179 
production on a European scale. The wide swath 
of the AVHRR and the high frequency of potential 
acquisition of data are both well adapted to 
Europe-wide vegetation monitoring. However, the 
pixel size is of the order of 1100x800 metres at 
nadir, which, given the characteristic size 
distribution of European fields, means that each 
pixel typically contains radiometric information 
from several fields, and usually more than one 
crop. The AVHRR cannot therefore be expected to 
produce accurate data on any single targeted 
crop, but will be used to give general 
indications of growing conditions. The methods 
of crop monitoring will need both current and 
historical data, the latter being necessary to 
establish long time-series with which to compare 
current data. 
To establish long time series, data will be drawn 
from several AVHRRs. In order to use the data 
from different AVHRRs in a continuous time series 
it is necessary to establish calibration 
coefficients which can be applied to the data as 
they are being processed. 
The passage of the radiation through the 
atmosphere alters the signal in characteristic, 
and partly predictable, ways. If no action is 
taken to account for them, changes in the 
concentration of atmospheric constituents over 
time and space, or changes in the atmospheric 
path length brought about by changes in scan 
angle, or changes in illumination due to changes 
in solar zenith angle, can make it difficult to 
use the data for crop monitoring. Some of the 
noise - in particular that due to illumination 
and to Rayleigh scattering and absorption by 
water molecules - can be corrected for by 
appropriate atmospheric models. 
At present, AVHRR data are not pre-processed to 
any internationally recognised standard at 
receiving stations in Europe (ESRIN 1986), and 
many users either pre-process the data themselves 
or pay for the services of specialised firms. 
The European Space Agency (ESA) is therefore 
setting up a network of receiving stations at 
which standard high-quality pre-processing 
software will be installed. At the same time the 
Joint Research Centre is developing software 
which accepts raw HRPT or LAC data from the AVHRR 
archives or on-line data from the NOAA series of 
satellites and corrects them geometrically, 
radiometrically and atmospherically. This 
software will be installed for operator- 
independent processing at the ESA receiving 
stations. 
The Pilot Project's version of the software will 
produce a corrected mosaic of Europe. Each 
mosaic will consist of a composite of several 
orbits (afternoon passes only are to be used) of 
a single day. This single-day five-channel 
mosaic of Europe will be archived, and will be 
used to construct various indices or temporal 
profiles. The software, developed under contract 
by Tecnodata Italia, is due for delivery and 
testing at the time of this conference. 
I NTERCAL I BRAT I OH OF RADIOMETERS 
The AVHRR on board TIROS-N and NOAA-6 to NOAA-11 
were designed for meteorological applications and 
not with land resource applications in mind. 
Until recently, the use of the entire historical 
archive for land applications had not been 
envisaged. The the need for radiometrically 
comparable data from the series of satellites was 
not anticipated. 
The visible and near infra-red channels are 
calibrated when the instrument is built 
(Lauritson e t a l . 1979), and are not calibrated 
in orbit. The only on-board standard to measure 
changes in these two channels is given by the 
Numeric Count (NC) on each rotation of the prism 
as the instrument measures the darkness of deep 
space at the end of the scan line (NC ). 
o 
The sensitivity of the instrument is known to 
change during operation (Teillet e t at. 1988). 
Furthermore, the responses in channel 1 and 
channel 2 of a single instrument vary 
independently. As a result, data in the visible 
and near infra-red channels (C h 1 and Ch2 
respectively) collected by one AVHRR are not 
directly compatible with those collected by 
another, or even by the same instrument at 
another date, and the sensitivity of vegetation 
indices depending on ratios or differences 
between these channels should not be assumed to 
remain constant over the life of a single 
radiometer or from one AVHRR to another. 
The only means of calibrating the instrument in 
orbit is to use data received from ground or 
atmospheric targets with known spectral 
properties. This exercise has been carried out 
for several dates in the lifetimes of AVHRR-6, -7 
and especially -9 (Kaufman and Holben 1989, 
Staylor 1989, Smith et al. 1988, Teillet et al. 
1988), but the dates used are distributed 
arbitrarily over the lives of the instruments, 
making the results difficult to use 
operationally. The JRC requires a consistent set 
of calibration coefficients covering the whole 
series of data that will be used in constructing 
the historical reference for agricultural 
monitoring. It has therefore established a 
contract with the Laboratoire d 1 Opt i que 
Atmosphérique of the University of Lille in 
France (LOA), to provide calibration coefficients 
for each 6 months of the life of AVHRR-7, -9 and 
-11. With this information it will be possible 
to create a long time series of correctly 
calibrated data. 
Action 2 of the Agriculture Project is required 
to provide yield indices and estimates of the 
quality of the current growing season in 
comparison with that of previous years. The 
estimates are therefore required to maintain a 
relative, and not absolute, accuracy. Ideally, 
in a given region, the yield of year n will be 
judged higher than that of year n-x only if it 
really is higher. However, in any small 
agricultural area, for most important crops, an 
error at the end of the season of 10% seems 
conventionally acceptable; that is, in about half 
of the cases, if the yield of year n-x is 10% (or 
less) higher than the true yield of year n, the 
estimate will place the yields in the wrong 
order. 
Given the many sources of error in the estimation 
of yields by remote sensing, and especially using 
AVHRR, the calibration should be as accurate as 
possible in order to reduce as much as possible 
this source of error. In the real world, 5% 
seems to be the limit of precision of any method 
of in-orbit calibration, and is the aim of this 
study. 
The study is also to provide a comprehensive 
review of single-channel calibration and inter-