Environmental Drivers of Baltic Sea Fishing – A LSTM Spatio-Temporal Study

Abstract

Climate change and anthropogenic influences are significantly altering marine ecosystems in the Baltic Sea, posing substantial challenges for fisheries. The Bornholm Basin, located in the central Baltic Sea, is a crucial fishing area for the region and has been experiencing shifts in fish distribution and abundance due to changing oceanographic conditions. These changes affect not only fish migration and spawning patterns but also the spatial and temporal dynamics of fishing activity. Understanding the environmental drivers behind these changes is key to developing sustainable fishery management strategies.
This thesis investigates how environmental variability affects fishing activity in the Bornholm Basin during February and March over a ten-year period (2014–2023). A Long Short-Term Memory (LSTM) model, a type of deep learning network designed for time series prediction, was trained to estimate weekly fishing density using environmental features. These features include sea water potential temperature, sea water salinity, sea surface height, eastward and northward water velocity, and upper mixed layer thickness. The model predictions were evaluated using R², MAE, and MSE, and feature importance was assessed using SHAP (SHapley Additive exPlanations) values.
The model achieved a moderate performance (R² = 0.54, MAE = 1.64), successfully capturing the general spatio-temporal trends in fishing activity, with peak densities occurring in the central Bornholm Basin. However, it consistently underpredicted high-density fishing events and overestimated low-density areas. SHAP analysis revealed that northward velocity, temperature, and sea surface height were the most influential features, indicating that geostrophic currents and ocean mixing processes strongly affect fish distribution. Salinity and mixed layer thickness were the least influential, likely due to the vertical stratification and limited mixing in the Baltic Sea.
The findings highlight the importance of integrating oceanographic data in predictive modeling of fishing activity, particularly under changing climatic conditions. Nonetheless, environmental factors alone do not fully explain observed patterns. Human influences, including fisher decision-making, vessel capabilities, economic constraints, and fisheries regulations—also play a central role in shaping fishing behavior. Consequently, future models should incorporate socio-economic and policy-related variables to improve predictive performance.
Future Directions
While the LSTM model provides a valuable tool for assessing the impact of environmental change on fishing dynamics, future research should explore model improvements through attention mechanisms, alternative algorithms, and extended temporal coverage. Moreover, engaging with stakeholders and integrating expert knowledge will be essential to enhance the applicability of findings for sustainable fishery management. Monitoring environmental changes and understanding their implications for fish distribution and fisher behavior are critical to ensuring the resilience of the Baltic Sea’s ecosystems and the communities that depend on them.

Popular Abstract

Climate, Currents and Catch: What Drives Fishing in the Baltic Sea? *
The Baltic Sea has long provided food security, employment, and cultural value for the countries that surround it. Today, however, fisheries in the region face several growing challenges. Climate change, overfishing, pollution, and shifts in dominant fish species are all contributing to declining stocks and changing fishing patterns. Understanding how environmental conditions affect fishing activity is essential for managing the sea’s resources sustainably.
This study focuses on the Bornholm Basin—an important fishing area in the central Baltic Sea—and investigates how oceanographic changes influence where and when fishing happens. To do this, I used a deep learning method called an LSTM model to predict weekly fishing activity during February and March from 2014 to 2023. The model used environmental data such as sea water temperature, salinity, surface height, ocean currents, and upper mixed layer thickness.
The results show that fishing activity was highest in the central Bornholm Basin and lowest in the eastern part. While the model successfully captured general patterns over time and space, it tended to underpredict high-density fishing areas. The model achieved a moderate correlation (R² = 0.54), with a mean absolute error of 1.64.
By analyzing which environmental features influenced the model’s predictions the most, northward water velocity, temperature, and sea surface height emerged as key factors. These are linked to physical processes like currents and up- and downwelling, which impact where fish are likely to be found. In contrast, salinity and upper layer thickness were found to be less important predictors, likely due to the stratified structure of the Baltic Sea.
The findings suggest that environmental conditions play a significant role in the spatial and temporal dynamics of fishing activity. However, it is also clear that human factors—such as fisher behavior, vessel capacity, economic decisions, and fisheries management policies—strongly influence where and when people choose to fish.
Looking ahead, improving model accuracy by testing alternative methods, incorporating more years of data, and expanding the environmental features could help better understand these dynamics. Importantly, integrating perspectives from stakeholders, fishers, and managers is key to developing strategies that can adapt to ongoing environmental change while supporting sustainable fisheries in the Baltic Sea.

• author: Kastens, Lilith
• supervisor: Pengxiang Zhao
• organization: Dept of Physical Geography and Ecosystem Science
• course: NGEM01 20251
• year: 2025
• type: H2 - Master's Degree (Two Years)
• subject: Earth and Environmental Sciences
• keywords: Physical Geography and Ecosystem Analysis, Fishing Activity, Baltic Sea, Deep Learning, Oceanography, LSTM, Climate Change
• publication/series: Student thesis series INES
• report number: 713
• language: English
• id: 9206590
• date added to LUP: 2025-06-27 14:00:52
• date last changed: 2025-06-27 14:00:52