Estimates of woody biomass and mixed effects improve isoscape predictions across a northern mixed forest

Abstract

Contemporary methods used to predict isotopic variation at regional scales have yet to include underlying distributions of the abundance of isotopic substrates. Additionally, traditional kriging methods fail to account for the potential influences of environmental grouping factors (i.e., random effects) that may reduce prediction error. We aim to improve upon traditional isoscape modeling techniques by accounting for variation in the abundances of isotopic substrates and evaluating the efficacy of a mixed-effects, regression kriging approach. We analyzed common moose forage from northeast Minnesota for δ13C and δ15N and estimated the isotopic landscape using regression kriging, both with and without random effects. We then compared these predictions to isoscape estimates informed by spatial variation in above-ground biomass. Finally, we kriged the regression residuals of our best-fitting models, added them to our isoscape predictions, and compared model performance using spatial hold-one-out cross validation. Isoscape predictions driven by uninformed and biomass-informed models varied by as much as 10‰. Compared to traditional methods, incorporating biomass estimates improved RMSE values by as much as 0.12 and 1.00% for δ13C and δ15N, respectively, while random effects improved r2 values by as much as 0.15 for δ13C and 0.87 for δ15N. Our findings illustrate how field-collected data, ancillary geospatial data, and novel spatial interpolation techniques can be used to more accurately estimate the isotopic landscape. Regression kriging using mixed-effects models and the refinement of model predictions using measures of abundance, provides a flexible, yet mechanistically driven approach to modeling isotopic variation across space.