Quadrat Size Research Articles

Support vector learning with quadratic programming and adaptive step size barrier-projection

Support vector learning with quadratic programming and adaptive step size barrier-projection

Effects of Prevailing Wind Direction on Spatial Statistics of Plant Disease Epidemics

A stochastic simulation model was used to study the effects of the strength of prevailing wind (W), the size/shape (Q) of sampling quadrats and their orientation in relation to the prevailing wind direction (D) on spatial statistics describing plant diseases. Spore dispersal followed a half‐Cauchy distribution with median distance μ, which depended on simulated wind speed. The relationship of spatial autocorrelation at distance k (ρk) to disease incidence (p) and distance was well described by a four‐parameter (α, β1, β2, β3) power‐law model; at a given p, ρk declined exponentially with distance. A total of 35 different quadrat sizes, ranging from 4 to 432 plants, were used to sample the simulated epidemics for estimating intraclass correlation (κ). The κ‐values decreased exponentially with increasing quadrat size; a binary power law model with three parameters (α1, β4, β5) successfully related κ to p. In general, the effect of W and D was greatest on the parameters α, β1, β2 and β3. The effect of W on α, β1, β2 and β3 depended critically on the spatial pattern of initial infected plants (Y); W had greatest effect for the random pattern. In contrast, the main effect of D and its interaction with W on the parameters α, β1, β2 and β3 were large and consistent over different initial conditions. Variations in α1, β4 and β5 were predominantly due to Y and Q. Only for β5 under the clumped pattern was the effect of W very large. For the parameters α1, β4 and β5 there was a large interaction among W, Q and D for the clumped and regular patterns. As expected, in general, the effect of D increased with increasing prevailing wind strength, quadrat size and quadrat length : width ratio. Using square quadrats reduced significantly the effect of W on the parameters α1, β4 and β5; however, the effect of W on β5 was still very large for the clumped pattern. Sampling perpendicular to the prevailing wind direction generally resulted in larger differences in the nine estimated parameters.

Spatial distribution of weeds in arable crops: are current sampling and analytical methods appropriate?

This paper reviews the literature concerning the spatial distribution of weeds; highlighting the limitations of our current sampling and analytical methodologies, and suggesting how these inadequacies can be addressed. Most research studies have used discrete sampling, i.e. weeds are counted within a quadrat, on a grid basis. Few have mapped weeds at a whole‐field scale, either with a resolution appropriate to spraying operations or key ecological processes. Statistical analyses used to describe the data can be divided into two main types, spatially implicit (also at the scale of the sampling unit) or spatially explicit, in which the location of individuals is included in the analyses. Spatially implicit methods can be strongly affected by quadrat size and mean density and are of doubtful benefit. More attention is required to address sampling resolution issues for spatially explicit methods. Our understanding of the formation and dynamics of spatial pattern, as well as predicting the consequences of site‐specific management, can be improved with models. Unfortunately, most models consider only newly expanding patches and appear incapable of predicting spatial distributions when an area has been fully invaded. More detailed biological information is required if models are to become more realistic and informative. We also need to ensure that we understand the spatial processes in the context of the whole field environment, to optimize the success of site‐specific weed management in the longer term.

Analysis of the spatiotemporal structure of rice sheath blight epidemics in a farmer's field

The spatiotemporal structures of two rice sheath blight epidemics were studied during two consecutive (rainy and dry) rice cropping seasons in a farmer's field in the Philippines. The amount of primary inoculum in the soil measured at transplanting was higher during the dry than during the rainy season, and showed nonaggregated spatial patterns in both seasons. Disease incidence (percentage tillers infected per hill) was monitored at weekly intervals on a (14 × 17) grid of rice hills 1 m apart, at the centre of adjacent, nonoverlapping elementary quadrats of 5 × 5 hills. The rate of disease increase at the onset of the epidemics was higher in the dry than in the rainy season, but was afterwards higher in the rainy than in the dry season. A C(α) test was used to test for overdispersion, and the beta‐binomial parameter θ was used to assess disease aggregation in quadrats consisting of one, two, four, eight and 16 elementary quadrats. Aggregation was detectable at all quadrat sizes throughout the epidemic, according to the C(α) test. θ was highest at epidemic onset and at the smallest (elementary) quadrat size, indicating a very strong aggregation of disease at the individual hill scale. θ increased until disease incidence reached 40%, then declined over time. θ values were generally smaller in the dry season. Semivariograms for both epidemics showed strong initial discontinuities, that is, large variance of disease incidence at the elementary quadrat scale. Some aggregation in the spatial distribution of disease incidence was indicated at higher scales in the rainy season, coinciding with the highest rate of disease increase. Semivariograms did not suggest any structure in the dry season, except for one date in the early epidemic stage. Spatiotemporal autocorrelation of disease incidence at 45 days after transplanting yielded very few variables that contributed significantly to describing the disease incidence and its increase in a given hill. In the rainy season these variables were disease incidence (or increase) in the nearest sampled neighbours, and disease incidence (or increase) observed 1 week earlier in the same hill. In the dry season only disease incidence (or increase) observed 1 week earlier in the same hill was significantly correlated with the current incidence (or its increase). The development of a spatial structure in distribution of disease is hypothesized to coincide with the spread of the pathogen from one hill to another, and with the shift from monocyclic to predominantly polycyclic processes in sheath blight epidemics. These results concur with previous reports in suggesting that management of rice sheath blight should first be directed against the leafborne, polycyclic phase of epidemics, rather than their initial phase.

