Spatial Variability of Microbial Processes in Soil—A Review

Spatial Variability of Microbial Processes in Soil--A Review T. B. Parkin* ABSTRACT Microbial transformations of fertilizersandpesticides in the surface soil havea direct impacton the massof the agrochemical that is susceptible to leachinglosses. Thus,ourgreatestpotentialfor controllingleachinglosses of agrochemicals is through the management of thesecompounds in the surfacesoil. Avarietyof strategieshave beenemployed to maximize the residencetimeof appliedchemical in thesurfacesoil, including: timingof application, formulation (e.g., slow-release fertilizers andencapsulated pesticides),andthe useof compounds thatmodify microbial activityin soil (e.g., nitrification inhibitors). Although thesestrategieshavemetwithsomesuccess, more precisequantification of the microbial transformations of agrochemicalsis requiredto aid the development of improved management strategies.Thehighspatialvariabilityexhibitedby many microbial processes, in many cases,precludes precisequantification. Agreater understanding of the factorscontributing to thevariabilityof microbial processes allowsfor improved estimation, as wellas forthe assessment of keydrivingvariablescontrollingmicrobial processes in soil. Thisarticlereviews severalaspectsof spatialvariability associatedwithmicrobial populations andprocesses.Thediscussion focuses onthescale at which variabilityis expressed, andthe soil andenvironmental variablesthat serveto controlvariabilityat eachscale. Implications for the development of newmanagement strategiesare also discussed, andfinally, somestatisticalconsiderations forcharacterizing variability arepresented. C ONCERNS over the environmental impact and efficacy of agrochemicals requires a reevaluation of agricultural chemical use. The Pesticides in Groundwater Data Base, developed by the USEPA(Williams et al., 1988) documents instances of groundwater contamination in 38 states as of 1988. In part, this maybe due to point source contamination at well heads; however, transport of field-applied pesticides and nitrate has also been identified as an important factor. Microbial transformations of fertilizers and pesticides in the surface soil influence the mass of the agrochemical, which is susceptible to leaching losses. Thus, management of these transformations in surface soils is a strategy for controlling leaching of agrochemicals to groundwater supplies. The development of alternative managementstrategies to reduce the risks of groundwater contamination is dependant upon having a detailed knowledgeof the factors that influence the fate of the applied agrochemicals. This includes an understanding of the important microbial processes influencing the fate of a particular chemical in soil. Whyfocus on variability? High spatial variability is a factor that often limits our ability to quantify the activity of microorganismsin soil. High variability also hinders the development of predictive relationships between the factors knownto impact a given microbial process. If our goal is to understand and ultimately predict and model the responses of microorganismsin nature, then we must have a quantitative understanding of the factors that inUSDA-ARS, National Soil Tilth Laboratory, 2150PammelDr., Ames,IA 50011.Received19 June 1992. *Corresponding author. Publishedin J. Environ.Qual. 22:409--417(1993). 409 fluence them. By focusing on variability, what we are really doing is addressing the interactions of factors impacting a given microbial process. Thus, characterization of variability has two important advantages: (i) improved quantification of the process/property under study, and (ii) improvedidentification of driving variables. Characterization of variability can be approached by considering three questions: 1. Howlarge is the variability? 2. Whatis causing the high variability? 3. Howdo we deal with the variability? Question 1 relates to quantification of the processes or population under study, as well as improved identification of variables controlling the populationor process. Answersto Question 2 yield insights into driving variables, which in turn aids in the developmentof predictive relationships betweendriving variables and the processes being investigated. The answer to Question 3 is primarily a statistical one and is dependent, in part, upon the answers for Questions 1 and 2. In addressing all three questions, statistical issues related to data analysis and interpretation must be considered. This article presents a general overview of past work on the spatial variability of microbial processes, with special attention given to the spatial scales of variability, the soil/environmental factors controlling variability at each scale, and a discussion of some of the tools or approaches to assess variability. Also, some practical statistical aspects related to the study and characterization of variability are discussed, and implications for the development of new management strategies are presented. SCALES OF VARIABILITY Spatial variability is manifestedat manydifferent scales. For purposes of discussion, four different scales can be defined: microscale, plot scale, field or landscape scale, and regional scale variability. Microscale Amongmicrobial ecologists, there is great interest to try to understand processes at the level at which they are occurring in nature. Studies focusing on microscale variability are usually performed to delineate the mechanisms driving microbial activity in nature. The microscale approach to the study of soil microbial processes is motivated by the fact that, in manycases the conditions experienced by soil organisms at the microscale are not reflected by measurements of these conditions made on bulk soil samples. For example, there have been numerous observations of anaerobic processes occurring in nonflooded soils where pore space oxygen contents approach 20%. Studies indicate that anaerobic microsites supporting high denitrification rates may be predominantly associated with particular organic matter in structureless soils (Parkin, 1987; Christensen et al., 1990) Abbreviations:SEM,scanning electron microscopy; UMVUE, uniformlyminimum variance unbiasedestimator. 