegetated, and the abundances of the 50 most common plant species; combined—combination of “plants” and “abiotic” variables; SOM—SOM %N and SOM C:N ratio; combined with SOM—combination of “plants,” “abiotic,” and “SOM”. Inorganic N = (NO3 + NH4+). See Appendix 2 in Supplementary material for more details on model specifications.

Figure 5

KCl-extractable NH4+ on July 30, 2008 versus July 9, 2008. Axis scales are logarithmic. The two measurement dates had 54 sampling locations in common; those shared locations are shown here. Canopy structure is shown by plotting symbol and the black line is a least squares linear regression for which the correlation coefficient r2 = 0.44.

In models of log-transformed NH4+ on July 30, 2008 that did not include the SOM %N or SOM C:N ratio, performance of models based on biotic factors, abiotic factors, or both together performed similarly, with r2 approximately equal to 30% (Table 2). Models based solely on abiotic factors (topographic or microclimatic) and models based solely on plant canopy structure performed poorly, but the interactions of these variables (“abiotic” models) produced better fits. Models based on two ordinated plant variables performed almost as well as models that included detailed plant species information (r2 = 22% and 26% for Jaccard-based and Canberra-based ordinations, respectively). In models of KCl-extractable NO3 values, there was more unexplained variation overall than for NH4+, with the best model having r2 = 33% (Table 2). None of the models of the July 2007 inorganic N values were effective in predicting variation in those values.

Models of net N mineralization and net nitrification made using the buried bag technique were also difficult to correlate with the biotic and abiotic factors that were measured, with most variation remaining unpredictable (Table 2). However, of the small amount of variation that was predictable, both of these processes were better correlated with biotic than abiotic factors, with the models based on individual plant species having r2 = 16 and 11%, respectively (Table 2). For net nitrification (but not for net N mineralization), a similar fit was achieved with the two ordinated variables based on the Jaccard index. For net N mineralization, the most useful predictive variables were the percent cover of Juncus drummondii, Caltha leptosepala, Trollius laxus, Ribes montigenum, and unvegetated ground. Models that combined biotic and abiotic factors did not perform substantially better than the models with biotic factors alone.

Like the buried bag measurements, the large majority of variation in resin-available NH4+ was uncorrelated with the variables measured in this study. The combined models that included SOM %N and SOM C:N provided the best fits, with r2 = 26% for NO3 and 17% for NH4+ (Table 2). Of the small amount of variation that was explained, plant canopy structure was an important predictor. Hot spots were more common under tree and shrub canopies, with 19% of measured values being hot spots of inorganic N below tree and shrub canopies and 8% of herbaceous areas being hot spots. Microclimatic variables were more effective than topographic variables in predicting resin-available NO3 (Table 1), but not as effective as the combination of all abiotic factors. Relatively low fits were provided by plant canopy structure alone (r2 = 8%) and negligible improvements were seen in models that included biotic variables in addition to abiotic variables. Although measurements of SOM %N were not made in the soil directly above the resins, interpolated measurements of SOM %N concentrations from the soil core composites were the most useful predictor of resin-available NH4+ (determined by permutations of predictor variables), and this relationship between SOM %N and resin-available NH4+ was most pronounced in herbaceous areas.

Neither the KCl-extractable inorganic N values nor the resin-available inorganic N values were well constrained by the identity of the plant canopy species above the sample (Figure 6). A range of possible values was observed beneath the canopies of particular plants. For example, the widespread Vaccinium sp. showed both some of the highest and lowest values for both KCl-extractable and resin-available inorganic N. The N-fixing Trifolium spp. showed a range of values. The highest KCl-extractable inorganic N value was below a Trifolium canopy, but the other three samples below Trifolium were not similarly high. Resin-available N below Trifolium canopies contained neither unusually high nor unusually low values. Hot spots occurred under a variety of plant species and under bare ground (Figure 6). Although plant species identity was not recorded for the samples of net N mineralization and nitrification, the two hot spots observed in the higher elevation dry meadows were both under Trifolium canopies.
Figure 6

Dot plots by plant species of log-transformed (ln(x + 1)) KCl-extractable inorganic N (NH4+ + NO3) (left panel) and resin-available inorganic N (right panel). Species are ordered from top to bottom by median. Soil cores with multiple species growing on them are shown multiple times in this plot for up to three species. Canopy structure is shown by plotting symbol. KCl-extractable inorganic N samples were collected on the morning of July 30, 2008. Resins were deployed from October 2006 to October 2007. Only species for which more than four samples were available (KCl extractions) or six samples (resins) are shown. Symbols representing shrub and tree locations for the resin samples have been positioned slightly above or below the open location symbols for visibility. The hot spot cutoff, Ch, is shown as a vertical line.

