Model development and comparison to predict softwood and hardwood per cent cover using high and medium spatial resolution imagery
Authors: Metzler, J. W.; Sader, S. A.
Source: International Journal of Remote Sensing, Volume 26, Number 17, 2005 , pp. 3749-3761(13)
Publisher: Taylor and Francis Ltd
Abstract:The recent availability of high spatial resolution multispectral scanners provides an opportunity to adapt existing methods and test models to derive spatially explicit forest type and per cent cover information at the Landsat pixel level. A regression modelling methodology was applied for scaling‐up high resolution (IKONOS) to medium spatial resolution satellite imagery (Landsat) to predict softwood and hardwood forest type and density (per cent cover) in a northern Maine study area. Regression relationships (63 different models) were developed and compared. The model variables included vegetation indices and several date (season) combinations of Landsat Enhanced Thematic Mapper Plus (ETM+) imagery (August, September, October and May). A model incorporating all variables from four dates of Landsat ETM+ imagery produced the highest coefficient of variation in predicting both softwood (0.655) and hardwood cover (0.66). The addition of vegetation indices with the six ETM+ reflected bands did not significantly improve or detract from the regression relationships for any of the multi‐date or single date models examined. A two‐date combination of October and May variables provided an acceptable (and arguably more cost‐effective) model as the adjusted R 2 value was 0.645 for softwood and 0.649 for hardwood. A significant result was that all single‐date models produced inferior results with a sharp drop in adjusted R 2 , compared with the multi‐date seasonal models. This research has demonstrated that the regression models including multi‐date variables produce good results and can provide spatially explicit forest type and stand structure data that has been difficult or infeasible to obtain from medium spatial resolution imagery using traditional classification methods.
Document Type: Research Article
Affiliations: Maine Image Analysis Laboratory, Department of Forest Management, University of Maine, Maine 04469‐5755
Publication date: 2005