Landscape-level patterns in photosynthetic marine fouling biofilm compositional heterogeneity as revealed by hyperspectral classification

Marine fouling biofilms typically have diverse community assemblages in which microalgae are strongly represented.  The visible light absorption properties of microalgal photosynthetic pigments typically drive the overall visible light reflectance spectra of these biofilms.  In some cases diagnostic spectral features can be used to infer algal taxonomy, while in mixed communities the overlapping pigment signatures of the constituent species often blur together.  In this study, we apply methods common in remote sensing approaches to spectral data to extract information from subtle variations in the reflectance spectra of mixed composition marine biofilms.  We demonstrate that marine biofilm community composition, as evidenced by their reflectance spectra, is both spatially heterogenous and spatially structured.

 

Visible-NIR hyperspectral images (3.3nm x 200 bands) of biofilms grown on 7.5cm x 7.5cm panels (n=9), immersed in a coastal marina at ~1m depth for 13 months, were captured with a benchtop line-scan imager.   The hyperspectral data were smoothed and transformed to consolidate the major aspects of spectral variability.  A novel active learning spectral classification method incorporating iterative spectral library building by k-means clustering and spectral angle mapping, followed by hierarchical clustering by spectral similarity, discovered more than 70 distinct spectral classes present in the biofilms.  Accordingly, the hyperspectral images of the fouling biofilms were converted to spatially explicit spectral class maps, where each class was assumed representative of a distinct community compositional mix.  Hyperspectral indexing calibrated to chl a surface area density was used to map biomass for the same images. 

 

Cross-tabulating the spectral class and biomass data, it was apparent that for these biofilms, different biomass density levels were consistently associated with specific community compositions (spectral classes.)  Only a small number of the possible classes were represented in the densest areas of biofilm, suggesting that these species composition mixes have a competitive advantage.  In contrast, the full diversity of class types was present in the low biomass areas. 

 

Our hyperspectral approach does not convey exact species composition, as would pooled metagenomic sampling or in-depth microscopy.  However it does allow for the examination of spatially explicit changes in biofilm composition at relatively large scales (the landscape), and so may be a useful tool in hypothesis generation, long term monitoring, and other environmental biofilm applications.