Fig. 4 | Scientific Data

Fig. 4

From: Historical insect disturbance maps from 1985 onwards for Canadian forests derived using earth observation data

Fig. 4

Illustration of the pixel-based method for detecting and classifying disturbances. From the Landsat NBR time series (blue x symbol), LandTrendr detects two starting break points (red points) of disturbance segments (red lines). A 10-year Landsat subsequence is extracted for each starting break point, which is then passed unto the TempCNN model for disturbance classification. Using the start and the end years of the LandTrendr segment and the classes predicted by TempCNN, a disturbance time series is created, with the type of disturbance for each year. In the case of rapid disturbances (wildfire, harvesting, windthrow), only the first year of the disturbance is retained in the time series. For pest disturbances, the information is preserved throughout the duration of the event, as defined by the LandTrendr segmentation.

Back to article page