Spectral Analysis and image classification of the dwarf shrub Empetrum nigrum (L.) by means of remote sensing data.

Abstract

Empetrum nigrum dominates in alpine and northern part of Norway. It is an allelopathic species that can reduce both productivity and biodiversity in ecosystems. The study focuses on identifying and determining different cover of E. nigrum by means of remote sensing data within two regions of Northern Norway, Ifjord in Finnmark and Troms areas. Field data were collected within 50 study points and E. nigrum was cover recorded. Field data were compared with Landsat 7 ETM+ and Landsat 8 OLI satellite images. Remote sensing is a practical and cost-effective tool to classify land cover and study vegetation changes when large areas are measured. Landsat images were chosen due to good cover, spatial resolution, free availability and its long history, which allows going back in time. Besides, small field plots were measured in the Troms area with ASD FieldSpec spectroradiometer in order to extract a spectral signature of E. nigrum, in coexistence with other common species. Several supervised and unsupervised classification algorithms were performed on the satellite data using the ENVI image processing software. It resulted that neither the specific features (evergreen appearance with a dense cover of tiny leaves with glands producing the allelopathic compound Batatasin-III, and its dominance over vast land areas) nor the ASD FieldSpec measurements of E. nigrum, were suitable for extracting a spectral signature of the species that made a good classification. However, the best result was obtained using a spectral unmixing classification applied to radiometric corrected images in both areas separately. A spectral library created by endmember determination from Landsat data was used for this classification. An evaluation (Pearson correlation), was made in both areas combined as well as in Ifjord study area and Tromsø study area separately. No-correlation between the ground truth data and the data extracted from the spectral unmixing analysis was observed when both areas where combined (r = 0.15). Whereas, a positive correlation appeared when analysing both areas separately (Tromsø, r = 0.55 and Ifjord r = 0.57). Landsat has several issues concerning to endmember determination, spatial, temporal and spectral resolution, as well as data acquisition problem. However, some solutions are proposed. So, it is concluded that Landsat is a good option for E. nigrum retrieval. The opinion is that future studies need to include these improvements or solutions, in order to achieve an E. nigrum classification with a higher correlation coefficient from Landsat imagery.