An Object-Based Image Analysis Method For Predicting Ecological Variables For Monitoring Macrozoobenthos In The Dutch Wadden Sea

The global trend in declining tidal flats due to sea level rise, climate change and anthropogenic factors requires monitoring and management to conserve the coastal ecosystem. Macrozoobenthos, the small creatures living in and on the flats, form an essential food source for birds and fish. Sediment characteristics are a dominant factor in their distribution. Since 2007 annual field surveys at a 500m grid (Synoptic Intertidal Benthic Survey, SIBES) produced a wealth of information, but these surveys are challenging due to low field accessibility and the vulnerability of ecosystems. To provide continuous maps and to include seasonal dynamics of the macrozoobenthos remote sensing offers a valuable data source. Spectral contrast between pixels on tidal flats is low, while spatial patterns reveal geomorphic features in turn related to sediment characteristics. The limitation of distinguishable features of the tidal flats with pixel-based spectral bands urges us to include spatial and geometric information in the analysis. Here we explore the value of Object-Based Image Analysis (OBIA). OBIA methods have the advantage of incorporating spatial context, textural properties and mutual relationships between neighbouring objects. Our study applied an OBIA method to Sentinel-2 images along with features extracted from a deep learning model, to map the sediment characteristics and macrozoobenthic properties in the Dutch Wadden Sea. A multi-resolution object-based bottom-up approach based on three parameters- scale, shape and compactness, segments the images of tidal flats into object entities based on the features extracted from the deep learning model and the spectral bands. The objects are used for the prediction of two sediment variables, median grain size and silt content, and of two ecological variables macrozoobenthic biomass and species richness in the tidal basins Pinkegat and Zoutkamperlaag in the Dutch Wadden Sea. In-situ measurements for the four variables are obtained at 500 m intervals from the SIBES field campaign for the years 2018, 2019 and 2020. Optimal segmentation parameters are selected for each of the four ecological variables based on the prediction results. The overall accuracy of the predictions using random forest regressor for all three years of median grain size and silt content is 51% and 49%, and for the benthos biomass and species richness, it is 31% and 35%. There is an improvement of 18-20% points in prediction results compared to pixel-based methods that use spectral bands alone and 10-15% points compared to pixel-based methods that use deep learning features along with spectral bands. These results show that the prediction methods based on the OBIA approach for these four ecological variables produced better accuracy than the pixel-based prediction technique. Mapping these four ecological variables provides us with more detailed prognoses about the Macrozoobenthos habitat, which can then be used to support decision-making for sustainable and adaptive conservation of the coastal ecosystem.