An approach using Dempster–Shafer theory to fuse spatial data and satellite image derived crown metrics for estimation of forest stand leading species

Leading species at the forest stand level is a required forest inventory attribute. Information regarding leading species enables the calculation of volume and biomass in support of forest monitoring and reporting activities. In this study, approaches for leading species estimation based upon very high spatial resolution (pixel sided <1 m) have been developed and implemented, with opportunities for improving attribute accuracy using data fusion methods. Over a study region located in the Yukon Territory, Canada, we apply the Dempster–Shafer Theory (DST) to integrate multiple resolutions of satellite imagery (including spatial and spectral), topographic information, and fire disturbance history records for the estimation of leading species.Among the data source combinations tested in the study, the QuickBird panchromatic combined with selected optical channels from Landsat-5 Thematic Mapper (TM) imagery provided the highest overall accuracy (70.4%) for identifying leading species and improved the accuracy by 3.1% over a baseline from a classification-tree based method applied on all data sources. Additional insights to the application of DST to fuse satellite imagery with ancillary data sources to map leading stand species in a boreal environment are also elaborated upon, including the range and distribution of training data and DST mass function establishment.

► Leading species at forest stand level is a required forest inventory attribute.
► Dempster–Shafer Theory (DST) used fuse satellite and spatial data.
► Fusion from DST improved classification accuracy upon classical method.
► Implications of differing data types and scales upon DST results elaborated upon.