The application of generalized linear mixed models to multi-level sampling for insect population monitoring

Iwao's quadratic regression or Taylor's Power Law (TPL) are commonly used to model the variance as a function of the mean for sample counts of insect populations which exhibit spatial aggregation. The modeled variance and distribution of the mean are typically used in pest management programs to decide if the population is above the action threshold in any management unit (MU) (e.g., orchard, forest compartment). For nested or multi-level sampling the usual two-stage modeling procedure first obtains the sample variance for each MU and sampling level using ANOVA and then fits a regression of variance on the mean for each level using either Iwao or TPL variance models. Here this approach is compared to the single-stage procedure of fitting a generalized linear mixed model (GLMM) directly to the count data with both approaches demonstrated using 2-level sampling. GLMMs and additive GLMMs (AGLMMs) with conditional Poisson variance function as well as the extension to the negative binomial are described. Generalization to more than two sampling levels is outlined. Formulae for calculating optimal relative sample sizes (ORSS) and the operating characteristic curve for the control decision are given for each model. The ORSS are independent of the mean in the case of the AGLMMs. The application described is estimation of the variance of the mean number of leaves per shoot occupied by immature stages of a defoliator of eucalypts, the Tasmanian Eucalyptus leaf beetle, based on a sample of trees within plots from each forest compartment. Historical population monitoring data were fitted using the above approaches.     

Effects of sampling error and temporal correlations in population growth on process variance estimators

Estimates of a population’s growth rate and process variance from time-series data are often used to calculate risk metrics such as the probability of quasi-extinction, but temporal correlations in the data from sampling error, intrinsic population factors, or environmental conditions can bias process variance estimators and detrimentally affect risk predictions. It has been claimed (McNamara and Harding, Ecol Lett 7:16–20, 2004) that estimates of the long-term variance that incorporate observed temporal correlations in population growth are unaffected by sampling error; however, no estimation procedures were proposed for time-series data. We develop a suite of such long-term variance estimators, and use simulated data with temporally autocorrelated population growth and sampling error to evaluate their performance. In some cases, we get nearly unbiased long-term variance estimates despite ignoring sampling error, but the utility of these estimators is questionable because of large estimation uncertainty and difficulties in estimating correlation structure in practice. Process variance estimators that ignored temporal correlations generally gave more precise estimates of the variability in population growth and of the probability of quasi-extinction. We also found that the estimation of probability of quasi-extinction was greatly improved when quasi-extinction thresholds were set relatively close to population levels. Because of precision concerns, we recommend using simple models for risk estimates despite potential biases, and limiting inference to quantifying relative risk; e.g., changes in risk over time for a single population or comparative risk among populations.     

Sensitivity analysis of North American bird population estimates

The Partners in Flight North American Landbird Conservation Plan provided estimates of population sizes for 448 landbird species using a multiplicative model. Input parameters in this calculation included the area of state × Bird Conservation Region polygons, area-specific mean Breeding Bird Survey counts circa 1995, and adjustment factors for the distance over which species may presumably be correctly counted, the assumed pairing of singing males with non-singing females, and variability in the propensity of birds to sing over the course of the survey day. I assessed the sensitivity of this population calculation to changes in the input parameters. I assessed both local and global sensitivity of the model to changes in the parameters with Monte Carlo one-at-a-time simulations and the Fourier amplitude sensitivity test (FAST). Monte Carlo simulations were an estimate of local model sensitivity whereas FAST estimated global model sensitivity, accommodating the potential shared variance between model parameters. Monte Carlo simulations suggested population estimates were 39% more sensitive to changes in the detection distance adjustment than to the other parameters; the other parameters were nearly equal in their contribution to model sensitivity. Conversely, FAST analysis determined that each of the input variables aside from the pair adjustment provided roughly equal contributions to variability in population estimates. The most efficient means for improving continental population estimates for birds surveyed by the Breeding Bird Survey will be through increased scrutiny of the species-specific distance detection and time-of-day adjustments and improved understanding in the spatial and temporal variability in the mean Breeding Bird Survey count.     