Inventory and Bioindicator Sampling: Testing Pitfall and Winkler Methods with Ants in a South African Savanna

The sampling efficiency and consistency of pitfall traps and Winkler samples for inventory, bioindicator and ecological studies in savanna habitats was compared using ants. Pitfall traps are often used for ant collecting while Winkler litter sampling has until now had rather limited use. We test Winkler sampling for the first time in a South African savanna. Pitfall traps were more efficient and productive than Winkler sampling for epigaeic ants, with a greater total species richness and higher abundance of ants recorded. Winkler samples contributed few additional species. The relative abundance of different sized ants was different with the two collection methods. Winkler sampling was found to catch greater numbers of smaller ants than pitfall trapping, whereas pitfall trapping caught more larger ants. The standard collecting Winkler quadrat size of 1 m2 did not perform as well as 2×\( - \frac{1}{2}\)m2 quadrats combined for one sample.

Effects of Quadrat Size and Time of Year for Sampling of Verticillium dahliae and Lesion Nematodes in Potato Fields

Six potato fields were randomly sampled using quadrat areas ranging from 10 m2 to 4,000 m2. A composite soil sample consisting of 20 soil cores was taken from each of 20 quadrats of each unit area. All samples were assayed for Verticillium dahliae and nematodes. The area in which a composite sample was taken had little effect on estimated mean and variance of V. dahliae until quadrat size was ≥1,000 m2. Then variances dramatically increased with size. With Pratylenchus spp., there was no systematic relationship between quadrat size and variance. Seven additional fields were divided into 2,000 or 4,000 m2 grids, and each grid was sampled in both spring and fall for V. dahliae and plant-parasitic nematodes (Pratylenchus spp. and Meloidogyne hapla). Density of V. dahliae was lower, and in some cases, degree of aggregation was higher, in the spring than in the fall. Sampling for V. dahliae in the fall required 2 to 7 composite samples to estimate mean density within 25% of the true mean, while in the spring, 10 to 19 samples were required. Assays of Pratylenchus spp. required a similar number of samples in spring and fall surveys, while M. hapla required fewer samples with spring than with fall sampling when estimating densities >20 second-stage juveniles per 100 cm3 of soil.

Effects of Quadrat Size and Shape, Initial Epidemic Conditions, and Spore Dispersal Gradient on Spatial Statistics of Plant Disease Epidemics

ABSTRACT The spatiotemporal spread of plant diseases was simulated using a stochastic model to study the effects of initial conditions (number of plants initially infected and their spatial pattern), spore dispersal gradient, and size and shape of sampling quadrats on statistics describing the spatiotemporal dynamics of epidemics. The spatial spread of disease was simulated using a half-Cauchy distribution with median dispersal distance mu (units of distance). A total of 54 different quadrat types, including 23 distinct sizes ranging from 4 to 144 plants, were used to sample the simulated epidemics. A symmetric form of the binary power law with two parameters (alpha, beta) was fitted to the sampled epidemic data using each of the 54 quadrats for each replicate simulation run. The alpha and beta estimates were highly correlated positively with each other, and their estimates were comparable to those estimated from observed epidemics. Intraclass correlation (kappa) was calculated for each quadrat type; kappa decreased exponentially with increasing quadrat size. An asymmetric form of the binary power law with three parameters (alpha (1), beta(1), beta(2)) was used to relate kappa to the disease incidence (p); beta1 was highly correlated to beta: beta1 approximately beta - 1. In general, initial conditions and quadrat size affected alpha, beta, alpha(1), beta(1), and beta(2) greatly. The parameter estimates increased as quadrat size increased, and the relationships were described well by a linear regression model on the logarithm of quadrat size with the slope or intercept parameters dependent on initial conditions and mu. Compared with initial conditions and quadrat size, the overall effects of mu and quadrat shape were generally small, although within each quadrat size and initial condition they could be substantial. Quadrat shape had the greatest effect when the quadrat was long and thin. The relationship of the index of dispersion (D) to p and quadrat size was determined from the alpha and beta estimates. D was greatest when p was 0.5 and decreased when p approached 0 or 1. It increased with quadrat size and the rate of the increase was maximum when p was 0.5 and decreased when p approached 0 or 1.

A comparison of sampling designs in a Hainan tropical rain forest

Different sampling strategies are simulated by changing quadrat size, quadrat shape, sample size and the arrangement of quadrats in a tropical rain forest of Hainan (South China). The simulation uses enumeration data of trees, and derived variables such as species richness, species importance, and species population density, to compare the efficiency of the sampling. The results verify that greater sampling efficiency is to be expected using systematic sampling than random sampling. Quadrat size has substantial influence on parameter estimation, but quadrat shape has negligible effect except when the quadrat is extremely long and narrow.

Relationships Between Several Quadrat-Based Statistical Measures Used to Characterize Spatial Aspects of Disease Incidence Data

ABSTRACT This article investigates the relationships between various statistical measures that are used to summarize spatial aspects of disease incidence data. The focus is on quadrat data in which each plant in a quadrat is classified as diseased or healthy. We show that spatial autocorrelation plays a central role via the mean intraclass correlation, rho, which is defined as the average correlation of the disease status of all pairs of plants within the quadrat. The value of rho determines the variance of the number of infected plants in the quadrat and, if this variable follows a beta-binomial distribution, the heterogeneity parameter of the beta-binomial distribution is directly related to the mean intraclass correlation. We consider in detail a model in which the spatial autocorrelation depends only on the distance between the plants. For illustration, we consider a specific autocorrelation model that was derived from simulated data. We show that this model leads, approximately, to the binary form of the power law relating the variance of the number of infected plants per quadrat to the mean. Using an approximation technique, we then show how the index of dispersion is related to quadrat size and shape. The index of dispersion increases with quadrat size. The rate of increase is dependent on quadrat shape, but the effect of quadrat shape is small in comparison to the effect of quadrat size. Finally, we note that if the spatial autocorrelation depends on the relative orientation of the plants, as well as the distance between them, there are connections with distance class methods.

Geostatistical Estimation Variance for the Spatial Mean in Two-Dimensional Systematic Sampling

Many ecologists use two-dimensional systematic sampling to estimate mean density of individuals over the domain sampled. They usually calculate the variance of the mean as if the sample were a simple random sample, using the unbiased estimator under this sampling design, that is, σ2/n. This practice leads to a selection bias, i.e., incorrect inclusion probabilities of population units in the sample are used in the estimator of variance of the mean. The magnitude of the bias varies with the underlying spatial autocorrelation structure. Design-based inference and model-based inference are two conceptual frameworks for tackling estimation of variance of the mean in a systematic sample. This paper reports use of the geostatistical estimation variance σ2E with the model-based approach. This variance of the overall spatial mean depends on the spatial autocorrelation structure of the data. We illustrate the method by considering the density of acorns fallen under a sessile oak during one season. Acorns were numbered in square quadrats of 0.25 m2; the data set was exhaustive. We drew nine one-start systematic samples from the whole population of quadrats. We computed semivariograms for the whole population and each sample and fitted them to exponential models without nugget. Using geostatistical theoretical results, we calculated a variance of the mean density of acorns by Monte Carlo integration. We show that our variance estimate depends on (1) the origin of the systematic sample, the central sample being the most accurate, (2) quadrat size, with a decrease in the variance when quadrat size increases, (3) the semivariogram model, (4) discretization of the domain used in Monte Carlo integration, and (5) the random number generator. Considering all sources of variation, the variance estimate we calculated ranged from ∼2 to 36 in our example, for an overall mean equal to 95 acorns per quadrat. Geostatistical variance mainly reflects the locations and size of the sampled quadrats, and the spatial autocorrelation function. In our example, using the variance estimator of a simple random sample leads to a high selection bias. The bias can be neglected only in the absence of significant spatial autocorrelation in the sample. Otherwise, we advocate a geostatistical model-based approach that accounts for spatial autocorrelation.

