Using a process-based dendroclimatic proxy system model in a data assimilation framework: a test case in the Southern Hemisphere over the past centuries

Tree-ring widths represent the most commonly used proxy to reconstruct the climate of the last millennium at high resolution, thanks to their large-scale availability. The approach often relies on a relationship between tree-ring width series and climate estimated on the basis of a linear regression. The underlying linearity and stationarity assumptions may be inadequate. Dendroclimatic process-based models, such as MAIDEN (Modeling and Analysis In DENdroecology), may be able to overcome some of the limitations of the statistical approach. MAIDEN is a mechanistic ecophysiological model that simulates tree-ring growth starting from surface air temperature, precipitation and CO₂ concentration daily inputs. In this study, we successfully include the MAIDEN model into a data assimilation procedure as a proxy system model to robustly compare the outputs of an Earth system model with tree-ring width observations and provide a spatially-gridded reconstruction of continental temperature, precipitation and winds in the mid to high latitudes of the Southern Hemisphere over the past centuries. More specifically, we evaluate the benefits of using process-based tree-growth models such as MAIDEN for reconstructing past climate with data assimilation compared to the commonly used linear regression. The comparison of the reconstructions with instrumental data indicates an equivalent skill of both the regression- and process-based proxy system models in the data assimilation framework. Nevertheless, the MAIDEN model still brings important advantages that could result in more robust reconstructions beyond the instrumental era. Moreover, improvements continuously made in such models or in their calibration procedure also offer encouraging perspectives. Important steps have thus been made to demonstrate that using a process-based model like MAIDEN as a proxy system model is a promising way to improve the large-scale climate reconstructions with data assimilation.