Riparian marshland composition and productivity mapping using IKONOS imagery

The Ontario Wetland Evaluation System (OWES) employs a visual assessment of wetland extent, composition and biomass as primary indicators in determining which wetlands should be considered provincially significant and subsequently protected. High resolution satellite remote sensing offers the potential to provide more quantitative analysis at greater spatial detail within a given wetland using spectral and spatial image information. In this study, Ikonos imagery was used to map vegetation composition and productivity in three riparian marshes located along the Rideau River near Ottawa, Ontario. Separability and correlation analyses aided in the selection of an optimum set of spectral and spatial data, which were used in classification tests, with training and validation data being randomly selected from a set of 107 field sites. Terrestrial and aquatic vegetation was classified comparing maximum likelihood (ML) and neural network classifications, with a ML classification using a Transformed Vegetation Index (TVI), visible bands, A2M5grn and CON5nir textures resulting in the highest accuracy of 88% and a Khat statistic of 0.72. For biomass mapping, a 1 m2 sample frame was used to collect green and senescent vegetation at 75 locations within the various vegetation classes. Dry biomass was modelled against the spectral and textural image measures using forward stepwise regression. Log green biomass modelled by a combination of texture and spectral variables provided the best result. The absolute error of this model in predictive biomass mapping was 213 g/m2 or approximately 40% of the mean field measured biomass.