Mapping forest vertical structure and disturbance characteristics with time series of gap-filled Landsat imagery

Background/Question/Methods

Forest vertical structure and disturbance history affect woody species composition and avian habitat.  Remote sensing of forest vertical structure is possible with lidar data, but lidar data are not widely and freely available. Landsat time series allow mapping of forest disturbance dates, but this process can be difficult in persistently cloudy regions with comparatively few scenes in the Landsat data record. In this study, we create a time series of cloud-gap-filled Landsat images and test if we can use it to map forest attributes important to avian habitat, including forest vertical structure, substituting time for vertical canopy space, as well as both disturbance type and age.

Results/Conclusions

With regression tree classification of a time series of Landsat and Advanced Land Imager (ALI) imagery for the island of Eleuthera, The Bahamas, we accurately mapped tropical dry forest height (RMSE = 0.9 m, R² = 0.84, range 0.6-7 m) and foliage height profiles, with a time series of Landsat and Advanced Land Imager (ALI) imagery for the island of Eleuthera, The Bahamas.  We also mapped disturbance type and age via decision tree classification of the image time series. Having mapped these variables in the context of studies of wintering habitat of an endangered Nearctic-Neotropical migrant bird, the Kirtland’s Warbler (Dendroica kirtlandii), we then illustrated relationships between forest vertical structure, disturbance type and counts of forage species important to the Kirtland’s Warbler. The ALI imagery and the Landsat time series were both critical to the result for forest height, which the strong relationship of forest height with disturbance type and age facilitated. Also unique to this study was that seven of the eight image time steps were cloud-gap-filled images: mosaics of the clear parts of several cloudy scenes, in which cloud gaps in a reference scene for each time step are filled with image data from alternate scenes. We created each cloud-cleared image, including a virtually seamless ALI image mosaic, with regression tree normalization of the image data that filled cloud gaps. We also illustrated how viewing time series imagery as red-green-blue composites of tasseled cap wetness (RGB wetness composites) aids reference data collection for classifying tropical forest disturbance type and age.