Exotic grasses and nitrate enrichment alter soil carbon cycling along an urban–rural tropical forest gradient

Study sites

The main site for this study was an urban–rural (hereafter ‘urban’) watershed in Puerto Rico, with additional reference forest sites in a more remote watershed (Fig. 1). In the urban Río Piedras watershed, six secondary forest stands were paired with nearby exotic grasslands. These sites were all within the San Juan Urban Long Term Research Area (ULTRA-Ex), and none of the sites were under active management. The six sites were in six subwatersheds (i.e., ‘watershed synoptics’) and represented all of the major forested areas of the urban watershed, spanning from the low-elevation urban core to higher elevation rural and suburban areas (Fig. 1). The urban watershed is in the subtropical moist forest life zone (sensu Holdridge et al., 1971), ranges in elevation from 0 to 220 m above sea level (masl), has mean annual precipitation (MAP) of 1750 mm, and has mean annual temperature (MAT) of 25.7 °C. Averages from 18 years of stream discharge data for the Rio Piedras show a distinct dry season (Dry) from February to March (0.3–0.37 m³ s⁻¹), an early wet season (Wet) from April to July (0.4–0.6 m³ s⁻¹), a peak in rainfall during the hurricane season (Hurricane) from August to December (0.7–1 m³ s⁻¹), and a transition back to drier conditions during January (0.5 m³ s⁻¹) (USGS 2015). Sixteen weeks of preliminary N deposition data in the mid-urban watershed (near site 3.0, Table 1) showed high variability in N deposition rates, which when scaled up could represent inputs ranging from 10 to 40 kg-N ha⁻¹ yr⁻¹ (USFS-IITF 2011, unpublished data). The urban watershed has also shown large river outputs of N compared with nearby rural watersheds (Ortiz-Zayas et al., 2006), suggesting substantial N deposition within the watershed.

The remote forest sites were located in the USFS Cubuy Annex (18°16′ N, 65°52′ W), which is also secondary forest in the subtropical moist forest life zone and has elevation from 300 to 500 masl, MAP of 2500 mm, and MAT of 26 °C (Silver et al., 2004). The remote watershed was used only to provide general reference comparisons of overall trends with the urban watershed. Rates of wet N deposition near the remote forest site were 3.9 kg-N ha⁻¹ yr⁻¹ (NADP/NTN 2013).

All forest stands were secondary regeneration following widespread agricultural abandonment in Puerto Rico starting in the 1930s (Kennaway & Helmer, 2007). Both the Rio Piedras and Cubuy forests contained novel mixtures of native and non-native tree species (Helmer, 2004; Lugo, 2004; Kennaway & Helmer, 2007; Cusack & McCleery, 2014). Abundant exotic grass species in the urban watershed included Panicum maximum Jac q. and Pennisetum purpureum Schumach., which are both African grasses and can establish in open areas (Lugo et al., 2001). Soils at all sites were volcaniclastic Oxisols (Beinroth, 1982; Huffaker, 2002).

Within the urban watershed, the site at watershed synoptic 0.5 (Fig. 1, Table 1) is nearest to the urban core in the Bosque San Patricio, which is managed by the Puerto Rico Department of Natural Resources and the Environment (DNRE). This site has had forest regeneration since the 1930s, although part of the forest was used as US military housing until the 1960s (DNRE 2003). The Bosque San Patricio also contains an area of exotic grassland. A larger forest fragment in the mid-watershed (3.0, Table 1) is in the Jardín Botánico of the University of Puerto Rico and has an adjacent area of exotic grass cover. Forest plots in this area have had secondary regeneration since the 1930s (Lugo et al., 2005). Site 4.3 was previously a private coffee plantation that was allowed to reforest starting in the 1930–1940s, with an adjacent empty lot of exotic grass cover. Stands 5.1 and 6.0 are on unprotected government land with natural forest regeneration of unknown age near Las Curias Reservoir and surrounding a power station, respectively, with adjacent exotic grassland areas. Site 5.0 is on private land belonging to the Betancourt family, with a nearby empty lot covered in exotic grasses.

