1174 SSSAJ: Volume 72: Number 4 • July–August 2008

WET

LAN

D S

OIL

S

Soil Sci. Soc. Am. J. 72:1174-1183doi:10.2136/sssaj2007.0354Received 27 Sept. 2007. *Corresponding author (sabgru@ufl .edu).© Soil Science Society of America677 S. Segoe Rd. Madison WI 53711 USAAll rights reserved. No part of this periodical may be reproduced or transmitted in any form or by any means, electronic or mechanical, including photocopying, recording, or any information storage and retrieval system, without permission in writing from the publisher. Permission for printing and for reprinting the material contained herein has been obtained by the publisher.

Wetlands and aquatic systems are complex in their biogeo-chemical structure and composition. This determines

their resilience and sensitivity to respond to internal and exter-nal stresses. Anthropogenic forces, such as nutrient infl ux, and natural events, such as hurricanes and fi re, may impact wet-lands at local or regional spatial scales. Biogeochemical cycling (e.g., N, P, C, and S cycles) and transport processes continu-ously change pedological and biological patterns of aquatic ecosystems. Noe et al. (2001) presented in a meta-analysis of how spatial patterns of soil P are interlinked to periphy-ton, vegetation, hydrologic manipulation, nutrient infl ux, and internal biogeochemical cycling in the Greater Everglades (?8500 km2), one unique and large subtropical wetland in Florida. Barbiéro et al. (2002) illustrated the complexity of the geochemistry of surface and groundwater in the Pantanal of Mato Grosso, Brazil. The health and functions of the Pantanal in the heart of South America (?165,000 km2), which is

the world largest freshwater wetland, has been threatened by anthropogenic impacts.

A large number of wetland studies have focused on the characterization of biogeochemical soil properties along expected gradients (transects) of impact (Qualls and Richardson, 1995; Reddy et al., 1998; Richardson and Qian, 1999; White and Reddy, 2000; Fisher and Reddy, 2001; Craft and Chiang, 2002). Other studies characterized spatial patterns of soil and water properties across wetlands at various spatial scales in dif-ferent geographic regions (Table 1). Grunwald et al. (2007b) pointed out that biogeochemical cycling within wetlands and aquatic systems can only be understood within a spatially explicit context considering the spatial autocorrelation and co-variation among different environmental properties. The spa-tial autocorrelation (or range) describes the relatedness among measurements, indicating that places close to one another tend to have similar values, whereas ones that are farther apart differ more (Webster and Oliver, 2001). In a subtropical freshwa-ter marsh in the Everglades it was demonstrated how sparse, site-specifi c observations were too simplistic to characterize the complex spatial patterns of soil P (Grunwald et al., 2007b) that were generated by various ecosystem processes (P infl ux, trans-formations, resuspension, transport, and chemical precipita-tion). Complex interactions between P and other constituents (organic matter, metals) and environmental factors (vegetation, hydrology) and external forces (e.g., fi re) continuously caused the co-evolution of soil patterns across this subtropical marsh (Grunwald et al., 2007b).

Questions remain that address the population size and density of observations that are required to characterize the

S. Grunwald* K. R. ReddySoil and Water Science Dep.Univ. of Florida2169 McCarty HallP.O. Box 110290Gainesville, FL 32611

Spatial Behavior of Phosphorus and Nitrogen in a Subtropical Wetland

Complex spatial patterns of biogeochemical properties such as soils are generated in wet-lands as a result of internal cycling of nutrients, natural forcing functions (e.g., hydrologic fl ow paths), and anthropogenic impacts (e.g., nutrient inputs). To preserve ecosystem health and the functions of a wetland ecosystem, it is critical to maintain a soil nutrient status that resembles natural reference conditions and to preserve biogeochemical variability. We investigated the spatial behavior of total P (TP) and total N (TN) in a subtropical wetland in Florida that historically has been impacted by nutrient infl ux. Our specifi c objectives were to: (i) assess the spatial autocorrelation structure and spatial variability of TP and TN across a subtropical wetland given a dense observation set (n = 266); and (ii) assess the information loss associated with underlying spatial autocorrelations and spatial variability in TP and TN under various scenarios of reduced sampling densities (n = 175–50). Both properties showed contrasting spatial metrics with long spatial autocorrelations for TP (7240 m) and much shorter ones for TN (1007 m). The contrasting spatial metrics of TP and TN along trajec-tories of sparser observation sets translated into different responses in spatial patterns that disintegrated dramatically for TN and less so for TP when compared with the reference set (n = 266). In this subtropical wetland, a sampling set of n > 75 (0.036 samples ha−1) for TP and n > 125 (0.0260 samples ha−1) for TN yielded satisfactory results to resemble spatial patterns of the reference set. This study provided insight into the spatial behavior and patterns of two important soil nutrients in wetlands.

Abbreviations: BCMCA, Blue Cypress Marsh Conservation Area; MPE, mean prediction error; TN, total nitrogen; TP, total phosphorus.

SSSAJ: Volume 72: Number 4 • July –August 2008 1175

Table 1. Overview studies that characterized the spatial distribution and variability of soil and water properties in wetlands.

