Barss, Joseph2025-04-092025-04-092025-04-08https://hdl.handle.net/10222/84937Species distribution models must account for spatial and temporal auto-correlation in ecological survey data. In this study, we considered a data set on lobster abundance collected by trawl survey programs in the Bay of Fundy area, and fitted a geostatistical generalized linear mixed model incorporating a Gaussian random field to account for spatial auto-correlation. We performed model selection using information criteria and 5-fold spatial block cross-validation. We then used the model’s predictions to produce an index of relative abundance, which displayed an increasing trend between 1995 and 2023. A Bayesian implementation of the model yielded similar results. In a simulation study, we showed that index estimates obtained by modelling standardized count data using the Tweedie distribution are reasonably accurate, and that estimates obtained using delta models are inconsistently biased. A second simulation study showed that combining data from two survey programs is appropriate when creating a model-based abundance index.enStatisticsSpecies distribution modelLobsterSpatiotemporal modelSpatial modelBay of FundyAbundance indexTMBFisheries scienceTrawl survey data