Discussion

Importance of Inorganic N Hot Spots

The results of this study show that pools and fluxes of inorganic N are highly variable across the alpine-subalpine ecotone and that landscape totals are strongly influenced by hot spots of activity. In the three KCl-extractable inorganic N data sets, as well as in the resin-extractable inorganic N data, 7–19% of values were identified as hot spots, and these hot spots accounted for 44–65% of the total observed inorganic N. In measurements of net N mineralization and net nitrification, 13 and 14% of the values were identified as hot spots. These findings match the results of other studies that have documented unequal distributions and high levels of spatial variation in soil inorganic N levels (Ghidey and Alberts 1999; Gonzalez and Zak 1994; Liptzin and Seastedt 2009; Mellert and others 2008; Robertson and others 1988, 1997; Stenger and others 1998; Yanai and others 2000). Although hot spots accounted for a large amount of inorganic N at our study site, the hot spots were not so dominant that they alone determined the size of landscape-level inorganic N pools. Thus, although inorganic N hot spots will complicate attempts to extrapolate from plot-sized samples or to use interpolation procedures such as kriging, measurement and modeling efforts cannot focus solely on hot spots at the expense of non-hot spot areas.

The presence of hot spots in the short-term incubation fluxes (net N mineralization and net nitrification) and the resin-available inorganic N values show that substantial spatial variation exists in inorganic N availability even when integrated over time, suggesting that hot moments do not occur in all landscape locations. This finding is consistent with other studies showing high levels of variation in net N cycling processes and in resin-available inorganic N (Fraterrigo and others 2005; Johnson and others 2010; Lajtha 1988; Yavitt and Wright 1996). However, in locations where hot spots do occur, the repeated measurements of KCl-extractable inorganic N suggest that there is spatial variability in their timing, with hot spots appearing in different locations on the three measurement dates. This type of temporal variation is not surprising given the variable timing of snowmelt and plant phenology within our study site. Taken together, the integrated measurements and the repeated short-term measurements suggest that both the characteristics of locations themselves and the time at which those locations are examined can play a substantial role in contributing to the observed variation in short-term inorganic N availability.

The moderate correlation between inorganic N availability on July 9, 2008 and July 30, 2008 and the moderate correlations with biotic and abiotic variables on those dates were consistent with other studies showing similar correlations with landscape features (Ettema and others 1998; Goovaerts and Chiang 1993). In contrast, the July 2007 measurements were not correlated with the July 9, 2008 and July 30, 2008 measurements and were not correlated with the biotic and abiotic variables in the random forest models. Although these poor relationships could be due to the slight differences in extraction methodology between the years, the more likely explanation is that these samples were collected over a 1-month period instead of during a single day, confounding spatial and temporal variation and obscuring the relationships between inorganic N and the biotic and abiotic variables that we measured. Interannual variation may also have played a role, as such variation has been observed in other ecosystems (Magill and others 2004; Yavitt and Wright 1996). A comparison of inequality levels in the resin-available and KCl-extractable inorganic N suggests that spatial variation is not magnified over time due to hot moments repeatedly occurring in the same locations. Such a magnification would result in higher levels of inequality as measured by the Gini index in resin-available inorganic N, but we instead observed similar levels of inequality in resin measurements and KCl-extractable inorganic N measurements. At time scales longer than 1 year, it is possible that differences among locations may be further reduced.

As shown by the Lorenz curves and Gini index values, SOM pools in the top 10 cm of mineral soil showed higher equality among values across the landscape than inorganic N pools. Furthermore, the hot spot cutoff that was calculated indicated that no hot spots were present for the SOM pools. This phenomenon has also been reported in many other studies of landscape heterogeneity in soil N cycling (Dick and Gilliam 2007; Rivero and others 2007; Stenger and others 2002; Wang and others 2007). Although hot spots of SOM N storage might be observed when soil depth, rock content, and bulk density are taken into account (Rodionov and others 2007), the results of this study provide a good example of the large difference in levels of inequality between SOM N and inorganic N pools in the rooting zone where N cycling is most active.

