HOME HELP FEEDBACK SUBSCRIPTIONS ARCHIVE SEARCH TABLE OF CONTENTS		QUICK SEARCH:   [advanced] Author: Keyword(s): Year:  Vol:  Page:

This Article Abstract Figures Only Full Text (PDF) Free Alert me when this article is cited Alert me if a correction is posted Services Similar articles in this journal Similar articles in Web of Science Alert me to new issues of the journal Download to citation manager Citing Articles Citing Articles via HighWire Citing Articles via Web of Science (2) Citing Articles via Google Scholar Google Scholar Articles by Grunwald, S. Articles by Lamsal, S. Search for Related Content PubMed Articles by Grunwald, S. Articles by Lamsal, S. GeoRef GeoRef Citation Agricola Articles by Grunwald, S. Articles by Lamsal, S. Related Collections Phosphorus Geostatistics Spatial Variability

SPECIAL SECTION: FROM FIELD- TO LANDSCAPE-SCALE VADOSE ZONE PROCESSES

Incorporation of Auxiliary Information in the Geostatistical Simulation of Soil Nitrate Nitrogen

^a Soil and Water Science Dep., Institute of Food and Agric. Sciences, Univ. of Florida, 2169 McCarty Hall, P.O. Box 110290, Gainesville, FL 32611-0290, USA
^b BioMedware, Inc., 516 N. State St., Ann Arbor, MI 48104, USA
^* Corresponding author (SGrunwald{at}ifas.ufl.edu)

Received 27 February 2005.

	ABSTRACT

TOP EXECUTIVE SUMMARY ABSTRACT INTRODUCTION MATERIALS AND METHODS RESULTS AND DISCUSSION CONCLUSIONS REFERENCES

 
In north-central Florida the potential risk for movement of nitrate into the aquifer is high due to the large extent of well-drained marine-derived quartz sand overlying porous limestone material coupled with high precipitation rates. Our objective was to estimate spatio-seasonal distributions of soil NO₃–N across the Santa Fe River Watershed in north-central Florida. We conducted spatially distributed synoptic and seasonal sampling (September 2003—wet summer/fall season, January 2004—dry winter season, May 2004—dry spring season) of soil NO₃–N. Prior distributions of probability for NO₃–N were inferred at each location across the watershed using ordered logistic regression. Explanatory variables included environmental spatial datasets such as land use, drainage class, and the Floridian aquifer DRASTIC index. These prior probabilities were then updated using indicator kriging, and multiple realizations of the spatial distribution of soil NO₃–N were generated by sequential indicator simulation. Cross-validation indicated that smaller prediction errors are obtained when secondary information is incorporated in the analysis and when indicator kriging is used instead of ordinary kriging to analyze these datasets characterized by the presence of extreme high values and a nonnegligible number of data below the detection limit. The NO₃–N values were lowest in September 2003 as a result of excessive leaching caused by large, intense tropical storms. Overall the NO₃–N values in January 2004 were high and could be attributed to fertilization of crops and pastures, low plant uptake, and low microbial transformation during the winter period. Despite seasonal trends reflected by the values of observed and estimated NO₃–N, we found areas that showed consistently high soil NO₃–N throughout all seasons. Those areas are prime targets to implement best management practices.

Abbreviations: ccdf, conditional cumulative distribution function • DL, detection limit • ESRI, Environmental Systems Research Institute • IK, indicator kriging • PI, probability interval • SFRW, Santa Fe River Watershed • SIS, sequential indicator simulation • SKlm, simple kriging with local mean • SRWMD, Suwannee River Water Management District • SSURGO, Soil Survey Geographic Database

	INTRODUCTION

TOP EXECUTIVE SUMMARY ABSTRACT INTRODUCTION MATERIALS AND METHODS RESULTS AND DISCUSSION CONCLUSIONS REFERENCES

 
WATER RESOURCES perform multiple functions supporting economic, ecological, and sociocultural activities. Multiple stresses are imposed on Florida's water resources by humanity (e.g., ecotourism, urban, recreation and agricultural activities) and nature (e.g., high-intensity rainfall-runoff events, and hurricanes). According to the Florida Department of Environmental Protection (Scott, 2002), Florida's aquifer system contains more than 8.33 x 10¹² m³ of fresh water and 90% of Florida's population relies on groundwater resources for its drinking water. The vadose zone provides an interface between land use activities and the Floridian (deep), Intermediate, and Surficial aquifer systems. In north-central Florida, the Floridian aquifer is confined (artesian) on the Highlands side but unconfined (nonartesian) on the Lowlands side, providing ample pathways for nutrients and contaminants to leach into the deep aquifer system (Fernald and Purdum, 1998). The potential risk for nutrient and contaminant movement in this region is high because of the large extent of well-drained marine-derived quartz sand overlying porous limestone material coupled with high annual precipitation rates (>1300 mm) (National Climatic Data Center, NOAA, 2005) that causes excessive leaching.