Study area† Size Type No. of sites (n)Upscaling method‡

Properties§ Reference

ha

Des Plaines River Wetlands Demonstration Project, IL

180constructed freshwater riparian marshes

15–16 (quadrants) in 4 separate wetlands in 1991¶

K SRP Mitsch et al. (1995)

WCA 1, Everglades, FL 59,000 wet prairie, sawgrass marsh 90 in 1991 K NH4–N, SRP, TP, TNNewman et al. (1997)

Coastal plain, GA 0.7 restored riparian wetland24 at 4 different times (11 and 2 in 1993 & 4 and 8 in 1994)

OK, P, T, UK BNP, iN, M, MR, SOM, TC, TN

Ettema et al. (1998)

Peckham Farm Research Area, Univ. of Rhode Island, Kingston

0.036 riparian forest soils300: 100 within 3 strata (drainage classes) at two times (5 and 11 in 1995)

KSM, Ox-Al, Ox-Fe, pH, SOM

Lyons et al. (1998)

Experimental farm of Kyoto University, Japan

0.5 paddy fi eld 91 sites in 1997, 50 in 1998 BKaP, C/N, EC, K, TMg, Na, pH, TC, TCa, TN

Yanai et al. (2000)

WCA 2A, Everglades, FL 44,800 sawgrass and cattail marsh 62 in 1990 and 1998 BK; IK TP DeBusk et al. (2001)

Floodplain of the Des Plaines River, IL

1.9–3.4constructed wetlands: 4 riparian wetlands separated by boundaries

112 sites in 1988, 85 in 1990 BKBP, K, OC, TCa, TMg, pH

Fennessy and Mitsch (2001)

Parts of Loura River Watershed, northwest Spain

1 fl oodplain 541 in 2000 OK, TK, TAl, TCa, TFe, TMg, RP, Si, TC Gallardo (2003)

Hula Valley and Jordan River runs, Israel

17,500 semiarid wetland 80 in 2000 CSGSDPS, Ox-Al, Ox-Ca, Ox-Al, Ox-P

Litaor et al. (2003)

WCA 2A, Everglades, FL 41,770 74 in 1990, 56 in 1998 CSGS, OK TP Grunwald et al. (2004)

ORWRP, OH0.81 and 0.88 (2 units)

marsh43 in 1993 and 1995, 127 in 2004

OKBP, Ca, K, Mg, pH, TC, TP

Anderson et al. (2005)

Coastal plain, NC 3.1–200created, restored, and natural wetlands

4 wetlands, 32 per site IDW s, SOMBruland and Richardson (2005)

WCA 3, Everglades, FL209,100(3 units) freshwater marsh 388

OK BD, iP, TAl, TC, TCa, TFe, TMg, TP

Bruland et al. (2006a)

Coastal plain, NC 3.1–200created, restored, and natural wetlands

4 wetlands 32 per site IDWDEA, NH4–N, NO3–N, OC

Bruland et al. (2006b)

WCA 1, Everglades, FL 55,900 freshwater marsh 131 OKBD, LOI, TAl, TCa, TFe, TiP, TMg, TP

Corstanje et al. (2006)

BCMCA, FL 4,800sawgrass, cattail, maidencane, coastal plain willow

266 OK, SMC biP, boP, MBP, TP Grunwald et al. (2006)

BCMCA, FL 4,800sawgrass, cattail, maidencane, coastal plain willow

266 PCK

APA, BD, BG; biP, boP, C/N, M, MBP, NH4–N, PEP, TC, TloC, TloN, TlP, TN, TP

Grunwald et al. (2007a)

WCA 2A, Everglades, FL 43,281 sawgrass and cattail marsh 111 OKBD, LOI, TAl, TC, TCa, TFe, TiP, TMg, TN, TP

Rivero et al. (2007b)

WCA-2A, Everglades, FL 43,281 sawgrass and cattail marsh 111 OK, CoK, RK ASTER, ETM, TP Rivero et al. (2007a)

Greater Everglades 825,000 sawgrass and cattail marsh1341 in 13 hydrologic units

OK, IDW, PCK

BD, LOI, TAl, TC, TCa, TFe, TiP, TMg, TN, TP

Grunwald et al. (2008)

† WCA, Water Conservation Area; ORWRP, Olentangy River Wetland Research Park; BCMCA, Blue Cypress Marsh Conservation Area.

‡ BK, block kriging; CoK, co-kriging; CSGS, conditional sequential Gaussian simulation; IK, indicator kriging; IDW, inverse distance weighting; K, kriging—type of kriging unknown; KS, kriging with strata; OK, ordinary kriging; P, pooling of semivariograms over sampling dates; PCK, principal component kriging; RK, regression kriging; SMC, spatial moving correlations; T, trend analysis; TS, trend surface analysis; UK, universal kriging.

§ aP, available P; APA, alkaline phosphatase activity; ASTER, Advanced Spaceborne Thermal Emission and Refl ection Radiometer satellite image; BD, bulk density; BG, beta-glucosidase activity; biP, bicarbonate-extractable inorganic P; BNP, bacterivorous nemadode populations; boP, bicarbonate-extract-able organic P; BP, Bray P1 extractable P; DEA, denitrifi cation enzyme activity; DPS, degree of P saturation; EC, electrical conductivity; EMT, Landsat Enhanced Thematic Mapper satellite image; iN, inorganic N; iP, inorganic P; LOI, loss on ignition; M, moisture; MBP, microbial biomass P; MR, micro-bial respiration; OC, organic C; Ox-Al, oxalate-extractable Al; Ox-Ca, oxalate-extractable Ca; Ox-Fe, oxalate-extractable Fe; Ox-P, oxalate-extractable P; PEP, peptidase activity; RP, redox potential; s, sand content; SOM, soil organic matter; SRP, soluble reactive P; TAl, total Al; TC, total C; TCa, total Ca; TFe, total Fe; TiP, total inorganic P; TloC, total labile organic C; TloN, total labile organic N; TlP, total labile P; TMg, total Mg; TN, total N; TP, total P.

