Phenology-guided saltcedar (Tamarix spp.) mapping using Landsat TM images in western U.S.

- We modeled the timing of saltcedar coloration using MODIS EoST data.
- We constructed composite images using the modeled timing of saltcedar coloration.
- Composite image captured across- and within-scene variation in saltcedar phenology.
- Saltcedar classification accuracy was improved using composite images.

Accurate distribution maps are essential to understand further and control the fast expansion of the introduced saltcedar (Tamarix spp.) in the western U.S. Remote sensing imagery, due to its broad spatial and temporal coverage, is well-suited for this task. More specifically, images acquired during late fall and early winter are used for saltcedar identification as the plant's foliage turns a unique orange-yellow color (coloration) during this time of a year, which makes them easily distinguishable from other plant species. However, saltcedar coloration usually lasts for only several weeks, and the timing of this phenological event varies across space. Therefore, without prior knowledge of saltcedar phenology, it is often challenging to select the remote sensing image collected at the optimal time of the year to capture the distinctive spectral signature of the coloring saltcedar. Moreover, the timing of saltcedar coloration also varies within a single remote sensing scene making it inappropriate to use an entire scene as the mapping unit. In our study, we found that the timing of saltcedar peak coloration and leaf drop were linearly correlated with each other and we were able to model this relationship using the MODIS End of Season-Time (EoST) product and Landsat TM images. Guided by a per-pixel model estimation of saltcedar coloration date, we constructed composite Landsat images at two study sites along the Rio Grande River such that each pixel in the composite images was acquired during saltcedar peak coloration. Classification results showed that the phenology-informed composite image was better for saltcedar identification than the single scene image as it was able to capture the within-scene variations in saltcedar phenology. In addition, given our lack of training data for the model, the successful classification results obtained at the Middle Rio Grande River site also demonstrated the potential of our proposed method to be applied for large-scale saltcedar mapping.