Training a neural network with a canopy reflectance model to estimate crop leaf area index

Abstract

This paper outlines the strategies available for estimating the biophysical properties of crop canopies from remotely sensed data. Spectral reflectance and biophysical data were obtained over 132 plots of sugar beet (Beta vulgaris L.) and in the first part of the paper the strength of the relationships between vegetation indices (VI) and leaf area index (LAI) are examined. In the second part, an approach is tested in which a canopy reflectance model is used to generate simulated spectra for a wide range of biophysical conditions and these data are used to train an artificial neural network (ANN). The advantage of the second approach is that a priori knowledge of the measurement conditions including soil reflectance, canopy architecture and solar position can be included explicitly in the modelling. The results show that the estimation of sugar beet LAI using a trained neural network is more reliable than the use of VI and has the potential to replace the use of VI for operational applications. The use of a priori data on the variation in soil spectral reflectance gave rise to a small increase in LAI estimation accuracy.

Acknowledgments

The data used in this paper were collected as part of the Biophysical Analysis Research Group Experiments (BARGE) and the authors acknowledge all the participants in the experiments. The work was supported by the Natural Environment Research Council (NERC), NERC Equipment Pool for Field Spectroscopy, Agriculture and Food Research Council (AFRC), British Council, Sugar Beet Research and Education committee (SBREC), Sheffield University, Programme National de Télédétection Spatiale (PNTS), Institut National de la Recherche Agronomique (INRA) and Service des relations étrangères.