¶ Average values were used during the growing season (May–September) 1991 at fi ve time periods; size of individual wetland was not provided.

1176 SSSAJ: Volume 72: Number 4 • July–August 2008

spatial variability and distribution patterns of biogeochemical properties within wetlands. Reddy and DeLaune (2008) argued that spatially explicit modeling of biogeochemical properties in wetlands at an appropriate spatial scale is critical for our under-standing of ecosystem processes. They suggested using biogeo-chemical indicators (Levels I, II, and III) that behave differently at micro, meso, and macro spatial scales to characterize wetland health. Knowledge of continuous spatial distributions and scal-ing behavior of biogeochemical soil properties in wetlands and aquatic systems, however, is still limited. Investigations that aimed at describing the spatial variability of biogeochemical soil properties have ranged from paddy fi elds (0.5 ha) to large wetlands (825,000 ha) (Table 1). High-intensity studies used sampling densities of 182.0 samples ha−1 (n = 91, Yanai et al., 2000), 8333.3 samples ha−1 (n = 300, Lyons et al., 1998), and 541 samples ha−1 (n = 541, Gallardo, 2003), while other stud-ies have relied on fewer observations to characterize spatial pat-terns (Mitsch et al., 1995; DeBusk et al., 2001; Bruland and Richardson, 2005). Regional investigations of biogeochemi-cal patterns in soils have used coarse sampling densities, with 0.0025 samples ha−1 (n = 111, Rivero et al., 2007a,b), 0.0019 samples ha−1 (n = 388, Bruland et al., 2006a), and 0.0016 samples ha−1 (n = 1341, Grunwald et al., 2008). Often costs and labor are prohibitive to use high sampling densities for sampling across larger wetlands. Coarser spatial studies often aim to elucidate on regional spatial patterns but fail to reveal fi ne-scale spatial patterns (compare Table 1).

Grunwald et al. (2007a) suggested distinguishing among three major groups of different biogeochemical indicator vari-ables in wetlands: (i) labile, fast-response properties with fi ne-scale spatial autocorrelation (e.g., total labile organic N and total labile organic C); (ii) stable, slow-response properties with regional spatial autocorrelation (e.g., total P); and (iii) proper-ties showing intermediate response (e.g., total C). Interestingly, the documented spatial autocorrelation range of biogeochemi-cal properties in wetlands and aquatic systems differed widely among studies. The range was relatively short, with values as low as 2 m for pH, 37 m for available P, and 49 m for K in paddy fi elds (Yanai et al., 2000), suggesting that observations should be taken at 2 m or fi ner intervals to capture the under-lying spatial biogeochemical patterns. Fennessy and Mitsch (2001) identifi ed a range of 300 m for Bray-extractable P in a fl oodplain in Illinois. In contrast, DeBusk et al. (2001) identi-fi ed long-range (regional) spatial autocorrelation structure for total P (TP) with 9.3 (1990) and 11.3 km (1998) in a sub-tropical wetland in Florida. The long range for TP in the top-soil was confi rmed in other studies at 6400 (1990) and 7550 m in Water Conservation Area (WCA)-2A (Grunwald et al., 2004), 7270 m in WCA-3AS and 13,750 m in WCA-3AN (Bruland et al., 2006a), double ranges with 3290 and 9978 m in WCA-1 (Corstanje et al., 2006), and 7468 m in WCA-2A (2003) (Rivero et al., 2007b). These fi ndings suggest that soil TP shows large (regional) spatial autocorrelations in subtropi-cal wetlands that requires less dense sampling to characterize spatial distribution patterns.

Other biogeochemical properties seem to vary across inter-mediate spatial scales, with spatial autocorrelations on the order of <1500 m (e.g., microbial biomass P, NH4–N, peptidase activity, and total labile organic C) (Grunwald et al., 2007a).

To avoid oversampling and excessive costs and labor, the aim is to identify an optimal spatial scale and sampling density for accurate characterization of biogeochemical patterns across aquatic and wetland ecosystems. Geostatistical analysis pro-vides metrics to quantify spatial variability and distribution patterns (Goovaerts, 1997; Webster and Oliver, 2001). This study underpins the importance of observing biogeochemical properties at an appropriate spatial scale and density.

Our specifi c objectives were to: (i) assess the spatial auto-correlation structure and spatial variability of TP and total N (TN) across a subtropical wetland given a dense observation set; and (ii) assess the information loss associated with underly-ing spatial autocorrelations and spatial variability in TP and TN under various scenarios of reduced sampling densities.

MATERIALS AND METHODSStudy Area

The study area comprised 4800 ha of freshwater wetland within the Blue Cypress Marsh Conservation Area (BCMCA), located in the headwater region of the St. Johns River in east-central Florida. The BCMCA is a subtropical marsh that has been impacted by nonpoint-source pollution, which began to be reduced in the mid-1990s. Since then, this ecosystem has undergone a natural succession of recov-ery. Ollila et al. (1995) and D’Angelo et al. (1999) documented the occurrence of nutrient-enriched and unimpacted zones within the marsh. Native vegetation in the study area is predominately a mosaic of sawgrass (Cladium jamaicense Crantz) and maidencane (Panicum hemitomon Schult.) fl ats, areas of scrub-shrub vegetation (e.g., the coastal plain willow, Salix caroliniana Michx.), cattail (Typha ssp.) marshes, and deep-water slough communities (e.g., white water lily, Nymphaea ssp.). Drying of the marsh permitted the expansion of woody vegetation (e.g., coastal plain willow) into areas previously occupied by her-baceous, wetland marsh plants. Since the early 1970s, coastal plain willow has been expanding from the southern to the northern part of the study area. Soils in the BCMCA are in the Histosol soil order (NRCS, 2007).