Landscape-scale urban factors were measured to characterize each site. Forest and grassland fragment sizes, distance to the urban center, and distance to the nearest major road were measured using Arc GIS 10 with a land-cover classification map for the area (Kennaway & Helmer, 2007) (Table 1). Forest fragments were defined as areas of continuous forest and were generally bounded by roads, grasslands, or urban cover. Four broad land-cover classes were modified from Kennaway & Helmer (2007) for the Rio Piedras watershed, and total area was calculated for each cover type, including high-density urban cover (2509 ha), low-density urban cover (2372 ha), grassland (782 ha), and forest (829 ha), leaving 185 ha as ‘other’ (waterways, wetlands, etc.).

Field collections

Soils were collected during the three dominant seasons from 2010 to 2011, including hurricane season (October 2010), the dry season (March 2011), and the early wet season (July 2011). For all measurements, soils were collected from paired grassland and forest sites on the same day. Soils were collected from 0 to 10 cm depth using a 2.5 cm diameter soil corer. This depth showed the largest changes in soil C cycling with N fertilization in remote forests in Puerto Rico (Cusack et al., 2011b). At each of the six paired exotic grassland/secondary forest sites, soils were collected from three 20-m transects under each of two cover types, for a total of 36 transects in the urban watershed. Along each transect, four soil cores were collected at each of four points (total of 16 cores per transect) and then pooled to give one representative sample for each of the 36 transects. Six forest stands were sampled similarly in the remote Cubuy forest. Bulk density was measured from 0 to 10 cm back from the undisturbed face of a 20-cm-deep pit in each site using a 6.5-cm-inner-diameter corer as described in Cusack (2013). Standing grass biomass was collected on each transect using a 323-cm² quadrant, dried at 60 °C until weight stabilized, and weighed. Grass heights at all sites ranged from 2 to 3 m.

Soil respiration and temperature were measured at all sites once during the wet season in July 2011 with a LiCor (Lincoln, NE, USA) 6400XT fitted with a soil CO₂ flux chamber. At each site, four 10-cm PVC collars were inserted directly into mineral soil along each transect leaving a 2-cm lip aboveground and allowed to equilibrate for 1 h prior to measurements. Paired forest and grassland plots were measured sequentially on the same day at approximately midday, avoiding heavy rain events.

Soil analyses

For soil nutrient analyses, fresh soils were extracted using 2 m potassium chloride (KCl) for ammonium (NH₄⁺), NO₃⁻, DOC, and total dissolved N (TDN) on the day of collection in Puerto Rico to minimize storage effects on mineral N pools (Turner & Romero, 2009) as described in Cusack (2013). Frozen extracts were shipped for analysis to the University of California – Los Angeles. Extractable NH₄⁺ and NO₃⁻ were analyzed using standard colorimetric techniques in 96-well plates (Rhine et al., 1998; Doane & Horwath, 2003). Analyses included standard curves, check standards, and four analytical replicates per sample. Mineral N levels were measured on a BioTek Instruments Inc. (Winooski, VT, USA), Synergy HT absorbance spectrometer. Extracted DOC and TDN were measured in duplicate together with standard curves and check standards on a Shimadzu TOC-L CSH, with a TNM-L unit for TN detection. Separate soil subsamples were air-dried and ground using a mortar and pestle for total C and N analyses, and grass biomass was ground on a Wiley Mill. Dry soils and grass biomass were analyzed for C and N content on a Costech Elemental Analyzer (Valencia, CA) using atropine and acetanilide as standards.

All microbial enzyme activity and biomass data reported here are for fresh pooled soil samples shipped overnight to the University of California – Los Angeles, and analyzed within 48 h. Standard methods for soil enzyme assays followed Sinsabaugh et al. (2003) with changes as described in Cusack (2013). Hydrolytic enzymes were measured using fluorometric assays on soils incubated at 27 °C for 0.5–8 h, with incubation times based on initial tests. Carbon acquisition was measured as cellulose-degrading β-glucosidase activity, N acquisition was measured as chitin-degrading N-acetylglucosaminidase (NAGase) activity, and P acquisition was measured as acid phosphatase activity. These enzymes were previously observed to have the highest activities for acquisition of C, N, and P in this watershed (Cusack et al., 2011a). Plates with 96-well plates were read on a BioTek Instruments Inc., Synergy HT absorbance/fluorescence spectrometer at 365 nm excitation/450 nm emission for fluorescence. Microbial biomass C and N were measured using a modified chloroform fumigation-slurry extraction as described in Fierer & Schimel (2003), similarly assuming an extraction efficiency of 0.45 (Beck et al., 1997; Dictor et al., 1998). Extracted microbial biomass C and N were compared to non-chloroform extractions on a Shimadzu TOC-L CSH, with a TNM-L.

