The development of 1901-2000 historical monthly climate models for Canada and the United States

Here we present the results of an intensive modelling effort that generated 100 years (1901-2000) of monthly precipitation, and minimum and maximum temperature grids for Canada and the United States-a total of 3600 grids. From these primary climate variables, a further 2900 grids were generated for 29 annual bioclimatic indices. Models were generated using the ANUSPLIN software package, which employs thin plate smoothing splines to develop elevation dependent spatially continuous climate surfaces from noisy weather station data. As a rigorous test of the predictive accuracy of the historical surfaces, we carried out validation runs in which climate station data were withheld from the model building process. The magnitude of the withheld errors appear reasonable, ranging from about 1 to 1.5 °C for temperature and 20-40% for precipitation on average. Predictive accuracy of the surfaces varied over time, with a modest improvement as the century progressed and the climate station network expanded. There was also considerable spatial variation in predictive accuracy across the continent. Not surprisingly, mountainous and coastal regions exhibited the highest errors for all climate variables, and error magnitude was closely related to the mean of the underlying climate variable. Though not developed specifically for climate change analysis, continental-scale trends calculated from our grids were comparable to those reported in the literature for several climate variables. The grids have already been used in a variety of applications and are available by contacting the senior author.