Field Data SetWe used a data set (n = 266) collected within the BCMCA in

March to April 2002 by the staff of the Wetland Biogeochemical Laboratory, Soil and Water Science Department, University of Florida. Each soil sample was the composite of two soil cores (0–10-cm depth) taken with a stainless steel core tube. Samples were sealed in plastic bags and stored on ice until received at the laboratory, where they were stored at 4°C until analysis. A targeted sampling design was used with irregular spacing on an approximate grid, with spacings of 31 and 420 m between sampling locations. Total soil P was determined using the ashing method (Anderson, 1976). Phosphorus content was analyzed using an automated colorimetric analysis (USEPA, 1993, Method 365.1). Total N was measured using a Carlo-Erba NA-1500 CNS Analyzer (Haak-Buchler Instruments, Saddlebrook, NJ). All analyses for this study followed National Environmental Laboratory Accreditation Conference quality control and quality assurance protocols.

Geospatial AnalysisSemivariograms were derived for TP and TN, respectively, using

the total data set collected at 266 observation sites. Variables were log transformed to better comply with stationarity assumptions and to stabilize variances, as suggested by Webster and Oliver (2001). Log-normal point ordinary kriging (LN-OK) (Goovaerts, 1997) was used

SSSAJ: Volume 72: Number 4 • July –August 2008 1177

to derive estimates of each soil variable at unsampled locations across a grid, using a 100-m resolution. The geostatistical analyses were per-formed in ISATIS (Geovariances Inc., Houston, TX). To evaluate the effect of using a reduced data set to represent the underlying spatial variability and distribution of TP and TN, we considered the fol-lowing criteria: (i) random selection of observations might be biased toward specifi c geographic regions, resulting in clustered or clumped subsets; and (ii) a one-time selection of observations from the total data set ignores the impact of sampling fl uctuations on prediction performances. Thus, subsets of 175, 150, 125, 100, 75, and 50 obser-vation sites from the total soil population (n = 266) were selected in jackknife mode using a random number generator. Jackknifi ng is a statistical method of numerical resampling based on deleting a por-tion of the original observations in subsequent samples. The average sampling density was 0.0365 samples ha−1 (n = 175), 0.0312 samples ha−1 (n = 150), 0.0260 samples ha−1 (n = 125), 0.0208 samples ha−1 (n = 100), 0.0156 samples ha−1 (n = 75), and 0.0104 samples ha−1 (n = 50). The average sampling density for the total data set of n = 266 was 0.0554 samples ha−1. The random selection procedure to identify each subset was repeated 50 times to account for sampling fl uctua-tions. For each selected subset, the semivariogram was computed and TP and TN predicted using LN-OK (Goovaerts, 1997). The 50 simu-

lations of TP and TN at each subset level generated cumulative distribution functions that were characterized by their mean and CV for each individual wetland pixel (100 by 100 m2). A validation set of 91 samples was used to assess model performance for

different subsets of 175, 150, 125, 100, 75, and 50 observation sites, using 50 iterations for each subset. Model performance was evaluated in validation mode for TP and TN using the mean prediction error (MPE) and RMSE (Webster and Oliver, 2001).

For TP, a cutoff value of 550 mg kg−1 was assumed, resembling historic conditions, and for each subset (n = 175, 150, 125, 100, 75, and 50), probabilities to exceed the cutoff value were calculated. We further assumed that the dense observation model (n = 266), called the reference model, was the most accurate among the considered soil populations. A misclassifi cation rate was calculated as the deviation between model and reference probabilities for each subset (n = 175, 150, 125, 100, 75, and 50) using ArcGIS 9.1 (Environmental Systems Research Inst., Redlands, CA).

RESULTS AND DISCUSSIONTotal P ranged from 349.6 to 1013.7 mg kg−1 with a

median of 602.8 mg kg−1. Total N showed a minimum of 17.6, maximum of 82.4, and median of 26.7 g kg−1 (Table 2). Semivariograms and estimated TP and TN are shown in Fig. 1. A spherical model was used to fi t the experimental semivar-iogram of TP with a nugget variance of 0.154, sill variance of

Table 2. Descriptive statistics of total P (TP) and total N (TN).

Variable Sites Min. Max. Mean Median SEM SD Skewness

TP, mg kg−1 266 349.6 1013.7 620.5 602.8 7.9 128.0 0.52TN, g kg−1 266 17.6 82.4 27.2 26.7 0.3 5.4 5.5

Fig. 1. Variography and predictions based on log-normal kriging using the dense soil population (n = 266) (experimental semivariogram, thin line, D1; fi tted semivariogram, thick line, M1): (a) semivariogram of log total P (TP); (b) prediction of TP; (c) semivariogram of log total N (TN); (d) predictions of TN.

1178 SSSAJ: Volume 72: Number 4 • July–August 2008

1.327, and range of 7240 m. While long-range spatial autocor-relation dominated TP, short-range spatial autocorrelation was found for TN. The experimental semivariogram for TN was fi tted using a spherical model with a nugget variance of 0.044, sill variance of 0.808, and range of 1007 m. Interestingly, the strength of the spatial relationships was stronger for TN than for TP, as indicated by the nugget/sill ratio. Total P and TN showed contrasting spatial patterns, with high TP in the south and east of the wetland and concentrated TN hotspots in the west and northeast (Fig. 1). A crescent-shaped area in the north-west of the BCMCA showed the lowest TP. Historical P inputs

from agricultural activities entered the marsh in the south, which is consistent with the observed spatial patterns of TP.

