Landscape Ecology

, Volume 28, Issue 4, pp 685–698

Ecosystem-scale spatial heterogeneity of stable isotopes of soil nitrogen in African savannas

Authors

    • Department of Earth SciencesIndiana University-Purdue University, Indianapolis (IUPUI)
    • Water Research Center, School of Civil and Environmental EngineeringUniversity of New South Wales
  • Gregory S. Okin
    • Department of GeographyUniversity of California, Los Angeles
  • Paolo D’Odorico
    • Department of Environmental SciencesUniversity of Virginia
  • Kelly K. Caylor
    • Department of Civil and Environmental EngineeringPrinceton University
  • Stephen A. Macko
    • Department of Environmental SciencesUniversity of Virginia
Research Article

DOI: 10.1007/s10980-012-9776-6

Cite this article as:
Wang, L., Okin, G.S., D’Odorico, P. et al. Landscape Ecol (2013) 28: 685. doi:10.1007/s10980-012-9776-6

Abstract

Soil 15N is a natural tracer of nitrogen (N) cycling. Its spatial distribution is a good indicator of processes that are critical to N cycling and of their controlling factors integrated both in time and space. The spatial distribution of soil δ15N and its underlying drivers at sub-kilometer scales are rarely investigated. This study utilizes two sites (dry vs. wet) from a megatransect in southern Africa encompassing locations with similar soil substrate but different rainfall and vegetation, to explore the effects of soil moisture and vegetation distribution on ecosystem-scale patterns of soil δ15N. A 300-m long transect was set up at each site and surface soil samples were randomly collected for analyses of δ15N, %N and nitrate content. At each soil sampling location the presence of grasses, woody plants, Acacia species (potential N fixer) as well as soil moisture levels were recorded. A spatial pattern of soil δ15N existed at the dry site, but not at the wet site. Woody cover distribution determined the soil δ15N spatial pattern at ecosystem-scale; however, the two Acacia species did not contribute to the spatial pattern of soil δ15N. Grass cover was negatively correlated with soil δ15N at both sites owing to the lower foliar δ15N values of grasses. Soil moisture did not play a role in the spatial pattern of soil δ15N at either site. These results suggest that vegetation distribution, directly, and water availability, indirectly, affect the spatial patterns of soil δ15N through their effects on woody plant and grass distributions.

Keywords

C3 plantsC4 plantsGeostatisticsIsotopeKalahariSavannasSoil δ15NSoil nitrogenStable isotopes

Introduction

Soil 15N is a natural integrator of soil nitrogen (N) cycling processes (Högberg 1997; Robinson 2001). Fundamentally, soil δ15N values are imposed by the isotopic compositions of N inputs (e.g., through N fixation and leaf litter decomposition), fractionations associated with N transformations (e.g., mineralization, nitrification and diagenesis) and N losses (e.g., denitrification and volatilization).

Factors influencing soil δ15N values can be classified as direct factors when they affect the soil δ15N through their impact on N inputs, transformations or losses. For example, vegetation affects the soil δ15N value through litter decomposition and N fixation. The fixed N in soil could either come from vegetation (e.g., N fixing trees or forbs) or from soil crusts especially in drylands (Evans and Ehleringer 1993; Aranibar et al. 2003). Soil N transformation processes such as ammonification, ammonia volatilization, coupled nitrification–denitrification will also have significant impact on soil δ15N values. It is worth noticing that some processes could be less important in certain conditions. For instance, if the pH values are low (e.g., soils are acidic), ammonia volatilization will be minimum. Besides the direct factors, the soil δ15N values depend also on other indirect factors such as soil topography (Garten and Van Miegroet 1994; Bai et al. 2009) and rainfall (Aranibar et al. 2004; Swap et al. 2004; Wang et al. 2010a). These factors are related to N transformations and losses through their impact on the soil water balance and can therefore indirectly affect soil δ15N values. For instance, soil moisture dynamics strongly control soil N transformations in drylands (Austin et al. 2004; Wang et al. 2009a). Because 15N is one component of total N, soil δ15N dynamics are also influenced by soil moisture variations.

