Work Package 2: Landscape Mapping
WP 2.1: Development of Advanced Automatic Classification Techniques
The aim of the WP is to improve the efficiency of RS data analysis by defining advanced classification techniques that can automatically identify the land-cover on the ground. The developed techniques will be exploited later in the WP2.2 and WP2.3.
A) Development of systems based on multisources and multiresolution data fusion for the analysis of high heterogeneity data (hyperspectral, multispectral, LIDAR).
B) Multiple classifier architectures will be developed and exploited for information fusion from different data sources. Multiclassifiers architectures in data fusion can significantly improve the quality of maps based on multiple data sources (Benediktsson and Kanellopoulos, 1999) and, consequently, a combination of different multi-classifier architectures will be investigated.
C) Different approaches for contextual analysis will be developed and applied (approaches based on mathematical morphology and Markov random fields(Moser et al., 2013)) in cases where the geometrical resolution is high enough to distinguish objects in the scene.
D) Processing, analysis and land cover classification of hyperspectral images will be improved. Active learning strategies (Crawford et. al., 2013) for efficient utilization of training samples will be developed to overcome the need of hyperspectral analysis to require a huge number of reference samples. The problem of nonlinear data structure modelling, which is noteworthy especially in land-cover classification applications, will be addressed by using Manifold-Learning based approaches (Ma and Crawford, 2010)
WP 2.2: Geological Mapping
The objective is to map (A) Volcanic units (B) Glacial units and (C) Fluvial units.
A) Volcanic units (IES, SPL, BGS, ZGIS: G.B.M. Pedersen)
1) Lava flow mapping in the Hekla area delineating lava flow boundaries. This is both done by automatic classification techniques developed in WP2.1 and through a semi-automated object-based image analysis (OBIA) approach using hierarchical landscape segmentation. The lava flow outlines will be used for lava volume estimation and LIDAR intensity maps will be used to investigate relative lava flow ages (Mazzarini et al. 2007). Furthermore, lava flow morphology will be mappedby the classification techniques developed in WP2.1. This includes mapping of lava flow roughness, ridges, channels, lobes and levees. Surface morphology of lava flows are controlled by parameters such as effusion rate, lava composition and subsurface topography, hence it provides essential information to understand emplacement mechanism of the lava flows (e.g., Walker, 1973; Pinkerton and Wilson, 1994; Griffith et al., 2003; Glaze and Baloga, 2007). Few automated mapping procedures have been developed (e.g., Tarquini et al. 2012) limiting the quantitative morphologic assessment of these parameters. Once the quantitative assessment of lava flow morphology is complete it will be correlated with available petrological and physical parameters (e.g., Griffith and Fink, 1992; Gregg and Fink, 1995; Harris and Rowland, 2001, 2009; Chevrel et al., 2013).
2) Tephra deposit mapping in the Hekla area by the classification techniques developed in WP2.1 using hyperspectral data to distinguish between the silicic and basaltic types of tephra.
3) Mapping móberg deposits in the Hekla and Öræfajökull area. Implement the morphometric characteristics and the approach developed by Pedersen and Grosse (2013, 2014, in review, in prep.) using OBIA and hierarchical landscape segmentation based on slope and curvature maps. In Iceland the móberg deposits cover 11200 km2 (Jakobsson and Guðmundsson, 2008) providing important information of the distribution and extent of the intraglacial volcanism.
B) Glacial units (IES, SPL: G. Aðalgeirdóttir & G.B.M. Pedersen).
Glaciers cover about 11% of the total area of Iceland (Björnsson and Pálsson, 2008) and they respond rapidly to climate change by reducing areaand volume (Björnsson et al., 2013). Themapping techniques from WP 2.1 will be used to map: (1) The glacial outlines of Öræfajökull. (2) The firn line at the end of summer, which can be interpretedas the Equilibrium Line Altitude (ELA) and which is closely related to the mass balance of the glacier (Björnsson and Pálsson, 2008). (3) The crevasse fields. Maps of crevasses have been produced for glaciers in Iceland (Guðmundsson, 2010) and a new crevasse mapcreated herefor Öræfajökull will benefit tourism as a safety asset.
C) Fluvial units (IES, ZGIS, SPL: G.B.M. Pedersen & D. Hölbling)
Different types of water bodies exist in the study areas including river systems and glacial lagoons (e.g.,the tourist attraction Jökulsárlón). These water bodies will be mapped using the mapping techniques in WP2.1. It is expected that a combined analysis of multi-sensor data
will lead to more accurate and comprehensive detection of flowing waters and standing water (e.g. Augusteijn and Warrender, 1998; Bourgeau-Chavez et al., 2009; Mwita et al., 2012).
WP2.3: Vegetation Mapping
The aim is to create a vegetation map by using the hyperspectral data, the LIDAR DEM and the classification techniques developed in WP 2.1. The map will serve two aims, i.e., (A) comparison to other vegetation mapping techniques and (B) vegetation analysis.
A) Comparison of mapping techniques. The outcome of the classification technique in WP2.1 will be compared to other vegetation mapping techniques used in Iceland, such as (i) the “conventional” vegetation mapping method (Guðjónsson, 2005) and (ii) the habitat type mapping (Magnússon et al., 2009).
B) Vegetation analysis. The extent and the distribution of the different vegetation types will be analyzedbased on the vegetation map and RS data. This will be used for answering questions about (i) the extent and spatial distribution of plant communities and/or species of ecological interest, such as woodlands (birch and willows), wetlands and invasive species (such as Alaska lupin), (ii) colonization patterns on the Hekla lavas, in front of the receding outlet glaciers of Öræfajökull and on the alluvial plains of Skeiðarársandur, i.e.,in areas where the age of the surfaces is known; (iii) the vegetation biomass in relation to the mapped plant communities, estimated by using vegetation indexes for carbon stock estimation (Ravindranath and Ostwald, 2008).