Taking the Pulse of Tropical Forests in and around UNESCO Workd Heritage Sites
Jonas, Ma´tÚ; Radoux, Julien; Defourny, Pierre
UniversitÚ catholique de Louvain, BELGIUM
Nowadays, UNESCO accounts for around 80 tropical forest sites among its World Heritage (WH) sites, as a support for REDD+ and biodiversity conservation. However, evaluating the state of conservation of each site every year is a big challenge because of the limited accessibility to these sites and their large area. While sustainably managing these sites requires good knowledge about forest state and changes occurring into them, only 15% of the WH sites were discussed in the WH Committee 2009 meeting.
In order to provide an efficient way to evaluate the state of conservation of the WH tropical forest sites and their neighborhood, an automated object based method was proposed and is implemented through the UNESCO-Watch project. In this project, high resolution remote sensing imagery is used in order to detect deforestation and forest degradation patterns.
The proposed method has been developed based on three different historical epochs (1990's, 2000's, 2010's). Images have been collected for 12 WH tropical forest sites located around the world from South America to Australia. A 20 km buffer region has been added to the official boundaries in order to account for potential external threats and leakages, and to evaluate if the protected area status had an impact on the land cover dynamic in and around the site. In total, the study areas cover around 170 000 sq km of tropical forests.
The method is divided into three main successive steps : i) preprocessing, which also includes automated cloud removal ii) multi-dates image segmentation and iii) wall-to-wall forest change detection. All these steps have been designed to be as much automated as possible so that it could become operational when high spatial and temporal resolution images from satellites like Sentinel-2 will be available.
Indeed, cloud cover probability is large in the tropics, so that high temporal resolution is arguably needed to compensate this effect. For instance, cloud free wall-to-wall coverage of the test sites could only be achieved at 90 percent by combining archived SPOT and Landsat images with a two year tolerance around each epoch. One of the main challenge of this project was therefore to achieve a very high cloud detection accuracy but to keep a reasonnable level of commission errors in order to avoid unnecessary gaps after the cloud removal.
A geographic object-based analysis framework has been implemented for the delineation of multidate object using SPOT and/or Landsat images. The segmentation parameters are automatically adjusted to be consistent with the minimum mapping unit. The attributes from these objects are then transmitted to the third epoch to evaluate its time trajectory.
The main challenge of the change detection method is its capability to evaluate change in any tropical forest sites, regardless the landscape, the vegetation or the degradation patterns we are facing to. The detection of outliers using statistical iterative trimming helps to identify the objects that are likely to have changed from one time period to another. Simple rule-based systems allow to increase the overall performance of the change detection by making sure that iterative trimming is used within its scope.
The final products are aggregated on larger changed polygons classified as deforestation, degradation, reforestation or forest regeneration. This information could help the UNESCO to more quickly identify endangered site.