Introduction and methods
Aquaculture can contribute to the worldwide demand for sustainably produced protein. With a yearly production of around 400 million kg in the European Union, the blue mussel (Mytilus edulis) is commercially important. New production areas need to be sought to upscale production and current areas might be used more efficiently. Computer models relating mussel growth to environmental variables such as food availability and temperature help understand the spatial and temporal variation in observed mussel growth.
Large scale, high resolution environmental data (chlorophyll-a concentration as proxy for food quantity, total suspended matter as modifier for food quality, irradiance, and temperature, all measured near the sea surface) were obtained from remote sensing (Sentinel-3 satellites). Models were trained and validated with a set of standardized measurements of monthly weight increment (grams total individual wet weight) of mussels on 12 mussel aquaculture plots in the Eastern Scheldt area, The Netherlands. These measurements were repeated yearly from 2018 to 2022.
Two model approaches were applied: (1) a data-driven (DD) model to find correlations between environment and measured mussel growth, unconstrained by forced relationships, and (2) a mechanistic Dynamic Energy Budget (DEB) model based on theoretical relationships between environment and physiology. Both models generated predicted weight-over-time series for a growing season (April – November) for the 12 locations in the Eastern Scheldt for several years. The models were trained and validated with leave-one-out cross validation whereby the location-year combination for validation is always different from the locations and years used for model training.
Results and Discussion
Comparing a data-driven (DD) model with a Dynamic Energy Budget (DEB) model yielded valuable insights in how blue mussels (Mytilus edulis) respond to their environment. The DD model was severely hindered by seasonal confounding of explanatory variables. Seasonal patterns in irradiance, sea-surface temperature and growth of the mussels were highly consistent across locations and all coincided with seasonal variation in growth. Consequently, the DD model was not able to distinguish between their effects. Contributions of chlorophyll-a concentration and total suspended matter were not statistically significant for the DD model, even though some unexpected correlations were revealed in the data. DD model predictions for the validation data were not significantly correlated to observed growth.
The DEB model was able to produce a small, but significant positive correlation between predicted growth and observed growth. This correlation was mainly due to variation in chlorophyll-a concentrations as spatial variation in temperature in the Eastern Scheldt is limited. Some of the spatial and temporal variation in observed mussel growth might be explained better after inclusion of another proxy for food quantity than chlorophyll-a (e.g. primary production), or a more appropriate proxy for food quality (e.g. species composition). Extrapolating temperature effects to novel locations with, for example, lower temperatures, may allow scenario testing for translocating mussels to promising new aquaculture locations.