159 
CIME2: A TOOLBOX FOR DEVELOPING EXPERT SYSTEMS IN THEMATIC MAPPING USING 
REMOTE SENSING AND GEOCODED DATA 
Catherine Mering 
L.I.A. O.R.S.T.O.M. 
72-74 route d’Aulnay 
93140 Bondy France 
ABSTRACT 
The traditional approach to the analysis of remotely sensed imagery has met with only limited acceptance. One 
possible cause is that the methods have generally be implemented directly from image processing algorithms 
and do not take into account contextual information coming from the thematic expertise: therefore the whole 
process cannot be completely reproduced. The aim of this work is not only to provide maps but to encode the 
whole process of building them up from remote sensing, geocoded data and ground-truth, according to expert 
knowledge and reasoning. 
In order to reproduce a mapping method one has to simulate the knowledge of an expert monitoring 
procedural-based data processing techniques. Therefore we have designed CIME2, which is a toolbox for 
developing expert systems in cartography using satellite images and geocoded data, such as a topographical 
data, DTM or illumination models. CIME2 is based on the simultaneous representation of this different know 
how: data processing techniques, geographical knowledge about physic and human characteristics of the area 
and more specialized knowledge such as altitudinal limits of vegetation and land use units. 
As the main purpose is to monitor and optimize the sequential activation of numerical treatment on remotely 
sensed and geocoded data, CIME2 provides the symbolic language for an expert to describe how to make a 
thematic map. 
The factual database of CIME consists in a hierarchy of objects that are pixels, regions (connected sets of pixels) 
and objects (sets of regions) described by numeric attributes provided by remotely sensed data (radiometry, 
vegetation index, textural index) and geocoded data (altitude, slope, illumination) and symbolic attributes that 
is the various elements of the taxonomy elaborated by the expert to produce the map. 
The procedural database consists in a hierarchy of numerical treatments described by symbolic attributes. 
The expert knowledge is represented as production rules providing contextual information and general control. 
With this toolbox, the expert writes a production system selecting appropriate data and algorithms to build a 
sequence of treatment on the data in order to produce the required map. He can build many sequences and 
select the best one according to a set of criterions such as minimization of ambiguity of classifications and the 
maximization of well classified pixels of training zones. 
CIME2 has been firstly utilized to produce a set of thematic maps of mountainous areas such as Nepal using 
MSS Landsat data, topographic data and an illumination model. But the techniques which has been developed 
could be generalized to other thematic mapping and other areas using database from different sources such 
as a predefined GIS and high resolution sensors such as SPOT and Thematic Mapper.