Detection of Non-stand-Replacing Cuttings Using Satellite Image Time Series With a Kalman Filter

Detection of non-stand-replacing disturbances is becoming highly important for forest monitoring. Methods to detect severe, stand-replacing disturbances are increasingly mature. On the other hand, non-stand-replacing (NSR) disturbances, though less severe, may cover large forest areas and thus pose an important monitoring problem, yet they remain difficult to detect in optical satellite image time series. The often very subtle change signals can usually be detected only for a brief time period, and the signal is often obscured by noise. In this study, we tested three detection algorithms that consider abruptness of and prior information about spectral changes to varying degrees. These detectors, in combination with a Kalman filter, were applied to detect both stand-replacing and NSR anthropogenic forest disturbances in six study sites throughout Finland. Clearcuts and heavy thinnings were easily detected without application of advanced detectors, but low intensity thinnings were detectable only by using both prior information and the assumption that changes are abrupt. The best detector achieved an F1-score of 0.72 for thinnings detection, compared to