410 J. ENVIRON.QUAL., VOL. 22, JULY-SEPTEMBER 1993 Table1. Denitrification rates associatedwith individualsoil aggregates,wormcasts, and small (<0.5 g) particulate organic C fragments. Microsite Soil aggregates Soil aggregates~" Hairy vetch Plant root Beetle carapace Wormcastings Wormcastings~ Denitrification -’ I~g N kg-ld Source 0.72 - 312 0 - 1200 242 - 2470 510 - 8100 2520 16.9 - 209 264 Sexstoneet ai., 1985 Seech and Beauchamp, 1988 Parkin, 1992, unpublished Parkin, 1987 Parkin, 1987 Parkin, 1992, unpublished Svenssonet al., 1986 Rates from48 h incubation of pooled (20 g) samples of aggregates ranging in size from <0.25 to 20 mm.Aggregates were water-saturated, and incubations performedunder an air headspace. Averagerate from pooled 5-g samples. Table2. Spatialrelationships of denitrification in a cornfield. Rangeof denitrification rates observedare reported.’~ Field position Denitrification g N h-a d-~ Bulk density -a gcm Row Interrow Wheel track 5 - 32 0 - 135 33 - 190 0.93 1.02 1.12 ~" Data from Doranet al. (1990). associated with the anaerobic zones existing in the aggregates of a structured soil (Sexstoneet al., 1985; Smith, 1990; Arah, 1990; Seech and Beauchamp, 1988). The factors controlling variability at the microscaleare dependant upon the physiological requirements of the organism or population of organisms under study. Therefore, the size of the microhabitat that defines the microscale is the physical and chemicalenvironmentdirectly adjacent to the microbial cell or microcolony. In this regard, it must be recognized that there is no fixed definition of the size of the microscale. Anorganism residing in soil maybe simultaneously exposed to a range of microenvironments. For example, a single fungal hyphae may span a number of zones having distinctly different chemical and physical characteristics. A conceptual model of soil aggregation presented by Tisdale and Oades(1982) defines the scales at which microorganismsare important in the aggregation process as being in the range of 2 to 2000 /zm. Thus, it is apparent that the size of the microscale is operationally dependantupon the specific microorganism or process of interest, and to some degree the tools available for study. Several tools or approaches have been applied to assess populations and activities at the microscale. These include: light microscopy, electron microscopy, microelectrodes, and a variety of physical separation procedures. Light microscopy has been extensively used as a tool for estimating microbial biomass in soils, or for gross determination of microbial diversity, such as fungal/bacterial ratios (Morganet al., 1991; Stamatiadis et al., 1990). Both transmitted light microscopy and fluorescent microscopy have been applied, although, coupled with fluorescent stains such as acridine orange or calcaflor, fluorescence microscopy has distinct advantages over transmitted light microscopy (Hobbie et al., 1977; Schmidt and Paul, 1982; Bitton and Gerba, 1984, p. 316-319; Postma and Altemuller, 1990). In addition to determination of microbial biomass, microbial activity has also been assessed through the use of metabolic stains Table3. Variabilityof denitrification andsoil propertiesas a function of landscapeposition in an Iowacorn field. Landscape position Hill top Midslope Pot hole Soil materiai Clarion Denitrification rate -’ ttg N kg-~d 8.3 Water NOa-N content Organic C mg/kg kg H20/kg 1.55 0.22 g/kg 0.21 Canisteo 16.6 1.52 0.21 0.29 Okeboji 70.5 5.17 0.29 0.55 Clarion, fine-loamy, mixed, mesic Typic Hapludolls; Canisteo, fineloamy, mixed(calcareous), mesic Typic Haplaquolls; Okeboji, fine, montmorillonitic, mesic CumulicHaplaquolls. such as fluorescein diacetate or indonitrotetrazolium (Ingham and Kein, 1984; Stamatiadis et al., 1990; Soderstrom, 1977; Zimmermanet al., 1978). Despite the relatively widespread use of light microscopy, there are several disadvantages. Direct observation and counting is a tedious process, and often results may be operator-dependant. Newtechnologies such as automated image analysis may be a solution to this problem (Morganet al., 1991); however, instrumentation to perform such analyses is expensive. Also, although general metabolic activity of organisms may be inferred through the use of metabolizable stains, precise function is impossible to assess. Finally, because of sample disruption during processing, light microscopyreveals little information concerning structural relationships between organisms and their habitat. Electron microscopy is one tool that can be used to visualize the spatial relationships of organisms in their environment. Scanning electron