Parameterization of multidimensional process-based tree-ring models: Why it is important?

<p>Improvement of our understanding of tree-growth processes and accurate interpretations of climatic signals in tree rings have recently become possible through the application of process-based models, e.g., Biome3, MAIDEN,  ASTANEA, CAMBIUM, PRYSM, VS-lite and others, which simulate tree growth based on non-linear effects of environmental conditions. The process-based Vaganov–Shashkin model (VS-model) is one such model which describes tree-ring formation as a result of multivariate affects of local climate (temperature, soil moisture and solar irradiance). As with most of the process-based models, the VS-model is a complex tool that requires a considerable number of model parameters that should be reasonably estimated for each forest stand. This leads to problem of accurate model parameterization, namely estimations of optimal values of the model parameters necessary to guarantee: (1) the best fit to the observed tree-ring measurements; (2) identification of the specific seasonal cell production and enlargement; (3) reasonable ecological interpretation in terms of processes involved in the model.</p><p>Based on differential evolution (DE) approach adopted to the model parameterization using the supercomputer facilities it was shown:</p><p>(1) a significant spatial variability of adjusted VS-parameter values (with corresponded ecological interpretation) that provide the best fit to the actual tree-ring chronologies from climatically contrasted sites distributed in the vast territories of Eurasia and as a result, the models ability to capture a significant diversity in non-linear tree-ring growth responses that are climatically induced,</p><p>(2) the high sensitivity of the models even for forest stands where mixed climatic signal affects on tree-ring growth during growing season,</p><p>(3) the high probability to obtain a "correct" model parameterization which explains up to 60% tree-ring variance by the climate forcing even for randomly generated "chronologies"  in case of incorrect usage of the calibration-verification strategy for multidimensional models.</p>