摘要：Data mining—the discovery of previously unknown information from a large collection of individual data sources—is becoming increasingly popular for scientific data archives. We describe an approach to data mining that uses spatial, temporal, and type constraints to obtain a broad list of data that are potentially related to a data set of interest. Tree- and spline-based multivariate regression and classification techniques are then used to identify functional relationships between the data. Expert knowledge is used to constrain and guide the model building and evaluation process. We demonstrate the approach by identifying relationships between indicators in a state of the Antarctic environment reporting database. Analyses of the fuel usage of electrical generators and boilers at Australia’s Davis station yielded fuel usage dependencies on air temperature and wind speed that were in good accordance with known physical processes. The phenomenon of periodic haul-outs of large numbers of leopard seals on Macquarie Island was related to anomalies in regional sea ice cover and sea surface temperature.