Several hot spots with very high magnitudes of disproportion were observed in KCl-extractable inorganic N pools, measurements of net N mineralization, and in resin-available inorganic N. In the KCl-extractable inorganic N measurements taken in July 2007, the inorganic N in just two soil cores accounted for 54% of the inorganic N measured in that data set. Reductions in these pools over the following 2-week period led to the observation of large reverse hot spots of net N mineralization as measured using the short-term incubations (buried bags). Hot spots with disproportion values above 10 such as these may be created by different processes than the more numerous hot spots that have disproportion values between 1 and 10. Although such high values are sometimes attributed to analytical error, we do not believe that is the explanation in this case. First, our use of analytical blanks has never indicated that such high error is possible. Second, the values were not isolated occurrences: multiple high resin values filled out the positive tail of a lognormal distribution, and the high values in the buried bag measurements were observed in both the initial core and the incubated core. Finally, there are several plausible explanations for the existence of these extreme hot spots. Animal excreta are a potential cause of hot spots (Bogaert and others 2000), and vertebrates are common in our study site (Armstrong and others 2001). Other disturbance events such as freeze–thaw cycles could also play a role in creating such extreme hot spots (Callesen and others 2007). Even without exogenous inputs or disturbance, it may be possible for a favorable convergence of environmental conditions and reactants (that is, the same processes likely controlling variation at lower levels) to create extreme hot spots of inorganic N through rapid mineralization (Aber and others 1993; Weintraub and Schimel 2005). These relatively extreme hot spots could play a substantial role in ecosystem function: periodic episodes of high N availability could affect competitive interactions, establishment opportunities, or at these high levels, even cause localized negative effects on plants (Britto and Kronzucker 2002).

Biotic and Abiotic Predictors of Inorganic N Hot Spots

Although we did observe some evidence for the hypothesis that plant species identities are useful predictors of variation in inorganic N availability, taken together our results do not provide strong support for this hypothesis. Random forest models of net N mineralization and nitrification as well July 9 NH4+ and July 30 NO3 measurements showed some improvement with the addition of data on plant species identities in the surrounding 1 m2 community. Furthermore, several hot spots observed in the KCl-extractable inorganic N measurements and the net N mineralization measurements were associated with N-fixing plants. However, random forest models of the other KCl-extractable pools and resin-available inorganic N data over the course of 1 year showed little improvement in fit with the addition of data on plant species identities in the surrounding 1 m2 community. Furthermore, the models based on ordinated variables (Jaccard and Canberra) suggest that the plant species-based models primarily made use of information about plant community instead of species identity. Finally, although a few species growing directly above the soil samples and the buried ion-exchange resins had some predictive value, we generally observed large variations in N pools and fluxes among cores or resins with the same canopy species.

The N-fixing Trifolium species (T. dasyphyllum and T. parryi) in our study site illustrates the extant but surprisingly weak influence of plant species identity on soil inorganic N availability that we observed. On July 30, 2008, the highest KCl-extractable NH4+ value observed within the study site was below a Trifolium canopy, which is probably not due to chance. However, four other Trifolium plants that were sampled that day had unremarkable values. The two net N mineralization hot spots present in the dry meadow community at the upper elevations were also associated with Trifolium. These data suggest that Trifolium, as we would expect, has the potential to create large local pools of bioavailable N, which in turn can create hot spots of KCl-extractable NH4+. However, Trifolium presence is by no means a consistent predictor of such hot spots, and hot spots can occur under other species as well.

Using the same short-term incubation technique to measure net N mineralization and net nitrification, Steltzer and Bowman (1998) observed large differences between Deschampsia caespitosa and Geum rossii. Although plant species identities did improve our random forest models of these fluxes, our measurements were not as strongly linked to plant species identity as were those in Steltzer and Bowman (1998). We did not detect strong effects of those particular species (D. caespitosa and G. rossii) and the overall poor fits of the random forest models show that large species differences were not sufficient to predict variation in inorganic N pools or fluxes at the landscape scale. This difference between the two studies may be due in part to different sampling techniques. Steltzer and Bowman (1998) measured rates in dense, monospecific stands of the dominant species, whereas samples in this study were haphazardly chosen and frequently contained multiple plant species even within the space of one 3-cm diameter soil core.