Quasi-Poisson vs. negative binomial regression: how should we model overdispersed count data?

Quasi-Poisson and negative binomial regression models have equal numbers of parameters, and either could be used for overdispersed count data. While they often give similar results, there can be striking differences in estimating the effects of covariates. We explain when and why such differences occur. The variance of a quasi-Poisson model is a linear function of the mean while the variance of a negative binomial model is a quadratic function of the mean. These variance relationships affect the weights in the iteratively weighted least-squares algorithm of fitting models to data. Because the variance is a function of the mean, large and small counts get weighted differently in quasi-Poisson and negative binomial regression. We provide an example using harbor seal counts from aerial surveys. These counts are affected by date, time of day, and time relative to low tide. We present results on a data set that showed a dramatic difference on estimating abundance of harbor seals when using quasi-Poisson vs. negative binomial regression. This difference is described and explained in light of the different weighting used in each regression method. A general understanding of weighting can help ecologists choose between these two methods.     

Estimating Population Size with Noninvasive Capture-Mark-Recapture Data

Abstract:  Estimating population size of elusive and rare species is challenging. The difficulties in catching such species has triggered the use of samples collected noninvasively, such as feces or hair, from which genetic analysis yields data similar to capture-mark-recapture (CMR) data. There are, however, two differences between classical CMR and noninvasive CMR. First, capture and recapture data are gathered over multiple sampling sessions in classical CMR, whereas in noninvasive CMR they can be obtained from a single sampling session. Second, because of genotyping errors and unlike classical CMR, there is no simple relationship between (genetic) marks and individuals in noninvasive CMR. We evaluated, through simulations, the reliability of population size estimates based on noninvasive CMR. For equal sampling efforts, we compared estimates of population size N obtained from accumulation curves, a maximum likelihood, and a Bayesian estimator. For a closed population and without sampling heterogeneity, estimates obtained from noninvasive CMR were as reliable as estimates from classical CMR. The sampling structure (single or multiple session) did not alter the results, the Bayesian estimator in the case of a single sampling session presented the best compromise between low mean squared error and a 95% confidence interval encompassing the parametric value of N in most simulations. Finally, when suitable field and lab protocols were used, genotyping errors did not substantially bias population size estimates (bias < 3.5% in all simulations). The ability to reliably estimate population size from noninvasive samples taken during a single session offers a new and useful technique for the management and conservation of elusive and rare species.     

Inference about the ratio of means from Negative Binomial paired count data

We derive some statistical properties of the distribution of two Negative Binomial random variables conditional on their total. This type of model can be appropriate for paired count data with Poisson over-dispersion such that the variance is a quadratic function of the mean. This statistical model is appropriate in many ecological applications including comparative fishing studies of two vessels and or gears. The parameter of interest is the ratio of pair means. We show that the conditional means and variances are different from the more commonly used Binomial model with variance adjusted for over-dispersion, or the Beta-Binomial model. The conditional Negative Binomial model is complicated because it does not eliminate nuisance parameters like in the Poisson case. Maximum likelihood estimation with the unconditional Negative Binomial model can result in biased estimates of the over-dispersion parameter and poor confidence intervals for the ratio of means when there are many nuisance parameters. We propose three approaches to deal with nuisance parameters in the conditional Negative Binomial model. We also study a random effects Binomial model for this type of data, and we develop an adjustment to the full-sample Negative Binomial profile likelihood to reduce the bias caused by nuisance parameters. We use simulations with these methods to examine bias, precision, and accuracy of estimators and confidence intervals. We conclude that the maximum likelihood method based on the full-sample Negative Binomial adjusted profile likelihood produces the best statistical inferences for the ratio of means when paired counts have Negative Binomial distributions. However, when there is uncertainty about the type of Poisson over-dispersion then a Binomial random effects model is a good choice.     

Including model uncertainty in estimating variances in multiple capture studies

We present bootstrap-based methods which incorporate model uncertainty in estimating variances in multiple capture studies. Each of our three methods has a specific set of properties, and we discuss when each method should be used. Our first method can be used in any multiple capture setting, but it gives an estimate of the variance conditional on the number of observed animals. Our other two methods yield estimates of the unconditional variance; they require good estimates of part or all of the specific probability model, respectively. Smoothed estimated cell probabilities are utilized by the latter method. We contrast the three methods on a real-life data set, and then conduct simulations for a simple setting. Finally, we detail the use of our methodology for specific settings and discuss adaptations for tag-return studies.     

Hierarchical spatial modeling for estimation of population size

Estimation of population size has traditionally been viewed from a finite population sampling perspective. Typically, the objective is to obtain an estimate of the total population count of individuals within some region. Often, some stratification scheme is used to estimate counts on subregions, whereby the total count is obtained by aggregation with weights, say, proportional to the areas of the subregions. We offer an alternative to the finite population sampling approach for estimating population size. The method does not require that the subregions on which counts are available form a complete partition of the region of interest. In fact, we envision counts coming from areal units that are small relative to the entire study region and that the total area sampled is a very small proportion of the total study area. In extrapolating to the entire region, we might benefit from assuming that there is spatial structure to the counts. We implement this by modeling the intensity surface as a realization from a spatially correlated random process. In the case of multiple population or species counts, we use the linear model of coregionalization to specify a multivariate process which provides associated intensity surfaces hence association between counts within and across areal units. We illustrate the method of population size estimation with simulated data and with tree counts from a Southwestern pinyon-juniper woodland data set.     

An EM algorithm for capture-recapture estimation

Analysis of capture—recapture data often involves maximizing a complex likelihood function with many unknown parameters. Statistical inference based on selection of a proper model depends on successful attainment of this maximum. An EM algorithm is developed for obtaining maximum likelihood estimates of capture and survival probabilities conditional on first capture from standard capture—recapture data. The algorithm does not require the use of numerical derivatives which may improve precision and stability relative to other estimation schemes. The asymptotic covariance matrix of the estimated parameters can be obtained using the supplemented EM algorithm. The EM algorithm is compared to a more traditional Newton-Raphson algorithm with both a simulated and a real dataset. The two algorithms result in the same parameter estimates, but Newton-Raphson variance estimates depend on a numerically estimated Hessian matrix that is sensitive to step size choice.     

Combining demographic and count-based approaches to identify source-sink dynamics of a threatened seabird.

Identifying source-sink dynamics is of fundamental importance for conservation but is often limited by an inability to determine how immigration and emigration influence population processes. We demonstrate two ways to assess the role of immigration on population processes without directly observing individuals dispersing from one population to another and apply these methods to a population of Marbled Murrelets (Brachyramphus marmoratus) in California (USA). In the first method, the rate of immigration (i) is estimated by subtracting local recruitment (recruitment from within the population due to reproduction) estimated with demographic data from total recruitment (f; recruitment from within the population plus recruitment from other populations) estimated using temporal symmetry mark-recapture models developed by R. Pradel. The second method compares population growth rates estimated with temporal symmetry models (lambdaTS) and/or population growth rates estimated from counts of individuals over multiple sampling periods (lambdaC) with growth estimates from a stage-structured projection matrix model (lambdaM). Both lambdaTS and lambdaC incorporate all demographic processes affecting population change (birth, death, immigration, and emigration), whereas matrix models are usually constructed without incorporating immigration. Thus, if lambdaTS and lambdaC are > or = 1 and lambdaM < 1, the population is sustained by immigration and is considered to be a sink. Using the first method, recruitment estimated with temporal symmetry models was high (f= 0.182, SE = 0.058), the mean adult birth rate, as estimated using the ratio of juveniles to > or = 1 year old individuals (observed during ship-based surveys) was low (bA = 0.039, SE = 0.014), and immigration was 0.160 (SE = 0.057). Using the second method, murrelet numbers in central California were stable (lambdaC = 1.058, SE = 0.047; lambdaTS = 1.064, SE = 0.033), but were projected to decline 9.5% annually in the absence of immigration (lambdaM = 0.905, SE = 0.053). Our results suggest that Marbled Murrelets in central California represent a sink population that is stable but would decline in the absence of immigration from larger populations to the north. However, the extent to which modeled immigration is due to permanent recruitment or temporarily dispersing individuals that simply mask population declines is uncertain.     