The spatial distribution of soil δ15N values therefore results from the integration of both direct and indirect factors. At watershed scales, soil δ15N has been found to be slightly higher on south facing slopes, compared with ridges, valley bottoms and north facing slopes in the northern hemisphere (Garten 1993). The soil δ15N values of extractable soil NH4+ and NO3, however, were much higher at the valley bottoms than other locations, because of the higher rates in mineralization and nitrification at valley bottoms (Garten 1993), which fractionated 15N and made the residuals enriched. The higher mineralization and nitrification rates were likely linked to higher water availability at the valley bottoms. At regional scale, soil δ15N values have been used to correlate with N deposition because soil δ15N increases when N deposition induced nitrification and nitrate loss increase (Pardo et al. 2007). At the global scale, the soil δ15N spatial distribution has been empirically related to climatic parameters such as mean annual temperature and mean annual precipitation (MAP, Amundson et al. 2003). Recent studies have explored the possibility of predicting foliar δ15N (closely related to soil δ15N) using remote sensed spectral data (Wang et al. 2007a; Elmore and Craine 2011), pointing out the possibility of direct estimate of large-scale δ15N variations. Spatial patterns of soil δ15N have rarely been investigated at the sub-kilometer to kilometer spatial scale (e.g., ecosystem-scale), especially with spatial-explicit samplings (vs. discreet sampling) (e.g., Bai et al. 2009; Cheng et al. 2009). The interpretation of such ecosystem-scale distribution of δ15N is partly complicated by the complex interactions and synergic effects between direct and indirect factors, whereby both vegetation distribution and soil moisture could affect δ15N patterns and these two factors could also interact with each other. Processes such as mineralization, nitrification and denitrification affect the δ15N spatial pattern at small scales. Because these processes are all directly or indirectly related to the presence of vegetation, at larger scales their overall effects may be investigated through a few aggregated indicators such as woody cover and grass cover. This approach simplifies the ecosystem-scale soil δ15N estimation without necessarily evaluating in detail all the processes involved in N cycling. The investigation of soil δ15N spatial patterns at the ecosystem-scale and the exploration of the underlying mechanisms are therefore critically needed to bridge the gap existing between δ15N measurements at different scales.

In the present study, we aimed to explore (1) whether there are spatial patterns in soil δ15N at a large spatial scale (e.g., sub-kilometer scale); and (2) how soil moisture and woody vegetation—particularly potential N-fixing species—interactively control the spatial distribution of soil δ15N at this scale (in case soil δ15N spatial patterns are observed). To this end, we used the Kalahari Transect (KT) as a model system. The KT is one of a set of International Geosphere–Biosphere Programme (IGBP) “megatransects” (Koch et al. 1995) identified for global change studies. There are several unique aspects of this transect making it ideal for the ecosystem-scale characterization of the spatial patterns of soil δ15N: (1) the soil substrate consists of Kalahari sands, which are consistent and homogeneous along the entire transect (Wang et al. 2007b); (2) the topography across the transect is flat with gentle slopes (<1 %), minimizing the interactive effects of rainfall and topography on soil moisture. (3) there are distinct differences in vegetation composition and structure along the transect. In the south (the dry end of the transect), it is a fine-leaf savanna with several potential N-fixers such as Acacia mellifera Benth. and A. luederitzii Engl. Vegetation in the north (the wet end) is dominated by broad-leaf woody species such as Burkea africana, Ochna pulchra, Bauhimia macrantha, which are not potential N-fixers. According to an earlier report, using foliar δ15N, C/N ratio and taxonomic classification as indication, there are potential N-fixing forbs and some potential N-fixing trees, such as Baphia, Indigophera, and Tephrosia at the wet end (Aranibar et al. 2004), but they are less common in the current study site. The mesic site considered in this study also does not exhibit N-fixing genera such as Dalbergia, Dichrostachys, Entada and Xeroderris that were found in Tanzania woodlands under similar rainfall regimes (Hogberg 1986). Soil crusts are also reported to fix N, but soil crust cover is low at the mesic site, probably due to anthropogenic disturbance based on field observations. The KT thus provides a species and rainfall contrasts to investigate ecosystem-scale soil δ15N spatial patterns along a precipitation gradient while minimizing confounding effects of soil heterogeneity. Three hypotheses were tested: (1) soil δ15N is spatially structured at both ends of the KT (dry vs. wet); (2) the spatial pattern of soil δ15N is influenced by woody plant distribution at both ends of the KT. The first two hypotheses are based on the results of previous studies showing a relationship between the spatial distribution of woody plants and total soil N (Okin et al. 2008) and based on the positive relationships between total N and δ15N in many non-N fixing dominant ecosystems (e.g., a global synthesis in Wang et al. 2009b); and (3) soil moisture affects the spatial distribution of soil δ15N. This is based on the modeling and field observations showing the effect of soil moisture on soil N dynamics (e.g., Wang et al. 2009a) as well as the positive relationships between total N and δ15N in non-N fixing dominant ecosystems (Wang et al. 2009b). Supposedly higher soil moisture will lead to higher mineralization and nitrification rates (Austin et al. 2004), since mineralization/nitration fractionate 15N, the residuals will be enriched.

