Automated Detection of Selective Logging Using SmallSat Imagery

We propose an automated processing workflow to detect and classify changes in bitemporal very high-resolution (VHR) SmallSat imagery. The workflow consists of two preprocessing steps: an image registration method with a cross correlation approach and, second, a radiometric normalization based on regression of automatically detected invariant pixels using the iteratively reweighted multivariate alteration detection (IR-MAD) method. The IR-MAD method also transforms pairs of registered images and detects the pixels of change. Finally, we distinguish different types of change using a clustering algorithm on the original spectral values as well as on the values resulting from different image transformations. We applied this workflow to SkySat images of Rennell Island (Solomon Islands) to detect selective logging and subsequent regrowth within a conservation area. We show how the use of VHR SmallSat images enables the detection of selective logging in the tropics, where significant cloud cover and the small extent of disturbances can prevent detection using data from traditional earth-orbiting platforms. The fine temporal resolution data from SmallSat constellations can generate enough valid observations to rapidly detect disturbances (even when cloud cover is frequent) and partial recovery. In addition, SkySat imagery provides sufficiently high spatial resolution to detect crown-level changes such as those from selective logging. Our change detection workflow is fast and highly accurate, and will be increasingly useful as the data flow from earth-observation sources increases.