GEOSTATISTICAL ESTIMATION VARIANCE FOR THE SPATIAL MEAN IN TWO-DIMENSIONAL SYSTEMATIC SAMPLING

Many ecologists use two-dimensional systematic sampling to estimate mean density of individuals over the domain sampled. They usually calculate the variance of the mean as if the sample were a simple random sample, using the unbiased estimator under this sampling design, that is, o2/n. This practice leads to a selection bias, i.e., incorrect inclusion probabilities of population units in the sample are used in the estimator of variance of the mean. The magnitude of the bias varies with the underlying spatial autocorrelation structure. Design-based inference and model-based inference are two conceptual frameworks for tackling estimation of variance of the mean in a systematic sample. This paper reports use of the geostatistical estimation variance cr2 with the model-based approach. This variance of the overall spatial mean depends on the spatial autocorrelation structure of the data. We illustrate the method by considering the density of acorns fallen under a sessile oak during one season. Acorns were numbered in square quadrats of 0.25 m2; the data set was ex- haustive. We drew nine one-start systematic samples from the whole population of quadrats. We computed semivariograms for the whole population and each sample and fitted them to exponential models without nugget. Using geostatistical theoretical results, we calculated a variance of the mean density of acorns by Monte Carlo integration. We show that our variance estimate depends on (1) the origin of the systematic sample, the central sample being the most accurate, (2) quadrat size, with a decrease in the variance when quadrat size increases, (3) the semivariogram model, (4) discretization of the domain used in Monte Carlo integration, and (5) the random number generator. Considering all sources of variation, the variance estimate we calculated ranged from -2 to 36 in our example, for an overall mean equal to 95 acorns per quadrat. Geostatistical variance mainly reflects the locations and size of the sampled quadrats, and the spatial autocorrelation function.

Modern pollen-representation of some boreal species on islands in a large lake in Canada

Studies of pollen source areas of closed-canopy sites are contradictory. Some authors found that closed-canopy sites mainly collect local pollen while others found more distant sources. This dichotomy might stem from the use of canopies of varying degrees of closure, and from variations in the pollen productivity of the local vegetation and the background pollen rain. Here, 30 islands were used to evaluate the pollen sources of closed-canopy sites. We compared pollen with the forest inventory in three quadrat sizes: 100, 400 m 2 and on the whole island. Regression analyses showed that most pollen of Picea spp., Pinus spp., and Betula spp. comes from within the 400 m 2 quadrat. Abies balsamea and Thuja occidentalis showed no relationship with vegetation in any of the quadrats considered, suggesting a more regional source. Insularity and island size are important factors influencing the pollen source area; correlations were stronger on islands located 1200 m from the nearest shore and on islands >50 ha. These results suggest that closed-canopy sites on islands may be useful in stand-level vegetation history reconstruction through pollen analysis, but that caution must be exercised in separating the local and regional signals.

A HEMISPHERIC ASSESSMENT OF SCALE DEPENDENCE IN LATITUDINAL GRADIENTS OF SPECIES RICHNESS

Considerable controversy surrounds the importance of historical, evolutionary, and ecological factors affecting continental patterns in species richness. Although the importance of area and latitude are both well documented, few attempts have been made to integrate their effects in a single model. Most studies have been conducted by superimposing grids on equal-area projection maps and counting the number of species occurring within grid cells (i.e., quadrats). Unfortunately, different grid-based studies use different quadrat sizes, making comparisons tenuous. We developed a hierarchical model to evaluate the degree to which area (based on a nested series of quadrats of five sizes, 1000–25000 km2) affects the latitudinal gradient in species richness. The model allows the relationship between latitude and area to be nonlinear and, in its simple form, evaluates how well species richness can be predicted by the additive influences of latitude and area. The complex model evaluates whether an area × latitude interaction accounts for significant additional variation in species richness above that in the simple model (i.e., assesses the scale dependence of the latitudinal gradient). For bats and marsupials, the simple model included only latitudinal effects and accounted for over half of the variation in species richness of each of the two taxa. The interactive effect was nonsignificant for each taxon, accounting for <0.1% of additional variation in species richness in each case. If other taxa or land masses produce similar relationships, then the form of the latitudinal gradient is relatively invariant with respect to area at 1000–25000 km2 scales, and comparisons among studies at this spatial scale are straightforward.

