Day 4

Detailed paper information

Back to list

Paper title Generating satellite-based aquatic vegetation maps for carbon storage estimation using Bayesian statistics
Authors
  1. Sakari Väkevä Finnish Environment Institute (SYKE) Speaker
  2. Sampsa Koponen Finnish Environment Institute (SYKE)
  3. Ari-Pekka Jokinen Finnish Environment Institute (SYKE)
  4. Andrea De Cervo University of Stockholm
  5. Thorsten Blenckner University of Stockholm
  6. Elina Virtanen Finnish Environment Institute (SYKE)
  7. Markku Viitasalo Finnish Environment Institute (SYKE)
Form of presentation Poster
Topics
  • A3. Biosphere
    • A3.06 Biomass monitoring
Abstract text Reed belts are an important subclass of aquatic vegetation as they represent some of the most important Blue Carbon ecosystems in the Baltic Sea. However, their extent has so far not been precisely mapped except in local field sampling experiments related to national inventory programmes. Differences in Normalized-Difference Vegetation Index (NDVI) have long been used as an indicator of vegetation in remote-sensed datasets and Earth Observation (EO). The differences are particularly large over coastal areas, where in the peak growth season (mid-to-late summer), uniformly vegetated areas such as reeds, sedges, rushes, and macrophytes have NDVI values around 0.7 ± 0.2 (1σ), whereas plain water has a relatively low NDVI, –0.2 ± 0.2 (1σ). In this work, Bayesian analysis is applied to identify areas of aquatic vegetation in monthly NDVI composites downloaded from the Sentinel-2 Global Mosaic (S2GM) service. These are then used as indicators for the occurrence of reed belts or other seasonal or permanent vegetation in coastal zones. The method is akin to naïve Bayes and outputs a value that is proportional to the probability of the pixel representing vegetation in water. The prior used is sensitive both to the NDVI and distance from shore; areas closer to the coastline are considered more likely to host aquatic vegetation. The method requires as its source datasets monthly composites of the NDVI from S2GM and a truthful sea mask, extractable from either national coastline layers or suitable land use classes from the Copernicus Coastal Zones data set.

The interpretation of aquatic vegetation has been carried out for the Finnish coast and two Swedish pilot areas in the south (Stockholm) and north (Piteå) in the context of project "Blue Carbon Habitats – a comprehensive mapping of Nordic salt marshes for estimating Blue Carbon storage potential –a pilot study", funded by the Nordic Council of Ministers. Training and test data were obtained from field-mapped reed outlines, and the probabilistic product was converted to a binary interpretation and sieved for too small areas or for areas too far from the shoreline. The ground truth of reed outlines aligns in general with the outlines inferred from EO, though the resolution (10 m) of the EO data limits the support near the shore. Unlike hypothesized, the posterior probability density of the Bayes product was not found to be strongly linked to species distribution nor the field-mapped reed belt density, and a different line of analysis will need to be carried out if these variables are to be predicted with remote-sensed observations.