microscopy (SEM) has produced dramatic photographs of microorganisms colonizing plant roots and particulate organic matter in soil (Foster et al., 1983; Campbell and Rovira, 1973). Although SEMprovides investigators with an indication of the microorganismsin their habitat, this technique does not allow determination of functional relationships between organisms and their habitat from such investigations. Another approach that has been applied with some success is that of direct measurements of processes at the microscale (Revsbech and Sorenson, 1990). Oxygen microelectrodes (sensing tip = 2/zM) have been inserted into intact soil aggregates, and anaerobic zones have been identified and mapped(Sexstone et al., 1985). The extent of anaerobiosis was related to measurementsof denitrification on the same soil aggregates. Microelectrodes for pH, NO3, CO2, and N20 have also been produced and used for characterization of microbial activities and the physico-chemical environment at the microscale (Revsbech and Sorenson, 1990). A group of techniques, here loosely classified as spatial separation or fraction methods, have been applied in an attempt to relate microbial populations and properties to the characteristics of the microhabitat (Hattori, 1988). The basic objective underlying these approaches is to isolate microenvironment so that measurementsof physicochemical environment can be better related to microbial activity. This approach has been applied to the determination of microbial populations and nitrogen fixation activity associated with individual pieces of straw (Harper and Lynch, 1984), C and N mineralization as- 411 PARKIN: MICROBIALPROCESSESIN SOIL sociated with soil microaggregates and particulate organic matter (Cambardella and Elliott, 1993), and microbial diversity within soil aggregates (Hattori, 1988). This approach has also been applied in the assessment of denitrification rates of a variety of soil microsites (Table 1). Methodsfor physical isolation mayinvolve the use of forceps or spatulas as in the above studies, or with more sophisticated techniques such as micromanipulators, sonicators, or ultramicrotomes (Cambardella and Elliott, 1992; Hattori, 1988; Gahoonia and Nielsen, 1991). Gahoonia and Nielsen (1991) developed a procedure to study rhizosphere processes in thin soil layers at various distances from plant roots. With this procedure, plants are grown in a double chamber system. After the root system develops, one of the chambersis removed, frozen briefly in liquid nitrogen, and then sliced into 0.2-mmsections with a refrigerated microtome.Using this technique gradients of bacterial numbersand ATPcontent in relation to proximity to plant roots have been determined (F. Eiland, 1992, personal communication). Advancements in methodologies for measurement of microbial processes at the microscale are currently under development. Binnerup and Serensen (1992) have recently reported on a bioassay technique for determination microgradients of NO 3 and NOzin the rhizosphere. Other emerging technologies, such as the use of fiber optics coupled with ELISAtechniques, may offer additional opportunities to investigate specific microbial processes at the microscale level in soil (Fitch and Gargus, 1985). Plot Scale Investigations of plot scale variability associated with microbial processes have focused primarily on evaluation of various treatment effects (e.g., residue, tillage, or fertilizer management effects). Less emphasis has been placed on characterization of variability within a plotsized unit. It is obvious that with row crop agricultural systems, distinct zones exist within small areas resulting from plant and managementeffects. The plant rhizosphere offers a habitat rich in organic nutrients, relative to the bulk soil, and populations of microorganisms in the rhizosphere are reported to be several orders of magnitude higher than in nonrhizosphere soils. In addition to direct effects of the plant rhizosphere on the distribution of microorganismswithin the soil profile, the plant canopycan also alter the soil environment. Recently, it has been reported that interception of rain water by a corn (Zea mays L.) canopy, and subsequent stemflow redirects up to 50%of total areal rainfall directly to the base of the corn plants (Parkin and Codling, 1990). However, the wetter conditions in the corn row are rapidly altered due to the rapid water uptake by corn plants (Zhai et al., 1990). Managementeffects such as compaction due to wheel traffic can impact microbial activity. Doranand coworkers (1990) reported higher denitrification activity in the compacted interrow area of corn plots as compared with nontraffic interrow areas or the inrow area (Table 2). this study, variation in denitrification response was attributed primarily to bulk density effects, and bulk density is clearly the dominant factor in controlling denitrification in the two interrow areas. However,the relative impact of compaction vs. plant root effects on denitrification cannot be assessed from this study. This is an important issue, as currently there is conflicting information on the impact of roots on denitrification. Earlier work reports that denitrification maybe enhanced in the rhizosphere (Smith and Tiedje, 1979; Brar, 1972; Garcia, 1975). However,recent reports in the literature suggest that there is no stimulatory effect of roots on denitrification, and in fact, due