Spectral Discrimination And Separability Analysis Of Agricultural Crops And Soil Attributes Using Aster Imagery

Abstract

There are growing interests on how Advanced Spaceborne Thermal Emission and Reflectance Radiometer (ASTER) image data could be used for various applications. ASTER offers improved spatial, spectral and radiometric resolutions. Hence, our purpose was to evaluate the utility of multispectral ASTER imagery in the discrimination and mapping of agricultural crops, soil and related land cover types. Four agricultural land cover attributes were specifically considered for spectral separability assessment and mapping: crop type, crop growth stages, soil colour and soil texture. Two scenes of approximately 60 km x 60 km ASTER Level 1B imagery, covering the eastern Darling Downs region, Queensland, Australia, were selected for this study. Acquired on 24 September 2001, the imagery covers extensive agricultural region encompassing the rural towns of Dalby, Brookstead, Jondaryan, Pittsworth and Millmerran. ASTER's visible and near infrared (VNIR) and shortwave infrared (SWIR) bands, as well as selected image transformation layers (e.g. ratios, vegetation indices, and principal components), were utilised. Training areas were selected and supported with field information, soil sampling and cropping details obtained by interview with farmers. The results of separability analysis indicated that ASTER data provided adequate spectral discrimination of crop types (wheat/barley vs. chickpea), and to some extent, crop growth stages. Interaction between cover attributes (e.g. same wheat but in an early stage or in a very wet patch) produced some classification errors. Despite that, ASTER's sensitivity to changes in bio-physical conditions indicate that these data are useful for mapping within-field variability where the focus is confined to a limited area. The three bands that produced the best average separability are the layers pertaining to vegetation indices: ratio NIR/R, Sqrt(NIR/R) and NIR-R. On the other hand, two-level broad classes of soil colour and texture were adequately mapped. Bands 2 (visible red), 8 (SWIR), and first principal component 1 (of bands 1 to 9) are the best layers to use for discriminating soil features. However, some spectral confusion between intermediate soil colour and texture class was observed. Thus, detailed level mapping of soil attributes using ASTER does not appear achievable. Nevertheless, ASTER's 15-m spatial resolution in the VNIR bands offers potential for mapping within-in field soil variability at relatively broad attribute classes.