Although we saw little evidence of a relationship between plant species identity or topography and inorganic N availability, we did observe that KCl-extractable NH4+ hot spots at our study site have a substantial exponential relationship with SOM N levels. This finding was unexpected because in many other studies this relationship has been less pronounced. For example, a study in sagebrush steppe found a weaker positive correlation (Spearman’s rank correlation coefficient of 0.13; Jackson and Caldwell 1993). A geostatistical study across a range of ecosystems inferred a connection between these variables based on ranges of spatial autocorrelation but did not report a direct correlation (Gallardo and others 2005). Several more studies have reported substantial correlations between soil inorganic N with similar variables such as canopy N, presence of tree canopy, or aboveground biomass, but did not report a direct exponential relationship as we observed here (Bengtson and others 2007; Cheng and others 2007; Gallardo 2003). The differences in the strength of this relationship between this study and others may lie in both inherent differences among ecosystems and methodological differences like timing of core collections and KCl extractions. For example, large numbers of soil cores are not often collected over a period of several hours and extracted immediately in the field.

The strong correlation between SOM %N and log-transformed NH4+ values in the same soil cores could be due to a number of reasons. A logical basis for an explanation is that as soil SOM levels increase, substrate limitation for microbes may be reduced, allowing mineralization levels to increase. However, a linear increase in substrate does not explain why NH4+ increases exponentially. The exponential increase means that in cores with higher quantities of SOM, not only was more NH4+ present, but a larger percentage of total soil N was in the inorganic N pool. If increased SOM N was associated with lower C:N ratios, this pattern might be explained as the efflux of unneeded N by microbes, but no such correlation was observed and NH4+ levels were instead positively correlated with soil C:N ratios. An alternative explanation is that the amount of SOM N in a soil core is correlated with the amount of labile material in the core. Cores that have larger amounts of labile material might stimulate exponential growth or activity in microbes, which could help explain this relationship.

Though variation in KCl-extractable NO3 was less predictable than NH4+ by the random forest models, the best predictor for NO3 was again SOM %N. SOM %N may have been the best predictor because it is correlated with NH4+, and NH4+ is the N source for the nitrification reaction that produces NO3. The additional unexplained variation in NO3 when compared with NH4+ suggests that the availability of NH4+ is not the only important factor controlling NO3 concentrations. In a similar effort to model extractable NO3 based on a time series in a conifer forest, throughfall N and either foliage N or organic layer C:N were found to be the most effective predictors (Kristensen and others 2004). Though we did not measure throughfall N, the identification of small-scale variation in organic N pools as the best predictors of soil NO3 is a finding similar to ours. The difficulty in predicting NO3 levels in soils has been suggested by other studies that have found very little spatial autocorrelation in NO3 values (Okin and others 2008). The high mobility of NO3 in soils may also contribute to its lack of spatial pattern across landscapes.

Like the measurements of net N mineralization and net nitrification, most of the observed variation in resin-available inorganic N was unpredictable despite the large number of potential predictors measured. Unlike a study in a tropical forest (Vitousek and others 2003), topography was not an effective predictor of resin-available inorganic N availability at our study site. Resin-available NO3 values were higher than NH4+ values and better fit by the random forest models, possibly due to the greater mobility of NO3 in the soil. For resin-available NO3, climatic factors (soil temperature and moisture) accounted for 13% of the variation, consistent with the importance of hydrologic transport in controlling resin-available NO3. The importance of SOM %N as a predictor in the combined random forest model of resin-available inorganic N (as determined by predictor variable permutations) provided some support for the positive correlation between SOM %N and NH4+ in the KCl-extractable pool. One recent study from Sierra Nevada forest soils also documented a correlation between resin-available inorganic N and SOM levels (Johnson and others 2010). Such a correlation at our site may have been clearer if SOM levels were measured at the same location and scale as the resin data instead of interpolated from nearby soil cores.

