Ordinary Cokriging was also performed using ArcMap’s
Geostatistical Analyst extension. Default settings were retained. The number
of neighbours to include was determined by comparing the RMSE from cross-validation
to determine the number of neighbours that minimized this value. From annual
data, the optimal number of neighbours to include was found to be four. This
was consistent among almost all other months; for the few for which this didn’t
hold true, the RMSE for four neighbours was very close to that for the “optimal”
number of neighbours (usually five). Only isotropic modeling was conducted as
the assumption of isotropy is commonly applied.
