Application of spectroscopial, hyperspectral and multispectral data to study wetlands in semi-arid environments (Central Spain)

Abstract

Sensitive ecosystems such as wetlands in a semi-arid environment of Central Spain are important indicators of environmental quality and biodiversity for an area dominated by human induced activities. These areas support an important ecological function for the variety of natural plant species as well as migrating and wintering waterfowl. The wetland areas are classified as saline and sub saline due to climatic and lithological conditions. However, rapid land use changes where natural and semi-natural vegetation is substituted for intensive agricultural cultivation, over exploitation of ground water, drying and artificial drainage of wetlands as well as altering and channeling of rivers; and introducing waste disposal have severely affected the wetlands and their natural functions. The objective of this study is to determine abiotic and biotic land surface changes within semi-arid wetland and surrounding upland areas using reflective spectroscopial field, hyperspectral and multispectral data. This is carried out with intensive field work on test plots obtaining soil samples, determining vegetation types and the related spectroscopial information from a hyper saline wetland and from an anthropogenic affected floodplain including surrounding upland areas. The information obtained from the field forms the basis to identify and compare the hyperspectral (DAIS 7915) and multispectral (Landsat TM, ETM+ and ASTER) data from various sensors. Image processing is performed with the DAIS 7915 data (acquired on 29 June 2000) using a minimum noise fraction transformation followed by a pixel purity index analysis. Applying multi dimensional scatter plots, image derived end members for the different surface cover components are determined and compared with field derived end members. A pool of field and image derived end members is created and selected end members with high-resolution information are spectrally resampled to multispectral Landsat TM (17 June 1987), ETM+ (28 June 2000), and ASTER (2 June 2002) data. In order to determine and monitor the changes of selected surface cover components, an unconstrained linear spectral unmixing model is applied to the multispectral data and individual fraction maps representing the spatial distribution of the different surface covers are generated. A change detection analysis is made from the resulting end member fraction maps obtained for the different dates (1987, 2000, 2002). The results show that a detailed methodology is required in order to determine surface cover changes for wetlands in semi-arid areas. The interpretation and statistical analysis of spectroscopial field data together with soil analysis form the basis to differentiate slight compositional changes of land surface covers within the wetland and upland areas. A high correlation is shown between spectral absorption features and soil analysis representing the mineralogy of gypsum, calcite and phyllosilicates, and soil properties such as iron oxide content, electrical conductivity and soil colour. The moisture influence is naturally a dominant feature in the wetland areas and serves to distinguish the drier areas leading to the upland soils. DAIS image-derived end members determine a series of surface components, which successfully delineates the wetland areas and differentiates the areas affected by human induced activities. Synthesizing and extrapolating high-resolution information to medium-resolution data has improved the resulting fraction maps obtained for selective surface cover components related to wetland and upland areas. The surface cover changes are particularly apparent in the areas where agricultural activities are pushing back the extension of the wetland areas where soils have lower salinity levels. This approach can therefore be extrapolated to similar medium-resolution data covering thus larger regional and temporal scales