Total rock-derived soil nutrients were measured from 0 to 10 cm to compare site characteristics for grasslands vs. forests and to include as predictors in statistical analyses. Air dried soils were prepared as above and then digested with concentrated HNO₃, 30% H₂O_2, and concentrated HCl at the US Forest Service International Institute of Tropical Forestry laboratory in Puerto Rico, in a modification of a standard digestion method (Luh Huang & Schulte, 1985). Total acid extraction values of aluminum (Al), calcium (Ca), iron (Fe), magnesium (Mg), manganese (Mn), potassium (K), sodium (Na), phosphorus (P), and sulfur (S) were determined using a Spectro Plasma Emission Spectrometer, Spectro, Kleve, Germany (model Spectro Ciros CCD – ICP). The certified reference material Montana Soil (NIST – 2711) was analyzed to verify the completeness of elemental recovery. Subsamples dried at 105 °C were used for moisture correction. These measures were conducted on soils collected during the wet season only, because total rock-derived nutrient content of soils is relatively stable over annual timescales.

Statistical analysis

Soil C and N pool sizes were analyzed for significant differences between exotic grassland and secondary forest sites over three seasons in the urban watershed. Repeated-measures multivariate analysis of variance (manova) was used to assess differences between cover types across the three time points, using post hoc tests to compare parameters for specific seasons or sites. We used preliminary backward stepwise linear regression analyses including all factors to identify potential predictors of soil C and N measures. We then included only potentially significant factors identified by these stepwise regressions in final linear models. Microbial biomass C and N, NO₃⁻, NH₄⁺, total mineral N, TDN, enzyme activities, soil moisture, pH, bulk density, and total rock-derived nutrients were included as covariates in initial stepwise tests. Final tests included manova for soil measures across the three seasons and ancova for wet season CO₂. We repeated these analyses to identify predictors of soil C (C concentrations, C stocks, DOC, CO₂), soil N (NO₃⁻, NH₄⁺, total mineral N), and enzyme activities. We assessed spatial variability in soil C and N (H1) by including distance to the urban core, distance to a major road, watershed synoptic, and elevation as predictors. We also calculated paired plot differences in soil C and N measures to assess the relationship between grassland and forest spatial variability. To assess effects of cover type and soil N on soil C (H2), we used the manova above to identify significant predictors of soil C and then ran post hoc Student's t-tests to compare between cover types. To explore seasonal patterns in exotic grasslands vs. secondary forests (H3), we used the manova above to identify when time was a significant factor, and then ran post hoc means separation Tukey HSD tests to compare biogeochemistry by season. To further assess the magnitude of seasonal changes in exotic grasslands vs. secondary forests, we calculated the percent change in soil properties for each individual plot from the dry season to the subsequent wet season (Fig. 1) and compared the averages of percent change. We used separate tests to compare the remote watershed forests with urban watershed secondary forests and grasslands. For site-level comparisons, we used n = 6 per vegetation type, and for initial stepwise regression analyses, we used the total observations for each parameter n = 36. Where necessary, values were log-transformed to meet assumptions of normality. These general linear analyses were used to test the three hypotheses individually and to explore specific relationships and trends.