The Suwannee Basin has the largest area of elevated NO₃–N concentrations in the Floridian aquifer system in the state of Florida (Suwannee River Water Management District [SRWMD], 2002). According to the groundwater monitoring program operated by SRWMD, NO₃–N was elevated in the Middle Suwannee River Watershed and along a NW-SE axis extending through the Santa Fe River Watershed (SFRW) (SRWMD, 2002). Springs in the Middle Suwannee River Watershed have NO₃–N concentrations ranging from 1.2 to 17.0 mg L^–1 (SRWMD, 2002). The groundwater in the Middle Suwannee River watershed flows toward the Suwannee River and affects the surface water quality of the Suwannee River via springs and seeps in the riverbed. Although the SFRW (3585 km²) covers only 13.8% of the Suwannee Basin (25 900 km²) it contributed 22.2% (2676 t NO₃–N) of the total NO₃–N loads in 2000 and 19.7% (2971 t NO₃–N) of the total NO₃–N loads in 2003 into the Gulf of Mexico (SRWMD, 2000, 2003). Katz et al. (1999) used ¹⁵N-NO₃ isotope tracers to track sources of nitrate in the Suwannee Basin. Their results indicated that nitrate most likely originated from a mixture of inorganic fertilizers and organic animal waste sources.

In this paper we adopt a geostatistical simulation approach for modeling the spatial distribution of soil NO₃–N in the SFRW. This approach is hybrid in that the prior probabilities of exceeding a series of NO₃–N threshold values are first derived from secondary information using ordered logistic regression (McCullagh, 1980), and these probabilities are updated using nonparametric geostatistics (i.e., indicator kriging). Goovaerts (1999), Heuvelink and Webster (2001), and McBratney et al. (2000, 2003) presented state-of-the-art geostatistical and hybrid modeling techniques for landscape-based modeling. McBratney et al. (2000) compared 24 different univariate and multivariate statistical, geostatistical, and hybrid models at field, subcatchment, and regional scales and found that hybrid geospatial techniques outperformed all other methods at all scales. They emphasized that mixed models that combine auxiliary environmental factors with sparsely sampled soil field data are well suited for regional predictive modeling. Other comparisons of numerous hybrid geostatistical applications were presented by Odeh et al. (1995) and Hengl et al. (2004).

Our objective was to estimate spatio-seasonal distributions of soil NO₃–N across the SFRW. Because of the expected spatio-temporal variation of NO₃–N, we conducted spatially distributed synoptic and seasonal sampling of soil NO₃–N. We used profile averaged soil NO₃–N field observations and auxiliary environmental spatial datasets such as land use, drainage class, DRASTIC Index, and other attributes to develop spatial probability trends for three different seasons (wet summer/fall, dry winter, dry spring seasons). The approach presented in this paper is the first application of ordered logistic regression to estimate the prior probabilities of occurrence in indicator kriging. Cross-validation was used to assess the prediction performances of the proposed approach and quantify the benefit over alternative approaches, such as ordinary kriging or univariate indicator kriging.

	MATERIALS AND METHODS

TOP EXECUTIVE SUMMARY ABSTRACT INTRODUCTION MATERIALS AND METHODS RESULTS AND DISCUSSION CONCLUSIONS REFERENCES

 
Study Area
The SFRW, which is part of the Suwannee Basin, is located in north-central Florida and covers an area of approximately 3585 km². The soils are predominantly sandy in texture. Dominant soil orders include Ultisols (36.7%), Spodosols (25.8%), Entisols (14.7%), and there is a minor extent of Histosols (2.0%), Inceptisols (1.1%), and Alfisols (1.0%) (Fig. 1 ). Land use is mixed, consisting of established pine (Pinus spp.) plantation (32.2%), wetlands (16.2%), upland forest (14.7%), improved pasture (14.0%), urban (8.8%), newly planted forest plantations (6.0%), crops (5.0%), rangeland (3.7%), and a variety of high intensity land uses such as tree groves, nurseries, dairies, and feeding operations (SRWMD and SJWMD, 1995). Two main physiographic regions in the watershed are the Gulf Coastal Lowlands and the Northern Highlands, which are separated from one another by the Cody Scarp (Schmidt, 1997). The underlying geology in the western part of the watershed is limestone, providing potential leaching pathways for soil NO₃–N into the aquifer. In contrast, the Cody Scarp is clayey, phosphate-rich, and dissects the watershed along a northwest–southeast axis. The eastern part of the watershed is dominated by sand-rich sediments with varying loamy subsoils (Brown et al., 1990; Randazzo and Jones, 1997). The elevation ranges from around 3 to 91 m above mean sea level (Fig. 1). By and large the land is level (0–2% slopes) or gently sloping and undulating (0–5% slopes), the major exception to this pattern being the moderately and strongly sloping land (5–12% slopes) along the Cody Scarp. The stream network shows a much higher density in the eastern part of the watershed, compared with the western part that is dominated by karst terrain and sinkholes (Fig. 1). The Santa Fe River flows underground for about 3 km bypassing the Cody Scarp. The mean annual precipitation based on 30 yr of records from 1971 to 2000 across the watershed was 1334 mm, and the mean annual temperature was 20.4°C (National Climatic Data Center, NOAA).