Materials and methods

Field sites and field samplings

The MAP along the KT ranges from less than 200 mm in the south to over 1,000 mm in the north. The physical parameters such as soil texture (>96 % of sand) and bulk density (around 1.4–1.5 g cm−3 along the whole transect) do not have significant variations along the KT (Wang et al. 2007b; Thomas et al. 2008). Two sites in the Botswana part of the Kalahari (Fig. 1) with different mean MAP (near the extreme MAPs along the KT) were chosen to compare the spatial patterns of soil δ15N and explore the underlying mechanisms of such patterns. The southern site, Tshane, is at the dry end of the transect (~365 mm MAP) and the northern site, Mongu, is wetter (~900 mm MAP). The Tshane site is classified as open savanna dominated by Acacia species such as A. luederizii and A.mellifera whereas the dominant grass species are Eragrostis lehmanniana and Schmidtia pappophoroides. The Mongu site is located in a woodland savanna dominated by tree species such as Brachystegia spiciformis Benth and the Eragrostis spp. grass species. The A. luederizii and A.mellifera species found in Tshane are potential N fixers, though their foliar δ15N reaches 8 ‰, suggesting that they may not rely heavily on atmospheric N as their major N source at the this site (Wang et al. 2010a). The sampling was conducted in February–March, 2005. The rainfall was 436 mm in Tshane, higher than the MAP of the site, and 564 mm in Mongu in 2005, lower than the MAP of the site. The 2005 rainfall data was from Tropical Rainfall Measuring Mission (TRMM), and was reported earlier (Wang et al. 2010b). Five individual trees and grasses were randomly selected at each site, foliar and root samples (0–30 and 30–50 cm for trees, 0–30 cm for grasses) were collected. For grasses, the whole individual was collected for foliar samples; for trees, 5–10 leaves were collected from each tree at 1.5 m height. The tree species in Tshane was A.mellifera and grass species were Schmidtia pappophoroides, Eragrostis lehmanniana, Stipagrostis suniplumis and Aristida congesta. The species at Mongu were not identified.
https://static-content.springer.com/image/art%3A10.1007%2Fs10980-012-9776-6/MediaObjects/10980_2012_9776_Fig1_HTML.gif
Fig. 1

Sampling locations and vegetation characteristics along the KT. The satellite images of each site are IKONOS (Tshane) and Quickbird (Mongu)

A 300-m line transect was set up at each site and surface soil samples (0–5 cm depth) were randomly collected along the transect (with an ~1 m interval on average, 220 samples from each site). The sampling locations were selected along the transect using a random number generator with uniform distribution. The irregular spacing of samples along the transect allowed us to calculate semivariograms relying on a relatively large number of points that are closely spaced. While performing the random soil sampling along each transect, the presence/absence (in binary form) of trees and shrubs (C3 vegetation) or of grasses and forbs (C4 vegetation) as well as presence/absence of bare ground were recorded for each sampling point. Soil crust covers were low at both sites, presumably due to disturbance of local villagers, and were not recorded. Surface soil moisture (0–5 cm) was also measured at each sampling point using Time Domain Reflectometry (TDR, Campbell Scientific Inc., Logan, USA). The resolution is about 1 % and varies for different soil conditions. Because of the onetime sampling, the soil moisture results are snapshot based and may not reflect the long-term values.

Laboratory analysis

All soil samples were oven dried at 60 °C in the laboratory, sieved to remove plant debris and homogenized using mortar and pestle. The soil C and N content was measured using an Elemental Analyzer (EA, Carlo Erba, NA1500). The nitrate concentrations (including both NO3 and NO2) were analyzed using Dionex ICS-2000 ion chromatograph (Dionex, Sunnyvale, CA) with an Ion Pac AS 18 anion exchange column. The soil C and N as well as nitrate concentrations were reported in an earlier study (Okin et al. 2008). The stable N isotope analyses of plant foliar, root samples and soil were performed using an optima isotope ratio mass spectrometer (IRMS) connected to the EA (GV/Micromass, Manchester, UK). Stable isotope compositions are reported in the conventional form:
$$ \delta^{15} {\text{N}}\left(\permille\right) = \left[ {{{\left( {^{15} {\text{N}}/^{14} {\text{N}}} \right)_{\text{sample}} } \mathord{\left/ {\vphantom {{\left( {^{15} {\text{N}}/^{14} {\text{N}}} \right)_{\text{sample}} } {\left( {^{15} {\text{N}}/^{14} {\text{N}}} \right)_{\text{standard}} - 1}}} \right. \kern-\nulldelimiterspace} {\left( {^{15} {\text{N}}/^{14} {\text{N}}} \right)_{\text{standard}} - 1}}} \right] \times 1000 , $$
(1)
where (15N/14N)sample and (15N/14N)standard are the isotopic ratios of the sample and standard, respectively. The standard is atmospheric N2. Reproducibility of these measurements is typically better than 0.2 ‰.

Statistical analysis

All data sets were tested for normality (SAS 9.1, SAS Inc., Cary, NC, USA) before performing conventional statistical and geostatistical analyses. Conventional statistics (i.e., mean and coefficient of variation (CV)) were calculated to investigate the overall variability for each analyzed variable. Geostatistical analyses were performed with the GS+ software (Version 7.0, Gamma design software, Plainwell, MI, USA) to determine for each site the spatial variation of soil δ15N, %N, NO3 + NO2, water content, and vegetation. Before the semivariogram computation, the data were tested for 1st or 2nd order trends (Davis 2002) and no trends were found.