Spatial patterns of TP may be linked to N cycling (White and Reddy, 2000), microbial community structure (Drake et al., 1996), and the productivity and community structure of macrophytes (Rivero et al., 2007b). Other processes that may play a major role for P retention include uptake and release by periphytons and microorganisms, sorption and exchange reactions with soils and sediments (Richardson, 1985), chemi-cal precipitation in the water column, and sedimentation and entrainment (Reddy et al., 1999). In the BCMCA, it was found that TP was positively correlated with labile inorganic P (r = 0.53) and labile organic P (r = 0.52), but less so with microbial biomass P (r = 0.21) (Grunwald et al., 2006). These fi ndings are unusual because eutrophication in marsh systems has generally been associated with increases in biomass content (Qualls and Richardson, 2000). In the BCMCA, however, the level of microbial biomass seemed to primarily respond to fac-tors other than TP enrichment. The fi ndings of Grunwald et al. (2006) in the BCMCA suggested that, in TP-enriched areas, the primary productivity is enhanced, which increased the sizes and turnover rates associated with different P pools.

Mean spatial patterns of TP generated using various subsets (n = 175–50) based on 50 iterations are shown in Fig. 2. Spatial patterns disintegrated slightly under reduced sample densities, although regional high and low TP values were prominent across all six maps. The spatial autocorrelation range for TP was relatively persistent across various sample subsets (n = 175, 150, 125, 100, 75, and 50) at 7293, 7267, 7296, 6080, 5730, and 5710 m, respectively (Table 3). Similarly, the TP nugget/sill ratio, which describes the strength of the spatial relation-ship, was preserved across various sample sets and ranged from 0.22 to 0.26. This suggests that TP was relatively insensitive

Fig. 2. Mean total P (TP) computed for subsets of 175, 150, 125, 100, 75, and 50 observation sites with 50 iterations.

Table 3. Pooled semivariogram metrics based on 50 iterations of log(total P) and log(total N) for different sets of observation sites.

Observation model (n)

Nugget Sill Nugget/sill Range

mLog(total P)

266 0.154 1.327 0.12 7240

175 0.320 1.202 0.26 7293

150 0.279 1.241 0.23 7267

125 0.294 1.203 0.24 7296

100 0.255 1.073 0.24 6080

75 0.232 1.062 0.22 5730

50 0.263 1.013 0.26 5710

Log(total N)

266 0.044 0.808 0.05 1007

175 0.035 0.853 0.04 1006

150 0.037 0.882 0.04 913

125 0.134 0.773 0.17 999

100 0.232 0.659 0.35 1425

75 0.438 0.734 0.60 502650 0.751 0.750 1.00 6210

SSSAJ: Volume 72: Number 4 • July –August 2008 1179

to different sampling densities. Even the sparser data sets of n = 75 and 100 seemed able to preserve the spatial patterns of TP across the wetland. Only very low sample densities (n = 50) were less able to resemble the spatial patterns found with denser samplings.

The disintegration of spatial patterns under reduced sam-ple sets was more severe for TN than for TP (compare Fig. 2 and 3). Total N showed heterogeneous spatial behavior based on denser sampling (e.g., n = 266 and 175) due to the short spatial autocorrelations of 1007 and 1006 m, respectively. The spatial autocorrelation range for TN for different sampling den-sities (175, 150, 125, 100, 75, and 50 m) changed dramatically from 1006 to 913, 999, 1425, 5026, and 6210 m, respectively (Table 3). The large spatial autocorrelation range of 6210 m (n = 50) generated the more muted, homogeneous spatial TN patterns when compared with the heterogeneous, fi ne-scale TN patterns shown for the range of 1006 m (n = 175). The nugget/sill ratio for TN was relative stable for observation sets between 266 and 125, with 0.05 to 0.17 indicating strong spatial rela-tionships, but started to increase dramatically for n < 125. This suggests that spatial relationships for TN were not represented adequately using smaller observations sets of n < 125. This was confi rmed by a dramatic increase in the TN range for n < 125 compared with denser TN sample sets.

The processes that may have contributed to generate TN spatial patterns include transport and transformations of N including enzyme hydrolysis, mineralization, nitrifi cation, NH4–N adsorption and desorption, NH3–N volatilization, denitrifi cation, and vegetative assimilation and decay (Reddy and DeLaune, 2008). Based on mechanistic N simulations in wetlands by Martin and Reddy (1997) under low-N-input con-ditions (51 g N yr−1), vegetative change accounted for −35.0 g

N yr−1, denitrifi cation for 27.2 g N yr−1, N soil change for −13.9 g N yr−1, and volatilization for 1.8 g N yr−1, resulting in a total N output of the wetland of 69.7 g N yr−1. In contrast, under high-N-loading conditions (2453 g N yr−1), vegetative change accounted for 44.7 g N yr−1, denitrifi cation for 651.5 g N yr−1, N soil change for 707.4 g N yr−1, and volatilization for 400.9 g N yr−1, resulting in total N output of 652.3 g N yr−1 for the wetland. Similarly, in the BCMCA, spatial TN patterns in areas impacted by N infl ux may have been controlled by soil N change, whereas areas low in N were probably dominated by denitrifi cation and vegetative change.