View larger version (70K): [in this window] [in a new window]   Fig. 1. GIS layers: soil orders (Soil Survey Geographic Database, Natural Resources Conservation Service); digital elevation model (National Elevation Dataset, USGS), and stream network (USGS and Florida Department of Environmental Protection).

Figure 2 shows the spatial distribution of observed soil NO₃–N values, which indicates persistently high values in the northeast for all sampling events. Soil NO₃–N was highest in January 2004, with a maximum value of 103.7 µg g^–1. Overall, the NO₃–N values in January 2004 could be attributed to fertilization of crops and pastures, low plant uptake, and low microbial immobilization and nitrification during the winter period. Observed NO₃–N values were lowest in September 2003 due to excessive leaching caused by large intense rainfall–runoff events. High NO₃–N values were measured in feedlots and tree groves, with a maximum value of 8.9 µg g^–1 (May 2004). Improved pasture showed low NO₃–N values in September 2003, with a mean of 1.2 and 2.9 µg g^–1 in May 2004, but a much higher mean in January 2004 of 11.4 µg g^–1. Throughout all sampling events, pine plantations showed very low NO₃–N measurements, with means throughout all seasons of 0.2 µg g^–1 and wetlands with 0.3 µg g^–1 (September 2003), 0.1 µg g^–1 (January 2004), and 0.5 µg g^–1 (May 2004).

View larger version (30K): [in this window] [in a new window]   Fig. 2. Location maps of data (i.e., profile average NO3–N, µg g–1) for the three sampling events.

Soft Indicator Kriging
The ccdf F(u₀;z|{Info}) can be modeled using two main categories of geostatistical algorithms, which are referred to as parametric or nonparametric depending on whether a parametric model (i.e., multiGaussian) is adopted for the random function Z(u) (for a review see Goovaerts, 2001). In this paper, a nonparametric approach, known as indicator kriging (IK), is adopted because the large proportion of data below the detection limit makes hazardous any transform to achieve a Gaussian distribution. This method requires a preliminary coding of each piece of information into a vector of indicators, defined for a set of K thresholds, with z_k discretizing the range of variation of the soil attribute. Two sets of indicator data can be created: hard indicators resulting from the coding of NO₃–N data and soft indicators derived from secondary information. Hard indicators, which take a value of 0 or 1, are computed at the sampled location u as:

[1]

Soft indicators j(u;z_k) are valued between 0 and 1, and they were calculated from the set of secondary variables (continuous and categorical) by performing ordered logistic regression using the proportional odds model (Procedure Logistic, SAS Institute, 1989). This model assumes that the impact of the different secondary variables is the same across thresholds (i.e., same regression coefficients) and only the intercept of the regression model is threshold-specific. In theory, a more complex cumulative logit model with nonproportional odds would have better predictive ability. Yet, the larger number of regression parameters associated with this model could not be estimated accurately (i.e., lack of convergence) given the small size of the dataset and the fact that most secondary variables are categorical and require the estimation of (S – 1) coefficients where S is the number of categories.

Soft indicators can be viewed as the local means of the hard indicator variable; in other words, the GIS data are used to inform on the regional trend of the NO₃–N data. At any location u A, the prior probability j(u;z_k) can be updated into a posterior probability I*(u;z_k) using the following simple kriging with local mean (SKlm) estimator (Goovaerts and Journel, 1995):

[2]

The weights (u;z_k) are the solution of the following system:

[3]

where C_R(u – u_ß;z_k) is the covariance function of the residual indicator variable R(u;z_k) = I(u;z_k) – j(u;z_k) for the vector joining the locations u and u_ß.

Although the kriging system (Eq. [3]) is written in terms of covariances, common practice consists of inferring and modeling the semivariogram rather than the covariance function. The experimental semivariogram for a given lag vector h is estimated as:

[4]

where N(h) is the number of data pairs within the class of distance and direction used for the lag vector h. A continuous function must be fitted to _R(h,z_k) to deduce semivariogram values for any possible lag h required by prediction algorithms, and also to smooth out sample fluctuations. In this study, the semivariograms were modeled using least-square regression (Pardo-Iguzquiza, 1999). All semivariogram models were bounded (i.e., reached a sill), and the covariance models were derived by subtracting the semivariance values from the sill value.