A semivariogram (SV hereafter) is a plot of a series of semivariance values (γ) against the corresponding lag distances (h). The semivariance γ at each h is defined as:
$$ \gamma \left( h \right) = \frac{1}{2N\left( h \right)}\sum\limits_{i = 1}^{N\left( h \right)} {\left[ {z\left( i \right) - z\left( {i + h} \right)} \right]^{2} }^{{}} , $$
(2)
where N(h) is the number of sample pairs separated by the lag distance h. z(i) is a value measured at location i and z(i + h) is a value measured at location i + h. A number of models are commonly used to fit the calculated values of the SV, γ(h). The model selection is based on two criteria: high R-square and the shape of the fitted model. Because in similar studies the spherical model was shown to provide an adequate representation of the spatial variability of soil data (Schlesinger et al. 1996; Su et al. 2006; Wang et al. 2007c), in this study the semivariograms of all the variables at the wet site (Mongu) were fitted by a spherical model. However, for the Tshane site the exponential model provides a better fit of the soil δ15N data both in terms of shape and R-square value. Therefore, the exponential model was used for all the variables at the dry site (Tshane) to facilitate the comparison within this site. A 75 m lag distance and 1 m uniform interval were used for semivariogram calculations.
The spherical model is defined as:
$$ \begin{aligned} \gamma \left( h \right) & = {\text{C}}_{ 0} + {\text{C}}\left[ {1.5\left( {h/{\text{A}}_{ 0} } \right) - 0.5\left( {h/{\text{A}}_{ 0} } \right)^{3} } \right]\;\;\;\;{\text{for}}\;h \le {\text{A}}_{ 0} \\ \gamma \left( h \right) & = {\text{C}}_{ 0} + {\text{C}}\;\;\;\;{\text{for}}\;h > {\text{A}}_{ 0} \\ \end{aligned} , $$
(3)
The exponential model is defined as:
$$ \gamma \left( h \right) = {\text{C}}_{ 0} + {\text{C}}\left[ {1 - \exp \left( { - h/{\text{A}}_{ 0} } \right)} \right] , $$
(4)
where C0 is nugget, A0 is range and h is lag distance interval.

Three SV parameters were derived and used in the analysis: nugget (C0), range (A0), the ratio of structure variance (C) and sill variance (C + C0) (C/(C + C0), thereafter). Nugget reflects either the variability at scales finer than data resolution or variability due to measurement or locational error. Range indicates the distance of spatial autocorrelation between data pairs. Samples separated by distances smaller than the range are correlated whereas samples separated by greater distance are independent. The value C/(C + C0) is the proportion of the total variance that is spatially structured. A high C/(C + C0) indicates that variability in the dataset is strongly structured (Brooker 1991; Wang et al. 2007c; Li et al. 2010). The uncertainties (95 % confidence limits) of the derived parameters were determined using the variance–covariance method of Pardo-Iguzquiza and Dowd (2001). The 95 % confidence limits of these parameters were calculated using a code written in Interactive Data Language (IDL, Exelis Visual Information Solutions, Boulder, CO, USA) that was applied to earlier studies (Li et al. 2008; Okin et al. 2008; Wang et al. 2009c) and is available on request.

The spatial correlations of all the parameters were calculated with “modified t test for correlation” using PASSAGE software package (http://www.passagesoftware.net/) developed by Rosenberg and Anderson (2011). This test corrects the degrees of freedom based on the amount of auto-correlation in the data (e.g., Wang et al. 2007c).

