Application and evaluation of the dendroclimatic process-based model MAIDEN during the last century in Canada and Europe

• Published in: Climate of the Past
• Main Authors: J. Rezsöhazy, H. Goosse, J. Guiot, F. Gennaretti, E. Boucher, F. André, M. Jonard
• Format: Article in Journal/Newspaper
• Language: English
• Published: Copernicus Publications 2020
• Subjects: envir; geo; Canada; taiga
• Online Access: https://doi.org/10.5194/cp-16-1043-2020; https://www.clim-past.net/16/1043/2020/cp-16-1043-2020.pdf; https://doaj.org/article/aa3f3dabbb734042a11c317698b4d514

Tree-ring archives are one of the main sources of information to reconstruct climate variations over the last millennium with annual resolution. The links between tree-ring proxies and climate have usually been estimated using statistical approaches, assuming linear and stationary relationships. Both assumptions may be inadequate, but this issue can be overcome by ecophysiological modelling based on mechanistic understanding. In this respect, the model MAIDEN (Modeling and Analysis In DENdroecology) simulating tree-ring growth from daily temperature and precipitation, considering carbon assimilation and allocation in forest stands, may constitute a valuable tool. However, the lack of local meteorological data and the limited characterization of tree species traits can complicate the calibration and validation of such a complex model, which may hamper palaeoclimate applications. The goal of this study is to test the applicability of the MAIDEN model in a palaeoclimate context using as a test case tree-ring observations covering the 20th century from 21 Eastern Canadian taiga sites and 3 European sites. More specifically, we investigate the model sensitivity to parameter calibration and to the quality of climatic inputs, and we evaluate the model performance using a validation procedure. We also examine the added value of using MAIDEN in palaeoclimate applications compared to a simpler tree-growth model, i.e. VS-Lite. A Bayesian calibration of the most sensitive model parameters provides good results at most of the selected sites with high correlations between simulated and observed tree growth. Although MAIDEN is found to be sensitive to the quality of the climatic inputs, simple bias correction and downscaling techniques of these data improve significantly the performance of the model. The split-sample validation of MAIDEN gives encouraging results but requires long tree ring and meteorological series to give robust results. We also highlight a risk of overfitting in the calibration of model parameters that ...