We also analyzed all of the above relationships simultaneously using structural equation modeling (SEM) path analysis to assess the relative importance of the main driving factors from each of the individual hypotheses (i.e., distance to the urban core, cover type, and seasonality). We used distance to the urban core, cover type, and seasonality as our exogenous factors, and soil NO₃⁻, moisture, microbial enzyme activity, and soil C concentrations as endogenous factors. Because of high autocorrelation among enzyme activities, we used only β-glucosidase activity in the model, which was also the most significant enzyme predictor in the general linear models. We used SEM modification indices to add or remove specific pathways from the model. An initial model included microbial biomass C as an endogenous factor, but this was removed because of poor model similarity to the saturated model. Season and cover type were rank-transformed and converted to binary data, respectively, for SEM. The final default model (i.e., our model) was tested for overall similarity to the saturated model (i.e., ideal model) using common metrics of comparison (i.e., no significant difference from saturated model, chi-square P > 0.05, GFI > 0.9, NFI > 0.9, RMSEA < 0.05). The model used 123 observations.

General linear models were performed using 11.0.0 jmp pro software (SAS Institute Inc., Cary, NC, USA), and SEM path analysis was performed using spss amos 23 software (IBM, Armonk, NY, USA). Statistical significance for all relationships and models was P < 0.05 unless otherwise noted, and means are reported ± one standard error.

Spatial trends in soil N and C in an urban watershed

Spatial trends for soil NO₃⁻ and C relative to urban features were strongest for exotic grasslands, with greater variability in secondary forests. Soil NO₃⁻ in both cover types was generally higher nearer to the urban core than in more suburban and rural sites, supporting H1 (Fig. 2). However, only grasslands had a clear and significant decline in soil NO₃⁻ with distance to the urban center (R² = 0.33, P < 0.05) and with increased subwatershed remoteness (i.e., watershed synoptic, R² = 0.37, P < 0.05). Soil NO₃⁻ in forests was more variable in the lower watershed and diminished more gradually in the upper watershed. Still, the most rural forests in the urban watershed had NO₃⁻ concentrations down near background levels found in the remote secondary forest (Fig. 2). Differences in soil NO₃⁻ for forest–grassland paired sites were positively related to distance to a major road (R² = 0.35, positive correlation, P < 0.05), showing greater accumulation of NO₃⁻ in forests vs. grasslands in more remote areas. Also, soil NO₃⁻ levels were not significantly correlated between paired forest–grassland sites, suggesting that distinct factors governed spatial variability in soil NO₃⁻ for the two cover types.

Spatial trends in soil C concentrations were positively related to patterns for soil NO₃⁻. Grassland soil C was negatively associated with distance to the urban center (R² = 0.28, P < 0.05) and with increased subwatershed remoteness (i.e., watershed synoptic, R² = 0.26, P < 0.05). Soil C concentrations in forests were more variable than in grasslands, but tended to be lowest in the upper watershed. Similar to NO₃⁻, there was not a significant correlation for soil C between paired forest–grassland sites.

Soil NO₃⁻ and C storage in exotic grasslands vs. secondary forests

Exotic grasslands had consistently lower soil C storage than urban secondary forests, which was linked to elevated soil respiration and higher decomposer enzyme activities, providing partial support for H2. On average, soil C concentrations were significantly lower in exotic grasslands than in secondary forests (Fig. 3a). There were also significantly higher CO₂ losses via soil respiration for grasslands vs. forests during the wet season measurement (Fig. 3b), and no difference in soil temperature (25.8 ± 0.4 °C in grasslands and 26.5 ± 0.2 °C in forests). Significant factors in the analysis of soil respiration included cover type, soil moisture, and total hydrolytic soil enzyme activities (P < 0.05). Interestingly, despite the higher soil respiration rates in grasslands, soil microbial biomass C and N during the wet season were lower in grasslands than in forests (Fig. 3c, Table 2), and microbial biomass C : N was significantly lower in grasslands (7.2 ± 0.5) versus forests (8.8 ± 0.6). C : N is a ratio and thus singular. Nonetheless, grasslands had higher β-glucosidase and NAGase enzyme activities relative to secondary forests (Fig. 3d, Table 2). Thus, β-glucosidase activity when normalized per unit of microbial biomass C was approximately twice as high in grasslands (4.1 ± 0.1 activity-nmol/μg-microbial C/hr) vs. secondary forests (2.0 ± 0.3, P < 0.05 activity-nmol/μg-microbial C/hr) for this time point, suggesting a fundamental difference in microbial function. Similarly, β-glucosidase, NAGase, and phosphatase enzyme activities when normalized to soil C content were much higher in exotic grasslands vs. secondary forests (Table 2). Overall, the higher microbial activity in grasslands vs. secondary forests was strongly associated with lower soil C stocks in grasslands.

