Near - Real Time Tropical Forest Cover Loss Detection with High-Resolution Satellite Data

Organised by Laboratory of Geo-information Science and Remote Sensing

Tue 27 August 2019 10:00 to 10:30

Venue Gaia, gebouwnummer 101
Room 2

By Julian Kremers

Forest cover loss remains a threat to tropical rainforest ecosystems. To prevent future forest cover loss in the tropics actionable insights are needed in real-time. Local authorities should see changes in the forest cover as soon as they occur. Such information on a large scale is unavailable despite the increasing public availability of satellite data for forest monitoring. PlanetScope, a novel constellation of shoebox-sized satellites, offers a new data source combining high spatial resolution with low revisit intervals. However, the data is not accessible in an analysis-ready format but requires intensive pre-processing.

This thesis research evaluates the usability of PlanetScope data for the detection of forest cover loss in near real-time. An analysis-ready data stack is created from PlanetScope satellite data. In addition, a methodology is developed to detect forest cover loss from such an image stack. Both, stack and method are tested in a scenario for forest cover loss detection for 2018 in northern Riau, a province on the island of Sumatra, Indonesia.

The co-registration algorithm AROSICS and the Planet tmask algorithm to mask out clouds are used to create the analysis-ready data stack. With the data and the developed methods changes in the forest cover in Riau can be detected with a producer’s accuracy of up to 90% and a user’s accuracy of up to 99%. An overall accuracy of 94% can be achieved. On average (median) it takes 12 days for a change to be detected and 13 days for a change to be confirmed. However, for a large part of the study area no data are available between October 2018 and January 2019.

Nonetheless, PlanetScope is a useful data source for forest cover loss detection in near-real time and could be used together with MODIS, Landsat, and Sentinel data to further improve the accuracy and timeliness of change detection in tropical forest monitoring.