Modeling of the Cumulative Conditional Distribution Function
Because the K probabilities are estimated individually (i.e., K indicator kriging systems are solved at each location) the following constraints, which are implicit to any probability distribution, might not be satisfied by all sets of K estimates:

[5]

[6]

All probabilities that are not between 0 and 1 are first reset to the closest bound, 0 or 1. Then, condition (Eq. [6], shown above) is ensured by averaging the results of an upward and downward correction of ccdf values (Deutsch and Journel, 1998).

Once conditional probabilities were estimated and corrected for potential order relation deviations, the set of K probabilities must be interpolated within each class (z_k,z_{k + 1}) and extrapolated beyond the smallest and the largest thresholds to build a continuous model for the ccdf. In this study, the resolution of the discrete ccdf was increased by performing a linear interpolation between tabulated bounds provided by the sample histogram of NO₃–N concentration (Deutsch and Journel, 1998).

Validation of the Prediction Models
Prediction errors, as well as the quality of the uncertainty models, were quantified using cross-validation. One NO₃–N observation is removed at a time, and the ccdf at this location is modeled using neighboring soil data and secondary information. The RMSE is computed as:

[7]

where the NO₃–N concentration at u is predicted using the mean of the ccdf there, z*_E(u), which is referred to as the E-type estimate (Goovaerts, 1997).

At any location u knowledge of the ccdf F(u;z|{Info}) allows the computation of a series of symmetric p probability intervals (PI) bounded by the (1 – p)/2 and (1 + p)/2 quantiles of that ccdf. For example, the 0.5 PI is bounded by the lower and upper quartiles [F^–1(u;0.25|{Info}), F^–1(u;0.75|{Info})]. A correct modeling of local uncertainty would entail that there is a 0.5 probability that the actual z value at u falls into that interval or, equivalently, that over the study area 50% of the 0.5 PI include the true value. Cross-validation yields a set of z measurements and independently derived ccdfs at the n locations u, allowing the fraction of true values falling into the symmetric p PI to be computed as:

[8]

where w(p_s) equals 1 if z(u) lies between the (1 – p)/2 and (1 + p)/2 quantiles of the ccdf and equals zero otherwise. The scattergram of the estimated, (p), vs. expected, p, fractions is called the "accuracy plot." Deutsch (1997) proposed assessing the closeness of the estimated and theoretical fractions with the following "goodness" statistic:

[9]

where w(p_s) equals 1 if (p_s) > p_s and equals 2 otherwise. The statistics are computed over S = 99 thresholds corresponding to the percentiles of the histogram of NO₃–N data. Twice the importance is given to deviations when (p_s) < p_s (inaccurate case), that is, the case where the fraction of true values falling into the p PI is smaller than expected.

Not only should the true value fall into the PI according to the expected probability p, but this interval should be as narrow as possible to reduce the uncertainty about that value. In other words, among two probabilistic models with similar goodness statistics one would prefer the one with the smallest spread (less uncertain). Different measures of ccdf spread can be used: variance, interquartile range, and entropy. Following Goovaerts (2001) the average width of the PIs that include the true value are plotted for a series of probabilities p. For a probability p the average width is computed as:

[10]

Sequential Indicator Simulation
Indicator kriging allows one to model the uncertainty prevailing at a specific location u (i.e., local uncertainty) and to estimate the value of the soil attribute z(u) using the mean of the probability distribution (E-type estimate) at that location. The map of E-type estimates is, however, very smooth and does not provide a realistic picture of the spatial variability present in the field. Also, there is no straightforward way to combine the measures of local uncertainty to quantify the uncertainty attached to the value of the variable over blocks of irregular size and much larger than the measurement support. For example, we were interested in estimating the average NO₃–N concentration within each subbasin of the watershed and quantifying the attached uncertainty.

Unlike kriging, the aim of stochastic simulation is not to minimize a local error variance but to reproduce statistics such as the sample histogram or the semivariogram model in addition to honoring the data values. Thus, stochastic simulation is increasingly preferred to kriging for all applications where the spatial variability of the measured field must be preserved, such as the delineation of contaminated areas (Desbarats, 1996) or the modeling of solute transport in the vadose zone (Vanderborght et al., 1997). In particular, ccdf for supports larger than the measurement support (e.g., remediation units or flow simulator cells) can be numerically approximated by the cumulative distribution of block simulated values that are obtained by averaging values simulated within the block (Saito and Goovaerts, 2003).

