(IX-B8, 2012
l - a program for
ta, Computers and
flood-watch from
n Space Research,
e E.T., Possingham
uacy in dynamic
ion of vegetation
Pantanal wetland.
rrent and potential
cisions on drought
view. Agricultural
R., 2011. Imagens
) Pantanal em áreas
ogas: resultados
de Sensoriamento
559-3566.
MBRASIL. Folha
plos, vegetaçäo uso
Janeiro. pp. 448.
ybrid approach to
ver transition from
from the Bolivian
ment, 115, pp. 353-
2007. Analysis of
ex data for crop
1s. Remote Sensing
(TS
2010/52614-4) for
o CAPES for the
iro de Almeida to
International Archives of the Photogrammetry, Remote Sensing and Spatial Information Sciences, Volume XXXIX-B8, 2012
XXII ISPRS Congress, 25 August — 01 September 2012, Melbourne, Australia
NDVI FROM ACTIVE OPTICAL SENSORS AS A MEASURE OF CANOPY COVER AND
BIOMASS
E.M. Perry", G.J. Fitzgerald", N. Poole’, S. Craig, A. Whitlock
? Department Primary Industries, Future Farming Systems Research, 110 Natimuk Rd. Horsham VIC 3401 Australia -
(eileen.perry, glenn.fitzgerald) 9 dpi.vic.gov.au
? Foundation for Arable Research, Lincoln 7640 NZ - poolen@far.org.nz
¢ Birchip Cropping Group, Birchip VIC 3483 Australia - simon@bcg.org.au
4 Precision.agriculture.com.au - Andrew @precisionagriculture.com.au
KEY WORDS: Agriculture, Farming, Vegetation, Sensor, Spatial, Real-time
ABSTRACT:
Commercially available proximal sensors are being used in precision agriculture to provide non-destructive, real-time spatial
information on ‘green biomass’ that may be of interest to the remote sensing community. The sensors are described as biomass
sensors, but questions remain on which canopy characteristics can be best estimated by the sensor measurements. In this study
Normalized Difference Vegetation Index (NDVI) measurements from active optical sensors were examined across multiple datasets,
representing different active optical sensors, different years, and different sites for wheat and barley. The NDVI values were
compared with spatially and temporally coincident measurements of fractional green cover and above ground biomass, expressed as
dry matter in kg ha’. Direct comparison of NDVI measurements from the different sensors for the same plots over a range of canopy
cover demonstrates differences for plot means of NDVI. Canopy fractional green cover values were well described by NDVI using a
linear model. Using all of the datasets, the linear regression of fractional green cover on NDVI yielded an r? of 0.71 and a standard
error of 0.12. NDVI measured by the sensors did not explain as much of the variance in dry matter as for fractional cover. Dry matter
was related to NDVI using a non-linear model. For all sensors, sites and dates with green biomass, the model was fitted with r^ value
of 0.27, and standard error of 2133 kg ha‘, The relationship between NDVI and dry matter was nearly linear at levels of biomass less
than 1000 kg ha''. Results for these datasets indicate that the active optical sensors may be a useful surrogate for fractional cover, but
not for above ground biomass.
1. INTRODUCTION
Commercially available proximal sensors are being used in
precision agriculture to provide non-destructive, real-time
spatial information on ‘green biomass’. These active optical
sensors utilise built-in light sources, pulsed to differentiate the
artificial signal from sunlight. This gives these sensors a
tremendous advantage over passive optical sensors (such as
ground-based spectrometers) in that they can be used under any
sky conditions, or even in complete darkness. The sensors are
described as biomass sensors, but questions remain on what
canopy characteristics the sensors best relate to. There is a
relevance to the remote sensing community in that these sensors
could potentially provide ground truth on fractional ground
cover and/or plant biomass. In this study Normalized Difference
Vegetation Index (NDVI; Rouse et al. 1973) measurements
from active optical sensors were examined across multiple
datasets, representing different active optical sensors, different
years, and different sites for wheat and barley grown under
rainfed conditions.
2. METHODS
Several sites and sensor combinations were used for this study:
*Small plot agronomic trials near Horsham Vic (36° 44° S
latitude, 142° 06’ E longitude) on various cultivars of wheat
(Triticum aestivum L) for the years 2007, 2008, and 2010. Two
sensor models were used, the Crop Circle ACS-210 and Crop
Circle ACS-470 (Holland Scientific Inc, Lincoln NE, USA).
: Corresponding author.
Large plot agronomic trials near Lubeck VIC (36° 46’ 42” S
latitude, 142? 30' 33" W longitude) on wheat (Triticum
aestivum L cv. Derrimut) for 2009 and 2010. Two sensor
models were used, the Greenseeker handheld (NTech Industries,
Ukiah CA, USA) and Crop Circle ACS-470 (Holland Scientific
Inc, Lincoln NE, USA).
ePaddock scale agronomy trials on eight paddocks near
Inverleigh VIC (144? 2" 17" W, 38? 9' 0" S) on wheat
(Triticum aestivum L cv. Derrimut) and barley (H. vulgare L.
cv. Gairdner) during 2009 and 2010. A Crop Circle ACS-470
was used to measure NDVI.
Sensor measurements were made throughout the growing
season for the Lubeck and Inverleigh trials, and until anthesis
(flowering) for the Horsham trials. Sensor measurements were
made with the sensor head situated 1 to 1.5 m above the canopy,
with the sensor field of view spanning across rows. The NDVI
values were either computed from the red and near infrared
bands (Holland Scientific sensors) or output directly from the
sensor (Greenseeker). Estimates of fractional green cover were
made using colour photographs following methods described in
Li el al. 2010 for each of the sensor data acquisitions. Above
ground biomass (dry matter in kg ha!) was determined by
biomass cuttings for the plots and paddock scale sample
locations at key crop growth stages that corresponded with
sensor acquisitions. Regression and non linear model fitting
were performed using GenStat (Payne et al. 2009).
