Dissertation Abstract
Spectroradiometry and remote sensing data interpretation for mapping the state and dynamics of tundra-taiga ecotone (case study of Khibiny Mountains, Kola Peninsula, Russia)
mikheeva, anna 2011
faculty of geography, lomonosov moscow state university (Russia), 139 pp.
Global changes are well indicated in the Arctic vegetation communities: the changes in species composition, tree coverage, advances of shrubs towards lichen tundra. The response of the vegetation to global changes varies depending on many factors and requires research at all levels from local to global.
This study aimed to answer two questions: i) to describe and map the current state and structure of tundra-taiga ecotone vegetation in test sites in Khibiny mountains, Kola peninsula, Russia on the basis of satellite image interpretation and ground data; ii) to investigate and map changes in the ecotone vegetation which occurred in these sites during the past 50 years, through remote sensing. Both questions were examined at two scale levels – large scale using very high resolution imagery (0,5-1 m) and regional scale with relatively high resolution imagery (15-30 m).
Several test sites were selected in Khibiny mountains at low gradient slopes to exclude exogenic processes and focus on climate processes influencing the ecotone vegetation. Four types of studies were performed as detailed below.
1Current state, large scale study. At the local scale the structure of the ecotone was mapped through automated classification of a QuickBird image (acquired on 28.06.2006) on the basis of combination of spectral and object-oriented approaches. This map can be used as a benchmark in the ecosystem dynamics studies.
2Dynamics, large scale study. Airborne image of 1958 was overlaid on the QuickBird image of 2006 with all necessary orthocorrections for the first test site and for the second airborne image of 1958 and WorldView-1 image of 2008 were overlaid in the same manner. Separate trees near the tree line were marked in both images for both test sites and then compared respectively. The elevation of tree line within test sites increased by an average of 30 m from 1958 to 2008, stand density increased by 3 times, the amount of undergrowth – by up to 10 times. The combination of remote sensing and ground-based data minimized the risk of under- or overestimating the tree line advance.
3Current state, regional scale study. Using spectral signatures defined for a Terra ASTER image for Tuliok Valley in Khibiny Mountains, and spectral unmixing techniques, we have produced a map of the current state of vegetation, which shows the quantitative projective cover (abundance) and species composition of the ecotone, and adequately describes heterogeneity of ecotone vegetation. In this classification we used the results of original experimental linear mixture modelling using ground spectroradiometric measurements. Typical ecotone components and their mixtures with known abundances were measured on the ground, using a handheld VNIR spectroradiometer and then estimated the accuracy of linear spectral unmixing for these components. through this experiment, we found typical spectral thresholds for recognizing each component in the image: stones and lichens can be recognized in subpixel mixtures when they occupy 86% and 57% respectively, green vegetation used in the experiment was recognizable after 43% abundance.
4Dynamics, regional scale study. It was demonstrated that dynamics rates in Khibiny mountains could not be mapped in relatively high resolution multispectral imagery (15-30 m) due to low width of the ecotone belt on mountain slopes, high degree of spectral mixing an thus low reliability of vegetation class determination.
This study aimed to answer two questions: i) to describe and map the current state and structure of tundra-taiga ecotone vegetation in test sites in Khibiny mountains, Kola peninsula, Russia on the basis of satellite image interpretation and ground data; ii) to investigate and map changes in the ecotone vegetation which occurred in these sites during the past 50 years, through remote sensing. Both questions were examined at two scale levels – large scale using very high resolution imagery (0,5-1 m) and regional scale with relatively high resolution imagery (15-30 m).
Several test sites were selected in Khibiny mountains at low gradient slopes to exclude exogenic processes and focus on climate processes influencing the ecotone vegetation. Four types of studies were performed as detailed below.
1Current state, large scale study. At the local scale the structure of the ecotone was mapped through automated classification of a QuickBird image (acquired on 28.06.2006) on the basis of combination of spectral and object-oriented approaches. This map can be used as a benchmark in the ecosystem dynamics studies.
2Dynamics, large scale study. Airborne image of 1958 was overlaid on the QuickBird image of 2006 with all necessary orthocorrections for the first test site and for the second airborne image of 1958 and WorldView-1 image of 2008 were overlaid in the same manner. Separate trees near the tree line were marked in both images for both test sites and then compared respectively. The elevation of tree line within test sites increased by an average of 30 m from 1958 to 2008, stand density increased by 3 times, the amount of undergrowth – by up to 10 times. The combination of remote sensing and ground-based data minimized the risk of under- or overestimating the tree line advance.
3Current state, regional scale study. Using spectral signatures defined for a Terra ASTER image for Tuliok Valley in Khibiny Mountains, and spectral unmixing techniques, we have produced a map of the current state of vegetation, which shows the quantitative projective cover (abundance) and species composition of the ecotone, and adequately describes heterogeneity of ecotone vegetation. In this classification we used the results of original experimental linear mixture modelling using ground spectroradiometric measurements. Typical ecotone components and their mixtures with known abundances were measured on the ground, using a handheld VNIR spectroradiometer and then estimated the accuracy of linear spectral unmixing for these components. through this experiment, we found typical spectral thresholds for recognizing each component in the image: stones and lichens can be recognized in subpixel mixtures when they occupy 86% and 57% respectively, green vegetation used in the experiment was recognizable after 43% abundance.
4Dynamics, regional scale study. It was demonstrated that dynamics rates in Khibiny mountains could not be mapped in relatively high resolution multispectral imagery (15-30 m) due to low width of the ecotone belt on mountain slopes, high degree of spectral mixing an thus low reliability of vegetation class determination.