Small Area Canopy Cover Change Estimation Using Sentinel 2 Data

Abstract

In recent decades, numerous studies on forest disturbance monitoring exploited remote sensing data, thanks also to the increased availability of open satellite data. Many studies made use of Sentinel 2 data to map large areas affected by destructive events. Often, the fine spatial resolution of Sentinel 2 was used in model construction, whereas detections were given for wider areas. This article studied the potential of Sentinel 2 data to detect minor crown cover changes or intense changes but at high spatial resolution. Two models were developed using random forest, one for detecting crown cover, to be applied at two points in time for change assessment, and the other for direct cover change detection. The models performed surprisingly well in detecting dense cover (> 80%) and change occurrence at 20 m pixel size. The results were confirmed by external validations (producer accuracy 71.4% and 95.7%). Finally, the models output was implemented in a simulation study to estimate areas that suffered crown cover reduction by a forest inventory approach. Based on our results, Sentinel 2 data are fruitfully usable for detecting 20 m size pixels where a cover change occurred and extremely useful for supporting activities aimed at assessing the intensity of such changes. Nevertheless, Sentinel 2 data alone are not enough to detect minor crown cover changes, and additional sources of data are required for this purpose.