Efficiency to assess climate-growth relationships: a comparative study between linear and non-linear methods.

Most of the current methods used for calibrating tree growth with climate can model only linear relationships. Recent studies demonstrated that artifi cial neural networks (ANN) are a promising method to detect nonlinear relationships. We present a comparative study on the effi ciency to assess the climate-growth responses using the ANN response function (ANNRF) and two “classical” methods: the correlation (COR) and linear response function (LRF). The results from the three methods, applied to three timberline coniferous species in the Italian Eastern Alps, are very coherent. COR and LRF have their strengths for being fast to compute, easy to replicate and to be interpreted and their major weakness in the linearity itself. ANNRF, instead, provide not only the sign and the intensity of growth responses but also their evolution along the variation range of the climatic parameters considered. They also allow the detection of thresholds or saturation limits providing a fi ner ecological interpretation of tree growth responses to climate. COR or LRF still represent a fundamental step to defi ne the boundaries of the processes, but ANNRF, computed with the best parameters, guarantee a better assessment of the complex non-linear interactions between trees and climate.