Contrary to our prediction, soil C and NO₃⁻ were significantly positively correlated overall (R² = 0.53, P < 0.05), with a somewhat weaker relationship in forests than in grasslands (R² = 0.52 vs. 0.63, respectively, P < 0.05). Also contrary to our prediction, soil NO₃⁻ was strongly negatively related to soil enzyme activities. Soil NO₃⁻ was the only significant biogeochemical factor in the analyses of normalized soil enzyme activities, with negative relationships between normalized β-glucosidase, NAGase, and phosphatase activities (R² = 0.13, 0.22, 0.09, respectively, P < 0.05), as well as with total hydrolytic enzyme activities (Fig. 4b, R² = 0.26, P < 0.05). Finally, and also contrary to our prediction, forests had higher soil NO₃⁻ concentrations than exotic grasslands, and forest soils were relatively more enriched in NO₃⁻ per unit of TDN compared to grasslands (Fig. 4a). Other soil nutrients did not show strong differences between cover types except for total K, which was significantly higher in grasslands than in urban forests (P < 0.05) (Table 3, Table S1). These data indicate greater accumulation of NO₃⁻ in secondary forests vs. paired grasslands across the urban–rural gradient and negative effect of NO₃⁻ enrichment on decomposer enzyme activities. Soil NO₃⁻ levels were positively linked to soil C levels via suppression of decomposer enzyme activities.

Seasonal variation in soil properties for forests and exotic grasslands

Exotic grasslands had somewhat less seasonality in soil moisture but more seasonality in soil NO₃⁻ compared to paired secondary forests, partially supporting H3. Assessing the percent change in soil moisture from the dry season to the subsequent wet season, forests had a 67 ± 7% increase in soil moisture, vs. a 42 ± 7% increase in grassland soil moisture (Fig. 1, P < 0.05). These seasonal shifts in soil moisture were significant for both cover types (i.e., percent change significantly different from 0), with generally drier soils in the dry vs. wet season, but overall wetter conditions in grasslands (Fig. 5a).

In contrast, exotic grasslands showed greater and much more consistent seasonality in soil NO₃⁻ than did secondary forests. Assessing the percent change in soil NO₃⁻ from the dry season to the wet season, forests had an average of −1 ± 20% shift in soil NO₃⁻ (n.s.) vs. grasslands which had a −54 ± 9% decline in soil NO₃⁻ (Fig. 1, P < 0.05). The lack of significant seasonality in forest soil NO₃⁻ resulted from high spatial variability in seasonal trends among forest sites (Fig. 1). Also, secondary forests had no seasonal shift in soil NO₃⁻ from the hurricane season to the dry season, whereas soil NO₃⁻ in grasslands nearly doubled during the dry season relative to both wet seasons (Fig. 5b). Overall, soil moisture was negatively but relatively weakly correlated with soil NO₃⁻ levels in exotic grasslands (R² = 0.27, P < 0.05), and there was no significant correlation in secondary forests. These data show a difference in the strength of seasonality for soil moisture and NO₃⁻ with highly variable temporal trends in forest soil NO₃⁻, and only a weak link between moisture and NO₃⁻.

Seasonal trends in decomposer enzyme activities generally tracked soil moisture. β-glucosidase and NAGase enzyme activities were significantly lower in the dry vs. wet season in both cover types, and seasonal shifts were greater in forests than in grasslands, similar to patterns for soil moisture. Assessing the percent change in β-glucosidase and NAGase activities from the dry season to the subsequent wet season, forests had averages of 48 ± 18% and 77 ± 14% increases, respectively, vs. 26 ± 21% and 40 ± 12% increases in grasslands (Table 2, P < 0.05). In contrast, phosphatase activities were relatively high during the hurricane and dry seasons in both cover types and declined 35 ± 3% in both during the final wet season (Table 3), following patterns for forest soil NO₃⁻. Thus, seasonality in soil moisture was most closely linked to seasonality in C and N acquisition activities, whereas temporal patterns in soil NO₃⁻ were positively linked to seasonality of P acquisition activity.

