Evaluation of alternative spatial models of vapour pressure in Canada

Spatial variations in the measurements of atmospheric humidity provide a basis for creating interpolated estimates of humidity variables such as vapour pressure. Here we compare several different thin plate spline models of 1961-90 monthly vapour pressure normals observed at Canadian climate stations including a model where observed vapour pressures were first adjusted to sea-level using standard barometric correction. Using a randomly-selected set of 30 stations withheld from the fitting procedure, we performed independent validations of each model. Percent errors ranged from 5.5% to 25.1%. All models in fact had good signals and would generally be considered as acceptable according to standard diagnostics. In general, observed data could be predicted with mean bias errors (MBE) of less than 10 Pa. The best model was a square root transformation of the vapour pressure field and a trivariate spline with a spatially varying dependence on elevation as determined by the data network. The maps generated as the final products used all data points except for 3 stations identified as potentially anomalous during the model-building phase.