Estimation of population parameters from aerial counts of North American mallards: a cautionary tale.

The accuracy of population estimates strongly interferes with our ability to obtain unbiased estimates of population parameters based on analyses of time series of population fluctuations. Here we use long-term data on fluctuations in the size of Mallard populations collected as part of the May Breeding Waterfowl Survey covering a large section of North America. We assume a log-linear model of density dependence and use a hierarchical Bayesian state-space approach in which all parameters are assumed to be realizations from a common underlying distribution. Thus, parameters for different populations are not allowed to vary independently of each other. We then simulated independent time series of aerial counts, using the estimated parameters and adding various levels of observation error. These simulations showed that the estimates of stochastic population growth rate and strength of density dependence were biased even when moderate sampling errors were present. In contrast, the estimates of the environmental stochasticity and the carrying capacity were unbiased even for short time series and large observation error. Our results underline the importance of reducing the magnitude of sampling error in the design of large-scale monitoring programs of population fluctuations.     

Design-based estimators and power analyses of trend tests for the proportion of fish that exhibit gross pathological disorders

Methods for estimating the proportion of fish that exhibit gross pathological disorders and for estimating the variance of these estimates are defined. The methods are for the situation in which a probability-based sampling design is used to collect fish for examination, but geographic locations (rather than individual fish) are assigned probabilities of being selected for sampling. To illustrate the use of the methods, they are applied to data collected during the 1992 EMAP- Estuaries sampling program in the Louisianian Province (i.e., the Gulf of Mexico). Separate estimates of the proportion of fish with gross pathological disorders are computed for demersal species, commercial species, pelagic species, and all species as one group. In addition, a test for trend in the proportion of fish that exhibit gross pathological disorders is defined, and analyses of the power of the test are presented. The power analyses are based on a general underlying model of the random distribution patterns of fish and the random process of catching fish. The power analyses also take into account the features of the sampling designs used for collecting fish. Component parameter estimates were computed using data from the 1992 EMAP-Estuaries sampling program in the Louisianian Province. Results from these analyses suggest that the EMAP-Estuaries sampling designs are capable of detecting a 0.15% change per year in the proportion of fish (all species groups combined) with gross pathological disorders in estuaries of the Louisianian Province over a 12-year period with a power of at least 80%. © Rapid Science 1998     

Traditional occupancy–abundance models are inadequate for zero-inflated ecological count data

Traditional occupancy–abundance and abundance–variance–occupancy models do not take into account zero-inflation, which occurs when sampling rare species or in correlated counts arising from repeated measures. In this paper we propose a novel approach extending occupancy–abundance relationships to zero-inflated count data. This approach involves three steps: (1) selecting distributional assumptions and parsimonious models for the count data, (2) estimating abundance, occupancy and variance parameters as functions of site- and/or time-specific covariates, and (3) modelling the occupancy–abundance relationship using the parameters estimated in step 2. Five count datasets were used for comparing standard Poisson and negative binomial distribution (NBD) occupancy–abundance models. Zero-inflated Poisson (ZIP) and zero-inflated negative binomial (ZINB) occupancy–abundance models were introduced for the first time, and these were compared with the Poisson, NBD, He and Gaston's and Wilson and Room's abundance–variance–occupancy models. The percentage of zero counts ranged from 45 to 80% in the datasets analysed. For most of the datasets, the ZINB occupancy–abundance model performed better than the traditional Poisson, NBD and Wilson and Room's model. He and Gaston's model performed better than the ZINB in two out of the five datasets. However, the occupancy predicted by all models increased faster than the observed as density increased resulting in significant mismatch at the highest densities. Limitations of the various models are discussed, and the need for careful choice of count distributions and predictors in estimating abundance and occupancy parameter are indicated.     

