Landsat remote sensing of forest windfall disturbance

a r t i c l e i n f o Knowing if a forest disturbance is caused by timber harvest or a natural event is crucial for carbon cycle assessments , econometric analyses of timber harvesting, and other research questions. However, while remote sensing of forest disturbance in general is very well developed, discerning between different types of forest disturbances remains challenging. In this work, we developed an algorithm to separate windfall disturbance from clear-cut harvesting using Landsat data. The method first extracts training data primarily based on Tasseled Cap transformed bands and histogram thresholds with minimal user input. We then used a support-vector machine clas-sifier to separate disturbed areas into 'windfall' and 'clear-cut harvests'. We tested our algorithm in the temperate forest zone of European Russia and the southern boreal forest zone of the United States. The forest-cover change classifications were highly accurate (~90%) and windfall classification accuracies were greater than 75% in both study areas. Accuracies were generally higher for larger disturbance patches. At the Russia study site about 60% of all disturbances were caused by windfall, versus 40% at the U.S. study site. Given the similar levels of accuracy in both locations and the ease of application, the algorithm has the potential to fill a research gap in mapping wind disturbance using Landsat data in both temperate and boreal forests that are subject to frequent wind events. Forests play an important role in the global carbon cycle and the provision of ecosystem services. Information on where and to what extent forest disturbances occur globally is thus a crucial necessity (Achard et al., 2002; Bonan, 2008). Remote sensing can provide accurate and timely information regarding forest disturbance in many ecoregions at scales ranging from local to global and at many different temporal resolutions (Achard et al. Enhanced Thematic Mapper Plus (ETM +) instruments have been used for many of these studies because of (1) the favorable combination of spatial, spectral and temporal resolution, (2) the free availability of In most forest disturbance mapping studies that utilize Landsat data, the derived change products only identify areas of 'forest disturbance', but do not discriminate among different types of disturbances The lack of attribution to the type of disturbance often makes it difficult to interpret forest disturbance maps, especially when these data are used as inputs to carbon budget assessments or econometric analyses. For example, many studies that seek to …