Process-based forecasting of near-term range shifts in marine species

Alexa Fredston and Malin L. Pinsky, Ecology, Evolution, and Natural Resources, Rutgers University, New Brunswick, NJ, Daniel Ovando, School of Aquatic and Fishery Sciences, University of Washington, Seattle, WA

Background/Question/Methods
Species around the globe are shifting their geographical ranges in response to climate change. Accurate, near-term prediction of their future distributions is critical for natural resource management and biodiversity conservation. At short time scales, transient dynamics of populations not in equilibrium with the environment may strongly influence species distributions, underscoring the need to explicitly model demography. We developed a process-based dynamic range model that estimates demographic rates, and the relationship between those rates and the environment, to forecast species range shifts in response to temperature change. This hierarchical Bayesian model uses only data on species’ occurrences and abundances to estimate parameters and simulate future states. We fitted this model to historical occurrence and abundance data from demersal trawl surveys conducted annually by the National Oceanic and Atmospheric Administration since the 1960s. We focused on four species of importance to fisheries in the mid-Atlantic region — summer flounder, shortfin squid, spiny dogfish, and gray triggerfish — and made forecasts for every year up to a decade (2009-2018). We then compared our annual forecasts to the real data from the testing interval, and to predictions from correlative ecological niche models.
Results/Conclusions
The dynamic range model simulated population dynamics including growth, reproduction, and dispersal. Some of these processes were modeled as temperature-dependent functions, which we fitted to in situ temperature data from the trawl survey and then forecast with outputs from a Regional Ocean Modeling System (ROMS) data product. All models converged and yielded reasonable estimates of key parameters. The dynamic range model outperformed a generalized additive model in explaining historical species distributions when tested on out-of-sample data. Incorporating stage-structured population dynamics improved forecast skill. Ongoing work will expand on model evaluation and comparison, and address the selection of appropriate observation models. This work is among the first applications to real data of a class of models that shows great promise in ecological forecasting: dynamic range models that can make mechanistic predictions about the future by estimating process rates from survey data. By explicitly modeling demographic processes, this study advances the ability to predict short-term range dynamics of species on the move.