Relative Contributions of Sampling Error in Initial Population Size and Vital Rates to Outcomes of Population Viability Analysis

Abstract:  We evaluated the relative contributions of sampling error (randomly chosen standard errors applied as 0–30% of parameter estimates) in initial population size and vital rates (survival and reproduction) to the outcome of a simulated population viability analysis for grizzly bears (  Ursus arctos ). Error in initial population size accounted for the largest source of variation (model II analysis of variance, F _25,5= 10.8, p = 0.00001) in simulation outcomes, explaining 60.5% of the variance. In contrast, error in vital rates contributed little to simulation outcomes ( F _25,5= 0.61, p = 0.70), accounting for only 2.4% of model variation. Reduced global variation in vital rates, as a result of independent random sampling of annual deviates for each parameter, likely contributed to the results. Errors in estimates of initial population size, if ignored in PVA, have the potential to leave managers with estimates of population persistence that are of little value for making management decisions.     

Inference about density and temporary emigration in unmarked populations

Few species are distributed uniformly in space, and populations of mobile organisms are rarely closed with respect to movement, yet many models of density rely upon these assumptions. We present a hierarchical model allowing inference about the density of unmarked populations subject to temporary emigration and imperfect detection. The model can be fit to data collected using a variety of standard survey methods such as repeated point counts in which removal sampling, double-observer sampling, or distance sampling is used during each count. Simulation studies demonstrated that parameter estimators are unbiased when temporary emigration is either "completely random" or is determined by the size and location of home ranges relative to survey points. We also applied the model to repeated removal sampling data collected on Chestnut-sided Warblers (Dendroica pensylvancia) in the White Mountain National Forest, U.S.A. The density estimate from our model, 1.09 birds/ha, was similar to an estimate of 1.11 birds/ha produced by an intensive spot-mapping effort. Our model is also applicable when processes other than temporary emigration affect the probability of being available for detection, such as in studies using cue counts. Functions to implement the model have been added to the R package unmarked.     

A power analysis for multivariate tests of temporal trend in species composition

Long-term monitoring programs emphasize power analysis as a tool to determine the sampling effort necessary to effectively document ecologically significant changes in ecosystems. Programs that monitor entire multispecies assemblages require a method for determining the power of multivariate statistical models to detect trend. We provide a method to simulate presence-absence species assemblage data that are consistent with increasing or decreasing directional change in species composition within multiple sites. This step is the foundation for using Monte Carlo methods to approximate the power of any multivariate method for detecting temporal trends. We focus on comparing the power of the Mantel test, permutational multivariate analysis of variance, and constrained analysis of principal coordinates. We find that the power of the various methods we investigate is sensitive to the number of species in the community, univariate species patterns, and the number of sites sampled over time. For increasing directional change scenarios, constrained analysis of principal coordinates was as or more powerful than permutational multivariate analysis of variance, the Mantel test was the least powerful. However, in our investigation of decreasing directional change, the Mantel test was typically as or more powerful than the other models.     

Determining Optimal Population Monitoring for Rare Butterflies

Abstract: Determining population viability of rare insects depends on precise, unbiased estimates of population size and other demographic parameters. We used data on the endangered St. Francis' satyr butterfly (Neonympha mitchellii francisci) to evaluate 2 approaches (mark–recapture and transect counts) for population analysis of rare butterflies. Mark–recapture analysis provided by far the greatest amount of demographic information, including estimates (and standard errors) of population size, detection, survival, and recruitment probabilities. Mark–recapture analysis can also be used to estimate dispersal and temporal variation in rates, although we did not do this here. Models of seasonal flight phenologies derived from transect counts (Insect Count Analyzer) provided an index of population size and estimates of survival and statistical uncertainty. Pollard–Yates population indices derived from transect counts did not provide estimates of demographic parameters. This index may be highly biased if detection and survival probabilities vary spatially and temporally. In terms of statistical performance, mark–recapture and Pollard–Yates indices were least variable. Mark–recapture estimates were less likely to fail than Insect Count Analyzer, but mark–recapture estimates became less precise as sampling intensity decreased. In general, count‐based approaches are less costly and less likely to cause harm to rare insects than mark–recapture. The optimal monitoring approach must reconcile these trade‐offs. Thus, mark–recapture should be favored when demographic estimates are needed, when financial resources enable frequent sampling, and when marking does not harm the insect populations. The optimal sampling strategy may use 2 sampling methods together in 1 overall sampling plan: limited mark–recapture sampling to estimate survival and detection probabilities and frequent but less expensive transect counts.     

