Albedo and LAI estimates from FORMOSAT-2 data for crop monitoring

REMOTE SENSING OF ENVIRONMENT

卷:113

Article

Bsaibes, Aline^1,2  Courault, Dorninique^1,2  Baret, Frederic^1,2  Weiss, Marie^1,2  Olioso, Albert^1,2  Jacob, Frederic³  Hagolle, Olivier⁴  Marloie, Olivier^1,2  Bertrand, Nadine^1,2  Desfond, Veronique^1,2  Kzemipour, Farzaneh^1,2 

[1] INRA, EMMAH, UMR1114, F-84914 Avignon, France

[2] Univ Avignon, EMMAH, UMR 1114, F-84914 Avignon, France

[3] IRD, UMR LISAH, F-34060 Montpellier, France

[4] CNES, F-31401 Toulouse, France

关键词: Albedo;    Leaf Area Index;    FORMOSAT-2 data;    Off-nadir single viewing;    Stepwise multiple regression;    Neural networks;    Wheat;    Meadow;    Maize;    Rice;   

DOI  :  10.1016/j.rse.2008.11.014

来源: Elsevier

【 摘 要 】

This paper aimed at estimating albedo and Leaf Area Index (LAI) from FORMOSAT-2 satellite that offers a unique source of high spatial resolution (eight meters) images with a high revisit frequency (one to three days). It mainly consisted of assessing the FORMOSAT-2 spectral and directional configurations that are unusual, with a single off nadir viewing angle over four visible-near infra red wavebands. Images were collected over an agricultural region located in South Eastern France, with a three day frequency from the growing season to post-harvest. Simultaneously, numerous ground based measurements were performed over various crops such as wheat, meadow, rice and maize. Albedo and LAI were estimated using empirical approaches that have been widely used for usual directional and spectral configurations (i.e. multidirectional or single nadir viewing angle over visible-near infrared wavebands). Two methods devoted to albedo estimation were assessed. based on stepwise multiple regression and neural network (NNT). Although both methods gave satisfactory results, the NNT performed better (relative RMSE=3.5% versus 7.3%), especially for low vegetation covers over dark or wet soils that corresponded to albedo values lower than 0.20. Four approaches for LAI estimation were assessed. The first approach based on a stepwise multiple regression over reflectances had the worst performance (relative RMSE=65%), when compared to the equally performing NDVI based heuristic relationship and reflectance based NNT approach (relative RMSE=34%).The NDVI based neural network approach had the best performance (relative RMSE=27.5%), due to the combination of NDVI efficient normalization properties and NNT flexibility. The high FORMOSAT-2 revisit frequency allowed next replicating the dynamics of albedo and LAI, and detecting to some extents cultural practices like vegetation cuts. It also allowed investigating possible relationships between albedo and LAI. The latter depicted specific trends according to vegetation types, and were very similar when derived from ground based data, remotely sensed observations or radiative transfer simulations. These relationships also depicted large albedo variabilities for low LAI values, which confirmed that estimating one variable from the other would yield poor performances for low vegetation cover with varying soil backgrounds. Finally, this empirical study demonstrated, in the context of exhaustively describing the spatiotemporal variability of surface properties, the potential synergy between 1) ground based web-sensors that continuously monitor specific biophysical variables over few locations, and 2) high spatial resolution satellite with high revisit frequencies. (C) 2008 Elsevier Inc. All rights reserved.

【 授权许可】

Free