Conclusion

The hypothesis that spatial variation in the timing of hot moments underlies observed spatial variation in inorganic N availability was partially supported. The ion-exchange resin data and the multiple measurements of KCl-extractable inorganic N showed that a substantial amount of spatial variation existed in the timing of hot moments. However, this variation did not result in hot moments occurring in all locations across the study site, at least over the course of 1 year. The hypothesis that spatial variation in inorganic N availability would be predictable based on plant species identity is not strongly supported by the data. Neither data from the 1 m2 quadrats nor the identities of plant species atop soil cores or resins were good predictors of either KCl-extractable inorganic N on three dates or resin-available N over the course of 1 year. Although a large amount of unexplained variation remained, the best observed predictor of variation in measurements of inorganic N was instead the size of the soil SOM N pool, suggesting that the quantity of organic matter may be an important determinant of inorganic N hot spot formation in alpine-subalpine ecosystems.

Notes

Acknowledgments

For funding, we thank NSF DGE 0202758, NSF DEB 0423662, NSF DEB 0808275, the John W. Marr Ecology Fund, and the Department of Ecology and Evolutionary Biology. For helpful suggestions on this manuscript, we thank Alan Townsend, Carol Wessman, Tim Seastedt, and Mark Williams, Michael Weintraub, and Zachary Rinkes. For their help in the field, lab, and/or planning stages of this project, we thank Courtney Meier, John Murgel, Carly Baroch, Brendan Whyte, Anna Lieb, Jaclyn Darrouzet-Nardi, Jeanette Darrouzet-Nardi, Chris Darrouzet-Nardi, Kathy Buehmann, Lisa Gerstenberger, David Knochel, Russ Monson, Stuart Grandy, Courtney Meier, Andy Thomspon, Todd Ackerman, and the numerous volunteers on the July 9th and 30th 2008 field days as well as during the snow sampling effort. We thank Chris Seibold and the Kiowa Lab assistants for help with nutrient analyses. Logistical support was provided the University of Colorado’s Mountain Research Station. Finally, we thank the reviewers, whose careful consideration and suggestions greatly improved this manuscript.

Supplementary material

10021_2011_9450_MOESM1_ESM.pdf (2.4 mb)
Supplementary material 4 (PDF 2,408 kb)

