Landsat time series-based multiyear spectral angle clustering (MSAC) model to monitor the inter-annual leaf senescence of exotic saltcedar

In the western United States, the rapid expansion of exotic saltcedar along riparian corridors has drastically altered landscape structures and ecosystem functions. Monitoring the geographical distribution and spatio-temporal dynamics of this invasive species is essentially critical to conduct the systematic restoration of affected riparian ecosystems. Previous studies indicated that the leaf senescence stage is the optimal time window to remotely monitor saltcedar distributions. Yet due to climate variability and anthropogenic forcing, the timing of saltcedar leaf senescence varies over space and time. Given that the saltcedar leaf senescence stage only lasts for a short temporal window (i.e., three or four weeks), pinpointing the appropriate Landsat image across years and locations without the expert knowledge is challenging. Remotely sensed phenological time series analysis provides a practical means to locate the leaf senescence date on a per-pixel basis. However, affected by temporal revisit rates and cloud contamination, Landsat time series can only capture very limited temporal segments of vegetation phenology. The lack of Landsat imagery throughout the year makes the conventional time series analysis difficult. In this study, we developed a multiyear spectral angle clustering (MSAC) model to monitor the inter-annual leaf senescence of saltcedar with limited Landsat imagery on a per-pixel basis. The MSAC model leverages the Landsat images across years to temporally predict the fall phenology of plant species in a single year, and constructs the synthesized time series of spectral signatures to estimate critical phenological transition dates. Results indicated that the MSAC model could guide the construction of the composite Landsat surface reflectance image to accommodate spatial and inter-annual variation in the timing of plant leaf senescence. The phenology-guided composite image from the MSAC model achieved a greater saltcedar mapping accuracy than any single Landsat image in 2004. The proposed MSAC model provides new insights in estimating phenological transition dates of plant species with limited Landsat imagery. It opens up unique opportunities to guide the selection of representative remotely sensed imagery on a per-pixel basis for repetitive saltcedar mapping over wide geographical regions.