Effective Population Size and Lifetime Reproductive Success

The mean and variance of lifetime reproductive success, E_LRS and V_LRS, influence the ratio of effective to census population size, N_e/N_c. Because the complete data needed to calculate E_LRS and V_LRS are seldom available, we provide alternatives for estimating N_e/N_c from incomplete data. These estimates should be useful to conservation biologists trying to compute the effective size of a censused population. An analytical approach makes assumptions regarding the process influencing offspring survival. We provide a method for examining the validity of those assumptions and show that particular violations can result in either over- or underestimates. When the assumptions are violated or when more data are available, we suggest estimating N_e/N_c using computer simulations of models based on individuals. We examine how such simulations can be used to estimate N_e/N_c using an individual-based model for Lesser Snow Geese ( Anser caerulescens ). We demonstrate that such estimates can be biased unless the simulations are based on complete cohorts and samples of known age. We show that because the estimate of N_e/N_c depends on the stage of the reproductive cycle used as a point of reference in the model, the census population size N_c must be based on the same stage to provide unbiased estimates of N_e.     

Can modeling improve estimation of desert tortoise population densities?

The federally listed desert tortoise (Gopherus agassizii) is currently monitored using distance sampling to estimate population densities. Distance sampling, as with many other techniques for estimating population density, assumes that it is possible to quantify the proportion of animals available to be counted in any census. Because desert tortoises spend much of their life in burrows, and the proportion of tortoises in burrows at any time can be extremely variable, this assumption is difficult to meet. This proportion of animals available to be counted is used as a correction factor (g0) in distance sampling and has been estimated from daily censuses of small populations of tortoises (6-12 individuals). These censuses are costly and produce imprecise estimates of go due to small sample sizes. We used data on tortoise activity from a large (N = 150) experimental population to model activity as a function of the biophysical attributes of the environment, but these models did not improve the precision of estimates from the focal populations. Thus, to evaluate how much of the variance in tortoise activity is apparently not predictable, we assessed whether activity on any particular day can predict activity on subsequent days with essentially identical environmental conditions. Tortoise activity was only weakly correlated on consecutive days, indicating that behavior was not repeatable or consistent among days with similar physical environments.     

Disentangling the effects of heterogeneity, stochastic dynamics and sampling in a community of aquatic insects

A dynamic and heterogeneous species abundance model generating the lognormal species abundance distribution is fitted to time series of species data from an assemblage of stoneflies and mayflies (Plecoptera and Ephemeroptera) of an aquatic insect community collected over a period of 15 years. In each year except one, we analyze 5 parallel samples taken at the same time of the season giving information about the over-dispersion in the sampling relative to the Poisson distribution. Results are derived from a correlation analysis, where the correlation in the bivariate normal distribution of log abundance is used as measurement of similarity between communities. The analysis enables decomposition of the variance of the lognormal species abundance distribution into three components due to heterogeneity among species, stochastic dynamics driven by environmental noise, and over-dispersion in sampling, accounting for 62.9, 30.6 and 6.5% of the total variance, respectively. Corrected for sampling the heterogeneity and stochastic components accordingly account for 67.3 and 32.7% of the among species variance in log abundance. By using this method, it is possible to disentangle the effect of heterogeneity and stochastic dynamics by quantifying these components and correctly remove sampling effects on the observed species abundance distribution.