Premium
Mapping the local variability of Natura 2000 habitats with remote sensing
Author(s) -
Feilhauer Hannes,
Dahlke Carola,
Doktor Daniel,
Lausch Angela,
Schmidtlein Sebastian,
Schulz Gundula,
Stenzel Stefanie
Publication year - 2014
Publication title -
applied vegetation science
Language(s) - English
Resource type - Journals
SCImago Journal Rank - 1.096
H-Index - 64
eISSN - 1654-109X
pISSN - 1402-2001
DOI - 10.1111/avsc.12115
Subject(s) - multispectral image , remote sensing , ordination , multispectral pattern recognition , geography , image resolution , cartography , environmental science , habitat , imaging spectroscopy , multidimensional scaling , ecology , computer science , hyperspectral imaging , mathematics , artificial intelligence , statistics , biology
Questions Can we map both discrete Natura 2000 habitat types and their floristic variability using multispectral remote sensing data? How do these data perform compared to full range imaging spectroscopy data? Which spectral and spatial characteristics of remote sensing data are important for accurate mapping of habitats and their variability? Location A mire complex in B avaria, southern G ermany. Methods To compare the performance of imaging spectroscopy and multispectral remote sensing data, airborne spectroscopy data ( AISA Dual) were spectrally and spatially resampled to the characteristics of two state‐of‐the‐art multispectral sensors ( R apid E ye and Sentinel‐2), resulting in three data sets with different spectral and spatial resolution. Based on the three data sets, we used a combination of field surveys, ordination techniques (non‐metric multidimensional scaling), as well as regression and classification techniques (Random Forests) to derive maps of the distribution of N atura 2000 habitat types and their compositional variability. Subsequently, we analysed effects of the spatial and spectral image resolution and spectral coverage on the mapping performance. Results Mire habitat types and their floristic composition could be accurately mapped with multispectral remote sensing data. In the case of accentuated floristic differences between habitats, the fits of the models for the three sensors differed only marginally. These effects and the importance of the spatial resolution are discussed. Conclusions The results are encouraging and confirm that multispectral data may allow the combined mapping of discrete habitats and their local variability. Still, questions with respect to the transferability of the approach to habitat types with less pronounced spectral differences, and with regard to bridging the gap between fine‐scale vegetation records and coarse resolution imagery remain open.