Phenology guided saltcedar mapping using Landsat TM images in western U.S.
The distinctive spectral characteristic of the invasive saltcedar (Tamarix spp.) during coloration has facilitated their detection in many studies using remote sensing images. However, the best calendar-date timing of acquiring images for saltcedar mapping usually cannot be generalized across or even within scenes due to spatial-temporal variations in saltcedar phenology. In order to account for these variabilities, we first build a model to estimate the timing of saltcedar coloration at a study site along the Lower Rio Grande River. Guided by the estimated coloration date, we then constructed a composite Landsat image of which each pixel was acquired during the period of saltcedar coloration so that the unique spectra signature of coloring saltcedar could be captured at a spatial scale of 30 m. Composite Landsat images were also constructed for another study site along the Middle Rio Grande Rive. Results of saltcedar classification using composite images were compared to those using single scene images at both study sites.
The results demonstrated that the timing of saltcedar peak coloration was linearly correlated with the end of season time. This relationship was then applied to derive the pixel-wise coloration date of saltcedar coloration of the composite images using MODIS End of Season-Time (EOST) data. The producer’s and user’s accuracy of saltcedar classification using composite images achieved 75% and 70% in 2005 at the Lower Rio Grande River site; 75% and 80%, and 75% and 70% in 2005 and 2011, respectively, at the Middle Rio Grande River site. Classification accuracies of the composite images at both study sites outperformed those of single scene classification results. Our results also indicated that the relationship between saltcedar coloration and its end of season time has the potential to be generalized across time and space. We believed that large scale, long-term saltcedar monitoring along western river corridors can be achieved using our proposed method.