The objective of stochastic simulation is to generate a set of simulated values {z^(l)(u); l = 1,...,L, u A} that reproduce some target statistics (i.e., histogram, semivariogram) conditionally to the finite set of z data {z(u); = 1,...,n} and the secondary information {y_j(u); j = 1,...,J, u A}. Like estimation, simulation can be accomplished using a growing variety of techniques that differ in the underlying random function model (multiGaussian or nonparametric), the amount and type of information that can be accounted for, and the computer requirements (Goovaerts, 1997). In this study, realizations were generated using sequential indicator simulation (SIS), which relies on the aforementioned indicator semivariogram modeling and soft indicator kriging. The SIS algorithm proceeds as follows:

1. Define a random path visiting each node u of the simulation grid only once.
   
2. At each location u, determine the ccdf using soft indicator kriging. The key difference with the implementation of indicator kriging is that not only the n original NO₃–N data, but also the previously simulated values, will be used as conditioning information for kriging.
   
3. Draw a simulated value from that distribution and add it to the data set.
   
4. Proceed to the next location along the random path, and repeat the two previous steps.
   
5. Loop until all N nodes are simulated.
   

The procedure is repeated using a different random path and set of random numbers to generate another realization.

	RESULTS AND DISCUSSION

TOP EXECUTIVE SUMMARY ABSTRACT INTRODUCTION MATERIALS AND METHODS RESULTS AND DISCUSSION CONCLUSIONS REFERENCES

 
Ordered Logistic Regression
For each of the three time periods, a proportional odds model was developed using 10 indicator classes as dependent variables. The construction of these cumulative logits required the choice of K = 9 thresholds. For each time period, the DL was selected as the first threshold. The proportion of observations below the DL ranged from 11.7% (May 2004) to 41.5% (January 2004) (see Table 1). The other eight thresholds were selected so that the data above the DL were distributed equally among the different threshold classes. The explanatory variables included four attributes (Fig. 3 ): drainage, land use, majority land use, DRASTIC index Floridian aquifer; and one cross product to capture the interaction between drainage and land use.

View larger version (55K): [in this window] [in a new window]   Fig. 3. GIS layers: drainage classes, Floridian aquifer DRASTIC index, and land use. (Drainage class code: VP = very poorly; SP = somewhat poorly; E = excessive; SE = somewhat excessive; MW = moderately well; P = poorly; W = well drained).

View larger version (95K): [in this window] [in a new window]   Fig. 4. (a) Map of the probability of exceeding a threshold of 0.582 µg g–1 in September 2003 estimated using the proportional odds model. (b) Map of the mean of the local prior distributions of probability, that is, a prior estimate of nitrate concentration (µg g–1). Polygons denote the limits of the drainage subbasins in the Santa Fee watershed.

Soft Indicator Kriging
Hard indicators were created according to Eq. [1] using the same nine thresholds as in the ordered logistic regression. Residuals were obtained by subtracting the soft indicators (i.e., probabilities estimated by the proportional odds model) from the colocated hard data, and the K corresponding indicator semivariograms were computed following Eq. [4]. Figure 5 shows the experimental indicator semivariograms with the model fitted for the data collected in January 2004. Any anisotropic (i.e., direction-dependent) variability was captured by the secondary information; hence, the omnidirectional semivariograms of the residuals were computed. For the larger thresholds in particular, the range was only a few kilometers, which means that most of the regional variability was explained by the secondary information and that the residual variability essentially reflects short-scale fluctuations.

View larger version (30K): [in this window] [in a new window]   Fig. 5. Omnidirectional semivariograms of indicator residuals computed at nine thresholds for the January 2004 sampling campaign. The solid line depicts the isotropic model fitted using weighted least-square regression.

View larger version (22K): [in this window] [in a new window]   Fig. 6. Scatterplot of estimated vs. observed NO3–N concentrations for each sampling period.

View larger version (28K): [in this window] [in a new window]   Fig. 7. Plots of the proportion of observed NO3–N data falling within probability intervals (accuracy plot) and the width of these intervals vs. the probability p. The goodness statistics measures the similarity between the expected and observed proportions in the accuracy plots (best if one). Open circles in the graphs of right column represent the width of probability intervals computed from the sample histogram (cdf).

View larger version (107K): [in this window] [in a new window]   Fig. 8. First two realizations of the spatial distribution of NO3–N concentrations generated using sequential indicator simulation for each period (units: µg g–1).

View larger version (77K): [in this window] [in a new window]   Fig. 9. Mean and standard deviation of the distribution of 100 realizations of NO3–N concentrations generated using sequential indicator simulation and aggregated within each subbasin (units: µg g–1).