to competition for water and NO3,the rhizosphere may support lower denitrification rates (Haider et al, 1987). In addition to plant root effects, bandedapplication of fertilizers and pesticides mayalso contribute to spatial heterogeneity of microbial processes at the plot level. A high degree of spatial variability associated with denitrification losses of fertilizer N has been observed in plots receiving banded application of organic waste material (Rice et al., 1988). In pasture systems, patchy inputs of N due to animal activity has also been observed to impact variability of N-flux activity (Limmerand Steele, 1983; Sherlock and Goh, 1983; Colbourn et al., 1987). Parkin and Shelton (1992) reported on distinct differences in carbofuran degradation activity in the interrow and intrarow areas of a corn field. Interactive effects of both microbial biomass and soil water content appeared to contribute to the observed differences in carbofuran degradation kinetics as well as to the positional differences observed. However, this study was unable to assess whether the observed increase in carbofuran degradation activity, in the intrarow areas, occurred in response to the banded application of carbofuran, or to increased C availability in the rhizosphere. Landscape Scale Variability associated with microbial processes has been investigated at the landscape scale in forested ecosystems (Davidson and Swank, 1986; Groffman and Tiedje, 1989; Robertson et al., 1988), a desert ecosystem (Peterjohn and Schlesinger, 1991), and in agricultural systems (Robertson et al., 1990; Pennocket al., 1992). The three primary factors controlling variability at the landscape level are soil type, surface topography, and water distribution. Soil type provides an integrative indication of factors such as texture, top soil depth, organic matter content, pH, and nutrient status. Surface topography is reflective of differing soil quality resulting from erosion, and reflects conditions such as topsoil depth and drainage. Both of these factors mayserve to provide indications concerningwater distribution at the landscape level. Heterogeneous distribution of water across a landscape mayoccur as a result of three factors. Spatial heterogeneity of rainfall inputs mayresult due to localized thunderstorms. Surface flow of water induced by topography and texture serves to redistribute water, and lateral flow of shallow groundwater that reemerges in localized depressions or riparian areas also contribute to the heterogeneity of water distribution. It is clear that at the landscape level these three driving variables do not operate independently. Thus, in somecases, it maybe difficult to isolate the primary driving variable impacting microbial populations or processes. Recent studies of a 32-ha agricultural field in central Iowa revealed that denitrification rate varied as a function of landscape position (Table 3). The soil was a Canisteo-Clarion-Nicollet association, and the field exhibited a typical prairie pothole topography in which shallow depressions frequently becomefilled with water as a re- J. ENVIRON.QUAL., VOL. 22, JULY-SEPTEMBER 1993 suit of water redistribution via surface flow and lateral flow of shallow groundwater. Intact soil cores were collected at three positions in the field and denitrification rates were measuredusing the acetylene inhibition method. Highest denitrification rates were observed in the shallow pothole depressions, and lowest rates were observed on the hilltop regions. It is likely that differences in soil type, soil NO3concentrations, organic C levels, and soil water content all contributed to the differences in denitrification rate observed(Table 3); however,it is difficult to isolate the primary causative agent. In this study landscape position appears to be useful as an integrative variable for water content and soil organic matter. These results are similar to a recent study that showed that landscape element, as defined by the surface topography, maybe a useful indicator variable for denitrification at the landscape scale (Pennock et al., 1992). Regional Scale An understanding of the regional scale variability of microbial processes is an important factor in the development of meaningful global models as well as for the development of general management recommendations. Twoissues that are of current concern related to regional scale estimation and prediction are global climate change, and atmospheric deposition of nutrients in relation to nutrient cycling (Vitousek et al., 1989; Mellilo et al., 1989). At the regional scale the key predictor variables are climatic factors, land use patterns, vegetative cover, and land surface characteristics. Althoughthese factors can be easily identified, relatively little information exists that relates these factors to the variability of microbial processes at the regional scale. Most of the work at the regional scale has been done with trace gas flux, such as CH4, N20, CO2, and NO. Nitric oxide flux rates from agricultural and natural ecosystems has been summarizedby Johansson et al. (1988) and expressed as a function of climatic region, land use pattern, and vegetative cover (Table 4). Fromthis synthesis it was observed that for cultivated soils N-fertilization was a key factor impacting flux, whereas for the Table 4. Nitric oxide emissions agricultural and natural ecosystems, modifiedfrom Johansson(1989). Climate/system