Remote sensing and the measurement of geographical entities in a forested environment. 2. The optimal spatial resolution

Abstract

The prime objective of this study was to propose and test a method to identify the optimal spatial resolutions for detection and discrimination of coniferous classes in a temperate forested environment. The approach is based on the paradigm that there is an intricate relationship between the definition and the measurement of geographical entities and implies the following steps: 1) a priori define the geographical entities under investigation, 2)determine an optimization criterion for the choice of a sampling system, 3) progressively aggregate data acquired from a fine spatial sampling grid, 4) apply the optimization criterion on the series of spatially aggregated data, and 5) verify the validity of the results obtained in relation to the goal of the study. Airborne MEIS-II data, acquired at 0.5 m in eight spectral bands of the visible spectrum, were used for the study. Fourteen forest classes, at the stand level, were defined on the basis of four attributes: species, density, height, and organization of the trees. Representative sites for each forest class were selected. From the center of each site, the spatial resolution of the original data was degraded to 29.5 m, with an increment of 1 m, using an averaging window algorithm. The intraclass variance was calculated for each forest class, at every spatial resolution and for the eight spectral bands. The minimal variance was used as the indicator of the optimal spatial resolution. To evaluate the importance of the optimal resolution for class discrimination, a bivariate test of variance was performed for each pair of forest class considered at their optimal spatial resolution. Profiles of spectral separability were also established in relation to the whole series of spatial resolutions. The results show that, for all coniferous classes and for the eight spectral bands considered in the study, there is a minimal value in intraclass variance that indicates the optimal spatial resolution for each class, varying between 2.5 m and 21.5 m. The optimal spatial resolution is primarily affected by the spatial and structural parameters of the forest stands. The analysis of variance between each pair of forest classes considered at their respective optimal spatial resolution reveals that all classes are significantly different in at least two spectral bands, except for 10 pairs. The spectral separability of the forest classes is at a maximum at, or very close to, their optimal spatial resolution. The study confirms the validity of the concept of optimal spatial resolution and proposes an original solution to the problem of the adequate scale of measurement for geographical entities.