temporal resolution (1/2 day), which makes it easier to acquire a
cloud free image in a certain time period.
A pilot study has been carried out to investigate the potential of
NOAA/AVHRR imagery to produce a geo-referenced land cover
data base. In particular improvements of the spatial accuracy of
the 10-minutes PELU were investigated. Test-sites were selected
in various agro-climatic regions in Europe: the Netherlands,
Eastern Spain, Austria and Western Poland (Mücher et al., 1996).
The responsibilities for Austria and Western Poland were taken
by the Austrian Research Centre in Seibersdorf, Austria and the
Institute of Geodesy and Cartography in Warsaw, Poland
respectively. Emphasis was laid on the classification experiments
at the test sites in the Netherlands and Eastern Spain.
Both original multi-spectral 10 bit NOAA/AVHRR data (5
channels) and derived Normalised Differential Vegetation Index
(NDVI) composites were used.
Land cover information from classified Landsat TM imagery was
used as training samples and as reference for checking the results
of the classification. Supervised classifications (maximum
likelihood) were performed multi-temporally on the multi-spectral
original NOAA/AVHRR data.
Thanks to the availability of high quality reference data, a solid
multi-temporal data set and the fact that the Netherlands are
characterised by intensive land use conditions and relatively large
homogeneous regions, the supervised classification results were
very satisfactory.
However, the individual classes can often not be distinguished in
regions where the land cover is heterogeneous and the transition
between the various land cover classes is vague. A good example
is the coastal region of Spain.
The results for the test site in Poland show that the accuracy of the
classification increases with the homogeneity of the
NOAA/AVHRR pixels. For Austria the classes grassland and
mixed forest were difficult to distinguish due to similar spectral
reflectance on the selected images. Here, the use of a DTM
improved the accuracy of the classification.
3. AGRICULTURAL APPLICATION
3.1 Introduction
A l-km land cover database can play a role in improving
agricultural statistics and in monitoring alarm situations like
drought. Although the pixel size of l-km does not allow
identification of individual crops, regions mainly in use for
agriculture can be discriminated. Crop specific information can be
obtained afterwards by applying high resolution satellite data. The
meaning of the l-km land cover database is to stratify
homogeneous regions and to extrapolate crop specific
information as derived from high resolution satellite data to obtain
information at national and continental level.
3.2 Problem description
A so-called Vegetation Index (VI) can be calculated. Comparing
VI time series with those of previous years or with other areas it is
possible to assess comparative yields and to search for drought
indicators at various levels: local, national and European. Various
methods have been developed to map regional transpiration from
scanned reflectance and temperature maps. In combination with
crop growth modelling, effects of drought periods on final crop
yield can be accounted for using transpiration maps as derived
from remote sensing images.
This study attempts to integrate crop growth modelling and
satellite remote sensing. Two tools which are different in many
aspects. Crop growth modelling is essentially very detailed, but for
regional applications data sets are difficult to collect. Assumptions
and generalisations are unavoidable. In the case of CGMS the
spatial resolution of weather variability is determined by the grid
size (50x50km) and soil variability by the EC soil map
(1:1000000). Simulations take place per combination of soil type,
weather grid cell and crop type. For each combination time series
of crop growth are produced with timesteps of 1 day. Because this
is an accumulating state variable, output at 10 day intervals is
sufficient to describe the whole agricultural season.
In the case of remote sensing the spatial and temporal resolutions
are contradicting. Platforms with a high overpass frequency (e.g.
NOAA-AVHRR, 2 images per day) have a course spatial
resolution (1x1km?). Sensors with high spatial resolution (e.g.
Landsat-TM 30x30m?) have a low overpass frequency (every 16
days). Also the type of data is different. Environmental and
management effects on the crop are lumped together in the
measurement. For a unique interpretation of remote sensing
derived parameters additional information about the underlying
processes is required.
This paper addresses two remote sensing techniques, moreover the
link with crop growth modelling is discussed. The first technique
relates Vegetation Indices (VI's) derived from satellite remote
sensing to Leaf Area Index (LAI) and ground cover. VI's are a.o. a
measure for green biomass, which is accumulated during the
growing season. Bouman (1995) discussed the use of VI's as
forcing function in crop growth modelling and as calibration tool.
The second technique is the evapotranspiration estimation based
on remote sensing derived surface reflectance, surface temperature
and VI's. In contrary with VI's, evapotranspiration is a state
variable, which does not allow interpolation in time between
measurements. Therefore as forcing only few estimated rates can
replace simulated values. In the case of evapotranspiration the link
of remote sensing with crop models is one of validation (absolute
values) and comparison of regional patterns.
3.3 Description of the test site.
This paper focuses on the MARS action 4 sample site of Seville,
Andalusia, Spain. For this site 4 SPOT images, 3 Landsat-TM
images and daily NOAA-AVHRR images are available for 1992.
All three sensor types are used to make vegetation index profiles.
Evapotranspiration maps of the area are only based on the
available Landsat-TM images, because of the necessary thermal
infra-red band. The resolution of the AVHRR sensor (1.1 km at
nadir) dictates that the NOAA vegetation profiles can not be
considered crop specific because of the small parcel size in the
Seville region. From the daily images 10-day maximum value
vegetation index composites are derived.
4. TIME SERIES OF VEGETATION INDICES
4.1 Applied Vegetation Indices
Optical reflectance of crops is determined by the interaction of
solar radiation with the crop canopy. From the late sixties, this
interaction process has been extensively studied and canopy
reflectance has been shown to be related to interesting crop
growth variables such as Leaf Area Index (LAI) and fraction
ground cover (e.g. Suits, 1972; Bunnik, 1978). For accurate
estimation of these variables, so-called Vegetation Indices have
been developed that are a combination of reflectance values in
visible and near-infrared wavelength bands, like the Normalised
Difference Vegetation Index (NDVI):
NDVI=(IR-R)/(IR+R) ) (D
144 International Archives of Photogrammetry and Remote Sensing. Vol. XXXII, Part 7, Budapest, 1998
whei
the e
Toc
the s
WD'
whei
soil 1
This
acco
LAI:
with
(199
WD
year:
tural
The
bion
sens
is le
cond
com
4.2
simt
4.2.1
data
Base
crop
conc
vege
imag
and
valu
two
betw
The
IT
are ¢
For
the €
Las
Figu
plot
