Subpixel measurement of mangrove canopy closure via spectral mixture analysis

摘要

Canopy closure can vary spatially within a remotely sensed image pixel, but Boolean logic inherent in traditional classification methods only works at the whole-pixel level. This study attempted to decompose mangrove closure information from spectrally-mixed pixels through spectral mixture analysis (SMA) for coastal wetland management. Endmembers of different surface categories were established through signature selection and training, and memberships of a pixel with respect to the surface categories were determined via a spectral mixture model. A case study involving DigitalGlobe’s Quickbird high-resolution multispectral imagery of Beilun Estuary, China was used to demonstrate this approach. Mangrove canopy closure was first quantified as percent coverage through the model and then further grouped into eight ordinal categories. The model results were verified using Quickbird panchromatic data from the same acquisition. An overall accuracy of 84.4% (Kappa = 0.825) was achieved, indicating good application potential of the approach in coastal resource inventory and ecosystem management.