Comparison of approaches for spatial interpolation of daily air temperature in a large region with complex topography and highly variable station density

This study compared 12 variations of regression-based and weighted-average approaches for interpolating daily maximum and minimum temperatures over British Columbia, Canada, a domain with complex topography and highly variable density and elevational distribution of climate stations. The approaches include simple extrapolation with elevation from the nearest climate station; nine variations of weighted-average methods employing three approaches to calculate lapse rates, two methods for station selection and three approaches for weight calculation; multiple linear regression using station coordinates as predictor variables; a method combining multiple regression and weighted averaging. Cross-validation for years with different densities and elevational distribution of climate stations showed varied mean prediction errors, which also depended on elevation and month. Methods that compute local lapse rates from the control points performed better for years for which there were a greater number of higher-elevation stations, which allowed for better estimation of lapse rates. The methods that involved specified lapse rates all performed similarly, indicating that the method for selecting control stations and for calculating weights have less effect on predictive accuracy than the method for accounting for elevation.