Results and discussion

Figure 2 depicts the ecosystem-scale variations of δ15N (A), soil %N (B), soil C/N (C), soil NO3 + NO2 (D) and soil moisture (E) at Tshane and Mongu. Soil δ15N is higher at the Tshane (dry) site with a mean value of 9.2 ‰, whereas the mean value at the Mongu (wet) site is 4.4 ‰ (Table 1). At Tshane the soil δ15N is higher under trees or shrubs than in bare soil or beneath grasses; similar results were found at Mongu but with smaller differences (Table 1). Unlike the soil δ15N, the soil N content exhibits an opposite pattern: it is lower at the Tshane (dry) site with a mean value of 0.02 % and higher at the Mongu (wet) site with a mean of 0.04 ‰ (Table 1). These results are in agreement with previous findings from this region (Swap et al. 2004; Aranibar et al. 2008; Wang et al. 2009d, 2010a). The higher soil δ15N values at Tshane generally indicate a more open N cycle (e.g., the ratio of N losses to internally cycled N is high) at the dry end, which likely result from higher N availability relative to N demand in more arid environments. δ15N values at the dry site were found to be around 14 ‰ in a 2000 sampling (Aranibar et al. 2004), higher than that in the current study (9.2 ‰) and than another independent 2005 wet season sampling (10.4 ‰ for under canopy and 9.7 ‰ for open canopy areas, Wang et al. 2010a). These differences may be due to the different sampling location among these studies since the heterogeneity is large at the dry site. Water availability may, however, play a more important role. The year 2000 was wet and followed the 1999 drought. The sampling year (2005) of the current study is also a wet year with rainfall value of 436 mm (MAP of the site is 365 mm). Since both sampling years are wetter than the MAP, the higher δ15N in 2000 is likely related to the dry-wet cycle in 1999 and 2000. It is possible that N gas loss is large in 2000 following heavy rainfall events because of the N buildup during the 1999 drought. The CV of soil δ15N is much lower at Tshane (24 %) compared to Mongu (51 %) (Table 1). The CVs of soil %N at the two sites are comparable with CV values of 65 and 61 % at Tshane and Mongu, respectively (Table 1); these results are similar to those reported for soil %C in an earlier study (Wang et al. 2009c). The lower soil δ15N CV value at the dry site is unexpected since resources are usually more heterogeneous at the dry end (Wang et al. 2007b). The lower spatial variability of soil δ15N at the dry end is probably due to the indistinct soil δ15N values between under canopy and open canopy areas (Table 1), as found in an early report with a non-spatial explicit sampling design (Wang et al. 2007b). The large CV in soil δ15N at the wet site (Mongu) is likely related to the rainfall situation of the sampling year (2005). The rainfall value (564 mm) is significantly lower than the MAP (about 900 mm). The drier condition at the wet site will make the water availability difference between open and under canopy areas larger, which will lead to higher δ15N difference as water availability largely determines δ15N in water-limited systems (Austin and Sala 1999; Aranibar et al. 2004; Wang et al. 2010a). The same principle but converse trend (higher rainfall and lower δ15N CV value) applies to the dry site, as the rainfall amount is larger than the MAP at the drier site. There was larger difference in NO3 + NO2 between under canopy and open canopy areas at the dry site compared with the wet site (Table 1), which is similar to the modeling study result in this region (Wang et al. 2009a). The soil moisture was higher at the wet site (~3 %) than the dry site (~0.7 %) but the differences between microhabitats were small at both sites (Table 1). The small difference at the dry site is unexpected, which is likely related to the fact that the low soil moisture (~0.7 %) of the site is close to the sensor resolution (~1 %).
https://static-content.springer.com/image/art%3A10.1007%2Fs10980-012-9776-6/MediaObjects/10980_2012_9776_Fig2_HTML.gif
Fig. 2

Ecosystem-scale variations of δ15N (a), soil %N (b), soil C/N (c), soil NO3 + NO2 (d) and soil moisture (e) at Tshane and Mongu. The %N, C/N and NO3 + NO2 data have been reported in Okin et al. (2008) and were used here to facilitate the interpretations of soil δ15N spatial pattern

Table 1

Mean value and CV of soil δ15N, %N, C/N, NO3 + NO2 (mg/g) and volumetric soil moisture (%) at Tshane (dry) and Mongu (wet)

Location

δ15N (‰)

 %N

C/N

NO3 + NO2 (mg/g)

Soil moisture (%)

Mean

CV

Mean

CV

Mean

CV

Mean

CV

Mean

CV

Tshane (dry)

 All samples

9.2

0.24

0.023

0.65

8.34

0.10

1.79

2.36

0.71

0.78

 Tree/shrub

9.9a

0.20

0.044a

0.42

8.18a

0.12

4.55a

1.37

0.58a

0.89

 Grass

8.9b

0.26

0.021b

0.37

8.35a

0.08

1.39b

2.27

0.65a

0.91

 Bare

9.3ab

0.20

0.019b

0.45

8.05a

0.08

1.24b

2.09

0.91a

0.58

Mongu (wet)

 All samples

4.4

0.51

0.039

0.61

17.82

0.10

3.44

1.03

3.25

0.20

 Tree/shrub

4.3a

0.49

0.041a

0.58

17.97a

0.11

3.13a

1.16

3.12a

0.15

 Grass

2.6b

0.74

0.032a

18.66a

10.10

3.00a

0.24

 Bare

4.8a

0.61

0.028b

0.66

17.88a

0.09

4.51a

0.68

3.17a

0.12

The %N, C/N and NO3 + NO2 data have been reported in Okin et al. (2008) and were used here to facilitate the interpretations of soil δ15N spatial pattern

The same letters indicate the same mean for different microhabitats (e.g., tree/shrub, grass and bare) within the same site at α = 0.05

Plant foliar δ15N follows similar trends as the soil δ15N, with higher mean values (8.3 and 5.3 ‰ for C3 and C4 plants respectively) at the Tshane (dry) site and lower mean values (2.5 and 2.1 ‰ for C3 and C4 plants respectively) at the Mongu (wet) site (Table 2). Root δ15N patterns are similar but the values are much lower than those of foliar and soil samples (Table 2). The pattern of lower plant δ15N than soil δ15N is similar to that found at the global scale (Amundson et al. 2003). The average foliar δ15N value at Tshane is higher than those reported for Namib Desert species with an average of 4.4 ‰ for N-fixing species (Schulze et al. 1991) and also higher than those found in a Tanzania savanna woodland (Hogberg 1986). The foliar δ15N in N-fixing plants has been shown to be close to atmospheric δ15N (0–2 ‰) (Shearer and Kohl 1986; Yoneyama et al. 1990). The average soil δ15N under two potential N-fixing species in this study is 9.2 and 10.1 ‰ for A. luederizii and A.mellifera, respectively. The high soil δ15N under the two Acacia species indicate that N-fixing may not be a major N source to this system, which is in agreement with earlier reports in southern Africa (Aranibar et al. 2004; Wang et al. 2010a, in press). Other studies have shown that some leguminous shrubs with nodules in their root systems do not appear to be active N fixers, though N fixation could occur in wetter years (Golluscio et al. 2006).
Table 2

