Monitoring of Plant Chlorophyll and Nitrogen Status Using the Airborne Imaging Spectrometer AVIS

Airborne hyperspectral remote sensing enables not only spatial
monitoring of vegetation cover, but also the derivation of
individual plant constituents such as chlorophyll and nitrogen
content. These are important parameters for optimised agricultural
management on a field basis through the possibility of spatially
differentiated fertilisation and for hydrological and vegetation
yield modelling. The use of existing airborne imaging spectrometers
is cost-intensive. Moreover, it is difficult to obtain these
sensors for multitemporal applications. The imaging spectrometer
AVIS (Airborne Visible/Near Infrared Imaging Spectrometer) was
built at the Chair of Geography and Geographical Remote Sensing of
the Ludwig Maximilians University Munich, Germany, to overcome
these difficulties. AVIS is designed as a cost-effective tool for
environmental monitoring using commonly available components. AVIS
enables the deployment of a hyperspectral sensor for both
scientific research and educational purposes. It is based on a
direct sight spectrograph coupled to a standard B/W CCD camera. The
signal received by the CCD is read out and sent via a frame grabber
card to a personal computer, where the data is stored on the hard
disc together with additional GPS data. The radiometric, spectral
and geometric properties of AVIS resulting from the calibration
procedure are summarised in Table 7-1. Table 7-1: AVIS
characteristics Parameter Description Spectral range 553-999nm
Spectral resolution 6nm Spectral sampling rate / resampling 2nm /
6nm Number of used bands 74 SNR 45dB (year 1999), 47dB (year 2000)
Spatial resolution 300 pixels per image line Spatial sampling rate
390 pixels per image line FOV 1.19rad IFOV across track 3.1mrad
IFOV along track 2.98mrad One aim of this thesis was to test the
potential of AVIS for the purpose of environmental monitoring,
especially of the chlorophyll and nitrogen status of plants. The
land cover types under investigation were grassland, maize ( Zea
mays L.) and winter wheat ( Triticum aestivum L.). Within this
scope, a total of 21 AVIS flights were carried out during the
vegetation periods of the years 1999 and 2000. The AVIS data were
preprocessed before analysis, including dark current and flat field
correction, resampling as well as atmospheric correction and
reflectance calibration. The test area chosen for the validation of
the AVIS data is located in the northern Bavarian foothills, 25km
southwest of Munich, Germany (48° 6’ N, 11° 17’ E). It is situated
between the Ammersee in the west and the Starnberger See in the
east. The municipalities Gilching and Andechs define the northern
and southern borders respectively. Within this area, three water
protection areas were chosen as test sites. In these test sites,
most of the farmers are under contract to the local agricultural
office “ Amt für Landwirtschaft” resulting in detailed management
data for each field. This data include useful information for the
interpretation of ground and AVIS data. Two weather stations of the
Bavarian network of agro-meteorological stations, namely No.72 (Gut
Hüll) and No.80 (Rothenfeld), are located in the test area and
provide information about precipitation, temperature and radiation.
Ten and thirteen stands were selected as test fields in 1999 and
2000 respectively, including three fields each of maize and wheat
in 1999 as well as three fields of maize and six fields of wheat in
2000. During both years, four meadows were investigated belonging
to the same plant community ( Arrhenatherion elatioris). The
meadows differ in the utilisation intensity (non-fertilised meadow
with one cut, meadow with one cut, meadow with rotational grazing
and meadow with four to five cuts). The ground truth campaigns
included weekly measurements of plant parameters, such as height,
dry and wet biomass, phenological stage, chlorophyll and nitrogen
content, as well as a photographic documentation for each field.
The chlorophyll and nitrogen measurements, which were derived from
the sampling on ground, are available in contents per area [g/m²]
and in contents per mass ([mg/g] for chlorophyll and [%DM] for
nitrogen). The former can be used to evaluate the photosynthetic
capacity or productivity of a canopy, which is an important input
parameter for hydrological or vegetation models; the latter may be
an indicator for plant physiological status or level of stress,
which is a valuable source of information for optimising field
management. The relationship between chlorophyll and nitrogen based
on the ground measurements showed that a differentiation of the
land cover types is necessary for significant correlation. When the
plant species are investigated separately, the chlorophyll and
nitrogen content per area are always highly correlated, especially
for chlorophyll a and total chlorophyll content (r²≥0.8). For all
investigated land cover types, the nitrogen and chlorophyll
contents per mass are uncorrelated. For wheat, the results improve
when the phenological state and the cultivar are considered
(r²>0.67). For maize, distinct variations in the chlorophyll
content per mass during the vegetation period reduced correlation
with these parameters. The use of a fitted chlorophyll trend curve
instead of the original measurements does not lead to a significant
improvement of the results. For grassland, no significant
correlation above r²=0.67 could be observed except for chlorophyll
and nitrogen, both per area, where a decreasing strength of
correlation could be monitored with increasing fertilisation level.
