PRE-PROCESSING 
Once the data arrive at S.C.O.T., they 
immediately handed over to IGN Espace, which 
the task of calibrating and correcting 
geometry of the data. 
are 
has 
the 
The geometric correction of the data is 
restricted to the site, which is wholly contained 
within the scene. SPOT data are corrected to a 
common map base by an orbital model followed by 
low-order polynomial. No satisfactory orbital 
model is available for Landsat, and many ground 
control points must be identified by an operator 
and a high-order polynomial must be used to 
correct TM data. Data from both sensors are 
corrected using a Digital Elevation Model (DEM) 
in areas of significant relief. 
Atmospheric correction is based on "5S" (Tanri e t 
al. 1986). Although most clouds can in principle 
be detected and distinguished from snow and ice 
automatically, detection of cloud shadows, fog, 
mist and some cirrus clouds depends on visual 
interpretation. In practice, the interpreters 
have to detect all such contaminants. 
Non-agricultural land (including water and urban 
areas, forest, and permanent pasture) is 
identified at the start of each year's work and 
masked from the images. 
INTERPRETATION 
Personnel 
The image interpretation depends on the expertise 
of three highly trained personnel with 
professional experience in agronomy and image 
interpretation. They were actively involved in 
the development of the software used to support 
their work. Their experience and the rules they 
use to make their judgements will be increasingly 
incorporated in the software. At present there 
are no plans to develop a full-blooded expert 
system, however. The presence of the three 
interpreters on the team can not be absolutely 
assured for the duration of the contract, and 
there is therefore provision for the training of 
supplementary interpreters, an important aspect 
given the dependence of the system on their ever- 
increasing expertise. 
Each interpreter is responsible for a fixed set 
of spatially distributed sites. This strategy 
means that each interpreter will gain experience 
of a group of increasingly familiar sites in a 
wide variety of agricultural regions and will get 
a feel for the quality of the current 
agricultural year over the whole of the 
Community. It also means that at least 
theoretically each interpreter will also be able 
to cover for an absent colleague if necessary. 
Given the present data volume and predictions of 
future needs once the system becomes operational, 
it seems possible that the present team of 
interpreters will be insufficient to maintain the 
necessary throughput. In this case a second team 
may be brought in to work in shifts. 
Activities 
Preparation of a site. Using the same rules 
as those laid down for the location of the ground 
segments in the sites, S.C.O.T. locates an 
additional 14 "image-segments" at each site. 
This increases the degrees of freedom in the 
regression estimator for the determination of the 
precision and bias of the classification from the 
image interpretation. No ground data are 
collected at these additional segments, but 
experience is built up here as it is over the 
ground segments. 
Data on slowly changing or permanent parameters, 
such as soil type, topography or general land 
use, are sometimes available as maps at suitable 
scales. These data, and official agricultural 
statistics (including crop lists, areas under 
crops, data on yield, crop calendars and growth 
cycles and typical rotations) are provided as 
background information for the image 
interpreters. In preparing a new site, the 
interpreter studies images collected the year 
before and becomes thoroughly familiar with the 
characteristics of the site. 
At the start of the first year of a new site the 
interpreters locate each segment and digitize all 
visible field boundaries within it. The set of 
boundary arcs are modified (boundaries added, 
deleted, or reoriented) if any changes become 
evident in subsequent images. The same set is 
used at the start of subsequent years, when the 
interpreter removes any boundaries that are not 
evident in the early images. 
Ancillary data. The contractor also accesses 
meteorological data collected at 160 synoptic 
European stations, and receives the ground data 
collected at the active sites in the previous 
crop year. 
Routine processing. When the new image of an 
existing site arrives, the interpreter first 
locates the segments and the field boundaries as 
described above. He or she then will then 
examine the three-colour composite or NDVI image 
and decide to what extent each field can be 
affected to a given class of land cover. His or 
her decision is based largely on the crop 
calendar and on information available from 
previous images of the same site. In images 
acquired early in the season, the class is 
normally very broad ("early vegetation" or "bare 
soil",, for example), but the process leads to a 
steady refinement of classification over the 
season. Access to a series of images at 
different dates helps to ensure the greatest 
possible number of opportunities for 
discriminating crops. The Action therefore 
depends critically on the repeated acquisition of 
imagery over the growing season. 
: 
»Pi§ 
' ■ 1