KNOWLEDGE-BASED SYSTEMS FOR COPING WITH CLOUDS 
David G. Goodenough, Dena Schanzer, and Michael Robson 
ABSTRACT 
The Canada Centre for Remote Sensing (CCRS) is developing a system of hierarchical experts for resource 
inventories, known as SHERI. The focus of the SHERI Project is to create knowledge-based systems for 
updating forest geographic information systems with remotely sensed imagery. This development is being done 
in cooperation with the Inventory Branch of the British Columbia Ministry of Forests (BCMOF). The 
production goal of BCMOF is to be able to update 1500,1:20,000 scale, forest maps per year. To achieve this 
goal, BCMOF must make use of remote sensing imagery containing some clouds (up to 15% of the scene). 
This paper deals with our knowledge-based systems for detecting clouds, smoke, and haze in Thematic Mapper 
imagery. The systems must also identify snow and water in the shadows of the clouds. It is not possible to 
interpret the forests directly cloud-covered. Therefore, these regions must be used to mark forest polygons 
which are not yet updated in the scene. The result is that the majority of a forest map may be updated, but 
there can be some forest polygons which were not updated because of clouds. These particular forest polygons 
are noted in the overall system in the event that cloud-free imagery covering these polygons becomes available. 
The paper describes the features used for cloud detection, the accuracies obtained, and the methods used to 
manage the agenda for future updates of the forest inventory. 
160