Major ecosystem observation programmes such as the Millennium Ecosystem Assessment (MEA), REDD (Reducing Emissions from Deforestation and forest Degradation) and subsequently REDD+, recommend the development of approaches capable of quantifying and spatialising ecosystem services to support the implementation of more appropriate environmental management practices and policies. Ecosystem service mapping could thus become an important tool for highpriority areas in terms of the environment. However, the approach still has a number of limitations, for example as regards carbon stocks in plant biomass. This ecological function was mapped on the scale of a 175 km2 locality in the Brazilian Amazon, to a spatial resolution of 30×30m. In order to quantify the carbon stocks, measurements of tree and shrub biomass in 45 different "points" were used together with geographical data obtained by remote sensing. To do so, two statistical methods were tested: the decision tree method and multiple linear regression. The statistical results from each of these methods are described here to show their advantages and disadvantages. Tests of the data adjustment quality of each model showed that while the decision tree method produces a better description of the role of explanatory variables, multiple linear regression is much more effective as a predictive tool as it gives a better picture of spatial variability for each type of land use. This method reveals terrain-specific phenomena on the scale of a single farm, thus allowing the result of an ecological process to be transcribed simply while also relating it to human activities. This study thus illustrates the importance of methodological choices in mapping a given process.
|Translated title of the contribution||Mapping carbon stocks in vegetation: Prospects for the spatialisation of an ecosystem service|
|Journal||Bois et Forets des Tropiques|
|Publication status||Published - 1 Dec 2013|