Temperate/coniferous Temperate/forest Temperate/grassland Tropical/savanna Tropical/rainforest Tropical/cloud forest Temperate/cropland Temperate~cropland Temperate/cropland Temperate/sward Temperate/cropland Subtropical/cropland Temperate/cropland Temperate/pasture Range Source -2 ng N m-2s Natural ecosystems 0.1 - 0.8 Johansson, 1987 0.2 - 4.1 Williams et al., 1988 Galbally and Roy, 1978 0.6 - 2.6 2 - 250 Johansson et al., 1988 9.2 - 16 Kaplan et al., 1988 0.1 - 2 Johansson et al., 1988 Agricultural/unfertilized 0.3 - 17 Johansson and Granat, 1984 0.003 - 67 Anderson and Levine, 1987 0.2 - 3.8 Williams et al., 1988 6.7J" Colbourn et al., 1987 Agricultural/fertilized 0.1 - 62 Johansson and Granat, 1984 -2 - 250 Slemr and Seiler, 1984 1.6 - 338 Williams et al., 1988 0 - 36 Colbourn et al., 1987 Weightedyearly average. natural ecosystems, NOflux was related to climatic regions, with highest fluxes occurring in tropical regions. Within a given climatic region, mixed results were observed with regard to the use of vegetative cover as an indicator variable. Patterns in NOflux in the tropical region appeared to be related to vegetative cover; however, in temperate regions, distinct differences as a function of vegetation were not observed. At this time there is not sufficient information to makeany generalizations concerning the utility of using climatic factors or vegetative cover as indicator variables. At all spatial scales statistics are the primarytools used in assessment and characterization of variability. The following discussion presents a brief outline of somebasic statistical considerations for characterizing variability. STATISTICAL CONSIDERATIONS The study of variability provides a mathematical or statistical frameworkthat is useful in elucidating both the interactions involved in controlling soil processes as well as estimating the magnitude of a given microbial process in soils. The function of statistics in this regard is twofold: estimation and comparison. Estimation is concerned with computation of a summary parameter, which indicates central tendency for the population. Typically, this summaryparameter is the mean, which is the center of gravity of the distribution, or the medianthat is the center of probability of the distribution. In addition to calculation of a summaryparameter estimation, some measureof the accuracy of the estimate is often required. This measure of accuracy usually takes the form of a standard deviation or a confidence interval. The issue of comparison, simple stated, is the determination of treatment effects. In controlled experimentation, comparisons are performed to aid in identification of driving variables. Two important benefits are achieved from detailed characterization of variability: (i) developmentof an optimal sampling design, and (ii) application of the appropriate statistical analysis for estimation and comparison procedures. In the development of a sampling design both the sampling pattern (in space or time), as well the sample number requirements must be considered. Both of these elements are impacted by the underlying nature of the variability. Gilbert (1987) provides an extensive discussion of sample design considerations relative to environmentalvariables. The following discussion deals, in greater detail, with the issue of appropriate statistical analysis relative to estimation and comparison problems. Estimation The use of appropriate statistical procedures for estimating the mean, median, and associated confidence intervals is not an issue if the variable understudy is normally distributed, as methodsfor analysis of normal data are well established. However, in many cases microbial processes have been reported to be lognormally distributed (Parkin and Robinson, 1992). If nonnormality exists, the standard procedures for estimation of the mean, median, and associated confidence intervals may be suboptimal. The arithmetic average of the sample values is always PARKIN: MICROBIALPROCESSESIN SOIL an unbiased estimate of the population mean, regardless of the form of the underlying frequency distribution. For asymmetric distributions, upper and lower confidence limits about the mean, if computed at the same probability level, will also be asymmetric. However,the formulaprovided in standard statistical texts for calculating confidence intervals for the meanof a Gaussian variable yields symmetriclimits; thus, the exact probability level of the calculated limits will differ markedly from the nominal alpha level at which the limits were computed (Parkin et al., 1990). Several estimators of the median of lognormally distributed data exist including: the sample median, the geometric mean, and a uniformly minimumvariance unbiased estimator (UMVUE).It has been recommended that "the preferred statistic for summarizingmicrobiological data is the geometric mean"(Greenberg et al., 1985). However, the geometric mean has been reported to be a biased estimator of the population median, with the magnitude of the bias being a function of the number of samples collected (Landwehr, 1978; Blackwood,1992). MonteCarlo evaluation of several estimators of the median and associated confidence limits reveal that the UMVUEor a bias-corrected form of the geometric mean are unbiased estimators of the population median over the range of sample sizes of 4 to 100 (Parkin and Robinson, 1993). Comparison There are two issues