Mixed effects: a unifying framework for statistical modelling in fisheries biology (original) (raw)
Related papers
A generalized linear mixed model analysis of a multi-vessel fishery resource survey
Fisheries Research, 2004
A generalized linear mixed model (GLMM) that treats year and spatial cell as fixed effects while treating vessel as a random effect is used to examine fishing power among chartered industry-based vessels and a research trawler, the FRV Miller Freeman, for bottom trawl surveys on the upper continental slope of U.S. West coast. A Bernoulli distribution is used to model the probability of a non-zero haul and the gamma distribution to model the non-zero catch rates of four groundfish species. The use of vessel as a random effect allows the data for the various vessels to be combined and a single continuous time-series of biomass indices to be developed for stock assessment purposes. The GLMMs fit the data reasonably well. Among the different models examined, the GLMM incorporating a random vessel × year effect had the smallest AIC and was thus chosen as the best model. Also, estimated random effects coefficients associated with the industry-based vessels and the FRV Miller Freeman for each year suggests that these vessels can be assumed to be from a common random effects distribution. These results suggest that combining data from the chartered industry-based vessels and from the research trawler may be appropriate to develop indices of abundance for stock assessment purposes. Finally, an evaluation of variances associated with abundance indices from the different models indicate that analyzing these data as a fixed effect GLM may underestimate the level of variability due to ignoring the grouped nature of tows within vessels. As such, use of a mixed model approach with vessel as a random effect is a reasonable approach to developing abundance indices and their variances.
Environmental variability and fisheries: what can models do?
Reviews in Fish Biology and Fisheries, 2008
This review is based on 58 climatefisheries models published over the last 28 years that describe the impacts of fishery pressure and environmental variability on populations and ecosystems and include basic principles of population dynamics. It points out that the incorporation of environmental factors in fishery models has already been done and is of great importance for future models used in the assessment of marine resources. The work is guided by the questions to what extent have these models a) enhanced our understanding of the interrelationships between the environment, the fishery and the state of the exploited resources and b) helped to improve the prediction of population dynamics and the assessment of marine resources. For each of the six most commonly used model categories a case study is critically analyzed. The problems of ''breaking relationships'' between environmental factors and the biological response used in models, the trade-off between model complexity (realism) and simplicity (data availability) and the potential of multivariate climate indices for forecasting ecosystem states and for use as proxies for combined models are discussed, as are novel non-linear and spatially explicit modeling approaches. Approaches differ in terms of model complexity, use of linear or non-linear equations, number of parameters, forecast time horizon and type of resource modelled. A majority of the models were constructed for fish and invertebrate stocks of the northeast Pacific and the epicontinental seas of the Atlantic, reflecting the advancement of fisheries science in these regions. New, in parts highly complex models and sophisticated approaches were identified. The reviewed studies demonstrate that the performance of fished stocks can better be described if environmental or climatic variability is incorporated into the fisheries models. We conclude that due to the already available knowledge, the greatly enhanced computer power, new methods and recent findings of large-scale climatic/oceanographic cycles, fisheries modeling should progress greatly in the coming years.
ICES Journal of Marine Science, 2008
Mikkonen, S., Rahikainen, M., Virtanen, J., Lehtonen, R., Kuikka, S., and Ahvonen, A. 2008. A linear mixed model with temporal covariance structures in modelling catch per unit effort of Baltic herring. -ICES Journal of Marine Science, 65: 1645-1654.
Canadian Journal of Fisheries and Aquatic Sciences, 2016
Estimating trends in abundance from fishery catch rates is one of the oldest endeavors in fisheries science. However, many jurisdictions do not analyze fishery catch rates due to concerns that these data confound changes in fishing behavior (adjustments in fishing location or gear operation) with trends in abundance. In response, we developed a spatial dynamic factor analysis (SDFA) model that decomposes covariation in multispecies catch rates into components representing spatial variation and fishing behavior. SDFA estimates spatiotemporal variation in fish density for multiple species and accounts for fisher behavior at large spatial scales (i.e., choice of fishing location) while controlling for fisher behavior at fine spatial scales (e.g., daily timing of fishing activity). We first use a multispecies simulation experiment to show that SDFA decreases bias in abundance indices relative to ignoring spatial adjustments and fishing tactics. We then present results for a case study i...
Evaluating drivers of spatiotemporal individual condition of a bottom-associated marine fish
A fish’s body condition is described by its weight given its length and is often positively associated with fitness. Atlantic cod (Gadus morhua) in the south-eastern Baltic Sea has experienced a drastic deterioration of its physiological status since the early 1990s to levels that compromise the growth or even survival of the population. Several hypotheses have been proposed (e.g., competition, hypoxia, lack of prey) and evaluated temporally using averages over large spatial scales (basin or population level). However, as these variables operate at local spatial scales and are heterogenous in space, it is important to evaluate the link between body condition and covariates on local scales as well. By applying a geostatistical model that includes spatially and spatiotemporally correlated random effects using Gaussian Markov random fields, we analyze the body condition of cod in the autumn (main feeding season) in relation to biotic and abiotic covariates at different spatial scales a...
