A Service Level Agreement between the subscriber and the service provider defines which products were provided by each service. The service provider specified the data, methods and processes used in the end products. Typical user-defined products included

  • forest cover and general land cover composition

  • biomass

  • degree of disturbance

together with the apparent changes of those variables from the early 1990’s until today. Some examples of products and original image mosaics are shown below.


Low-resolution examples of selected typical deliverables from two study sites.

Forest monitoring on five study sites

The empirical forest monitoring was performed on five study sites for the local users: the Chiapas state in Mexico, a sub-region of Colombia, the entire Guyana, a selected site in the Democratic Republic of the Congo (DRC), and the territory of Fiji islands. The image data were available since the early 1990’s from Landsat Thematic Mapper, Spot 4 and Spot 5, RapidEye, GeoEye, Kompsat 2, Quickbird 2, ERS-1&2, Envisat ASAR, ALOS PALSAR and AVNIR-2 instruments, depending on the site and the target year of image interpretation.

Forest cover and biomass products of good quality

Two-stage sampling approaches were developed for the forest cover assessment. The sampling ensures, when comprising the whole area of interest, provision of objective statistical data with confidence intervals on the variables, relevant to REDD. The sampled data were also used to assess the accuracy of the wall-to-wall mapping and to correct the bias in the map estimates. Very fine resolution satellite data were used to collect the reference sample for the accuracy assessment.

Generally, (semi-)automated classification approaches based on high resolution (10 – 30 m) optical satellite imagery could distinguish forest land from other land with overall accuracy levels close to or above 90%. The classification accuracy exceeded 90 % in areas with high forest cover and limited anthropogenic influence. The overall accuracy was somewhat lower when considering all six IPCC classes (accuracy between 60-70%) than in forest and non-forest classification. For example, the separation of grassland from cropland remained problematic.


Elements of production: (a) Landsat TM image 1994, (b) RapidEye image 2009, (c) change map 1994-2009 with forest removal shown in red, (d) forest biomass map 2009, with maximum prediction 184 t/ha. Area size 2.7 km by 2.9 km.

Easy user access through WebGIS

Users can access the delivered products using the specially developed WebGIS platform. The ReCover Toolbox 1.0 is a web application that can be used with browser. Registered users can access and browse the ReCover project products with the purpose of their visualization and download.


An example of user interface for the developed WebGIS, with selected product visualized on the map.