related to comparison. The first is what to compare, and the second deals with how to perform the comparison. For skewed data the mean and median occupy different positions on the frequency distribution; thus, a choice exists concerningthe use of the mean or median as the location parameter to use in conducting hypotheses tests. The choice of the appropriate location parameter is critical, as it can affect the conclusions drawnfrom the data. Fewguidelines exist concerning the validity of focusing on the mean or median as a summarystatistic. Often the median or the geometric mean is arbitrarily chosen over the meanbecause of the resistance of these two estimators to the influence of extreme values typically observed with lognormal sample data. This rationale is not alwaysvalid. Selection of the appropriate location parameter is also a function of the nature of the question being asked. The answers to two related questions can help resolve whether the mean or median should be the estimator of choice: (i) Are the sample units themselves of intrinsic interest? and (ii) Is an estimate of the mass of a particular constituent or an average rate of a given transformation required? The issue of the intrinsic significance of the sample unit is of interest because of the impact of sample volume on the value of the median. The central limit theorem predicts that, regardless of the form of the underlying distribution, the distribution of sample meansapproaches normality as the number of samples used in computing the mean increases. Computer simulation experiments indicate that bulking samples drawn from a lognormal distribution results in increasing values of the median (Parkin, 1990). Through the process of averaging, the population of bulked samples approaches symmetry, and the value of the median approachesthe value of the mean. 413 This effect of sample bulking on estimates of the median has been observed in studies of bacterial populations colonizing leaf surfaces (Hirano et al., 1982). In natural systems, if the variable of interest is randomlydispersed, collection of large samples has the sameeffect as bulking or pooling small samples. The value of the population median is, therefore, functionally dependent upon the sample volume collected. This implies that unless the sample itself has significance, the median should not be the estimator of choice. This rationale has been applied in the use of a median estimator for quantification of bacterial populations associated plant canopies and root systems. In a study of epiphytic bacterial populations on leaf surfaces the median was chosen as the relevant summarystatistic (Hirano et al., 1982) because the samples themselves (i.e., the individual plant leaves) had significance. In a similar vein the geometric meanhas also been used as a measure of central tendency for populations of bacteria in the rhizosphere (Loper et al., 1984) of individual plant rhizospheres. In both cases estimators of the median were chosen over the mean because the samples units (i.e., individual leaves or individual plant root systems) had identity and significance. In soil systems, if the sample units have no intrinsic identity, and, in essence, are defined by the size of the samplingtool available to the investigator, the meanshould be used as an estimator of the massof a given constituent or the average transformation rate of a given process. There have been several reports of environmental pollution exhibiting positively skeweddistributions. Gilbert (1987) defines the total massof pollutant at a site as the "’inventory" of the pollutant. If the inventory of the pollutant is the desired summaryvariable, then the median is the wrong estimator of location since it will systematically underestimate the total mass of material at the contaminatedsite. In the area of microbial ecology, often what is desired is an estimate of the magnitudeof a given microbial process in a particular environment. For example, soil denitrification in agricultural systems maybe an important mechanismof fertilizer N loss. Measurements of denitrification exhibit highly skewedfrequency distributions. Since the individual soil samples(soil cores or chambers) have no inherent significance, the median, which estimates the center of probability of the distribution, is not the location parameter of choice. Rather, what is desired is an estimate of N loss from a given field; thus, the mean, whichis an estimator of the center of mass if the distribution, is the appropriate location parameter (Parkin, 1990). Regardless of whether the mean or median is chosen, the appropriate hypothesis testing machinery must be employed in making the comparison. Application of standard analysis of variance procedures requires several assumptions concerning the underlying error structure of the data, and amongthese is the assumption of normality. The effects of violations of the assumption of normality on the efficacy of parametric statistical tests, such as the t-test, are well known(Hey, 1938; Cochran, 1947). Nonnormalitywill influence the ability of a statistical test to perform at the stated alpha level. Nonnormality will also affect the powerof a statistical test to detect differences when real differences in the underlying populations actually exist. Two commonprocedures have 414 J. ENVIRON.QUAL., VOL. 22, JULY-SEPTEMBER 1993 been recommended when the normality assumption has been