Notably, there was no seasonal variation in DOC, in stark contrast to trends for soil moisture, NO₃⁻, and enzyme activities (Fig. 5c). Rather, extractable DOC closely reflected total soil C stocks, with higher levels in secondary forests vs. exotic grasslands. Soil C concentrations and content did not vary seasonally, nor did soil N concentrations or TDN (Table 3). Thus, total soil organic matter and dissolved organic matter were stable through time in each cover type, suggesting a temporal decoupling in the processes driving soil C vs. NO₃⁻ levels.

Other factors showed some seasonal variation. Seasonal trends for microbial biomass C and N were opposite of enzyme activities, with the highest overall biomass during the dry season and an interaction with cover type (P < 0.1), such that microbial biomass was lower in grasslands than in forests only during the hurricane and dry seasons (Table 2). Interestingly, soil bulk density also changed seasonally, with higher bulk density in the dry vs. wet season (Table 3), and a negative correlation with soil moisture (R² = 0.41, P < 0.05). Aboveground grass biomass was marginally lower during the dry season, with 2.2 ± 0.4 kg m⁻² in the dry season and 3.6 ± 0.7 kg m⁻² in the wet seasons (P = 0.07). Foliar C and N concentrations in grass biomass did not vary seasonally and averaged 42.5 ± 0.3% C and 1.2 ± 0.06% N across sites.

Urban soil properties vs. remote forests

Comparing the urban watershed sites to the remote Cubuy secondary forests showed general differences in soil properties. The urban forests had lower soil moisture, NH₄⁺, DOC, TDN, and soil C and N concentrations than the remote forests, and higher soil NO₃⁻, bulk density, and pH (Table 3). Notably, remote forests had much lower enrichment in NO₃⁻ relative to TDN than urban forests (Fig. 4a). Soil respiration rates in the remote forest during the wet season were similar to overall rates in the urban sites (Fig. 3b), whereas soil temperatures were lower (23.3 ± 0.1 °C). The remote forest also had higher microbial biomass C and N, and higher overall soil enzyme activities than the urban sites, although normalized enzyme activities were similar (Table 2). Interestingly, the C : N and C : P enzyme ratios were lower in the remote forest than in urban sites, while N : P enzyme ratios were similar (Table 2), suggesting that N and P were more scarce relative to C for decomposers in the remote vs. urban sites. Remote forest parameters were generally more similar to urban secondary forests than to exotic grasslands, but the largest differences were at the watershed scale. These data provide further evidence for general NO₃⁻ enrichment in the urban watershed relative to more remote sites.

Relative importance of urbanization, cover type, and seasonality

Testing the three hypotheses simultaneously in SEM path analysis revealed that seasonality was a fairly weak predictor of soil biogeochemical properties relative to cover type and distance to the urban center (Fig. 6). Assessing path coefficients between pairs of factors, distance to the urban center was significantly correlated with all of the endogenous factors in this analysis, including a negative relationship with NO₃⁻, further supporting H1 (Table 4). Cover type was also significantly related to NO₃⁻ and soil moisture, while the direct path from cover type to soil C was not significant. Rather, soil NO₃⁻ was the strongest direct predictor of soil C. Thus, the effect of cover type on soil C appeared to be mediated by changes in soil NO₃⁻, partially supporting H2. Interestingly, season was only a marginally significant predictor of soil moisture (P = 0.1) and NO₃⁻ (P = 0.06) in these analyses (Fig. 6, Table 4), suggesting that although seasonality was significant when analyzed alone (above), the effects of human disturbance (i.e., cover type and distance to the urban center) were much stronger drivers of urban soil C and N levels. Overall, the final default model was not significantly different from the saturated model (df = 4; χ² = 2.883, P = 0.578; GFI = 0.994; NFI = 0.989; RMSEA < 0.001), indicating that our model was close to an ideal fit for the data. The full model predicted 66% of the variability in soil C concentrations and 45% of the variability in soil NO₃⁻ (Fig. 6).

