This study proposes a new automated and completely image based approach, named CRASh, for the extraction of vegetation variables from hyperspectral remote sensing data. The focus of the approach lies on detecting spatial variations in leaf area index, leaf chlorophyll content, leaf dry matter content, and leaf water content. Detecting anomalies in these important indicators of crop development and stress supports agricultural management decision making at a local to regional scale.
CRASh relies on the inversion of physically based radiative transfer models, which allows it to exploit the entire spectral feature space and to adapt to changing sun and observation geometry and to local background reflectance and atmospheric properties. The approach incorporates an automated land cover classifier, which facilitates optimizing model inversion to specific land cover types and reduces the under-determined and ill-posed nature common to remote sensing observations of vegetated surfaces. To the latter purpose, also a new regularization technique, based on regression analysis between heuristic spectral vegetation indices and radiative transfer model simulations, was introduced.
CRASh was validated at three spatial levels and various sensor configurations, ranging from field spectrometer measurements, over airborne HyMap imaging spectrometer data takes, up to multi-angular observations of CHRIS aboard the PROBA satellite. Biometrical validation measurements took place on intensively used meadows and pastures near Lake Waging-Taching in Upper-Bavaria and on cotton (Gossipium Hirsutum) in The Khorezm region of Uzbekistan. When no a priori knowledge on land cover was assumed, CRASh showed a significant gain in accuracy and stability compared the conventional approaches that were tested. Both the hyperspectral and multi-angular data dimensions provided additional information compared to multi-spectral mono-directional observations, leading increased retrieval accuracy.
«
This study proposes a new automated and completely image based approach, named CRASh, for the extraction of vegetation variables from hyperspectral remote sensing data. The focus of the approach lies on detecting spatial variations in leaf area index, leaf chlorophyll content, leaf dry matter content, and leaf water content. Detecting anomalies in these important indicators of crop development and stress supports agricultural management decision making at a local to regional scale.
CRASh relie...
»