ReCover: A Concept for Tropical Forest Assessment for REDD
Häme, Tuomas1; Sirro, Laura1; Cabrera, Edersson2; Haarpaintner, Jörg3; Enssle, Fabian4; Hämäläinen, Jarno5; de Jong, Bernardus6; Paz Pellat, Fernando7; Pedrazzani, Donata8; Reiche, Johannes9
1VTT Technical Research Centre of Finland, FINLAND; 2IDEAM, COLOMBIA; 3Norut, NORWAY; 4ALU-Freiburg, GERMANY; 5Arbonaut Oy, FINLAND; 6ECOSUR, MEXICO; 7COLPOS, MEXICO; 8GMV, SPAIN; 9WU, NETHERLANDS
The ReCover project under the Framework Program 7 of the European Union develops beyond state-of-the-art service capabilities to support fighting deforestation and forest degradation in the tropical region in the context of the REDD process (Reducing Emissions from Deforestation and Forest Degradation). The ReCover project specifically contributes to the reduction of errors in the estimation of terrestrial carbon balance that result from uncertain rates of tropical deforestation. This is achieved by developing and implementing satellite image based methods for the monitoring of tropical forests. The need for effective use of these techniques is motivated by the fact that many developing countries lack human resources and funding for detailed forest inventories.
The services are based on the service level agreements. This ensures a close cooperation with the service providers and users in Mexico, Guyana, Colombia, DRC, and Fiji. The satellite data applied include ERS-2 SAR, Envisat ASAR, ALOS PALSAR and AVNIR-2, Terra MODIS, RapidEye, Landsat and several VHR satellite data. Target variables are IPCC compatible land cover classes and their changes as well as biomass and degradation.
Evaluation of the reliability of the estimation is one of the key focuses of ReCover. For this purpose, a statistical concept that applies two phase sampling was developed. The first phase sample is a wall-to-wall map with data of 10 - 30 meter resolution. The second phase is a sample within VHR images that are collected from 1-3 % of the area of interest. The VHR data sample provides reliable information on the accuracy of the wall-to-wall mapping. VHR data are also used to train the models in wall-to-wall mapping but the training data are not used for the accuracy assessment.
A separate study in which different service providers produced the land cover class estimates with the same training data in Mexico was conducted. This gave information on the sensitivity of the estimation to the applied analysis method and image data type. The estimations were assessed with a statistical VHR sample that was not in the possession of the service providers. A party that did not conduct the estimation was responsible for the testing.
The paper presents the main results of ReCover project that started in November 2010 and will be completed in October 2013. Also the outlook beyond the ReCover project will be presented. Sentinel 2 is foreseen to be an ideal basic data source for REDD. Sentinel 2 data should be augmented with SAR data on cloudy regions and for change and with VHR data for model training and accuracy assessment.