The spatial behavior of TP and TN was very different across the BCMCA, which suggests that biogeochemical indi-cator variables be separated based on their spatial variability and behavior (Grunwald et al., 2007a). This has implications for the observation density and population size required to map a given biogeochemical indicator variable. For TN, a much higher sampling density was required to capture spatial patterns than for TP. Webster and Oliver (2001, p. 90) pin-pointed that “100 observations might be acceptable in some circumstances and 144 are likely adequate (to characterize soil spatial variability), at least with normally distributed isotro-pic data. Semivariograms computed from samples of 225 will almost certainly be reliable, and samples over 400 are extrava-gant.” Chilès and Delfi ner (1999) considered experimental semivariograms not reliable if calculated from <50 observation data pairs. The small number of observations (n < 50) used in numerous wetlands studies (e.g., Mitsch et al., 1995; Ettema et al., 1998; Bruland and Richardson, 2005) may be of concern; however; the researchers acknowledged the considerable dif-fi culties associated with collecting large numbers of samples in

Fig. 3. Mean total N (TN) computed for subsets of 175, 150, 125, 100, 75, and 50 observation sites with 50 iterations.

1180 SSSAJ: Volume 72: Number 4 • July–August 2008

spatially explicit designs at remote, wet, and densely vegetated sites (Bridgham et al., 2001).

As expected, the validation showed larger MPE and RMSE values with decreasing sampling density (n = 266–50) for TP and TN (Table 4); however, the RMSE increased only 18% for TP but 390% for TN for the reduced sample set (n = 50) when compared with the dense reference set (n = 266). Such large RMSE and MPE errors for TN for small sample sets of n < 125 are neither desirable nor acceptable due to large uncertainties in predictions. This was confi rmed by the large CVs for TN for various subset scenarios (n = 175–50) when compared with CVs for TP that appear much lower (compare Fig. 4 and 5). The CV maps describe the uncertainty of predictions based on 50 iterations for various subset samples (n = 175–50).

To restore wetlands, often threshold values are used to indicate nutrient enrichment (Reddy and DeLaune, 2008). In

particular, oligotrophic subtropical wetlands are P limited and respond to P enrichment (Rivero et al., 2007b). This is also true for the BCMCA, where nutrient infl ux has dramatically changed vegetation patterns, with non-native species domi-nating nutrient-enriched areas. To assess the impact of biased assessment of TP spatial patterns, the probabilities of exceeding a cutoff TP value of 550 mg kg−1 were evaluated using vari-ous sample subsets (n = 175–50) (Fig. 6). Overall, probabilities with small sample densities (n < 100) showed smaller coverage of low probabilities (<0.25), larger coverage of intermediate probabilities (0.25–0.75), and smaller coverage of large prob-abilities (>0.75) compared with the dense sample set (n = 266). These under- and overestimations in exceedence probabilities from subsets (n = 175–50) to the reference set (n = 266) are shown in Fig. 7. These maps illustrate the magnitude of devia-tions between the reference set, which was assumed most accu-rate, and sample subsets (n = 175–50). Some of the over- and underestimations were dramatic, as indicated in dark red and dark green (Fig. 7).

The implications for biased TP predictions (i.e., large overestimations) are noteworthy. Misleading spatially explicit predictions of TP or TN that provide base data for wetland restoration programs involve much risk. For successful restora-tion and impact assessment, it is pertinent to aim at preserving spatial biogeochemical patterns and diversity across wetlands and aquatic systems. Thus, it is not only important to assess the magnitude (e.g., maximum values of TP and TN) but to assess the spatial distribution patterns and spatial diversity. Optimized observation schemes for wetlands and aquatic sys-tems are needed that provide insight into the spatial behavior of various biogeochemical patterns that take into account short- and long-range variability of properties. One-dimensional

Table 4. Validation results: a validation set of 91 samples was used to assess the mean prediction error (MPE) and root mean square prediction error (RMSE) for the dense soil population (n = 266) and subsets of 175, 150, 125, 100, 75, and 50 observation sites based on 50 iterations.

No. of observations (n)

Total P Total N

MPE RMSE MPE RMSE

—— mg kg−1 —— ——— g kg−1 ——266† −1.8 94.7 0.07 5.7

175 −2.0 95.8 2.0 6.7

150 −2.9 97.6 3.1 10.3

125 −4.1 101.9 5.2 12.4

100 −6.0 101.4 7.5 17.3

75 −6.5 104.8 8.3 24.850 −7.6 111.5 10.6 27.9

† Cross-validation.

Fig. 4. Coeffi cients of variation for total P (TP) for subsets of 175, 150, 125, 100, 75, and 50 observations with 50 iterations.

SSSAJ: Volume 72: Number 4 • July –August 2008 1181

Fig. 5. Coeffi cients of variation for total N (TN) for subsets of 175, 150, 125, 100, 75, and 50 observations with 50 iterations.

Fig. 6. Probabilities to exceed total P (TP) of 550 mg kg−1 computed for subsets of 175, 150, 125, 100, 75, and 50 observation sites with 50 iterations.

1182 SSSAJ: Volume 72: Number 4 • July–August 2008

investigations that analyze biogeochemical properties at a few sites may be too simplistic and fail to incorporate lateral bio-geochemical interactions with adjacent locations.

CONCLUSIONSThis study characterized the spatial variability of soil TP

and TN across a subtropical freshwater marsh at different sam-pling densities. The disintegration of spatial patterns of soil TP and TN across a subtropical wetland in Florida was illustrated along trajectories of dense (n = 266) to sparse (n = 50) observa-tions. The two properties showed contrasting spatial metrics, with long spatial autocorrelations for TP (7240 m, n = 266) and much shorter ones for TN (1007 m, n = 266). Total P and TN also different in terms of the strength of the spatial rela-tionship, as indicated by the nugget/sill ratio, which was twice as high for TP as for TN. Spatial patterns of TP were much more resilient across different observation densities compared with TN. The contrasting spatial metrics of TP and TN along trajectories of sparser observation sets (n = 175–50) translated into different responses in spatial patterns that disintegrated dramatically for TN and less so for TP when compared with the reference set (dense set, n = 266 or 0.0554 samples ha−1). In this subtropical wetland, a sampling set of n > 75 (or 0.0156 samples ha−1) for TP and n > 125 (or 0.0260 samples ha−1) for TN yielded satisfactory results to resemble the spatial pat-terns of the reference set. Yet larger observation densities are desirable to quantify spatial biogeochemical patterns across wetlands.

