2005 Salt Lake City Annual Meeting (October 16–19, 2005)

Paper No. 5
Presentation Time: 2:30 PM


KOCH, Magaly, INZANA, Jennifer and EL-BAZ, Farouk, Center for Remote Sensing, Boston University, 725 Commonwealth Ave, Boston, MA 02215, farouk@bu.edu

Satellite remote sensing techniques are very useful tools for detecting and mapping terrain features that may indicate the presence of surface and/or subsurface water in arid lands. One of the features that serves as direct evidence of favorable groundwater occurrence is vegetation. Monitoring its spatial and temporal distribution (seasonal/cyclical or progressive change) may provide useful clues about the type of water resources that are being tapped, i.e., runoff water, shallow groundwater or irrigation water from deeper aquifers. Knowing the extent and type of vegetation cover and its changes over time can help in determining 1) rates of evapotranspiration, 2) amount and type of water resources used in agriculture (all year irrigation or seasonal water resources), and 3) potential water-bearing structures or buried paleochannels that may serve as preferential flow paths for subsurface water.

In this paper, we use Hyperion data to extract the spectral signatures of various types and stages of vegetation patches (irrigated, non-irrigated, semi-natural mountain, coastal vegetation) to improve the results of classification of multispectral images such as those produced by the ASTER instrument. ASTER images have the advantage of providing spatial and temporal coverage of extended areas, which are needed for differentiating cyclical from progressive changes. In order to translate the hyperspectral information into a lower spectral resolution image, we adopted the following approach: 1) selection of same image acquisition dates, 2) conversion of DN to reflectance values and removal of the haze component, 3) application of a NDVI mask to limit the collection of spectral signatures (endmembers) to vegetated areas, 4) derivation of vegetation endmembers from the hyperspectral data and translation to the multispectral resolution using a spectral filter, 5) application of two classification procedures: the Spectral Angle Mapper and the unsupervised ISODATA classification method, 6) validation of the classification results using field observation (photographs) as well as a high resolution image (IKONOS) covering part of the data set, and finally 7) evaluation of the two classification results in terms of detecting and characterizing vegetation type and condition as related to water resources availability.