A Hemispheric Assessment of Scale Dependence in Latitudinal Gradients of Species Richness

Considerable controversy surrounds the importance of historical, evolutionary, and ecological factors affecting continental patterns in species richness. Although the importance of area and latitude are both well documented, few attempts have been made to integrate their effects in a single model. Most studies have been conducted by superimposing grids on equal-area projection maps and counting the number of species occurring within grid cells (i.e., quadrats). Unfortunately, different grid-based studies use different quadrat sizes, making comparisons tenuous. We developed a hierarchical model to evaluate the degree to which area (based on a nested series of quadrats of five sizes, 1000–25000 km2) affects the latitudinal gradient in species richness. The model allows the relationship between latitude and area to be nonlinear and, in its simple form, evaluates how well species richness can be predicted by the additive influences of latitude and area. The complex model evaluates whether an area × latitude interaction accounts for significant additional variation in species richness above that in the simple model (i.e., assesses the scale dependence of the latitudinal gradient). For bats and marsupials, the simple model included only latitudinal effects and accounted for over half of the variation in species richness of each of the two taxa. The interactive effect was nonsignificant for each taxon, accounting for <0.1% of additional variation in species richness in each case. If other taxa or land masses produce similar relationships, then the form of the latitudinal gradient is relatively invariant with respect to area at 1000–25000 km2 scales, and comparisons among studies at this spatial scale are straightforward.

Influence of interatomic correlation effects on short-range order in hexagonal close-packed polycrystalline alloys

The first x-ray diffraction procedure suitable for the investigation of short-range order in polycrystalline alloys having a hexagonal close-packed (hcp) lattice is developed on the basis of the theory proposed by M. A. Krivoglaz for diffuse x-ray scattering by hcp single-crystal alloys. This procedure takes into account specific details of the crystal structure of the hcp lattice, in particular, the presence of two atoms in the unit cell and the close radii of the individual coordination spheres. The realistic character of the procedure is demonstrated experimentally in the example of hcp Mg-Dy and Mg-Tb alloys. The new procedure is used to calculate, for the first time, the modulating functions of linear and quadratic size effects in Mg-Dy and Mg-Tb alloys. It is shown that the size-effect modulating functions for coordination spheres with close radii have different characters, and their inclusion in diffuse scattering sets the stage for solving the problem of calculating the short-range order parameters on these spheres.

Comparative Ordination of Low Arctic Vegetation Recovering from Disturbance: Reconciling Two Contrasting Approaches for Field Data Collection

Multivariate statistics are applied to data on low arctic flora and vegetation in an effort to compare and interpret information derived from contrasting field methods normally considered to be incompatible. Data were collected using two of the most widespread field techniques currently in use in the circumpolar North: (1) standardized sampling, using a point frame-based protocol; and (2) more traditional phytosociological methods. The comparison is based on simultaneous objective (quantitative) and subjective (qualitative) sampling of identical stands of vegetation at and north of the latitudinal treeline in the Yamal Region of Northwest Siberia. The species composition of vegetation on anthropogenic primary surfaces is emphasized, but undisturbed tundra was also sampled. One hundred and sixty-five quadrats (1 × 1 m) were analyzed together with 33 relevés (5 × 5 m) using detrended correspondence analysis (DCA). The ordination results revealed very good correlation within anthropogenic vegetation stands. However, in ordination diagrams derived from samples within undisturbed tundra, samples were clearly separated based on quadrat size because many non vascular taxa were missed in the large quadrats.

Comparative Ordination of Low Arctic Vegetation Recovering from Disturbance: Reconciling Two Contrasting Approaches for Field Data Collection

Multivariate statistics are applied to data on low arctic flora and vegetation in an effort to compare and interpret information derived from contrasting field methods normally considered to be incompatible. Data were collected using two of the most widespread field techniques currently in use in the circumpolar North: (1) standardized sampling, using a point frame-based protocol; and (2) more traditional phytosociological methods. The comparison is based on simultaneous objective (quantitative) and subjective (qualitative) sampling of identical stands of vegetation at and north of the latitudinal treeline in the Yamal Region of Northwest Siberia. The species composition of vegetation on anthropogenic primary surfaces is emphasized, but undisturbed tundra was also sampled. One hundred and sixty-five quadrats (1 × 1 m) were analyzed together with 33 relevés (5 × 5 m) using detrended correspondence analysis (DCA). The ordination results revealed very good correlation within anthropogenic vegetation stands. However, in ordination diagrams derived from samples within undisturbed tundra, samples were clearly separated based on quadrat size because many non vascular taxa were missed in the large quadrats.