violated. Theseare: (i) first transform for normality, or (ii) apply nonparametricstatistical methods(Snedecor and Cochran, 1967, p. 123-125). However, the consequences of implementing these two approaches are not typically considered. In the preceding discussion it was observed that with lognormally distributed variables, a choice of location parameters exists, and that the choice of the appropriate location parameter must be consistent with the objectives and methodologies of the problem under study. After the choice of the appropriate location parameter has been made, consideration must be given to the statistical methods used at the hypothesis testing stage. It is imperative that the experimenter whohas to analyze positively skewed data understand what is being compared when In-transformed data or nonparametric procedures are used. This concern was addressed in a recent study on the efficacy of five hypothesis testing techniques applied to samples drawn from two different populations (Parkin, 1993). In this study, five statistical procedures for detecting differences between samples drawn from lognormal populations were evaluated: (i) t-test on untransformeddata, (ii) t-test on In-transformed data, (iii) nonparametric Mann-Whitneytest, (iv) median confidence interval overlap method, and (v) a mean confidence interval overlap method. The tests were evaluated with regard to Type I error rate by comparing batches of samples drawn from the same lognormal population. Also, the powerof the statistical tests to detect differences between two batches of samples drawn from different lognormal populations was evaluated over a range of population variances and sample sizes (n = 4 to 100). It was found that the t-test on In-transformed data, the nonparametric test, and the median confidence interval overlap methodwere only sensitive to sample differences when the underlying populations differed with regard to their medians. The other two tests (t-test on untransformed data and mean confidence limit overlap) were sensitive to differences in population means. It was recommended in this study that the t-test on In-transformed data be used when the median is the location parameter of interest, and the meanconfidence limit overlap method should be used to detect differences in means. IMPLICATIONS FOR MANAGEMENT Howdoes characterization of the spatial variability aid in the development of new management strategies? A detailed characterization of variability results in improved estimation and comparison at all spatial scales. Knowledgeof the factors controlling variability at the microscale will lead to a mechanistic understanding of how microorganisms interact with their environment. Process level information at the microscale is crucial to improving our estimation and predictive capabilities at the larger scales. Understandingmicroscale processes will enhance our ability to manipulate microbial populations and activities in nature. At the plot scale, knowledgeof variability of microbial populations and processes will enable the developmentof better fertilizer-pesticide application strategies. For example, with regard to N fertilization, it may be possible to devise a schemewherethe fertilizer is placed in a zone accessible to plant roots but not susceptible to losses by denitrification or leaching. Also, it is clear that information of plot-scale variability is critical to the problemof estimation and prediction at larger scales. At the landscape scale, knowledgeof the variability may allow for the application of variable management strategies through the use of satellite geopositioning systems (Ambuel et al., 1991). Such systems would allow for the micromanagement of fertility, tillage, or pest control operations based on landscape element. It is clear that the application of new agricultural management strategies such as satellite-based positioning systems require additional information on the spatial variability of microbial processes at the landscape level. An understanding of the regional scale variability of microbial processes is required for the development of meaningful global models as well as for the development of general managementrecommendations. A critical aspect relates to the integration of information from the microscale up to the landscape or regional scale. The integration of results across different scales is often not straightforward. As we move from the smaller scale to the larger scale, predictor variables becomemore integrative in nature. It therefore becomesincreasingly more difficult to identify the key driving variables, primarily because at the larger scales the driving variables are interrelated. At each scale, quantification and characterization of driving variables poses a different set of problems. Microscale studies are difficult to do, but, because by definition they attempt to spatially isolate or identify the driving variables, interpretation of the results of such studies is straightforward. At the plot, scale stochastic models, which account for the variability, maybe a potentially useful tool for constructing relationships between soil variables (Addiscot and Wagenet,1985). Parkin and Robinson (1989) developed a stochastic model predict meandenitrification rate along with an estimate of the associated variability. Geostatistics is another tool that has been applied at the plot scale to perform spatial interpolations. Cokfigingis