View larger version (49K): [in this window] [in a new window]   Fig. 10. Average variance across all 100 realizations of the spatial distribution of nitrate values simulated within each subbasin (units: (µg g–1).

Florida dairies need year-around forage systems that prevent loss of N to groundwater from waste effluent sprayfields (Woodard et al., 2002). Dairies in northern Florida are under governmental mandate to use manure nutrients while preventing excessive leaching. Most are involved in the year-round production of forage crops in manure effluent sprayfields. According to Woodard et al. (2002), forage crops remove large quantities of nutrients from the soil because, unlike grain production systems, nearly all aboveground biomass is removed during harvest. They found that forage systems in bermudagrass [Cynodon dactylon (L.) Pers.] with N concentrations in the plant tissue of 18.1 to 24.2 g kg^–1 were more favorable than corn (Zea mays L.)–forage sorghum [Sorghum bicolor (L.) Moench]–rye (Secale cereale L.) systems with N concentrations of 10.3 to 14.7 g kg^–1 in removing soil N. Therefore, the bermudagrass system was more effective at preventing NO₃–N leaching.

Fertilization rates and timing are critical to avoid excessive N loads in water bodies. Bakhsh et al. (2001) emphasized that residual soil NO₃–N from fertilization depends on the time of application as well as the tillage management practice. Katz et al. (1999) documented that long-term trends of elevated NO₃–N in spring waters of the Suwannee Basin followed the increase in fertilizer use. Therefore, the documentation of land use, land use shifts, and management practices is critical to make inferences about soil NO₃–N and N observed in surface waters and wells.

Wetlands (>16% of watershed) and sites with shallow water tables (flats and flatwoods) comprise large areas of the watershed. Although wetlands are endpoints of flowpaths accumulating material from upland areas, they did not show elevated soil NO₃–N values. This is likely due to the fact that denitrification rates in these Florida wetlands were high (White and Reddy, 2003) and remove N from the watershed by emitting nitrous oxides into the atmosphere.

Soil NO₃–N in pine plantations and upland forest was low for several reasons. First, N fertilizer use is low compared with other crop production systems. An intensive N fertilization practice for forest plantations might apply 50 kg N ha^–1 once in the first 5 yr and as much as 200 kg N ha^–1 once during the following 15 yr. While fertilization is common, it is not necessarily as intensive as indicated above, nor is it universally applied. Second, forest ecosystems impose a closed nutrient cycle on a landscape. Nitrogen deficiencies are commonly most pronounced by age 10 in these pine plantations because much of the bioavailable N is tied up in the forest floor as N is immobilized during decomposition. Third, forest plantations continue to absorb nutrients from the soil during the winter (Comerford et al., 1987) leaving less available for leaching. Fourth, leaching of N from the crowns of pine plantations is minimized as the needles in the crown senesce because much of the N is retranslocated from these needles back into the tree crown and used to support the next season's growth. It has been estimated that approximately 50% of the total N content of senescing needles is removed from the needle before it falls and begins to decompose (Dallatea and Jokela, 1994; Aerts, 1996). All these processes minimize N use, maintain N on the site, and leave less for leaching.

	CONCLUSIONS

TOP EXECUTIVE SUMMARY ABSTRACT INTRODUCTION MATERIALS AND METHODS RESULTS AND DISCUSSION CONCLUSIONS REFERENCES

 
We presented a geostatistical approach that integrated sparse field observations of soil NO₃–N with auxiliary spatial environmental datasets. The approach was innovative in its use of ordered logistic regression to estimate prior probabilities from secondary information, which were later incorporated in soft indicator kriging. The characterization of seasonal and spatial patterns of NO₃–N in the vadose zone within the SFRW will enable managers to minimize adverse impacts on the environment. Three different typical seasons were considered to acknowledge seasonal trends in soil NO₃–N due to climate and land use management activities. The most pronounced spatial patterns in soil NO₃–N were found for the January sampling event, with maximum values >100 µg g^–1. Thus, in Florida soil sampling during winter months may be considered the most representative period to highlight differences among land uses, soils, and geographic locations within this watershed.

Despite seasonal trends reflected by the values of observed and estimated NO₃–N, we found areas that showed consistently high soil NO₃–N in January, May, and September. Those areas should be targeted by best management practices to reduce N loads. Since this watershed is characterized in some areas having karst terrain combined with high annual precipitation, attention has to be given to land use activities performed within these "high risk areas" that potentially serve as short-circuits for NO₃–N entering the aquifer.

