Skip to main content

Can regime shifts in reproduction be explained by changing climate and food availability?

Cite this dataset

Tirronen, Maria; Depestele, Jochen; Kuparinen, Anna (2023). Can regime shifts in reproduction be explained by changing climate and food availability? [Dataset]. Dryad.


Marine populations often show considerable variation in their productivity, including regime shifts. Of special interest are prolonged shifts to low recruitment and low abundance which occur in many fish populations despite reductions in fishing pressure. One of the possible causes for the lack of recovery has been suggested to be the Allee effect (depensation). Nonetheless, both regime shifts and the Allee effect are empirically emerging patterns but provide no explanation about the underlying mechanisms. Environmental forcing, on the other hand, is known to induce population fluctuations and has also been suggested as one of the primary challenges for recovery. Yet, traditional stock-recruitment models used in fisheries management have been time-invariant and considered only density-dependence. In the present study, we build upon recently developed Bayesian change-point models to explore the contribution of food and climate as external drivers in recruitment regime shifts, while accounting for density-dependent mechanisms (compensation and depensation). Food availability is approximated by the copepod community. Temperature is included as a climatic driver. Three demersal fish populations in the Irish Sea are studied: Atlantic cod (Gadus morhua), whiting (Merlangius merlangus), and common sole (Solea solea). We demonstrate that while spawning stock biomass undoubtedly impacts recruitment, abiotic and biotic drivers can have substantial additional impacts, which can explain regime shifts in recruitment dynamics or low recruitment at low population abundances. Our results stress the fact that traditional abundance-based stock-recruitment models are not sufficient to capture variability in fish recruitment.


The recruitment models were fitted to the empirical data using the Bayesian online change-point detection method (BOCPD), combined with simulation-based filtering. The data were divided into segments by calculating their most likely segmentation (MLS). The methods are described more in detail in the article and its supporting material.

Usage notes

The file environmental_predictors.Rmd describes data pre-processing.

The folder BOCPD provides scripts for fitting the change point models to the empirical data. Consider parallel computing for efficiency. The empirical data are not adjusted, consider lags, when needed (see the article and its supplementary material).

The results data enable model comparison based on the negative log marginal likelihood (nlml).


Academy of Finland, Award: 317495

European Research Council, Award: COMPLEX-FISH 770884

H2020 Societal Challenges, Award: 101000318