Abstract
The arange and biomass distribution of marine fish species offer insights into their underlying niches. Quantitative data are rare compared to occurrences and remain underused in species distribution models (SDMs) to explore realized niches-the actual space occupied by a species shaped by abiotic and biotic factors. Local densities drive differences in species contributions to ecological processes and ecosystem function rather than through presence alone. If a species growth rate is strongly controlled by macro-environmental conditions, then predicting geographical abundance or densities should be possible. We collated 20 years (2001-2020) of standardized scientific bottom trawl data to fit several versions of hierarchical generalized additive models using biomass (kg km(-2)) of four dominant demersal species (Common dab, European flounder, European plaice, Atlantic cod) within yearly and seasonal (winter and autumn) time windows. Covariates were represented with trawl-level geographic information (position, depth) and high-resolution oceanographic features. This work illustrates species-specific spatiotemporal biomass patterns across two decades and demonstrates superior predictive performance with seasonally variable smoothing terms, revealing seasonally different responses to oceanographic predictors. Firstly, we find relative stasis in Common dab biomass which is linked to the macro-environmental salinity gradient in the western Baltic Sea but with different temperature responses across seasons. Secondly, we show both European flounder and plaice have increased in biomass in the western Baltic Sea with different seasonal relationships to bottom temperature, and that flounder switches between salinity conditions based on season during spawning/feeding periods. Lastly, both juvenile and adult Atlantic cod life stages are shown to have declined most significantly in the Bornholm Deeps and the Gdańsk Deeps. For cod, we conclude that biomass was less reliably predicted in comparison to the other major Baltic demersals studied here, warranting dynamic fishing covariates as a formerly major commercial fishing target. These models approach more dynamic species distribution models and are increasingly valuable to constrain uncertainties in biogeographic forecasting which often rely on annually-averaged response curves, occurrence data, and suitability maps which rarely discriminate between areas of high and low biomass areas in space and time.