Spatial trends in urban soil N enrichment

Atmospheric N deposition is the most likely source contributing to the relatively large soil NO₃⁻ pools in these urban-proximate ecosystems, similar to other urban sites (Fenn et al., 2003; Martinelli et al., 2006). The strong decline in grassland soil NO₃⁻ with distance to the urban center in particular suggests an N deposition gradient. Previous work in the Puerto Rican urban forests showed that they were enriched in NO₃⁻ and depleted in NH₄⁺ compared to two remote primary forests (Cusack, 2013), like the comparison here with secondary remote forests, indicating external inputs of NO₃⁻ to urban ecosystems. Grassland soil NO₃⁻ levels were significantly lower than in secondary forests, contrary to our predictions and opposite of a temperate urban study that showed higher soil NO₃⁻ in grasslands vs. forests (Groffman et al., 2009). Here, canopy scrubbing of atmospheric pollution by forest canopies [i.e., higher effective N deposition (Clark et al., 1998; Schulze, 1989)] could contribute to the higher soil NO₃⁻ that we observed in secondary forests vs. grasslands. Forest fragment size was previously shown to be a strong predictor of soil NO₃⁻ levels among these urban forests (Cusack, 2013), suggesting that larger canopies ‘scrub’ more N from atmospheric pollution. This mechanism could explain why soil NO₃⁻ levels were higher but also more variable among forest sites, whereas exotic grasslands showed more predictable declines in soil NO₃⁻ with distance to the urban center.

Other factors may also have contributed to the variability in soil mineral N levels among sites and between cover types. For example, there are exotic N-fixing tree species distributed unevenly among the forest fragments, which have been associated with elevated soil N levels (Cusack & McCleery, 2014), and could help explain some of the variability in soil NO₃⁻ among forest stands. Comparing the two cover types, higher soil moisture in grasslands vs. forests could also have led to greater leaching losses or denitrification of NO₃⁻ (Corre et al., 2014), reducing grassland soil NO₃⁻ pools. However, soil moisture and NO₃⁻ were only weakly correlated here. Similarly, a lack of relationship between soil moisture and redox-sensitive nutrient concentrations was observed in a remote forest moisture manipulation in Puerto Rico (Hall et al., 2013). Together, these data suggest that drivers of temporal patterns in tropical forest soil NO₃⁻ are complex and are somewhat decoupled from soil moisture. It appears likely that the higher soil NO₃⁻ in secondary forests vs. exotic grasslands observed here resulted from both the aboveground structural properties of forests, as well as soil processes.

Effects of exotic grasslands and N deposition on soil C

Our results show that the spread of non-native grasses in tropical forest regions is likely to drastically reduce soil C storage. Here, differences in normalized enzyme activities and microbial biomass C : N point to a compositional and/or metabolic shift in the decomposer community between the two cover types, potentially driving lower soil C in grasslands. The higher respiration rates per unit of microbial biomass in the grasslands in particular indicate reduced microbial carbon use efficiency (CUE). This result for grasslands is somewhat contradictory to a recent review which suggested that microbial CUE increases with lower C : N ratios of litter and with higher soil fertility, due to a shift in resource stoichiometry to more closely match requirements of microbial biomass (Manzoni et al., 2012). In these urban Puerto Rican sites, grass tissues had relatively low C : N ratios of ~35, which according to the above review should lead to high microbial CUE. While not measured directly, litter C : N ratios in the urban forests were most certainly higher than in the grasslands, and likely similar to litter C : N ratios of ~41 in lower elevation remote Puerto Rican forests (Cusack et al., 2009). In terms of soil fertility, grasslands had higher soil K, but lower soil N availability and lower apparent N deposition levels than did secondary forests. Rather than a simple shift in CUE due to changes in resource availability, the apparent reduction in microbial CUE in grasslands vs. secondary forests could reflect a broader shift in microbial community composition toward more fast-growing, low-efficiency microbes (Blazewicz et al., 2014), which may result from the greater availability of more labile C compounds produced by grasses vs. woody species (Filley et al., 2008; Liao & Boutton, 2008). Similar to our results, a temperate study found lower soil C and microbial biomass C levels in urban grasslands vs. forests (Groffman et al., 2009). Our data strongly suggest that a change in the microbial processing of soil C is responsible, at least in part, for smaller soil C stocks in exotic tropical grasslands vs. urban secondary forests.