These results lead to the conclusion that the chlorophyll and
nitrogen contents per mass of the investigated land covers are
decoupled when the compensation point for effective photosynthesis
is exceeded. Beyond this limit the nitrogen in the plants is no
longer incorporated into chlorophylls, but mainly into proteins,
alkaloids and nucleic acids, whereas the proteins especially are
used for internal storage of nitrogen. The derivation of the
chlorophyll and nitrogen content of the plant leaves on a mean
field basis was conducted using three hyperspectral spectral
approaches, namely the hyperspectral NDVI (hNDVI), the Optimised
Soil Adjusted Vegetation Index OSAVI as well as the relatively
unknown Chlorophyll Absorption Integral CAI. The multispectral
NDVITM was simulated as established reference. The results of the
derivation of both chlorophyll and nitrogen content of plants with
the investigated approaches depend strongly on a priori knowledge
about the canopies monitored. In general, the use of contents per
area rather than contents per mass has been found more suitable for
the investigated remote sensing applications. A significant
correlation between any index and the chlorophyll or nitrogen
content for the whole sample size could not be derived. The optimal
spectral approach for derivation is species-dependent, but also
dependent on the cultivar. The chlorophyll and nitrogen level of
the plants under observation as well as their temperature
sensitivity mainly caused this dependence. The NDVITM, hNDVI and
OSAVI became insensitive for high chlorophyll content above about
1g/m² (1.5mg/g) chlorophyll a and 0.2g/m² (0.4mg/g) chlorophyll b,
respectively. A saturation of the indices was also found for
nitrogen content above 2.5g/m². The saturation limit of nitrogen in
percentage of dry matter could be rated at about 4%. The positive
correlation between the indices and this parameter for wheat leads
to insensitivity at values above this limit, while the negative
correlation for maize results in saturation for values below 2.5%.
The CAI is not affected by saturation as much as the other spectral
approaches, leading to higher coefficients of determination,
especially for contents per area. The CAI becomes insensitive at
chlorophyll contents per area above 2g/m². The results lead to the
assumption, that the flattening and narrowing of the chlorophyll
absorption feature at 680nm most probably causes the saturation of
the NDVITM, hNDVI and OSAVI. The ratios are directly affected by an
increase in reflectance in the red wavelength region. The high
correlations between the CAI and contents per area can be ascribed
to the fact that the CAI is based on an integrated measurement over
an area and therefore is less affected by an increase of
reflectance in the red wavelengths. The CAI probably becomes
insensitive at the point where the narrowing of the absorption
feature leads to a shift of the red edge position towards the blue
wavelength region. This saturation limit lies at approximately 2g
chlorophyll per m². In contrast, the chlorophyll content per mass,
which indicates the plant’s physiological status or level of
stress, could be estimated more accurately using spectral indices
such as hNDVI and OSAVI, especially for wheat. The low correlations
derived for maize are caused by its higher temperature dependence,
leading to daily variations in the chlorophyll content per mass.
The chlorophyll and nitrogen contents of the grassland canopies
could not be derived with the spectral approaches investigated.
When the meadows were investigated separately, correlations could
only be found between the CAI and the chlorophyll content per area
for the most intensely utilised meadow (four to five cuts), which
on the one side is characterised by the highest level of
fertilisation, but on the other side is affected by the highest
nutrient offtake. The low potential of the investigated indices can
be mainly assigned to the fact that the chlorophyll and nitrogen
values of the meadows mostly exceeded the saturation limits of the
applied indices. The possibility of deriving chlorophyll and
nitrogen accurately enough to map within field heterogeneities was
discussed on the basis of a wheat field, which was analysed
separately at three sampling points for chlorophyll and nitrogen
content. The approaches found to be most suitable for the parameter
estimation of wheat were applied. The CAI was used for the
estimation of the chlorophyll content per area and mass as well as
for the nitrogen content per area. The hNDVI was applied to
estimate the canopy’s nitrogen content per mass. Both approaches
were able to reproduce the chlorophyll contents of the different
sampling points accurately enough to derive the differences between
the measurement points when the saturation limits were not
exceeded. Beyond these limits the index values decreased with
increasing measurement values. The spatial pattern of the nutrient
supply was discussed by comparing nitrogen pattern images, which
were derived from CAI measurements acquired in 2000 with the yield
measurement map of the same field. The phenological stage of stem
elongation (EC 30) turned out to be most suitable for the
derivation of the nitrogen pattern. On the one hand, the crop
condition at these stages determine yield and on the other hand the
nitrogen pattern images were able to map the inner field patterns
of nitrogen supply. After anthesis the nitrogen images can map
areas with different degrees of maturity. Therefore they can be
used for the monitoring of maturity stages for the determination of
the most favourable harvest date. As described here, AVIS is still
in its early stages. It has the potential to become a
costeffectiveAVIS2, which covers the spectral range of 400-900nm,
has been in commercial use since 2001. tool for the monitoring of
the environment. A modification of AVIS, namely