North American forest disturbance mapped from a decadal Landsat record

Long-term vegetation dynamics associated with climatic changes can be assessed using Advanced Very High Resolution Radiometer (AVHRR) red and near-infrared re fl ectance data provided that the data have been processed to remove the effects of non-target signal variability, such as atmospheric and sensor calibration effects. Here we present a new method that performs a relative calibration of re fl ectance data to produce consistent long-term vegetation information. It is based on a simple biological framework that assumes that the position of the vegetation cover triangle is invariant in re fl ectance space. This assumption is in fact an intrinsic assumption behind the commonly used Normalised Difference Vegetation Index (NDVI) and is violated when the NDVI is calculated from inadequately corrected re fl ectance data. In this new method, any temporal variability in the position of the cover triangle is removed by geometrically transforming the observed re fl ectance data such that two features of the triangle — the soil line and the dark point — are stationary in re fl ectance space. The fraction of Photosynthetically Active Radiation absorbed by vegetation (fPAR; 0.0 – 0.95) is then calculated, via the NDVI, from calibrated re fl ectances. This method was tested using two distinct, monthly AVHRR products for Australia: (i) the coarse-resolution, fully calibrated, partially atmospherically corrected PAL data (1981 – 1994); and (ii) the fi ne-resolution, fully calibrated, non-atmospherically corrected HRPT data (1992 – 2004). Results show that, in the 20-month period when the two datasets overlap (1992 – 1994), the Australia-wide, root mean square difference between the two datasets improved from 0.098 to 0.027 fPAR units. The calibrations have produced two approximately equivalent datasets that can be combined as a single input into time-series analyses. The application of this method is limited to areas that have a wide-enough variety of land-cover types so that the soil line and dark point are evident in the cover triangle in every image of the time-series. Another limitation is that the methodology performs only bulk, relative calibrations and does not remove the absolute effects of observation uncertainties. The simplicity of the method means that the calibration procedure can be easily incorporated into near-real-time operational remote-sensing environments. Vegetation information produced using this invariant-cover-triangle method is expected to be well suited to the analysis of long-term vegetation dynamics and change.