Detecting Structural and Growth Changes in Woodlands and Forests: The Challenge for Remote Sensing and the Role of Geometric-Optical Modelling

Vegetation structure can be defined as the vertical and horizontal distribution of plant material and individual plants within a plant community. The structure of forests and woodlands is sometimes regarded merely as a factor for determining the function of total biomass, leaf area or leaf area index (LAI). However, in vegetation dynamics as well as in resource and habitat assessments, structure is a key factor. In the same way, biomass, vegetation floristic association and LAI are routinely derived from remotely sensed data, while structural effects are largely ignored or regarded as a nuisance. In semiarid woodlands in particular, high-image variance due to crown spacing and shadowing is a mirror of the high spatial variance and gappy structure that characterise this type of land surface. Since the amount of shadowing depends on sun and view angles as well as on the proportion and size of the plants, a range of reflectance values can be obtained for exactly the same land surface viewed at different times and using different view perspectives. This effect is generally called the Bidirectional Reflectance Distribution Function (BRDF). Interpreting significant temporal trends in the structure and composition of vegetation requires the application of physical models — both to interpret remotely sensed data and to understand the elements creating the vegetation dynamics. Detecting these trends assumes particular importance in assessing the impacts of global climate change. The remote sensing measurement models discussed here belong to the class of geometric-optical, or GO, models for vegetation types in which the crowns of the upper stratum (trees or shrubs) are not touching, as occurs in open forests, woodlands and shrublands. These vegetation types occupy a large proportion of the earth’s surface. For example, in the Murray-Darling Basin in Australia, these systems make up nearly 75% of an area comprising 1 million sq. km. To illustrate structural change, the dynamics that occurred in an area of woodland vegetation over a 10-yr period were simulated and expressed by changes in the height and cover of trees, shrubs and grasses and the cover of bare soil. A vegetation dynamics model (RESCOMP) based on availability of water, light and nutrients was used to simulate the structural changes. Over this period, the shrubs increased dramatically (from 2% to 20%), tree cover fluctuated slightly and grass or bare soil fluctuated dramatically depending on seasonal rainfall conditions. These data form the vegetation input into the GO model. We focused on shrub invasion, and asked whether reflectance data (even assuming effective atmospheric correction) could detect the increase in shrub cover in the woodland. While total cover changes were clearly detectable, trends in shrub cover as opposed to tree cover could not be identified using only spectral data. We propose that separation could be achieved by quantifying and monitoring BRDF effects for pure woodlands, pure shrublands and mixtures of trees and shrubs; modelling changes in image texture for a range of pixel sizes; or both. Ultimately, the underlying relationships among spatial variance, ecological condition and future dynamics in visible/near-infrared, thermal and radar images seem to indicate that image texture and variance are the tools that can best monitor vegetation dynamics and structure for spatially variable vegetation types. At this time, however, BRDF seems to providethemostaccessible operational tool for monitoring structural changes at multiple scales.