Optimizing spatial image support for quantitative mapping of natural vegetation

High spatial resolution images allow us to assess the spatial structure and variability of natural vegetation characteristics. In order to optimize quantitative image analysis, we looked for the optimal spatial resolution to map natural vegetation in a Mediterranean environment. The optimal spatial resolution was then related to the spatial characteristics of the scene and the mapped parameters. Based on the relation between airborne hyperspectral imagery and field dataset consisting of 227 measured plots of Leaf Area Index (LAI) and aboveground biomass, we found an optimal pixel size of 55 and 95 m for the mapping of LAI and aboveground biomass, respectively. The optimal spatial resolution we found is determined by the spatial structure of the mapped parameters on the one hand, and by the effects of shading patterns and gaps in the canopy on the other. The spatial properties of scene elements influencing the optimal spatial resolution can be detected effectively using average local variance functions and variograms, provided that these are studied at a wide spatial range. The use of an optimally sized mapping unit instead of the original 5 m pixels resulted in an improvement of mapping accuracy of 7 to 17%. It is therefore recommended that the image support used is considered carefully in all quantitative mapping projects based on remotely sensed imagery.