Detecting subpixel deciduous components to complement traditional land cover classifications in Southwest Finland
2015 (English)In: International Journal of Applied Earth Observation and Geoinformation, ISSN 1569-8432, E-ISSN 1872-826X, Vol. 42, 97-105 p.Article in journal (Refereed) Published
To ensure successful conservation of ecological and cultural landscape values, detailed and up-to-datespatial information of existing habitat patterns is essential. However, traditional satellite-based and rasterclassifications rely on pixels that are assigned to a single category and often generalized. For many frag-mented key habitats, such a strategy is too coarse and complementary data is needed. In this paper,we aim at detecting pixel-wise fractional coverage of broadleaved woodland and grassland componentsin a hemiboreal landscape. This approach targets ecologically relevant deciduous fractions and com-plements traditional crisp land cover classifications. We modeled fractional components using a k-NNapproach, which was based on multispectral satellite data, assisted by a digital elevation model and acontemporary map database. The modeled components were then analyzed based on landscape struc-ture indicators, and evaluated in conjunction with CORINE classification. The results indicate that bothbroadleaved forest and grassland components are widely distributed in the study area, principally orga-nized as transition zones and small patches. Landscape structure indicators show a substantial variationbased on the fractional threshold, pinpointing their dependency on the classification scheme and grain.The modeled components, on the other hand, suggest high internal variation for most CORINE classes,indicating their heterogeneous appearance and showing that the presence of deciduous components inthe landscape are not properly captured in a coarse land cover classification. To gain a realistic perceptionof the landscape, and use this information for the needs of spatial planning, both fractional results andexisting land cover classifications are needed. This is because they mutually contribute to an improvedunderstanding of habitat patterns and structures, and should be used to complement each other.
Place, publisher, year, edition, pages
2015. Vol. 42, 97-105 p.
Subpixel fractions, k-NN modelling, Remote sensing, Landscape heterogeneity, Ecotones, Key habitat mappinga
Other Earth and Related Environmental Sciences Physical Geography
Research subject Physical Geography
IdentifiersURN: urn:nbn:se:su:diva-118827DOI: 10.1016/j.jag.2015.05.011ISI: 000362061200010OAI: oai:DiVA.org:su-118827DiVA: diva2:840053