There are numerous implications of our fi ndings for suc-cessful assessment of nutrient enrichment, functionality, and restoration of an impaired wetland ecosystem. Besides accurate

assessment of nutrient status (TP, TN, and various other bio-geochemical response variables), it is pertinent to characterize spatial behavior and distribution patterns of biogeochemical properties. It is critical to preserve biogeochemical variabil-ity (or pedodiversity) to maintain ecosystem health and func-tions of a wetland system. Observation schemes that measure the spatial variability of biogeochemical properties at a suit-able spatial scale and density of measurements are needed. Geostatistical metrics such as spatial autocorrelations for vari-ous biogeochemical properties can guide this endeavor. This study provided insight into the spatial behavior and patterns of two important soil nutrients in wetlands that may guide future investigations. More research is needed to gain insight into the spatial behavior of various other biogeochemical indi-cator variables.

ACKNOWLEDGMENTSThe data collection was supported by a grant from the USEPA, Grant no. R-827641-01. We thank J. Prenger, Matt M. Fisher, R. Corstanje, and Y. Wang for assistance in sampling and analytical analysis of soil samples.

REFERENCESAnderson, C.J., W.J. Mitsch, and R.W. Nairn. 2005. Temporal and spatial

development of surface soil conditions at two created riverine marshes. J. Environ. Qual. 34:2072–2081.

Anderson, J.M. 1976. An ignition method for determination of total phosphorus in lake sediments. Water Res. 10:329–331.

Barbiéro, L., J.P. de Querioz Neto, G. Ciomai, A.Y. Sakamotot, B. Capellari, E. Fernandes, and V. Valles. 2002. Geochemistry of water and ground water in the Nhecolandia, Pantanal of Mato Grosso, Brazil: Variability and associated processes. Wetlands 22:528–540.

Bridgham, S.D., C.A. Johnston, and J.P. Schubaurer-Berigan. 2001. Phosphorus sorption dynamics in soils and coupling with surface and

Fig. 7. The dense soil population of total P (TP) was used as reference and model probabilities for subsets of 175, 150, 125, 100, 75, and 50 ob-servation sites with 50 iterations subtracted to derive deviation maps.

SSSAJ: Volume 72: Number 4 • July –August 2008 1183

pore water in riverine wetlands. Soil Sci. Soc. Am. J. 65:577–588.Bruland, G.L., S. Grunwald, T.Z. Osborne, K.R. Reddy, and S. Newman.

2006a. Spatial distribution of soil properties in Water Conservation Area 3 of the Everglades. Soil Sci. Soc. Am. J. 70:1662–1676.

Bruland, G.L., and C.J. Richardson. 2005. Spatial variability of soil properties in created, restored, and paired natural wetlands. Soil Sci. Soc. Am. J. 69:273–284.

Bruland, G.L., C.J. Richardson, and S.C. Whalen. 2006b. Spatial variability of denitrifi cation potential and related soil properties in created, restored and paired natural wetlands. Wetlands 26:1042–1056.

Chilès, J.-P., and P. Delfi ner. 1999. Geostatistics: Modeling spatial uncertainty. Wiley-Interscience Publ., New York.

Corstanje, R., S. Grunwald, K.R. Reddy, T.Z. Osborne, and S. Newman. 2006. Assessment of the spatial distribution of soil properties in a northern Everglades marsh. J. Environ. Qual. 35:938–949.

Craft, C.B., and C. Chiang. 2002. Forms and amounts of soil nitrogen and phosphorus across a longleaf pine–depressional wetland landscape. Soil Sci. Soc. Am. J. 66:1713–1721.

D’Angelo, E.M., T.C. Oña, R. Corstanje, and K.R. Reddy. 1999. Selected biogeochemical properties of impacted and unimpacted marsh sites in the Upper St. Johns River Basin. Rep. 98B368. St. Johns River Water Manage. District, Palatka, FL.

DeBusk, W.F., S. Newman, and K.R. Reddy. 2001. Spatial and temporal patterns of soil phosphorus enrichment in Everglades Water Conservation Area 2A. J. Environ. Qual. 30:1438–1446.

Drake, H.L., N.G. Aumern, C. Kuhner, C. Wahner, A. Griesshammer, and M. Schmittroth. 1996. Anaerobic microfl ora of Everglades sediments: Effects of nutrients on population profi les and activities. Appl. Environ. Microbiol. 62:486–493.

Ettema, C.H., D.C. Coleman, G. Vellidis, G. Lowrance, and S.L. Rathbun. 1998. Spatiotemporal distributions of bacterivorous nematodes and soil resources in a restored riparian wetland. Ecology 79:2721–2734.

Fennessy, M.S., and M.J. Mitsch. 2001. Effects of hydrology on spatial patterns of soil development in created riparian wetlands. Wetlands Ecol. Manage. 9:103–120.

Fisher, M.M., and K.R. Reddy. 2001. Phosphorus fl ux from wetland soils affected by long-term nutrient loading. J. Environ. Qual. 30:261–271.

