Answers are grounded in this paper and cite their sources.
Ecological modeling study demonstrates that accounting for sampling-depth dispersion improves copepod abundance estimates, indicating that standard models may produce biased environmental inferences.
Zooplankton form the trophic link between marine primary production and higher consumers and play a central role in carbon export, making reliable estimates of their abundance essential for understanding marine ecosystem dynamics in a changing world. However, zooplankton datasets are frequently employ different sampling depth intervals (i.e., bin widths, defined as the vertical extent sampled per tow). Because wider bins integrate over larger water volumes, they yield abundance estimates with differing precision which, if unaccounted for, introduce heteroscedasticity that can bias inference. We compared statistical frameworks for modeling calanoid copepod abundance in waters surrounding the Kerguelen Islands as a function of temperature and mean depth. Specifically, we evaluated a linear model fitted to log-transformed response data and generalized linear models (GLMs) assuming log-normal, Gamma, and inverse Gaussian error distributions. Model selection based on generalized Akaike information criterion (GAIC) identified the inverse Gaussian GLM as the best-supported framework (GAIC = − 3068.25). We then extended that model by explicitly parameterizing dispersion as a function of normalized bin width to account for heterogeneity in measurement precision. This approach addresses an important but often overlooked source of heteroscedasticity in marine ecological datasets. Incorporating a bin-width-dependent dispersion structure improved model fit (GAIC decreased from − 3068.25 to − 3084.61) and altered covariate inference. The temperature-by-depth interaction was statistically significant in the constant-dispersion model but became non-significant when dispersion heterogeneity was included (p = 0.0023 to 0.0657), whereas depth remained a robust predictor. For datasets with strictly positive, strongly right-skewed abundance and sampling protocols that create predictable differences in observation precision, direct distributional models with an explicit dispersion component may yield more reliable inference than a log-transformed Gaussian model. This conclusion is limited to the four candidate frameworks and the single-year dataset evaluated here and requires validation in other regions, years, and taxa.
No takes yet. Share an insight, caveat, or question.
Egorova et al. (2026) studied this question.
Synapse has enriched 8 closely related papers on similar clinical questions. Consider them for comparative context: