A procedure for model-assisted climate impact assessment is developed. The approach combines data from observations and atmospheric general circulation models (GCNs), and provides the basis for a potentially valuable means of using information derived from GCMs for climate impact assessments on local scales. The first component of this procedure is an extension of the 'climate inverse' method of Kim al. (1984). Daily mesoscale temperature and precipitation values are stochastically specifed on the basis of observational data representing the average over an area corresponding to a GCN grid element. Synthetic local data sets generated in this manner resemble the corresponding observations with respect to various spatial and temporal statistical measures. A method for extrapolation to grid-scale 'scenarios' of a changed climate on the basis of control and experimental integrations of a GCM, in conjunction with observational data, is also presented. The statistical characteristics of daily time series from each of these data sources are portrayed in terms of the parameters of a multivariate time-domain stochastic model. Significant differences between the model data sets are applied to the corresponding parameters derived from the observations, and synthetic data Bets representing the inferred changed climate are generated using Monte-Carlo simulations. The use of the procedure is illustrated in a case study. The potential climatic impacts of a doubling of atmospheric carbon dioxide concentrations on three important North American grain cropping regions is investigated using two 'physiological' crop models. Although the specific results must be interpreted with caution, they are moderately optimistic and demonstrate possible means by which agricultural production may adapt to climatic changes. / Graduation date: 1987