Due to the current, relatively weak, process-based knowledge and methodological setup, the distributional issues are unfortunately neglected in Baltic fish stock assessments and ecosystem-based approaches to management. Further, while management advice is issued for cod, herring and sprat, this is not the case for flatfishes in the Baltic Sea as yet. INSPIRE will fill in both these major gaps in the Baltic fisheries research by advancing our knowledge in patterns and processes of spatial distribution of fish, proposing survey design to resolve the habitat requirements of different life-history stages of commercially/ecologically important fish, as well as carrying out pilot analytical stock assessments for flounder in the Baltic.
To this end, interdisciplinarity in the BONUS INSPIRE Project is interpreted as merging cutting-edge knowledge from different fields of natural sciences, putting into perspective
- ecosystem/fish pilot field surveys involving fishers.
- physiology (of early life history stages).
- habitat modeling.
- statistical approaches to analyse long-term data for different ecosystem features.
- application of molecular methods to analyse condition of fish larvae.
- otolith microchemistry investigations.
- advanced ecosystem modeling.
- spatially explicit fish stock assessment methods.
Video of flounder trying to escape a beach seine in the INSPIRE project (45 sec.)
The philosophy of the BONUS INSPIRE Project is, that only combinations of at least three of these disciplines allow to resolve each of the key questions of the project. Learn more about the key questions of the project
For example, combination (i) of biochemical data, larval survival estimates and drift modeling allows to estimate the importance of the Northern Baltic Sea for sprat reproduction; (ii) otolith micro-chemistry, eco-physiology and habitat modeling allows to distinguish areas important for reproduction of cod and flatfish. The results will not only be made available but readily be implemented in stock assessment, and ecosystem-based management of fisheries including MSFD indicators.
The pre-requisite for most of the work in the BONUS INSPIRE Project is a novel sampling scheme for juvenile flounder and cod (see figure, aiming at fine-scale data on habitat conditions and spatial distribution (catch rates and their probability densities). A special value is the close collaboration with regional fishers for surveying, making their profound knowledge available to the scientists in the project, and increasing the participatory spirit in the project. Also acoustic sampling is intensified beyond its current state in the Baltic, and small-scale studies for selected issues of small scale spatial heterogeneity using video plankton recorder and Triaxus (multi-purpose towing platform for ecosystem profiling, including hydrographics and multi-beam acoustics) complete the set of innovative measurement techniques.
Schematic map of gillnet transsects for habitat mapping and modelling of juvenile flounder and cod. Click on the map to see larger version.
Estimation on the spatial distribution of a structured population is usually performed by fitting surfaces to observed abundances of different stages. Such an approach ignores the information available in the correlation among stages due to the growth of individuals between stages. In INSPIRE, the spatial distributions of fishes will be analysed using the Log Gaussian Cox Process (LGCP). In contrast to other spatial models of the distribution of fish, LGCP avoids problems with zero observations and includes the spatial correlation between observations. It is therefore possible to predict and interpolate unobserved densities at any location in the area. This is important for obtaining unbiased estimates of stock concentration and other measures depending on the distribution in the entire area. Incorporating growth, in the BONUS INSPIRE Project we will simultaneously estimate the spatio-temporal distribution and the size-structure of a population. The method will be applied using empirical data sampled during the BONUS INSPIRE Project sampling activities, but also available data from ICES databases. Particular attention is going to be paid to the correlation structure of the sizes within each sample due to the clustering of individuals with similar size. Environmental preferences, for example oxygen or temperature tolerances, will be included in the model as covariates. We will estimate growth, mortality and reproduction, which lead to a prediction of the size-structure of the population. After model estimation, any aspect of the temporal and spatial population dynamics as well as the sampling process can be probed. This will allow for tracking the movement of a single cohort in space and time, but also for predicting the risk of by-catch of small individuals e.g. in a spatial management context.
The BONUS INSPIRE Project will also make use of regression techniques to describe the relationship between fish population spatial data (relative densities or absolute abundances) and environmental variables, such as abiotic habitat features and densities of interacting species. In addition to exploring species-habitat relationships, the models will be used to make map predictions in order to identify essential fish habitats. Regression techniques may include linear and non-parametric regressions. Particular focus will be given, due to their flexibility, to the non-parametric Generalized additive models (GAM) and Random Forest (RF). GAM and RF are able to account for unbalanced design in the sampling between years, seasons, geographical locations and depths. Moreover, they can handle zero inflation and over-dispersion, which are typical features of fish or fishery data.
Modified versions of GAM will also be used to study the distribution of fish populations under the occurrence of complex interactions. In particular, models able to account for abrupt changes in the spatial distribution of a species over time or environmental regimes will be implemented, as Threshold GAMs (TGAMs). The BONUS INSPIRE Project will also use Geographical Weighted Regressions (GWR) and variable-coefficient GAMs to address the problem of spatially-variable relationships. These two last approaches may be applied to investigate fish-environment responses that change over space.
Random forest is an ensemble learning method for classification and regression, where a large number of decision trees are built and responses are predicted based on aggregated results from all trees. It is a recent method in ecology, which has been shown to perform excellently in capturing species-habitat relationships in Baltic Sea settings. In comparison to traditional regression tree methods, the main advantages of RF is that it produces more accurate predictions and is easier to use as it requires no pruning.
In addition to the statictical techniques described, the population dynamic consequences of spatio-temporal shift will be modelled on ICES Sub-divison spatial scale, which is the relevant scale for implementation into stock assessment and management scenarios. For this purpose, the BONUS INSPIRE Project will apply and further developed the spatial SMS. SMS is a stochastic multispecies model describing stock dynamics of interacting stocks linked together by predation. It operates on annual or seasonal time steps. The model consists of sub-models of survival, fishing mortality, predation mortality, survey catchability and stock-recruitment. SMS uses maximum likelihood to estimate parameters and the total likelihood function consists of four terms related to observations of international catch at age, survey CPUE, stomach contents observation, and a stock-recruitment (penalty) function. Specifically for the Baltic Sea, the advantage of SMS is that the food selection sub-model is able to account for the observed changes in herring and sprat weight at age, which co-occurred with the increase of sprat abundance during the 1990s. An extended SMS model with area-dependent predation mortality for cod, herring and sprat has been developed and will be applied in rhe BONUS INSPIRE Project. ICES Sub-division based values for predation mortalities of herring and sprat were derived in the hindcast SMS by accounting for the distributions of cod, herring and sprat when estimating the prey-specific consumption rates of cod.