This project builds upon the TOPS architecture to provide an agent-based interface for ecological forecasting.

A software agent is sophisticated computer program to which a person can delegate tasks. We often delegate tasks to humans, such as insurance agents, travel agents, or personal assistants. In limited domains, we have developed the technology that enables people to delegate certain tasks to computers. In the case of data management, access, and display, IMAGEbot coupled with a natural language interface (NLI), constitutes such an agent, where a person can issue a high-level, potentially ambiguous request in English and expect the program to figure out how to satisfy the request using the available resources, such as SQL [32] queries, ecological models and image-processing operations.

For example, a user asks for a display showing areas of high fire risk in Montana (figure). The query is ambiguous, since it does not specify for when the fire risk is needed. The agent asks for clarification, and the user specifies “tomorrow.” Based on this query, the agent builds a map of fire potential. This requires many operations, which the agent must plan and execute. To determine fire potential, the agent needs, for example, data on how much dry vegetation will be present and how windy it will be. There is no direct data on dry vegetation, but this can be estimated based on land cover and soil moisture. After feeding the available data into the appropriate models, the agent constructs an image using a default color map, in which low values are shown in red. The user finds this display counter-intuitive, and tells the agent that high fire areas should be red. The agent complies, operating on the data product already produced rather than repeating the same time-consuming sequence of data-processing steps, and quickly presents an updated image. The user then asks to see areas of high risk in which homes are present, and where firefighting equipment is located relative to the at-risk homes.