In contrast to the effects of exotic grasses, it appears that NO₃⁻ enrichment in these urban ecosystems suppressed decomposer enzyme activities and may have had a positive effect on soil C storage, potentially contributing to the higher soil C levels in forests vs. grasslands. Effects of N fertilization on soil microbial activity in remote mature tropical forests have been mixed, with some studies observing increased soil respiration (Cleveland & Townsend, 2006), and others observing declines in soil respiration and increased soil C stocks with elevated N (Mo et al., 2008; Cusack et al., 2011b). In theory, increased N availability should promote the production of microbial enzymes that acquire C and P from organic matter (β-glucosidase and phosphatase, respectively), and suppress enzymes that acquire N (NAGase) (Sinsabaugh & Moorhead, 1994). However, we saw an overall decline in total hydrolytic enzyme activities with elevated soil NO₃⁻, suggesting that N enrichment in urban-proximate ecosystems may suppress decomposition. Thus, there appear to be interacting effects of cover type and NO₃⁻ accumulation on soil C cycling.

Seasonal variation in urban soil biogeochemistry

While canopy capture of N deposition may help explain forest vs. grassland spatial patterns in soil NO₃⁻, temporal patterns are likely related to seasonal shifts in microbial activity. Both cover types had generally high soil NO₃⁻ during the dry season. Nitrate accumulation during dry periods is common in tropical seasonal sites and has been attributed to declines in microbial denitrification and reduced plant uptake (Yavitt et al., 1993; Miambo et al., 2007). While the overall dry-to-wet season pattern for NO₃⁻ was similar in both cover types, secondary forests also showed surprisingly high levels of soil NO₃⁻ during the hurricane season preceding the dry season. This unexpected seasonal trend could be related to episodic increases in soil nutrients in these tropical forests following large storms during the hurricane season, which transfer nutrient-rich green biomass from trees to soils (Scatena et al., 1993; Silver et al., 1996; Heartsill-Scalley et al., 2010). In general, temporal and spatial trends in soil NO₃⁻ for secondary forests were complex, whereas grassland soil NO₃⁻ and enzyme activities more clearly followed predicted seasonal effects.

The lack of a relationship between seasonality in soil NO₃⁻ and DOC levels was surprising. Instead, DOC levels closely reflected total soil C stocks. Global trends have shown a positive relationship between NO₃⁻ accumulation and DOC levels, with these two solutes linked via decomposition rates and microbial biomass requirements for C and N (Taylor & Townsend, 2010). If NO₃⁻ accumulation depended primarily on mineralization of OM followed by nitrification in these soils, then we would have expected a strong positive relationship of NO₃⁻with DOC (DOC production from incomplete decomposition of OM), or a strong negative relationship of NO₃⁻ with DOC (evidence for complete decomposition of C during mineralization) (Kalbitz et al., 2000). For example, N fertilization in remote Puerto Rican forests showed that decomposer activity was positively related to DOC production (Cusack et al., 2010, 2011b). Here, the seasonality of soil enzyme activities did indicate a shift in C cycling, but DOC levels did not reflect this. These results may suggest that salt-extractable DOC is not as microbially responsive a pool of C as expected, and/or that DOC production in these systems is decoupled from NO₃⁻ levels. A decoupling between soil NO₃⁻ and DOC seems particularly likely if NO₃⁻ levels most strongly reflect N deposition rather than mineralization and nitrification processes.

As indicated by the SEM path analysis, seasonality in general did not emerge as a dominant driver of soil biogeochemistry at these sites, with spatial effects of cover type and the urban core providing clearer controls on soil dynamics. Overall, our results suggest that exotic grasses and atmospheric N deposition may have opposite or interacting effects on tropical soil C stocks. Net changes in soil C storage across urbanizing tropical landscapes are likely to vary according to the spread and establishment of exotic grasslands, local N deposition rates, and interacting urban effects on decomposer communities.