Assessing prediction accuracy in a regression kriging surface of Pinus radiata outerwood density across New Zealand

Validation data (n = 200) found the final model accounted for 70% of the variance in outerwood density. The model included three significant (P < 0.0001) variables - stand age, planting decade, and mean annual air temperature - and a significant (P < 0.0001) isotropic spherical kriging model that accounts for spatial covariance in the data. Mean annual air temperature was included in the model as a positive linear relationship, whereas the natural log of age was fitted with a quadratic function that increases with age. Planting decade was added to the model as a categorical variable that indicates that, relative to stands planted since 1970, wood density in those planted between 1940 and 1970 was about 15 kg m−3 higher after adjusting for site and age effects. Outerwood density varies widely across New Zealand with highest values in the warmer northern latitudes of the upper North Island and in the north of the South Island. Conversely, cooler southern latitudes and higher elevations had lowest outerwood density values. Prediction accuracy at different levels of scale was established using the validation dataset allowing users to estimate the error associated with wood density predictions at different scales.