	ACKNOWLEDGMENTS

 
This project was supported by the USDA grant no. 2002-00501 funded through the "Nutrient Science for Improved Watershed Management" program. This research was supported by the Florida Agricultural Experiment Station and approved for publication as Journal Series no. R-10791. We thank Dr. Mark W. Clark for his extension efforts, Dr. Donald A. Graetz and Dr. Randy B. Brown for their advice on planning this project, and Wade Hurt for characterizing soils at each field location. We also thank the public and private land owners that provided us access to their property to collect soil samples.

	REFERENCES

TOP EXECUTIVE SUMMARY ABSTRACT INTRODUCTION MATERIALS AND METHODS RESULTS AND DISCUSSION CONCLUSIONS REFERENCES

 

• Aerts, R. 1996. Nutrient resorption from senescing leaves of perennials: Are there general patterns? J. Ecol. 84:597–608.[CrossRef]
• Bakhsh, A., R.S. Kanwar, D.L. Karlen, C.A. Cambardella, T.B. Bailey, T.B. Moorman, and T.S. Colvin. 2001. N-management and crop rotation effects on yield and residual soil nitrate levels. Soil Sci. 166:530–538.[CrossRef]
• Brown, R.B., E.L. Stone, and V.W. Carlisle. 1990. Soils. p. 35–65. In R.L. Myers and J.J. Ewel (ed.) Ecosystems of Florida. University of Central Florida Press, Orlando, FL.
• Comerford, N.B., A.V. Mollitor, and W. McFee. 1987. Late season changes in fascicle nutrient content, weight, and phosphorus uptake by slash pine. Soil Sci. Soc. of Am. J. 51:806–808.[Abstract/Free Full Text]
• Dallatea, F., and E.J. Jokela. 1994. Needlefall returns and resorption rates of nutrients in young intensively managed slash and loblolly-pine stands. For. Sci. 40:650–662.
• Desbarats, A.J. 1996. Modeling spatial variability using geostatistical simulation. p. 32–48. In S. Rouhani et al. (ed.) Geostatistics for environmental and geotechnical applications. ASTM STP 1283. ASTM, Philadelphia, PA.
• Deutsch, C.V. 1997. Direct assessment of local accuracy and precision. p. 115–125. In E.Y. Baafi and N.A. Schofield (ed.) Geostatistics Wollongong '96. Kluwer Academic Publishers, Dordrecht, The Netherlands.
• Deutsch, C.V., and A.G. Journel. 1998. GSLIB: Geostatistical software library and user's guide. 2nd ed. Oxford Univ. Press, New York.
• Environmental Systems Research Institute. 2005. ArcGIS. Vers. 8.3. ESRI, Redlands, CA.
• Fernald, E.A., and E.D. Purdum (ed.) 1998. Water resources atlas of Florida. Institute of Science and Public Affairs, Florida State University, Tallahassee, FL.
• Goovaerts, P. 1997. Geostatistics for natural resources evaluation. Oxford Univ. Press, New York.
• Goovaerts, P. 1999. Geostatistics in soil science: State-of-the-art and perspectives. Geoderma 89:1–45.[CrossRef][Web of Science]
• Goovaerts, P. 2001. Geostatistical modelling of uncertainty in soil science. Geoderma 103:3–26.[CrossRef][Web of Science]
• Goovaerts, P., and A.G. Journel. 1995. Integrating soil map information in modelling the spatial variation of continuous soil properties. Eur. J. Soil Sci. 46(3):397–414.[CrossRef]
• Hengl, T., G.B.M. Heuvelink, and A. Stein. 2004. A generic framework for spatial prediction of soil variables based on regression-kriging. Geoderma 120:75–93.[Web of Science]
• Heuvelink, G.B.M., and R. Webster. 2001. Modelling soil variation: Past, present, and future. Geoderma 100:269–301.[CrossRef][Web of Science]
• Katz, B.G., H.D. Hornsby, J.F. Bohlke, and M.F. Mokray. 1999. Sources and chronology of nitrate contamination in spring waters, Suwannee River Basin, Florida. Water Resour. Invest. Rep. 99-4252. USGS, Tallahassee, FL.
• Keeney, D.R., and D.W. Nelson. 1982. Nitrogen—Inorganic forms. p. 649–660. In A.L. Page et al. (ed.) Methods of soil analysis. Part 2. Agron. Monogr. 9. ASA and SSSA, Madison, WI.
• McBratney, A.B., M.L. Mendonça Santos, and B. Minasny. 2003. On digital soil mapping. Geoderma 117:3–52.[CrossRef][Web of Science]
• McBratney, A.B., I.O.A. Odeh, T.F.A. Bishop, M.S. Dunbar, and T.M. Shatar. 2000. An overview of pedometric techniques for use in soil survey. Geoderma 97:293–327.[CrossRef][Web of Science]
• McCullagh, P. 1980. Regression models for ordinal data (with discussion). J. R. Stat. Soc. Ser. B 42:109–142.
