Fertilization is a common practice to increase the productivity and the stand value in the southeastern US. The decision to fertilize a given site is driven by site characteristics and the expected magnitude of response. To determine the magnitude, forest researchers typically rely on fertilization trials established throughout the region of interest and derive growth equations to reflect an increase in either site index or volume. Such equations lack an explicit spatial prediction component. To bridge this gap, we developed a modeling framework that explicitly evaluates the likelihood of a fertilization response as a binary process and the magnitude of such response as a separate model. The methodology relies on the non-parametric interpolator thin plate spines. To test the efficacy of this framework, both percent volume and dominant height response to repeated fertilizer treatments were estimated using data from long-term research trials in Georgia. Several environmental covariates were evaluated on their ability to reduce the models’ root mean square error and account for more of the variation in percent gain from fertilization regimes. Results showed that the inclusion of such covariates improved the model performance and reduced errors associated with interpolation. Thresholding expected responses from fertilization treatments allows practitioners to evaluate the probability of achieving a given response.

This article is published and distributed under the terms of the Oxford University Press, Standard Journals Publication Model (https://academic.oup.com/journals/pages/open_access/funder_policies/chorus/standard_publication_model)