Marine Ecology Progress Series, 2014
Evaluating the effects of marine reserves on exploited species can be challenging because they occur within a context of natural spatial and temporal variation at many scales. For rigorous inferences to be made, such evaluations require monitoring programmes that are replicated at appropriate scales. We analysed monitoring data of snapper Pagrus auratus (Sparidae) in northeastern N ew Zealand, comprised of counts from baited-underwater-video surveys from inside and outside 3 marine reserves. Surveys were replicated at many levels, including areas inside and outside of marine reserves at 3 locations in 2 seasons, over a period of up to 14 yr, in an unbalanced design. The Bayesian modelling approach allowed the use of some familiar aspects of ANOVA, including mixed models of fixed and random effects, hierarchically nested structures, and variance decomposition, while allowing for overdispersion and excess zeros in the counts. Model selection and estimates of variance components revealed that protection by marine reserves was by far the strongest measured source of variation for relative densities of legal-sized snapper. The size of the effect varied across years among the 3 reserves, with relative densities between 7 and 20 times greater in reserves than in nearby areas. Other than the reserve effect, the temporal factors of season and year were generally more important than the spatial factors at explaining variation in counts. In particular, overall relative densities were ~2 to 3 times greater in autumn than in spring for legal-sized snapper, although the seasonal effect was also variable among locations and years. We consider that the Bayesian generalised linear mixed modelling approach, as used here, provides an extremely useful and flexible tool for estimating the effects of management actions and comparing them directly with other sources of spatial and temporal variation in natural systems.
Fisheries Research, 2010
Modeling growth is required in many ecological studies and stock assessment applications, but most fish and shellfish growth analyses focus on the estimation of average parameters, which do not provide a complete description of the growth of members of a population. We investigated individual and spatial variation of growth in striped clams (Ameghinomya antiqua) from San Jose Gulf (Argentine Patagonia) using series of growth ring measurements obtained from individual clams from seven populations ("longitudinal data"). Data showed a clear geographical pattern, with two clusters of locations corresponding to domains separated by a thermal front. In the West Domain circulation is dominated by strong tidaldriven eddy flushing, temperature is lower during the growth season, and nutrient concentration and primary productivity are generally higher; circulation is sluggish in the East Domain. West of the front (i) growth rate of small clams and maximum growth rate are highest, and (ii) individual growth rate tends to peak at a smaller size and at a younger age. Evidence of an inflection point in growth rate prompted use of the Richards model, which has been frequently applied to benthic invertebrates for that reason. The model, however, had structural limitations and failed to fit the sharp inflection point. Differences in average parameter values, on the other hand, captured the variability between populations. While all individuals followed the same general growth pattern, there was high variability in individual growth profiles. Accounting for this variability through random effects in all growth parameters affected the estimated average parameters: predicted growth increments at size were larger initially, and the trend reversed after a certain size. Within-individual autocorrelation was not significant, a benefit of using growth increments instead of size-at-age data. We discuss between populations variation in relation to mesoscale environmental gradients, the use of mixed-effects models to analyze longitudinal data, and the implications of our results for stock assessment and management.
2007
I would first like to thank my advisor, Dr. Kenneth Rose, for eight years of his glorious tutelage. Kenny provided excellent training in every aspect of science-from development and programming of models, on his creative approach to science, to successful collaboration, communication and writing. Working with Kenny has been demanding at times because of his high expectations, advanced solutions and problem solving methods, and remarkable attention to detail, but that is exactly what makes Kenny such a productive and respected scientist. It has been a joy to learn from and work with him. He has made me a better scientist. I thank my committee members, Drs. Jim Cowan, James Geaghan, Megan Lapeyre, Barry Moser, and Greg Stone, for their guidance and practical insights. I enjoyed interacting with each one. I would like to extend an added thanks to Jim Cowan for my summer at LUMCON, his encouragement, generosity, and ideas. Thanks to Drs.
Fisheries Research, 2004
Catch-per-unit-effort (CPUE) data have often been used to obtain a relative index of the abundance of a fish stock by standardizing nominal CPUE using various statistical methods. The theory underlying most of these methods assumes the independence of the observed CPUEs. This assumption is invalid for a fish population because of their spatial autocorrelation. To overcome this problem, we incorporated spatial autocorrelation into the standard general linear model (GLM). We also incorporated into it a habitat-based model (HBM), to reflect, more effectively, the vertical distributions of tuna. As a case study, we fitted both the standard-GLM and spatial-GLM (with or without HBM) to the yellowfin tuna CPUE data of the Japanese longline fisheries in the Indian Ocean. Four distance models (Gaussian, exponential, linear and spherical) were examined for spatial autocorrelation. We found that the spatial-GLMs always produced the best goodness-of-fit to the data and gave more realistic estimates of the variances of the parameters, and that HBM-based GLMs always produced better goodness-of-fit to the data than those without. Of the four distance models, the Gaussian model performed the best. The point estimates of the relative indices of the abundance of yellowfin tuna differed slightly between standard and spatial GLMs, while their 95% confidence intervals from the spatial-GLMs were larger than those from the standard-GLM. Therefore, spatial-GLMs yield more robust estimates of the relative indices of the abundance of yellowfin tuna, especially when the nominal CPUEs are strongly spatially autocorrelated. (T. Nishida). fish stock. The nominal (observed) CPUEs are affected by changes of year, season, area of fishing and various environmental factors. Many statistical methods have been used to 'standardize' them to account for such variations. These include the general or generalized linear models (hereafter referred to as the standard-GLM), general additive models (GAM), neural networks (NN), regression trees (RT), and others (ICCAT, 0165-7836/$ -see front matter
Focused model selection for linear mixed models with an application to whale ecology
Annals of Applied Statistics
A central point of disagreement, in certain long-standing discussions about a particular whaling dataset in the Scientific Committee of the International Whaling Commission, has directly involved model selection issues for linear mixed effect models. The biological question under discussion is associated with a clearly defined parameter of primary interest, i.e. a focus parameter, which makes model selection with the Focused Information Criterion (FIC) more appropriate than other selection methods. Since the existing FIC methodology has not covered the case of linear mixed effects models, this article sets up the required framework and develops the necessary formulae for the relevant FIC. Our new criterion requires the asymptotic distribution of estimators derived for a given candidate linear mixed model, but with behaviour examined under a wider linear mixed model. These results, needed here to build our FIC, also have independent interest.