Resumen
The attribution of forest disturbances to disturbance agents is a critical challenge for remote sensing-based forest monitoring, promising important insights into drivers and impacts of forest disturbances. Previous studies have used spectral-temporal metrics derived from annual Landsat time series to identify disturbance agents. Here, we extend this approach to new predictors derived from intra-annual time series and test it at three sites in Central Europe, including managed and protected forests. The two newly tested predictors are: (1) intra-annual timing of disturbance events and (2) temporal proximity to windstorms based on prior knowledge. We estimated the intra-annual timing of disturbances using a breakpoint detection algorithm and all available Landsat observations between 1984 and 2016. Using spectral, temporal, and topography-related metrics, we then mapped four disturbance classes: windthrow, cleared windthrow, bark beetles, and other harvest. Disturbance agents were identified with overall accuracies of 76?86%. Temporal proximity to storm events was among the most important predictors, while intra-annual timing itself was less important. Moreover, elevation information was very effective for discriminating disturbance agents. Our results demonstrate the potential of incorporating dense, intra-annual Landsat time series information and prior knowledge of disturbance events for monitoring forest ecosystem change at the disturbance agent level.