Summary of statistical parameters of plant δ15N, %N and C/N at Tshane (dry) and Mongu (wet)

Location

δ15N (‰)

%N

C/N

Mean

CV

Mean

CV

Mean

CV

Tshane (dry)

 C3 leaves

8.3

0.16

3.17

0.079

14.12

0.072

 C3 roots (0–30 cm)

5.5

0.25

1.74

0.20

19.39

0.092

 C3 roots (30–50 cm)

2.0

0.26

1.70

0.12

26.31

0.14

 C4 leaves

5.3

0.26

2.27

0.26

20.74

0.22

 C4 roots (0–30 cm)

1.1

1.1

0.82

0.33

59.80

0.27

Mongu (wet)

 C3 leaves

2.5

0.25

2.37

0.070

21.94

0.051

 C3 roots (0–30 cm)

2.0

0.97

1.60

0.20

28.75

0.35

 C3 roots (30–50 cm)

−1.3

−0.52

1.08

0.13

45.14

0.15

 C4 leaves

2.1

1.1

0.84

1.06

28.35

0.042

 C4 roots (0–30 cm)

0.9

0.8

0.68

0.40

59.95

0.036

There appear to be well-defined spatial patterns for soil δ15N at Tshane but not at Mongu. At Tshane the range of autocorrelation is similar for soil δ15N and woody existence (Table 3; Fig. 3), with estimated values of 15.4 and 15.0 m, respectively (Fig. 3a, c). At Mongu, the spatial structure of soil δ15N is consistent with that of a nugget model (random pattern). Using spherical model we determine a range of autocorrelation in woody cover of 1.3 m at Mongu, although with a low R2 of 0.06 (Fig. 3b, d), indicating a random distribution. The C/(C + C0) values for soil δ15N is lower than that of woody cover (Fig. 3a, c) at the dry site, indicating that the degree of spatial dependence of soil δ15N is less than that of woody cover.
Table 3

Range (m) and proportion of structured variance (C/(C0 + C)) for variograms from both sites

Location

Analyte

Range (m)

C/(C0 + C)

M

L

U

M

L

U

Tshane (dry)

δ15N

15.40

15.08

15.72

0.38

0.34

0.41

%N

19.50

19.16

19.84

0.34

0.32

0.37

NO3 + NO2

4.31

4.14

4.48

0.01

−0.05

0.08

Mongu (wet)

δ15N

Nugget model

%N

4.77

4.59

4.95

0.64

0.59

0.68

NO3 + NO2

Nugget model

“M” denote the mean jackknife estimate. “L” and “U” denote the lower and upper bounds of the 95 % confidence estimates, respectively. The %N, and NO3 + NO2 data have been reported in Okin et al. (2008) and were re-analyzed here using exponential model for Tshane to facilitate the comparison with soil δ15N spatial pattern

https://static-content.springer.com/image/art%3A10.1007%2Fs10980-012-9776-6/MediaObjects/10980_2012_9776_Fig3_HTML.gif
Fig. 3

Semivariograms of δ15N at Tshane (a), Mongu (b), and semivariograms of woody cover at Tshane (c) and Mongu (d) along the KT. The summary of semivariogram model parameters is reported under the curve. Proportion structural variation (C/(C + C0)) is used as an index of the magnitude of spatial dependence

At the dry site (Tshane), the spatial patterns of soil δ15N are in agreement with those of woody plant presence, whereas the spatial distribution of soil δ15N is different from that of snapshot soil moisture measurements, as suggested by the different ranges of autocorrelation of these two variables (Figs. 3, 4). The range of autocorrelation of soil moisture is 344 m (Fig. 4a), while the range of autocorrelation of soil δ15N is 15.4 m, which indicates that snapshot soil moisture results do not explain the spatial structure of soil δ15N at the dry site. At the wet site, soil moisture does not exhibit a spatial structure as suggested by the nugget model (Fig. 4b). In addition, the correlation between soil moisture and soil δ15N is not significant (Table 4), indicating that the snapshot soil moisture do not affect the spatial distribution of soil δ15N at the wet site either. We should stress, however, that these soil moisture measurements were taken as snapshot of the soil water content at the time of soil sampling. Even though these data capture the higher soil moisture existing at the wet site (3.25 vs. 0.71 % at the wet and dry end, respectively, Table 1), we cannot exclude that more significant within-site relations might be found between soil moisture and soil δ15N if longer-term soil moisture measurements were available. Therefore, caution should be used before concluding that soil moisture does not contribute to the spatial patterns of soil δ15N. Long-term records (e.g., a whole growing season) of soil moisture monitoring or the geochemical indicators reflecting a long-term soil water status such as carbonate-rich soil horizon layer depth may help assess whether a relationship with the spatial distribution of soil δ15N exists. In fact, process-based modeling work has shown that soil moisture plays an important role in regulating the N cycles in drylands in general (D’Odorico et al. 2003) and in the Kalahari in particular (Wang et al. 2009a). Because 15N is more sensitive than the total N changes, it is expected that soil moisture spatial distribution will affect the soil δ15N spatial patterns.
https://static-content.springer.com/image/art%3A10.1007%2Fs10980-012-9776-6/MediaObjects/10980_2012_9776_Fig4_HTML.gif
Fig. 4

