Comparison of physically based and empirical models to estimate corn (Zea mays L) LAI from multispectral data in eastern Canada

Abstract

Leaf area index (LAI) is a key input for many ecological models such as crop and carbon and nitrogen models. The LAI patterns measured in situ are time consuming and expensive and could be substituted by remote sensing technology. The objective of this study was to evaluate the possibility to map LAI of corn fields in eastern Canada using airborne multispectral imagery. Four models were locally optimized, tested, and compared. The first two methodologies relied on simple empirical relationships between LAI and the normalized difference vegetation index. The third methodology was a physically based model assuming that spectral vegetation indices are proportional to the fraction of photosynthetically active radiation absorbed by the green vegetation canopy. The last methodology relied on neural network (NN) models using different channels of the multispectral images as input. No studies combining airborne remote sensed data and NN for corn LAI simulation at regional or field scales are known by the authors. Ground measurements of LAI in five experimental sites at two dates in 2011 and 2012 were used to optimize and evaluate the models. Even though performances of the standard methods were improved by local optimisation, an NN model built with the near-infrared and red channels provided more accurate LAI simulations. All models performed better at LAI values below 3, but scattering at high LAI values was less pronounced with the NN model.