Assessing the Reliability of Species Distribution Model Projections in the Face of Climate and Ecosystem Regime Shifts: Small Pelagic Fishes in the California Current Ecosystem

Species distribution models (SDMs) are a commonly used tool, which when combined with earth system models, can project changes in organismal occurrence, abundance, and phenology under climate change. An often untested assumption of SDMs is that relationships between organisms and the environment are stationary. To evaluate this assumption, we used a >60-year time series to examine whether patterns of habitat use by ichthyoplankton of four small pelagic fishes in a coastal upwelling area remained steady across climate regime shifts, changes in secondary productivity, and break-points in time series of spawning stock biomass. Generalized additive models were constructed for each phase of these regimes using temperature, salinity, dissolved oxygen, and mesozooplankton volume as predictors of ichthyoplankton occurrence. We assessed non-stationarity based on changes in six metrics: 1) variables included in SDMs; 2) whether a variable exhibited a linear or non-linear form; 3) rank order of deviance explained by variables; 4) responsiveness of fishes to a variable; 5) preferred range of environmental variables used by fishes; 6) response curve shape. Across all species and regimes, non-stationarity was ubiquitous, affecting at least one of the six indicators. Rank order of environmental variables and the preferred range of oceanic conditions were the indicators most subject to change. Non-stationarity was most common among regimes defined by shifts in zooplankton productivity. Similarly, the relationship between fishes and mesozooplankton was more likely to change across regimes, whereas the relationship between fishes and temperature was relatively stable. Respectively, sardine, chub mackerel, anchovy, and jack mackerel exhibited non-stationarity across 83%, 75%, 61%, and 58% of indicators. These widespread changes in how fishes utilize their environment suggest that non-stationarity may hamper our ability to reliably project how species respond to future climate change.