References

  1. Aber JD, Magill A, Boone R, Melillo JM, Steudler P, Bowden R. 1993. Plant and soil responses to chronic nitrogen additions at the Harvard Forest, Massachusetts. Ecol Appl 3:156–66.CrossRefGoogle Scholar
  2. Armstrong DM, Halfpenny JC, Southwick CH. 2001. Vertebrates. In: Bowman WD, Seastedt TR, Eds. Structure and function of an alpine ecosystem: Niwot Ridge, Colorado. Oxford (NY): Oxford University Press. p 128–56.Google Scholar
  3. Bengtson P, Basiliko N, Prescott CE, Grayston SJ. 2007. Spatial dependency of soil nutrient availability and microbial properties in a mixed forest of Tsuga heterophylla and Pseudotsuga menziesii, in coastal British Columbia, Canada. Soil Biol Biochem 39:2429–35.CrossRefGoogle Scholar
  4. Bogaert N, Salomez J, Vermoesen A, Hofman G, Van Cleemput O, Van Meirvenne M. 2000. Within-field variability of mineral nitrogen in grassland. Biol Fertil Soils 32:186–93.CrossRefGoogle Scholar
  5. Bowman WD, Theodose TA, Schardt JC, Conant RT. 1993. Constraints of nutrient availability on primary production in 2 alpine tundra communities. Ecology 74:2085–97.CrossRefGoogle Scholar
  6. Breiman L. 2001. Random forests. Mach Learn 45:5–32.CrossRefGoogle Scholar
  7. Britto DT, Kronzucker HJ. 2002. NH4+ toxicity in higher plants: a critical review. J Plant Physiol 159:567–84.CrossRefGoogle Scholar
  8. Burns DA. 2003. Atmospheric nitrogen deposition in the Rocky Mountains of Colorado and southern Wyoming—a review and new analysis of past study results. Atmos Environ 37:921–32.CrossRefGoogle Scholar
  9. Callesen I, Borken W, Kalbitz K, Matzner E. 2007. Long-term development of nitrogen fluxes in a coniferous ecosystem: does soil freezing trigger nitrate leaching? J Plant Nutr Soil Sci 170:189–96.CrossRefGoogle Scholar
  10. Cheng X, An S, Chen J, Li B, Liu Y, Liu S. 2007. Spatial relationships among species, above-ground biomass, N, and P in degraded grasslands in Ordos Plateau, northwestern China. J Arid Environ 68:652–67.CrossRefGoogle Scholar
  11. Darrouzet-Nardi A. 2010. Landscape heterogeneity of differently aged soil organic matter constituents at the forest-alpine tundra ecotone, Niwot Ridge, Colorado, U.S.A. Arct Antarct Alp Res 42:179–87.CrossRefGoogle Scholar
  12. Dick DA, Gilliam FS. 2007. Spatial heterogeneity and dependence of soils and herbaceous plant communities in adjacent seasonal wetland and pasture sites. Wetlands 27:951–63.CrossRefGoogle Scholar
  13. Eno CF. 1960. Nitrate production on the field by incubating the soil in polyethylene bags. Soil Sci Soc Proc 24:277–89.CrossRefGoogle Scholar
  14. Ettema CH, Coleman DC, Vellidis G, Lowrance R, Rathbun SL. 1998. Spatiotemporal distributions of bacterivorous nematodes and soil resources in a restored riparian wetland. Ecology 79:2721–34.CrossRefGoogle Scholar
  15. Fenn ME, Baron JS, Allen EB, Rueth HM, Nydick KR, Geiser L, Bowman WD, Sickman JO, Meixner T, Johnson DW, Neitlich P. 2003. Ecological effects of nitrogen deposition in the western United States. Bioscience 53:404–20.CrossRefGoogle Scholar
  16. Fraterrigo JM, Turner MG, Pearson SM, Dixon P. 2005. Effects of past land use on spatial heterogeneity of soil nutrients in southern Appalachian forests. Ecol Monogr 75:215–30.CrossRefGoogle Scholar
  17. Gallardo A. 2003. Effect of tree canopy on the spatial distribution of soil nutrients in a Mediterranean Dehesa. Pedobiologia 47:117–25.CrossRefGoogle Scholar
  18. Gallardo A, Covelo F. 2005. Spatial pattern and scale of leaf N and P concentration in a Quercus robur population. Plant Soil 273:269–77.CrossRefGoogle Scholar
  19. Gallardo A, Parama R, Covelo F. 2005. Soil ammonium vs. nitrate spatial pattern in six plant communities: simulated effect on plant populations. Plant Soil 277:207–19.CrossRefGoogle Scholar
  20. Gallardo A, Parama R, Covelo F. 2006. Differences between soil ammonium and nitrate spatial pattern in six plant communities. Simulated effect on plant populations. Plant Soil 279:333–46.CrossRefGoogle Scholar
  21. Gastwirth JL. 1972. The estimation of the Lorenz curve and Gini index. Rev Econ Stat 54:306–16.CrossRefGoogle Scholar
  22. Ghidey F, Alberts EE. 1999. Temporal and spatial patterns of nitrate in a claypan soil. J Environ Qual 28:584–94.CrossRefGoogle Scholar
