By Erika Romijn
Abstract
To safeguard rivers from flooding, river floodplains in the Netherlands have been enlarged to accommodate the water discharge. Many floodplains however are also developed for nature rehabilitation, resulting in natural processes of vegetation succession. Remote sensing is seen as an important tool to map and monitor the vegetation structure in order to provide river managers with up-to-date information on hydraulic roughness of the vegetation.
The objective of this study was to develop a methodology for monitoring the location and structure properties of vegetation structure types in a river floodplain ecosystem using multi-directional hyperspectral data. In this study data were used from the CHRIS sensor onboard the PROBA satellite acquired in 2005 over the test site Millingerwaard, a river floodplain ecosystem along the river Waal in the Netherlands. CHRIS data are particularly suitable for mapping vegetation structure because of its high spatial resolution (~17m*17m), spectral coverage (18 bands from 400 nm to 1050 nm) and angular sampling (5 viewing angles).
First, the CHRIS images were classified into eight major land use classes, using different classification techniques. Best results were obtained from the CHRIS nadir classification, using the maximum likelihood classifier and multiple angular images for determining the regions of interest.
After classification, relevant vegetation structure properties such as leaf area index (LAI) and fractional cover (fCover) were quantified on a pixel-by-pixel basis by using the canopy reflectance model FLIGHT. FLIGHT is a physical based radiative transfer model that simulates canopy bidirectional reflectance by using Monte Carlo ray tracing. LAI and fCover maps were computed through model inversion of the CHRIS data for the three main classified vegetation structure types “herbaceous”, “shrubs” and “forest”. All three vegetation classes were modelled as a 1D turbid medium, the forest class was additionally modelled with explicit 3D canopy geometry. The outcomes were validated with in situ LAI and fCover measurements that were collected using hemispherical photography and TRAC measurements.
Although physically simplified, the 1D modelling approach provided superior results compared to the 3D approach, probably due to the less extensive parameterization process. LAI and fCover maps were computed for the CHRIS viewing angles in nadir direction, backscattering direction (view zenith angle of -36°) and forward scattering direction (view zenith angle of +36°). The backscattering direction gave the best results, and showed most variation in LAI and fCover values. Further research is needed in order to find out if the inferred vegetation structure maps can be related to hydraulic roughness values and thereby provide relevant information to river managers.
Keywords: CHRIS/PROBA, multi-angular, vegetation structure types, maximum likelihood (ML) classification, Leaf area index (LAI), fractional cover (fCover), radiative transfer (RT) model, FLIGHT