Gallardo, A. 2003. Spatial variability of soil properties in a fl oodplain forest in northwest Spain. Ecosystems 6:564–576.

Goovaerts, P. 1997. Geostatistics for natural resources evaluation. Oxford Univ. Press, New York.

Grunwald, S., R. Corstanje, B.E. Weinrich, and K.R. Reddy. 2006. Spatial patterns of labile forms of phosphorus in a subtropical wetland 10 years after a sustained nutrient impact. J. Environ. Qual. 35:378–389.

Grunwald, S., K.R. Reddy, S. Newman, and W.F. DeBusk. 2004. Spatial variability, distribution and uncertainty assessment of soil phosphorus in a south Florida wetland. Environmetrics 15:811–825.

Grunwald, S., K.R. Reddy, T.Z. Osborne, and G.L. Bruland. 2008. Soil and vegetative patterns and their variability in space across the Greater Everglades. J. Environ. Qual. 37(4):(in press).

Grunwald, S., K.R. Reddy, J.P. Prenger, and M.M. Fisher. 2007a. Modeling of the spatial variability of biogeochemical soil properties in a freshwater ecosystem. Ecol. Modell. 210:521–535.

Grunwald, S., R.G. Rivero, and K.R. Reddy. 2007b. Understanding spatial variability and its application to biogeochemistry analysis. p. 435–462.

In D. Sarkar et al. (ed.) Environmental biogeochemistry: Concepts and case studies. Elsevier, Berlin.

Litaor, M.I., O. Reichman, M. Belzer, K. Auerswald, A. Nishri, and M. Shenker. 2003. Spatial analysis of phosphorus sorption capacity in a semiarid altered wetland. J. Environ. Qual. 32:335–343.

Lyons, J.B., J.H. Görres, and J.A. Amador. 1998. Spatial and temporal variability of phosphorus retention in a riparian forest soil. J. Environ. Qual. 27:895–903.

Martin, J.F., and K.R. Reddy. 1997. Interaction and spatial distribution of wetland nitrogen processes. Ecol. Modell. 105:1–21.

Mitsch, J.W., J.K. Cronk, X. Wu, and R.W. Nairn. 1995. Phosphorus retention in constructed freshwater riparian marshes. Ecol. Appl. 5:830–845.

Newman, S., K.R. Reddy, W.F. DeBusk, Y. Wang, G. Shih, and M.M. Fisher. 1997. Spatial distribution of soil nutrients in a northern Everglades marsh: Water Conservation Area 1. Soil Sci. Soc. Am. J. 61:1275–1283.

Noe, G.B., D.L. Childers, and R.D. Jones. 2001. Phosphorus biogeochemistry and the impact of phosphorus enrichment: Why is the Everglades so unique? Ecosystems 4:603–624.

NRCS. 2007. Soil data mart. Available at soildatamart.nrcs.usda.gov/ (verifi ed 7 May 2008). NRCS, Washington, DC.

Ollila, O.G., K.R. Reddy, and L. Keenan. 1995. Labile and non-labile pools of phosphorus in surface water and soils in the upper St Johns River basin. Rep. 94D179. St Johns River Manage. District, Palatka, FL.

Qualls, R.G., and C.J. Richardson. 1995. Forms of soil phosphorus along a nutrient enrichment gradient in the northern Everglades. Soil Sci. 160:183–198.

Qualls, R.G., and C.J. Richardson. 2000. Phosphorus enrichment affects litter decomposition, immobilization, and soil microbial phosphorus in wetland mesocosms. Soil Sci. Soc. Am. J. 64:799–808.

Reddy, K.R., and R.D. DeLaune. 2008. Biogeochemistry of wetlands. CRC Press, Boca Raton, FL.

Reddy, K.R., R.H. Kadlec, E. Flaig, and P.M. Gal. 1999. Phosphorus retention in streams and wetlands: A review. Crit. Rev. Environ. Sci. Technol. 29:83–146.

Reddy, K.R., Y. Wang, W.F. DeBusk, M.M. Fisher, and S. Newman. 1998. Forms of soil phosphorus in selected hydrologic units of the Florida Everglades. Soil Sci. Soc. Am. J. 62:1134–1147.

Richardson, C.J. 1985. Mechanisms controlling phosphorus retention capacity in freshwater wetlands. Science 228:1424–1427.

Richardson, C.J., and S. Qian. 1999. Long-term phosphorus assimilative capacity in freshwater wetlands: A new paradigm for maintaining ecosystem structure and function. Environ. Sci. Technol. 33:1545–1551.

Rivero, R.G., S. Grunwald, and G.L. Bruland. 2007a. Incorporation of spectral data into multivariate geostatistical models to map soil phosphorus variability in a Florida wetland. Geoderma 140:428–443.

Rivero, R.G., S. Grunwald, T.Z. Osborne, K.R. Reddy, and S. Newman. 2007b. Characterization of the spatial distribution of soil properties in Water Conservation Area-2A, Everglades, Florida. Soil Sci. 172:149–166.

USEPA. 1993. Methods for the determination of inorganic substances in environmental samples. Environ. Monitoring Syst. Lab., Cincinnati, OH.

Webster, R., and M.A. Oliver. 2001. Geostatistics for environmental scientists. John Wiley & Sons, New York.

White, J.R., and K.R. Reddy. 2000. Infl uence of phosphorus loading on organic nitrogen mineralization of Everglades soils. Soil Sci. Soc. Am. J. 64:1525–1534.

Yanai, J., C.K. Lee, M. Umeda, and T. Kosaki. 2000. Spatial variability of soil chemical properties in a paddy fi eld. Soil Sci. Plant Nutr. 46:473–482.