FORECAST - Forest Cartography using very high resolution SaTellite data

You are here

Context and objectives

One of the main objectives of this project is to transfer the research results of the scientific world in VHR satellite image processing in forest mapping to the private sector and to develop operational procedures for image processing and map production. The Catholic University of Louvain-La-Neuve (Unit of Environmetry and Geomatics and Unit of Water and Forests) and the private operator I-MAGE Consult have joined their efforts in this way, and different organisations1 – one for each of the three sites selected – will be requested as end users in the project for the definition of the technical requirements and the validation of the product/service to be developed.

Project outcome

Expected scientific results

The technological transfer was conducted in such a way that I-Mage consult is now able to respond to some main requirements of forest managers thanks to very high resolution satellite images. Two end user products, namely forest stand delineation and extraction of stand parameters from existing vector database, were achieved through operational methodologies.
Considering the required planimetric quality of the final products and the processing costs, Toutin’s orbital model and polynomial function with RPC file were the most appropriate for SPOT 5 and IKONOS respectively. In all cases, the use of SRTM digital elevation model was suitable.
The planimetric quality of automated forest stand delineation mostly depended on the geometry of acquisition and on the spatial resolution. Multispectral IKONOS images with zenith angles lower than 15 degree were most appropriate for 1:20000, while PAN + XS SPOT 5 images were a much cheaper alternative when smaller scales were tolerated. The new generalization algorithm helped improving precision and enhancing visualization while conserving topology.
Fuzzy rule decision tree gave more than 90 % overall accuracy with 5 forest classes. In terms of forest stands characteristics, texture parameters were only successful in coniferous stands and species recognition was hindered by high variance due to age and stand structure. The use of a stereo imagery was very efficient in visual interpretation but resulted in 10% of errors with the automated height extraction.