Modeling Cheatgrass using Quickbird Imagery

Cheatgrass (Bromus tectorum) is of growing concern among farmers, ranchers, and researchers in the intermountain west due to its ability to alter rangeland vegetation communities and their associated fire regimes. Quickbird multi-spectral imagery (spatial resolution 2.4 mpp) was used to detect and model cheatgrass presence in southeastern Idaho. Training sites were created using 2003 field data. The four bands of Quickbird imagery were processed using IDRISI software and various vegetation indices were created using the red and near-infrared bands of the Quickbird imagery. Principal Components Analysis (PCA) was performed using eight vegetation indices and the four original bands of data. Redundant data was minimized using results from PCA. Maximum likelihood classification was then used to classify the imagery to model the presence of cheatgrass. The accuracy of the resulting model was assessed using an error matrix and Kappa statistic. An overall accuracy of 73% was achieved through this process.