| Forest satellite scenes are difficult to interpret and efforts to automate their analyse have not, until now, provided tools for operational level tasks such as forest inventory, that often require implicit instead of explicit information. In this study, we have examined the process of human image interpretation and have come to understand that an interpreter has the capacity for a reasoning that integrates data derivatives from multiple sources with knowledge from one or more domains. Artificial intelligence has widely contributed to this understanding in developing an impressive series of methods able to simulate different human cognitive processes. Among the most recent, Bayesian networks, a type of mix of artificial neural networks and expert systems, have piqued our interest. We have explored their utility on a study area of approximately 8400 km2 to the north of Quebec City, to extract arborescent composition of forest cover from Landsat TM-5 satellite images, map data and ground survey plots. (Abstract shortened by UMI.)... |
