Earth Observation to Enhance Forest Certification
Forest certification is a process that fosters sustainable forest management by empowering consumers to purchase sustainable timber products and allowing timber supply chain members to prove the sustainability of their operations and products.
A major point of criticism towards certification procedures is that they rely on document-based assessments to evaluate forest management practices and do not explicitly monitor logging activities, environmental impacts or track individual timber supplies.
The goal of the EO-EnForCe project was to develop a web-based prototype for a forest monitoring system that records legal and illegal logging within and around certified forest areas and is operated as an on-demand and globally available service. In the course of the project it became evident that full development towards an interactive on-demand service was not feasible in the foreseen project run time. Nonetheless, the key components of the processing chain have been implemented, but without providing an interactive user interface and online processing capabilities. The project results are demonstrated for three pre-selected areas of interest (AOIs) representing certified forest areas.
The project is funded under the “Austrian Space Applications Programme ASAP” of The Austrian Research Promotion Agency (FFG).
EO-EnForCe Forest Monitoring Demonstrator
The three demonstration sites are located in Indonesia on the island of Borneo. The island is crossed by the equator and features a tropical climate. In the past decades, more and more tropical rain forest has been destroyed, very often in order to replace it with palm oil plantations. Satellite-based Earth observation (EO) provides the means to map and document this large-scale deforestation and generally collect information about land use and land cover (LULC). In order to independently confirm that timber is harvested only from legal concession areas, the ability to map harvested areas with high spatial accuracy and reasonable timeliness is required.
For each demonstration site, several output layers have been computed offline and integrated into a web-based viewer for display. Two vector layers show the outlines of the EO data coverage and the limits of the concession area including a 2 km buffer. The concession and buffer area may be queried in order to show disturbance statistics for the respective polygon. There are four groups of raster layers:
(1) Initial forest parameters – This group contains a forest mask representing the state at the end of the year 2021 as well as a classification into primary and secondary forest. Please note that the secondary forest class also includes oil palm plantations, although these are thematically not considered as forest.
(2) EO-EnForCe results – Disturbance maps produced in the project. Two processing systems have been implemented, one based on Sentinel-1 synthetic aperture radar (SAR) imagery and the other based on Sentinel-2 optical imagery. The required EO input data has been acquired using the SentinelHub service. For the demonstrated use case, both systems extracted disturbance patches with a minimum size of 0.1 ha (minimum mapping unit, MMU) which occurred in the year 2022. Both systems generate a 0/1 mask, where 1 represents forest cover change areas within the initial forest layer. The analysis of Sentinel-1 data is based on temporally filtered SAR backscatter information in VV and VH polarization. The workflow extracts the minimum backscatter over the selected time period. A backscatter threshold is applied to extract areas with low backscatter values that are likely to present non-forest areas. Also the Sentinel-2 processing system uses temporal filtering, but in a different way. First, historic imagery (i.e. preceding the monitoring period) is used to train mathematical models representing the image content and its expected variations over the year. These mathematical models are dynamically updated with new information and can be used to create dynamic forecasts in the form of synthetic images. New deforestation patches are detected by evaluating the differences between forecast and observed image. The third layer in this group is a combination of the SAR and optical system outputs.
(3) Global Forest Watch (GFW) products – For comparison, additional deforestation maps available at the GFW Open Data Portal (https://data.globalforestwatch.org/) have been integrated into the viewer. The RAdar for Detecting Deforestation system (RADD, Reiche et al., 2021) uses Sentinel-1 data to extract disturbance patches with a MMU of 0.1 ha. Individual pixels may be classified as low or high confidence (suspected vs. confirmed alert). GFW also produces a product called Integrated Deforestation Alerts, where results from the RADD and GLAD-L (based on Landsat, Hansen et al., 2016) systems are combined. The original data format has been simplified in order to harmonize it with the project results.
(4) Filtered Sentinel-2 data for 2022 – The output of the optical time series filtering algorithm, gap-less and cloud free synthetic images generated in a 30-day interval. The image synthesizer uses all imagery available up to and including the generation target date in order to produce a best estimate of the current ground state. However, a temporal lag must be considered.