期刊名称：ISPRS Annals of the Photogrammetry, Remote Sensing and Spatial Information Sciences
印刷版ISSN：2194-9042
电子版ISSN：2194-9050
出版年度：2006
卷号：XXXVI Part 7
出版社：Copernicus Publications
摘要：For optimal management of river floodplains in the Netherlands monitoring of natural vegetation succession and hydrodynamic processes is essential. A key biophysical parameter to monitor floodplains is vegetation biomass. Not only because it influences the hydraulic resistance determining the discharge capacity of the floodplains, but also indicating species diversity and habitat heterogeneity in the floodplains. The objective of this study is to investigate the feasibility of mapping above-ground biomass and plant functional type distribution of heterogeneous canopies in river floodplains using imaging spectroscopy. We establish linear predictive models between vegetation indices derived from airborne imaging spectrometer (HyMap) data and field measurements of biomass (n = 21). Image and field data were acquired during a field campaign in the summer of 2004 in the Millingerwaard, a river floodplain situated along the river Waal in the Netherlands. Results for broad-band and narrow-band derived VIs (e.g., NDVI, SAVI, WDVI and RSR) and a multivariate approach using PLS were compared using a cross-validation procedure to assess the prediction power of the regression models. Results showed that regression models could be improved when differences in vegetation structure were taken into account. Therefore, regression models were developed for individual plant functional types (grassland, mixed herbaceous, shrub and softwood forest). Vegetation biomass maps for the Millingerwaard were prepared in two steps. First a classification of plant functional types was made using mixture tuned match filtering (MTMF). In a second step, the best regression models were inverted and used to map the spatial distribution of the vegetation biomass. The results demonstrate the necessity to use a PFT based approach for biomass assessment, improving the quality of the prediction significantly over conventional approaches
