ľ¹Ï¸£ÀûÓ°ÊÓ

Tidal flats are some of the most dynamic and ecologically rich coastal areas on Earth, supporting migratory birds, fish, and diverse benthic communities. However, these fragile ecosystems are increasingly threatened by climate change and coastal development. Protecting them requires continuous monitoring, yet traditional fieldwork is costly and limited in scope. This thesis explores how satellite remote sensing combined with modern artificial intelligence can improve the way we monitor and understand tidal flats. 

Using deep learning, specifically a method called autoencoders, the research extracts hidden patterns from satellite imagery that go beyond what the human eye or standard analyses can detect. These features allow more accurate predictions of key ecological indicators, such as sediment properties and benthic communities. When combined with Object-Based Image Analysis (OBIA), which groups pixels into meaningful spatial units, the approach provides richer insights into species distributions and environmental changes. 

The thesis also investigates how well these models work across different times, datasets, and locations. Results show that while predictions transfer well across seasons and datasets, applying models to entirely new locations remains a challenge. Seasonal analysis revealed that sediments and benthic communities are relatively stable, while chlorophyll-a (a measure of primary productivity) shows strong seasonal cycles. 

By integrating deep learning, OBIA, and remote sensing, this research offers new, cost-effective ways to monitor tidal flats. The findings highlight both the promise and the limitations of AI-driven approaches, paving the way for smarter conservation strategies in these vulnerable coastal environments.