• National Climatic Data Center (NOAA). 2005. Web Climate Services. [Online] Available at www.ncdc.noaa.gov/oa/climate/stationlocator.html (accessed 8 Aug. 2005, verified 4 Nov. 2005).
• Odeh, I.O.A., A.B. McBratney, and D.J. Chittleborough. 1995. Further results on prediction of soil properties from terrain attributes: Heterotopic cokriging and regression-kriging. Geoderma 67:215–226.[CrossRef][Web of Science]
• Pardo-Iguzquiza, E. 1999. VARFIT: A Fortran-77 program for fitting variogram models by weighted least squares. Comput. Geosci. 25:251–261.[CrossRef]
• Randazzo, A.F., and D.S. Jones. 1997. The geology of Florida. Univ. Press of Florida, Gainesville.
• Sabesan, A. 2004. Geo-spatial assessment of the impact of land cover dynamics and distribution of land resources on soil and water quality in the Santa Fe River Watershed. M.S. thesis. University of Florida, Gainesville.
• Saito, H., and P. Goovaerts. 2003. Selective remediation of contaminated sites using a two-level multiphase strategy and geostatistics. Environ. Sci. Technol. 37:1912–1918.[Medline]
• SAS Institute. 1989. SAS/STAT user's guide. Version 6. 4th ed. Vol. 2. [Online] Available at www.id.unizh.ch/software/unix/statmath/sas/sasdoc/stat/a-index.htm (accessed 18 July 2005, verified 6 Oct. 2005). SAS Inst., Cary, NC.
• Schmidt, W. 1997. Geomorphology and physiography of Florida. p. 55–75. In A.F. Randazzo and D.S. Hones (ed.) The geology of Florida. Univ. Press of Florida, Gainesville.
• Scott, T.M. 2002. Florida's springs in jeopardy. Geotimes 47(5):16–20.
• SRWMD and SJRWMD (Suwannee River Water Management District and St. Johns River Water Management District. 1995. Remote sensing derived land use according to the Florida Land Use and Cover Classification System (FLUCCS) [Online]. Available at www.fgdl.org (accessed 18 July 2005, verified 6 Oct. 2005). Florida Geographic Data Library, Univ. of Florida, Gainesville.
• SRWMD (Suwannee River Water Management District). 2000. Surface water quality and biological monitoring. WR01–02. SRWMD, Tallahassee, FL.
• SRWMD (Suwannee River Water Management District). 2002. Groundwater quality report. WR-02/03–02. SRWMD, Tallahassee, FL.
• SRWMD (Suwannee River Water Management District). 2003. Surface water quality and biological monitoring. WR02/03–03. SRWMD, Tallahassee, FL.
• Vanderborght, J., D. Jacques, D. Mallants, P.H. Tseng, and J. Feyen. 1997. Analysis of solute redistribution in heterogeneous soil: II. Numerical simulation of solute transport. p. 283–295. In A. Soares et al. (ed.) GeoENV I—Geostatistics for environmental applications. Kluwer Academic Publishers, Dordrecht, The Netherlands.
• White, J.R., and K.R. Reddy. 2003. Nitrification and denitrification rates of Everglades wetland soil along a phosphorus-impacted gradient. J. Environ. Qual. 32:2436–2443.[Abstract/Free Full Text]
• Woodard, K.R., E.C. French, L.A. Sweat, D.A. Graetz, L.E. Sollenberger, B. Macoon, K.M. Portier, B.L. Wade, S.J. Rymph, G.M. Prine, and H.H. van Horn. 2002. Nitrate removal and nitrate leaching from forage systems receiving dairy effluent. J. Environ. Qual. 31:1980–1992.[Abstract/Free Full Text]

This article has been cited by other articles:

				D. L. Corwin, J. Hopmans, and G. H. de Rooij From Field- to Landscape-Scale Vadose Zone Processes: Scale Issues, Modeling, and Monitoring Vadose Zone J., March 8, 2006; 5(1): 129 - 139. [Abstract] [Full Text] [PDF]

This Article Abstract Figures Only Full Text (PDF) Free Alert me when this article is cited Alert me if a correction is posted Services Similar articles in this journal Similar articles in Web of Science Alert me to new issues of the journal Download to citation manager Citing Articles Citing Articles via HighWire Citing Articles via Web of Science (2) Citing Articles via Google Scholar Google Scholar Articles by Grunwald, S. Articles by Lamsal, S. Search for Related Content PubMed Articles by Grunwald, S. Articles by Lamsal, S. GeoRef GeoRef Citation Agricola Articles by Grunwald, S. Articles by Lamsal, S. Related Collections Phosphorus Geostatistics Spatial Variability