The effect of overstorey proteas on plant species richness in South African mountain fynbos

SummaryTwo South African mountain fynbos sites, similar in drainage, elevation, slope angle, slope aspect and soil type but with differing fire histories, were studied to measure how the effect of high densities of overstorey proteas in one fire cycle affects the α‐diversity levels of the plant community in the following fire‐cycle, how their repeated absence due to several short fire‐cycles affects their species richness and finally, at what spatial scale such patterns are most appropriately measured. High prefire canopy cover percentages and densities of overstorey proteas increase the postfire α‐diversity of understorey species. In addition, the increase in species richness observed occurred for all higher plant life history types present. At sites where one or more short fire cycles resulted in the repeated absence of overstorey proteas, the number of plant species present in the understorey was lower than at a site where overstorey proteas persisted. These results are dependent on the spatial scale at which the α‐diversity of understorey species is measured. At small quadrat sizes (&lt; 5 m2), overstorey proteas decrease the number of understorey species present, while at larger quadrat sizes (100 m2) higher species richness is observed. The contradiction in conclusions when α‐diversity is measured at different spatial scales can be attributed to the patchiness of fynbos communities. Overstorey proteas play an important role in maintaining the patchiness component of fynbos communities by diminishing the effect of understorey resprouting species, making available regeneration niches for the maintenance of plant species richness. Where small quadrats are used, the effect of patchiness on the dynamics of the mountain fynbos community is lost. Thus, it is the fire history prior to the last fire and how it affects overstorey proteas that is important in the determination of α‐diversity levels in mountain fynbos plant communities.

Quadrat Sample Precision and Cost with a High-vacuum Insect Sampling Machine in Cotton Ecosystems

Accurate and reliable insect samplers must be used in an effective manner if predators are to be used in the integrated pest management (IPM) decision-making process for complex ecosystems; therefore, optimum sampling procedures were evaluated for a high-vacuum insect sampling machine (Insectavac) for use in cotton, Gossypium hirsutum L. A total of 15 adult and immature insect groups was collected from 3 cotton fields, 12 times over a 3-yr period with an Insectavac. These data were used to evaluate the effect of quadrat size and number on the cost and precision of sampling in a cotton ecosystem. We sampled 340, 30.5-m quadrats end to end and 250, 61.0-m quadrats were sampled in a randomized complete block design over a 3-yr period. Data were analyzed using analysis of variance (ANOVA) and jackknife procedures. Variances obtained from the jackknife procedure were essentially the same as those obtained from the ANOVA. We concluded that sampling biases caused by patchy or clumped distributions of insects were not significant and the resulting estimates were unbiased. Because the insect populations were not significantly clumped, nothing was gained by the use of the jackknife procedure. Large quadrats resulted in significantly greater sampling variability because the extreme values from uneven distributions were higher than with smaller quadrat sizes. Advantages of small quadrats include a greater number of quadrats sampled for similar labor input which results in more degrees of freedom, better coverage of habitats within a field, and better representation of average insect densities.

Scale of Disturbance and the Structure of a Temperate Fish Guild

Although it is acknowledged that the spatial scale of sampling and experimentation is important in assessing the degree of stability of a community, multiple scales are rarely incorporated into the same study. We compared the persistence and resilience of a temperate blennioid fish guild at three scales spanning an order of magnitude. Replicate permanent square quadrats of three sizes (4 m2, 25 m2, 100 m2) were delimited within contiguous tracts of reef in two habitats: broken reef and macroalgal forest. Two forms of disturbance were applied to the treatment quadrats. Direct disturbance (fish removal) was applied to treatments in the broken reef habitat. Indirect disturbance (macroalgal removal) was applied to treatments in the macroalgal forest. Both experiments identified several scale-dependent effects. Large quadrats were less variable than smaller quadrats, probably due to reduced edge effects due to movement of fish in and out of the quadrats. The smallest quadrats did not adequately represent the available substratum types within the reef zone. This interacted with species-specific behavioral factors such as mobility and habitat association to consistently bias the blennioid composition within the smallest quadrats toward dominance of the most sedentary blennioid species. Fish responded to disturbance in different ways depending on the stage of their life cycle. Adult fish responded at scales within their motility range of only a few meters. In contrast, newly settling fish responded to the habitat at larger scales. Quadrat size also determined which assemblage dynamics could be measured. Within-year changes in guild composition due to species-specific settlement timing and survivorship could be measured only at larger scales. We conclude that different scales of monitoring and manipulation will determine what can be perceived by the researcher as well as how the organism will respond to disturbance.