a geostatistical technique that capitalizes upon the relationship between correlated variables to provide improved predictions at unmeasured points (Vauclin et al., 1983). At the landscape scale, coupling models that contain soil characteristics as driving variables with Geographical Information Systems have been applied. Loague and Green (1988) advanced the concept of using soil taxonomyand soil survey information for development of mapsof pesticide attenuation and retardation factors. Hall and Olson (1991) point out that it is critical in the prediction of variability through the use of landscape elements should be related to surface and subsurface water movement,because water redistribution is the primary causative agent of variability at the landscape scale. In contrast to the use of distinct soil mapunits, a gradient approach for extrapolation over landscapes is another option (Robertson, 1987). This approach involves the use of techniques such as geostatistics to characterize the spatial structure of the variability. In this vein, it has been suggested that geographical referenced data be combined with mechanistic models to produce maps of trace gas flux (Robertson et al., 1989). Advancements in the area of spatial statistics are being continually made. Yates and Yates (1988) proposed the PARKIN: MICROBIAL PROCESSES IN SOIL use of disjunctive kriging as a management decision tool, and used this technique for the development of probability maps of virus concentrations in groundwater. Recently, multivariable indicator kriging, an extension of classical geostatistics, has been proposed as a technique for integrating soil variables for the purpose of producing landscape maps of soil quality (Smith et al., 1993). Many commercial software packages are available that perform geostatistical analyses; however, care must be used in the application of such analyses, because violations of the assumption of second-order stationarity (Journel and Huijbregts, 1978) is often a possibility at the landscape scale. For integration of plot or landscape data to the regional scale, some of the same techniques described above are applicable. Potential exists for classification based on slowly changing topographic and vegetative characteristics (Maison et al, 1989). Then, continuously varying factors such as seasonal climatic changes or anthropogenic factors can be used to modify the empirical relationships between classification type, and the microbial process. Note that this approach will require information about temporal variations as well. This article has not dealt with temporal variations in microbial processes, but it should be recognized that for many microbial processes temporal variation can be several times greater than spatial variation. CONCLUSIONS The major technological advances of the past 50 yr have dramatically influenced agricultural production. Increased cultivation, use of chemical fertilizers and pesticides, monoculture production, and use of higher-yielding varieties have increased crop yields by two- to threefold since the 1940s. However, with this increased production, there have been associated problems, including: erosion of topsoil, increased contamination of ground and surface waters, and loss of productivity in some areas. These adverse effects demand a change from past practices to sustainable practices that maintain or increase production levels, while at the same time preserving the soil base and minimizing adverse environmental effects. In the development of alternative agricultural practices, the role of soil microorganisms cannot be ignored. Microorganisms are key players in nutrient cycling reactions, soil aggregate formation, and crop disease processes. Thus, management and prediction of microbial processes and populations must be an integral part of emerging sustainable agricultural systems. A feature common to measurements of many soil processes is a high degree of associated variability. In this article, an effort has been made to illustrate the importance of characterizing variability with regard to estimation, hypothesis testing, and data interpretation. Also, it should be recognized that variability contains information concerning the underlying factors controlling natural processes. An increased understanding of the factors contributing to the high variability of soil microbial processes (such as the shape of the frequency distributions) should have a direct impact on our understanding of the factors that are important in controlling these processes and our ability to predict these processes. The 415 variance associated with a given process is as much a property of the processes as the mean. Increased concern over the impact of anthropogenic activities on the environment depends on an increased understanding of fundamental mechanisms controlling microbial processes. It is clear that improved estimation through better characterization of variability will aid in this process. However, additional tools are needed, such as techniques for assessing microbial processes at the scale at which they are occurring in soil, as well as methods for determination of biomass of microorganisms carrying out specific reactions in soil. Advancements such as these will lead to better understanding of microbial processes in nature. 416 J. ENVIRON. QUAL., VOL. 22, JULY-SEPTEMBER 1993 STEFAN ET AL.: FUTURE LAKE WATER QUALITY MODELING 417