Forest structural complexity in a temperate hardwood forest: A geomatics approach to modelling and mapping indicators of habitat and biodiversity .

摘要

Remote sensing has been widely used for modelling and mapping individual structural attributes within forests, however, knowledge of the multivariate nature of structural complexity, which is of specific interest as an indicator of forest habitat and biodiversity, is lacking. This research presents methods and results describing the development of geomatics-based indicators of forest structure, which are spatially continuous, extensive and repeatable. Two distinct, but related, structurally-based indicators derived from high resolution airborne imagery and topographic information were developed: (1) modelling and mapping forest structural complexity, and (2) the detection and mapping of the spatial distribution of dead wood. A Redundancy Analysis (RDA) was used to develop an image-based Structural Complexity Index (SCI) representing structural complexity as measured on the ground. An extensive set of image spectral, spatial, and object-based variables, along with topographic variables, were tested as predictors of structural complexity. The SCI, as a general gradient of structural complexity, accounted for 35% of the original variance in the field data. The model was applied spatially to map the SCI across the entire study area within Gatineau Park, Quebec. Field validation of the extreme conditions (high and low complexity areas) showed the map to be ∼ 80% accurate. Tests using simulated 60 cm and 1 m imagery showed potential for scaling up the RDA modelling procedure to be used with coarser resolution imagery, however map validation at these resolutions was somewhat inconclusive and further investigation using lower resolution airborne imagery and possibly high resolution satellite imagery is required. Additionally, a semi-automated method for detecting and mapping dead wood was investigated, with field validation of detected objects having an accuracy of 94%, and control sites, or areas with no detectable dead wood, showing an accuracy of 90%. The methods presented in this research can help to advance remote sensing research for forest structure modelling and mapping, and specifically in a temperate hardwood forest. Further, they could potentially be adapted and applied to different forest types and used for enhancing forest inventories by providing methods for reporting on habitat and biodiversity levels.