Underlying the majority of remotely-sensed data analysis is the assumption that geographical phenomena, such as rivers, heather-moors and the dynamics associated with such objects, can be adequately detected and identified through the use of spectral and other visual information alone. There is a common misconception that any major deficiencies of quantitative analyses are "hardware problems": that by increasing the spectral, spatial, radiometric and temporal resolutions of sensors, geographical phenomena will be identified with similarly increasing accuracy and reliability. This, however, is an unrealistic viewpoint. This thesis has developed a prototype of an automated system based on the principle that by considering the "real-world" properties of the land, a more effective and robust analysis of its dynamic nature can ensue. SYMOLAC is an automated SYstem for MOnitoring LAnd Cover based upon theories of artificial intelligence. It has been developed within a specifically designed hybrid software environment called ETORA, an Environment for Task-Orientated Analysis. This prototype environment allows SYMOLAC to utilise disparate sources of spatial data, to reason with both quantitative and qualitative knowledge, to model disparate domain uncertainties, and to exploit the functionality of third-party software components. Unlike standard approaches, it allows an automated analysis to focus on each particular domain task and how it may best be performed with the available data, knowledge and software resources. The detection of forest felling and the subsequent update of the Land Cover of Scotland (1988) dataset forms the initial application of SYMOLAC. It is concluded that the system's approach is flexible, extensible and adaptable, and demonstrates one way in which satellite imagery can offer <I>potential </I>to the future monitoring of complex land cover change without the need for human intervention.