  23. Gonzalez OJ, Zak DR. 1994. Geostatistical analysis of soil properties in a secondary tropical dry forest, St-Lucia, West-Indies. Plant Soil 163:45–54.Google Scholar
  24. Goovaerts P, Chiang CN. 1993. Temporal persistence of spatial patterns for mineralizable nitrogen and selected soil properties. Soil Sci Soc Am J 57:372–81.CrossRefGoogle Scholar
  25. Groffman PM, Butterbach-Bahl K, Fulweiler RW, Gold AJ, Morse JL, Stander EK, Tague C, Tonitto C, Vidon P. 2009. Challenges to incorporating spatially and temporally explicit phenomena (hotspots and hot moments) in denitrification models. Biogeochemistry 93:49–77.CrossRefGoogle Scholar
  26. Grubbs FE. 1950. Sample criteria for testing outlying observations. Ann Math Stat 21:27–58.CrossRefGoogle Scholar
  27. Hastie T, Tibshirani R, Friedman JH. 2001. The elements of statistical learning: data mining, inference, and prediction: with 200 full-color illustrations. New York: Springer.Google Scholar
  28. Hoaglin DC, Mosteller F, Tukey JW. 1983. Understanding robust and exploratory data analysis. New York: Wiley.Google Scholar
  29. Jackson RB, Caldwell MM. 1993. Geostatistical patterns of soil heterogeneity around individual perennial plants. J Ecol 81:683–92.CrossRefGoogle Scholar
  30. John R, Dalling JW, Harms KE, Yavitt JB, Stallard RF, Mirabello M, Hubbell SP, Valencia R, Navarrete H, Vallejo M, Foster RB. 2007. Soil nutrients influence spatial distributions of tropical tree species. Proc Natl Acad Sci USA 104:864–9.PubMedCrossRefGoogle Scholar
  31. Johnson DW, Glass DW, Murphy JD, Stein CM, Miller WW. 2010. Nutrient hot spots in some Sierra Nevada forest soils. Biogeochemistry 101:93–103.CrossRefGoogle Scholar
  32. Kristensen HL, Gundersen P, Callesen I, Reinds GJ. 2004. Throughfall nitrogen deposition has different impacts on soil solution nitrate concentration in European coniferous and deciduous forests. Ecosystems 7:180–92.CrossRefGoogle Scholar
  33. Lajtha K. 1988. The use of ion-exchange resin bags for measuring nutrient availability in an arid ecosystem. Plant Soil 105:105–11.CrossRefGoogle Scholar
  34. Liaw A, Wiener M. 2002. Classification and regression by randomForest. R News 2:18–22.Google Scholar
  35. Liptzin D, Seastedt TR. 2009. Patterns of snow, deposition, and soil nutrients at multiple spatial scales at a Rocky Mountain tree line ecotone. J Geophys Res Biogeo 114. doi:10.1029/2009JG000941.
  36. Magill AH, Aber JD, Currie WS, Nadelhoffer KJ, Martin ME, McDowell WH, Melillo JM, Steudler P. 2004. Ecosystem response to 15 years of chronic nitrogen additions at the Harvard Forest LTER, Massachusetts, USA. For Ecol Manag 196:7–28.CrossRefGoogle Scholar
  37. McClain ME, Boyer EW, Dent CL, Gergel SE, Grimm NB, Groffman PM, Hart SC, Harvey JW, Johnston CA, Mayorga E, McDowell WH, Pinay G. 2003. Biogeochemical hot spots and hot moments at the interface of terrestrial and aquatic ecosystems. Ecosystems 6:301–12.CrossRefGoogle Scholar
  38. Meier CL, Bowman WD. 2008. Links between plant litter chemistry, species diversity, and below-ground ecosystem function. Proc Natl Acad Sci USA 105:19780–5.PubMedCrossRefGoogle Scholar
  39. Mellert KH, Gensior A, Gottlein A, Kolling C, Rucker G. 2008. Variation in soil nitrate concentrations in two n-saturated norway spruce forests (Picea abies (L.) karst.) in southern Bavaria. Water Air Soil Pollut 187:203–17.CrossRefGoogle Scholar
  40. Moorhead DL, Sinsabaugh RL. 2006. A theoretical model of litter decay and microbial interaction. Ecol Monogr 76:151–74.CrossRefGoogle Scholar
  41. Okin GS, Mladenov N, Wang L, Cassel D, Caylor KK, Ringrose S, Macko SA. 2008. Spatial patterns of soil nutrients in two southern African savannas. J Geophys Res Biogeo 113. doi:10.1029/2007JG000584.
  42. Pennock DJ, Vankessel C, Farrell RE, Sutherland RA. 1992. Landscape-scale variations in denitrification. Soil Sci Soc Am J 56:770–6.CrossRefGoogle Scholar
  43. R Development Core Team. 2007. R 2.6.0: A Language and Environment for Statistical Computing. Vienna, Austria: R Foundation for Statistical Computing.Google Scholar