Semivariograms of soil moisture at Tshane (a), Mongu (b), and semivariograms of grass cover at Tshane (c) and Mongu (d) along the KT. The summary of semivariogram model parameters is reported close to the curve

Table 4

Correlations of soil δ15N, %N, C/N, NO3 + NO2, volumetric soil moisture, woody distribution, grass distribution and Acacia tree distribution (Tshane only) at Tshane (dry) and Mongu (wet)

 

δ15N (‰)

%N

C/N

NO3 + NO2 + NO2

Moisture

Woody

Grass

ALE

AME

ALE/AME

Bare

Tshane (dry)

 δ15N (‰)

1

 

 %N

0.29**

1

 

 C/N

–0.04

0.11

1

 

 NO3 + NO2

–0.04

–0.08

0.08

1

 

 Moisture

–0.0008

–0.02

–0.07

0.10

1

 

 Woody

0.25**

0.52***

0.19*

−0.07*

–0.16

1

 

 Grass

–0.20**

–0.42**

0.07

−0.09

–0.11

–0.31***

1

 

 ALE

–0.01

0.47***

0.35***

0.04

0.13

0.42***

–0.20*

1

 

 AME

0.10

0.06

–0.03

−0.03

–0.04

0.29***

–0.11

–0.08

1

 

 ALE/AME

0.05

0.43***

0.28***

0.02

0.02

0.53***

–0.24**

0.79***

0.54***

1

 

 Bare

–0.003

–0.13*

–0.14*

–0.0099

0.13

–0.28***

–0.53***

–0.12

–0.08

–0.15*

1

Mongu (wet)

 δ15N (‰)

1

 

 %N

–0.08

1

 

 C/N

–0.32***

0.22**

1

 

 NO3 + NO2

–0.02

0.08

–0.05

1

 

 Moisture

0.02

–0.02

–0.007

0.20

1

 

 Woody

–0.07

0.14

0.04

0.002

–0.03

1

 

 

 Grass

–0.18*

–0.12

–0.019

–0.06

–0.10

0.08

1

 

 Bare

0.12

–0.15*

–0.056

0.0026

0.030

–1.0***

–0.078

1

The correlations were calculated using a modified t test that corrects the degrees of freedom based on the amount of auto-correlation in the data (e.g., Wang et al. 2007c). The %N, C/N and NO3 + NO2 data have been reported in Okin et al. (2008) and were used here to facilitate the interpretations of soil δ15N spatial pattern

ALE: Acacia luederitzii, AME: Acaciamellifera

P < 0.05, ** P < 0.01, *** P < 0.001

There is a significant negative spatial correlation between grass cover and soil δ15N both at the dry and at the wet site with correlation coefficients of −0.20 and −0.18 at Tshane and Mongu, respectively (Table 4). This negative correlation is caused by the lower foliar δ15N input from grasses under the grass canopy areas since grass foliar δ15N is lower than woody species (Table 2). There is a significant negative spatial correlation between soil C/N ratios and soil δ15N at the wet site with a correlation coefficient of −0.32 (Table 4), which is likely caused by the higher foliar and root C/N ratios and lower foliar and root δ15N in grasses compared with woody species (Table 2). Such a negative correlation between soil C/N and soil δ15N does not appear at the dry site. The pattern of C/N and δ15N in root and foliar samples between under canopy and open canopy areas is similar to that of the wet end, but other factors such as wind and water move soil and debris around and homogenize the landscape at the dry site, resulting in non-different soil δ15N and C/N values between under canopy and open canopy areas (Table 1). This is probably the reason why a negative correlation between soil C/N and soil δ15N is not found at the dry site. There is also a significant positive spatial correlation between soil %N and soil δ15N at the dry site with a coefficient of 0.29 (Table 4), which is likely caused by the lower foliar and root %N and lower foliar/root δ15N in grasses compared with woody species at the dry site (Table 2). Such a positive correlation between %N and soil δ15N does not appear at the wet site because there is no difference in foliar δ15N values between trees and grasses at the wet end (Wang et al. 2010a). There is no spatial correlation between NO3 + NO2 and soil δ15N both at the dry and wet site (Table 4), indicating these anions alone could not represent N transformation processes in these systems. Bare ground is negatively correlated with woody cover at both sites and with grass cover only at the dry site. There is no spatial correlation between bare ground and soil δ15N but negative relationships exist for bare ground and soil %N at both sites (Table 4). Together with the fact that %N spatial autocorrelation range is larger than that of soil δ15N (Table 3), these results indicate that different mechanisms are responsible for the spatial patterns of soil δ15N and soil %N.

