Hydrological data that are input into catchment/regional scale water balance models will depend, not only on limitations in component estimators (i.e., evapotranspiration (ET) and discharge (D)), but on how accurately precipitation (P) and estimator input data are predicted over the landscape. Spatially predicting total precipitation (P) and the temperature and radiation data required for many ET estimators is a non-trivial task, especially for regions where climate stations are sparsely spaced and the terrain and surface attributes highly heterogeneous. Moreover, the lack of site ET data makes verifying water balance models extremely difficult. This study estimates yearly water balances (based on monthly 30-yr normals) for several catchments in the southern Rocky Mountains/foothills of Canada. The water balance model has the form: Change in water storage = P-ET-D. Actual stream flow data provided estimates of D. Changes in water storage were considered negligible. Monthly ET was estimated from the Thornthwaite method and Priestley-Taylor model using alpha values representing monthly surface covers (i.e., rangeland and forests). Minimum and maximum temperature data were used in the estimation of net radiation and the slope of the saturation vapor pressure curve. The temperature and P data were spatially modeled using "hard" station data and "soft" secondary data (estimated from climate vs topographic variable regressions) in several geostatistical interpolation/extrapolation techniques; these techniques were, i) simple and ordinary kriging, ii) kriging regression residuals, iii) simple kriging with exhaustive mean estimates, iv) cokriging with one nonbias constraint, v) collocated cokriging, and vi) kriging with external drift. The independent variables used to estimate the "soft" primary climate data were, elevation, northing, windward-leeward locale relative to continental divide, slope aspect, slope gradient, and visibility measures. The accuracy of the spatially distributed temperature and precipitation data were assessed using cross-validation procedures and mapped representations of results. Catchment estimates of P and ET were also evaluated by comparing P-ET with observed D.