USING MULTITEMPORAL DATA ANALYSIS AND GEOMORPHOLOGY TO PREDICT SITES OF CHEATGRASS INVASION

Hyperspectral data has become increasingly popular with remote sensing scientists due to the vast spectral information it contains in comparison to multispectral data. The disadvantages of hyperspectral data include the difficulty in processing and the large expense of the imagery. Temporal stacking of multispectral data (e.g. Landsat) can provide an inexpensive alternative to hyperspectral data when used as a monitoring tool for features with seasonal variation (e.g. vegetation, snow cover, soil moisture). Relative to a number of remote sensing algorithms, temporal stacking can, in a rough mathematical sense, effectively increase the ‘spectral’ resolution of the system. This study combines three Landsat scenes to produce a temporal data stack which is then processed similar to a hyperspectral data cube. Cheatgrass (Bromus tectorum) occurrences are distinguished based on their temporal differences in the stack. Judicious selection of a good hyperspectral analysis technique (e.g. linear spectral unmixing or matched filtering) helps delineate features of interest which are not possible by conventional multispectral classification techniques. Information about the landscape and geomorphology (e.g. topography and soils) of the study area are used to further refine the classifications. Recent research indicates that soil type (soil texture and chemistry), elevation, and aspect all influence cheatgrass occurrence. These ancillary data sets are processed in combination with the temporal data stack. The techniques outlined in this study are applicable to studies in environmental geology, hydrology, and geomorphology. For example, this technique can be applied to monitor hazardous waste and pit sites, and to monitor changes in snow cover extent and relative soil moisture.