To further explore the determinant factors of the spatial patterns of soil δ15N, the spatial distribution of two potential N-fixing Acacia species (A. luederizii and A.mellifera) were compared with that of soil δ15N. The range of autocorrelation of A. luederizii (Fig. 3a, 5a) is closer to that of soil δ15N (12.3 vs. 15.4 m, respectively), whereas the autocorrelation range of A. mellifera is substantially different (2.7 m, Fig. 3a, 5b). However, the distribution of both Acacia species have no significant correlation with that of soil δ15N (Table 4), thereby strengthening the argument that N-fixation is not a major N source in this system, and therefore it does not affect the spatial pattern of soil δ15N. It also indicates that A. mellifera plays a minor role in shaping up the soil δ15N spatial patterns. The woody cover distribution is positively correlated with soil δ15N, as well as with soil %N (Table 4). Together with the fact that no correlation exists for potential N-fixing trees, these results indicate that woody plants affect the spatial distribution of soil δ15N not through N-fixation. The autocorrelation ranges of soil δ15N and woody existence are similar, namely, 15.4 and 15.0 m, respectively (Fig. 4), while that of %N is 19.50 m (Table 3). The larger autocorrelation range for %N compared with woody existence indicates the plant root effect on %N since in water-limited systems root distribution is generally larger than the canopy size. It also indicates that factors influencing δ15N are confined within tree canopy, but factors influencing %N are exceeding the canopy (e.g., root distribution). The factors influencing δ15N include physical processes such as canopy shading, litter input, evaporation reduction, as well as chemical processes such as the enhancement of mineralization, nitrification and denitrification. Litter inputs may also inhibit the development of soil crusts thereby causing a decrease in N fixation by cyanobacteria. The similar autocorrelation ranges of δ15N and woody existence, however, indicate that these factors can be lumped together at ecosystem spatial scale. This result thus indicates that—in dry environments where the spatial distribution of woody vegetation determines the soil δ15N spatial distribution—predicting ecosystem-scale patterns of soil δ15N can be achieved through the quantification of woody vegetation spatial patterns.
https://static-content.springer.com/image/art%3A10.1007%2Fs10980-012-9776-6/MediaObjects/10980_2012_9776_Fig5_HTML.gif
Fig. 5

Semivariograms of Acacialuederitzii alone (a), Acacia mellifera alone (b) and the combination of the two (c) at Tshane. The summary of semivariogram model parameters is reported on the right

Summary

Contrary to our first hypothesis, spatial patterns of soil δ15N only exist at the dry site (Tshane), but not at the wet site (Mongu), which may result from less spatially structured woody plant distribution at the wet site. Consistent with our second hypothesis, based on the similar spatial distribution of woody vegetation and soil δ15N, it is found that woody cover distribution affects soil δ15N at the dry site; however, the two potential N-fixing species do not contribute the spatial pattern for soil δ15N. This is explained by the very low N input from N-fixation under dry conditions suggested by the high foliar and soil δ15N values. This result also indicates that woody vegetation affects the spatial distribution of soil δ15N through N inputs and other transformation processes, not through N-fixation. Grass cover has negative correlations with the spatial pattern of soil δ15N spatial pattern both at the dry and at the wet site because of the lower foliar δ15N values of grasses compared with woody species. Contrary to our third hypothesis, soil moisture did not exhibit any control on the spatial pattern of soil δ15N both at the dry and wet site, presumably because soil moisture measurement were available only as “snapshots” and the time scale of soil moisture dynamics are different from those of δ15N. Whereas previous studies showed a strong rainfall control on vegetation distribution and tree-grass interactions along the KT (Caylor et al. 2003; D’Odorico et al. 2007; Aranibar et al. 2008; Wang et al. 2010a), this study shows that both tree and grass distributions affect the spatial patterns of soil δ15N at the dry end of the KT. The combined results suggest that water availability indirectly affects the spatial distribution of soil δ15N through its effects on the distribution of woody vegetation and grass vegetation at the dry site.

Acknowledgments

The project was supported by NASA-IDS2 (NNG-04-GM71G). We greatly appreciate the field assistance and technical assistance from Natalie Mladenov, Lydia Ries, Beth Caylor, Junran Li, Dikitso Kolokose, O.G.S.O. Kgosidintsi, Thoralf Meyer and Kebonyethata Dintwe. Lixin Wang also acknowledges the financial support from Vice-Chancellor’s postdoctoral research fellowship of University of New South Wales. We thank three anonymous reviewers for their constructive comments.

Copyright information

© Springer Science+Business Media B.V. 2012