Edzer Pebesma, ifgi; Lubia Vinhas, INPE.

Change detection from imagery data is often performed conditional to time, meaning that time snapshots are classified independently and compared afterwards. This makes it hard to evaluate the statistical properties of the changes found. We will explore possibilities for assessing change from image time series based on unconditional, joint analysis and classification of the series, and compare this to the conditional approach. We will also look into the different requirements with respect to the collection of ground truth data that arise from both approaches. The application domain is deforestation in the Amazon area.

to be presented at: 2009-10 Program on Space-time Analysis for Environmental Mapping, Epidemiology and Climate Change; Opening Tutorials & Workshop September 13-16, 2009