COMPARISON BETWEEN REMOTE SENSING CLASSIFICATION AND STOCHASTIC SIMULATION TECHNIQUES FOR PREDICTING AREAS OF WEED INVASION

The objective of this study is to develop a procedure for mapping and predicting the spatial occurrence of cheatgrass (Bromis tectorum) based on a comparison of remote sensing and geostatistical techniques. Mapping invasive plants like cheatgrass using remotely sensed data is becoming popular; however similar spectral signatures with surrounding vegetation and large soil reflectances often make it difficult to successfully delineate the target weed. Supplementary techniques may be resorted to in order to achieve the desired result. Analysis of multitemporal Landsat imagery, along with elevation and soil stratification techniques, are being evaluated in this study to improve prediction of cheatgrass occurrences in southeast Idaho. The geostatistical technique of sequential indicator simulation is also being evaluated to predict sites of cheatgrass invasion; simulation reproduces patterns of spatial continuity along with local variability, based on ground-truth data and its geostatistical autocorrelation structure. This study compares the results from both techniques and ways in which the results from one could be used to improve the performance of the other.