Climate‐informed models benefit hindcasting but present challenges when forecasting species–habitat associations

Although species distribution models (SDMs) are commonly used to hindcast fine-scale population metrics, there remains a paucity of information about how well these predict future responses climate. Many conventional SDMs rely on spatially-explicit but time-invariant conditions quantify distributions and densities. We compared status quo ‘static' with more climate-informed 'dynamic' assess whether the addition time-varying processes would improve performance and/or forecast skill. Here, we present two groundfish case studies from Bering Sea – high latitude system that has recently undergone considerable warming. relied statistics (R2, % deviance explained, UBRE or GCV) evaluate for presence–absence, numerical abundance biomass arrowtooth flounder Atheresthes stomias walleye pollock Gadus chalcogrammus. then retrospective skill testing near-term Retrospective enables direct comparisons between forecasts observations through process fitting forecasting nested submodels within given time series. found inclusion covariates improved hindcasts. However, dynamic either did not decreased relative static SDMs. This is likely result rapidly changing temperatures ecosystem, which required environmental were outside range observed values. Until additional model development allows fully predictions, (or persistence models) may serve as reliable placeholders, especially when anomalous anticipated. Nonetheless, our findings demonstrate support use rather than selecting priori based their ability species–habitat associations in past.