A comparison of two statistical methods for spatial interpolation of Canadian monthly mean climate data

Two methods for elevation-dependent spatial interpolation of climatic data from sparse weather station networks were compared. Thirty-year monthly mea n minimum and maximum temperature and precipitation data from regions in we stern and eastern Canada were interpolated using thin-plate smoothing splin es (ANUSPLIN) and a statistical method termed 'Gradient plus Inverse-Distan ce-Squared' (GIDS). Data were withheld from approximately 50 stations in ea ch region and both methods were then used to predict the monthly mean value s far each climatic variable at those locations. The comparison revealed lo wer root mean square error (RMSE) for ANUSPLIN in 70 out of 72 months (thre e variables for 12 months for both regions). Higher RMSE for GIDS was cause d by more frequent occurrence of extreme errors. This result had important implications for surfaces generated using the two methods. Both interpolato rs performed best in the eastern (Ontario/Quebec) region where topographic and climatic gradients are smoother, whereas predicting precipitation in th e west (British Columbia/Alberta) was most difficult. In the latter case, A NUSPLIN clearly produced better results for most months. GIDS has certain a dvantages in being easy to implement and understand, hence providing a usef ul baseline to compare with more sophisticated methods. The significance of the errors for any method should be considered in light of the planned app lications (e.g., in extensive, uniform terrain with low relief, differences may not be important). (C) 2000 Elsevier Science B.V. All rights reserved.