ESTIMATION OF LEAF AREA INDEX USING GROUND SPECTRAL MEASUREMENTS
OVER AGRICULTURE CROPS: PREDICTION CAPABILITY ASSESSMENT OF
OPTICAL INDICES
D. Haboudane * *, J. R. Miller^, N. Tremblay *, E. Pattey “, P. Vigneault®
* Laboratoire d'expertise et de recherche en télédétection et géomatique, UQAC, 555, Boul. Université, Chicoutimi,
Quebec, G7H 2B1, Canada - Driss_Haboudane(@ugac.ca
Department of Physics and Astronomy, York University, Toronto, Ontario, M3J 1P3, Canada - jrmiller@yorku.ca
* Horticultural Research and Development Centre, Agriculture and Agri-Food Canada, 430, Boul. Gouin, St-Jean-sur-
Richelieu, Quebec, J3B 3E6, Canada - tremblayna@agr.gc.ca
- " Agriculture and Agri-Food Canada, K. W. Neatby, 960, Carling Ave., Ottawa, Ontario, K1A 0C6, Canada -
patteye@agr.gc.ca
Commission VII, WG VII/1
KEY WORDS: Hyperspectral, Precision agriculture, Modelling, Algorithm development, Spectral indices, LAI
ABSTRACT:
Leaf area index (LAI) is a key canopy descriptor that is used to determine foliage cover, and predict photosynthesis and
evapotranspiration in order to assess crop yield. Its estimation from remote sensing data has been the focus of many investigations in
recent years. In this context, we have used ground measured reflectances to study the potential of spectral indices for LAI prediction
using remotely sensed data. LAI measurements and corresponding ground spectra were collected over four years (2000, 2001, 2002
and 2003) for three crop types (corn, beans, and peas) in a study area at Saint-Jean-sur-Richelieu, near Montreal (Quebec, Canada).
Hence, a set of vegetation indices were assessed in terms of their linearity with LAI variation, as well as their prediction ability for a
range of crops types. Predictive equations have been developed from ground measured data, and then applied to airborne CASI
hyperspectral images acquired over agricultural fields of corn, wheat, and soybean grown during summer 2001 (former greenbelt
farm of Agriculture and Agri-Food Canada, Ottawa). The results demonstrated that while indices like NDVI suffer from saturation at
medium and high LAI values others like MSAVI2 and MTVI2 result in significantly improved performances. Evaluation of
predictions revealed excellent agreement with field measurements: values of CASI-estimated LAI were very similar to the measured
ones.
1. INTRODUCTION
Remote sensing is seen as an important tool to provide missing
or inappropriate information for the achievement of sustainable
and efficient agricultural practices. Assessment of crop leaf area
index (LAI) and its spatial distribution in agricultural
landscapes are of importance for addressing various agricultural
issues such as: crop growth monitoring, vegetation stress, crop
forecasting, yield predictions, and management practices.
Indeed, LAI is a canopy biophysical variable that plays a major
role in vegetation physiological processes, and ecosystem
functioning. Its retrieval from remotely sensed data has led to
the development of various approaches and methodologies for
LAI determination at different scales and over diverse types of
vegetation canopies (Baret and Guyot, 1991; Daughtry et al,
1992: Chen et al., 2002; Haboudane et al., 2004; etc.). While
some studies have used model inversions (Jacquemoud et al.,
2000), and spectral unmixing (Hu et al., 2002; Peddle and
Johnson, 2000; Pacheco ef al., 2001), others have expended
considerable effort to improve the relationships between LAI
and optical spectral indices (Spanner et al., 1990; Chen and
Cihlar, 1996; Fassnacht ef al., 1997; Haboudane et al., 2004).
Even though some spectral indices have shown satisfactory
correlation with LAI, studies have demonstrated that those
* Corresponding author.
108
indices were as well very sensitive to other vegetation variables
such as canopy cover, chlorophyll content, and absorbed
photosynthetically active radiation (Broge and Leblanc, 2000;
Broge and Mortenson, 2002; Daughtry et al., 2000; Gitelson et
al, 2001; Haboudane et al., 2002). Furthermore, farmers are
concerned with controlling the spatial variability within
agricultural fields, aiming to improve farm productivity and to
reduce input (fertiliser) costs. To this end, various precision
agriculture technologies and tools have been developed during
the recent years (Moran et al., 1997). Their primary goal is to
help scientists and farmers better manage agricultural fields
through the use of spatially-variable application rates that are
based on localised plant growth requirements and deficiencies
(Cassel et al., 2000). Hence, crop LAI status at any particular
stage in the growth cycle can be a consequence of several crop
and soil variables, such as soil condition, nutrient imbalances,
and disease. Its spatial heterogeneity can be used as an indicator
of the crop condition resulting from vegetation response to soil
properties and specifically nutrients . availability for given
weather conditions.
In this context, the objectives of the present study were (1)
to use ground-measured spectra to establish relationships
between ground-measured LAI and selected spectral indices,
(ii) to assess the potential of these indices for LAI predictions,
Internationa
UE EE
and (iii) t
CASI (Cor
images an
from those
2.1 Grou
Predictior
The study
Research
Food Can
known as
where vari
Intensive
growing S
collect grc
crop grow
(corn, bea
commerci:
with diff
monitor te
LAI, grou
emphasis
earliest, M
Spectr
wavelengt
spectrorad
A white S
NH) was
radiance r
reference
ground ta
the non-id
Robinson
were colle
conditions
Crop
determine
using the
an area m
latter was
well as to
during the
2.2 CAS
Validatio
The study
Greenbelt
three suc
were gro
well as o
consisted
various ı
fertilizatic
over a re
recommet
were thus
technique
agricultur
water def
grid of ge