  44. Rivero RG, Grunwald S, Osborne TZ, Reddy KR, Newman S. 2007. Characterization of the spatial distribution of soil properties in water conservation area 2A, Everglades, Florida. Soil Sci 172:149–66.CrossRefGoogle Scholar
  45. Robertson GP, Huston MA, Evans FC, Tiedje JM. 1988. Spatial variability in a successional plant community—patterns of nitrogen availability. Ecology 69:1517–24.CrossRefGoogle Scholar
  46. Robertson GP, Klingensmith KM, Klug MJ, Paul EA, Crum JR, Ellis BG. 1997. Soil resources, microbial activity, and primary production across an agricultural ecosystem. Ecol Appl 7:158–70.CrossRefGoogle Scholar
  47. Rodionov A, Flessa H, Grabe M, Kazansky OA, Shibistova O, Guggenberger G. 2007. Organic carbon and total nitrogen variability in permafrost-affected soils in a forest tundra ecotone. Eur J Soil Sci 58:1260–72.CrossRefGoogle Scholar
  48. Schimel JP, Bennett J. 2004. Nitrogen mineralization: challenges of a changing paradigm. Ecology 85:591–602.CrossRefGoogle Scholar
  49. Seastedt TR, Bowman WD, Caine TN, McKnight D, Townsend A, Williams MW. 2004. The landscape continuum: a model for high-elevation ecosystems. Bioscience 54:111–21.CrossRefGoogle Scholar
  50. Sibbeson EA. 1977. Simple ion-exchange resin procedure for extracting plant available elements from soil. Plant Soil 46:665–9.CrossRefGoogle Scholar
  51. Steltzer H, Bowman WD. 1998. Differential influence of plant species on soil nitrogen transformations within moist meadow Alpine tundra. Ecosystems 1:464–74.CrossRefGoogle Scholar
  52. Stenger R, Priesack E, Beese F. 1998. Distribution of inorganic nitrogen in agricultural soils at different dates and scales. Nutr Cycl Agroecosyst 50:291–7.CrossRefGoogle Scholar
  53. Stenger R, Priesack E, Beese F. 2002. Spatial variation of nitrate-N and related soil properties at the plot-scale. Geoderma 105:259–75.CrossRefGoogle Scholar
  54. Strayer DL. 2005. Challenges in understanding the functions of ecological heterogeneity. In: Lovett GM, Jones CG, Turner MG, Weathers KC, Eds. Ecosystem function in heterogeneous landscapes. New York: Springer. p 411–25.CrossRefGoogle Scholar
  55. Townsend AR, Asner GP, Cleveland CC. 2008. The biogeochemical heterogeneity of tropical forests. Trends Ecol Evol 23:424–31.PubMedCrossRefGoogle Scholar
  56. van der Loo MPJ. 2010. Distribution based outlier detection in univariate data. The Hague: Statistics Netherlands. pp 1–14.Google Scholar
  57. Vitousek PM, Howarth RW. 1991. Nitrogen limitation on land and in the sea—how can it occur. Biogeochemistry 13:87–115.CrossRefGoogle Scholar
  58. Vitousek PM, Aber JD, Howarth RW, Likens GE, Matson PA, Schindler DW, Schlesinger WH, Tilman DG. 1997. Human alteration of the global nitrogen cycle: sources and consequences. Ecol Appl 7:737–50.Google Scholar
  59. Vitousek P, Chadwick O, Matson P, Allison S, Derry L, Kettley L, Luers A, Mecking E, Monastra V, Porder S. 2003. Erosion and the rejuvenation of weathering-derived nutrient supply in an old tropical landscape. Ecosystems 6:762–72.CrossRefGoogle Scholar
  60. Wang LX, Mou PP, Huang JH, Wang J. 2007. Spatial heterogeneity of soil nitrogen in a subtropical forest in China. Plant Soil 295:137–50.CrossRefGoogle Scholar
  61. Weintraub MN, Schimel JP. 2005. The seasonal dynamics of amino acids and other nutrients in Alaskan Arctic tundra soils. Biogeochemistry 73:359–80.CrossRefGoogle Scholar
  62. Yanai J, Lee CK, Umeda M, Kosaki T. 2000. Spatial variability of soil chemical properties in a paddy field. Soil Sci Plant Nutr 46:473–82.Google Scholar
  63. Yavitt JB, Wright SJ. 1996. Temporal patterns of soil nutrients in a Panamanian moist forest revealed by ion-exchange resin and experimental irrigation. Plant Soil 183:117–29.CrossRefGoogle Scholar

Copyright information

© Springer Science+Business Media, LLC 2011

Authors and Affiliations

  • Anthony Darrouzet-Nardi
    • 1
    • 2
    • 3
  • William. D. Bowman
    • 1
    • 3
  1. 1.Department of Ecology and Evolutionary BiologyUniversity of Colorado at BoulderBoulderUSA
  2. 2.Department of Environmental SciencesUniversity of ToledoToledoUSA
  3. 3.Mountain Research StationInstitute of Arctic and Alpine Research, University of ColoradoBoulderUSA

Personalised recommendations