When a tree falls in the woods, can a satellite image tell you why?
A picture may be worth 1000 words, but when they're used to make policy decisions, or to explain ecological phenomena, they need to tell a consistent story. Matthias Baumann's work in the temperate forest zone of European Russia attempts to push the boundaries of land-use/land-cover change detection by focusing on disturbance processes through spectral and pattern recognition. 'Ultimately we want remote sensing to produce useful maps' says Matthias. Ecologists may use maps of landcover change to determine how carbon budgets change over time, whereas economists may use the same data to determine the extent and value of year-to-year timber stock. In both of these cases simply knowing that landcover has changed only gets the end-user so far. For instance, when economists consider all disturbances anthropogenic this may influence the results of an analysis to determine the best predictors of timber harvest (such as proximity to roads, or population density). Such results may then lead to changes in timber harvest policy. This trickle-down effect may result in suboptimal timber practices simply due to a lack of information in the map. Matthias's goal then, has been to develop techniques to further classify disturbance causes, which can be readily built upon established spatial analysis processing chains. In his most recent research he has developed a decision-making algorithm to differentiate between human-caused deforestation (i.e. timber harvest) and natural disturbance events such as windfall. This algorithm uses a linear transformation of tassel-cap components and thresholding to separate the two causes of disturbance. Through cross-validation, Matthias has demonstrated his method to be 80% accurate in choosing between timber harvest and windfall.