Testing the performance of dendroclimatic process-based models at global scale with the PAGES2k tree-ring width database

Abstract

Tree-rings are one of the most commonly used proxies for reconstructing past climates at annual resolution. The climate information is generally deduced from tree-rings using statistical relationships, but the assumed linearity and stationarity may be inadequate. Process-based models allow for non-stationarity and non-linearity; however, many challenges are associated with their application for global scale reconstructions. In this study, we aim to test the feasibility of using the mechanistic model MAIDEN at the global scale for paleoclimate data assimilation based reconstructions by applying it to the PAGES2k tree-ring width database. We also compare its performance with the simpler model VS-Lite, often used in global applications. Both models are skillful in terms of calibration and verification correlations for a similar number of sites (63 and 64 for VS-Lite and MAIDEN, respectively). VS-Lite tends to perform better for sites where the climate signal in tree-rings is strong and clear.

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

“…Therefore, this variable is used for the comparison with the observed TRW. In practice, as previously discussed in Rezsöhazy et al (2020Rezsöhazy et al ( , 2021, the comparison of the tree-growth observations with the forest carbon accumulation simulated by MAIDEN implies some limitations. For instance, Gennaretti et al ( 2018) computed a wood biomass index directly comparable to what MAIDEN simulates.…”

“…2.3.1). MAIDEN was calibrated at the 12 SH sites from PAGES2k2017 in Rezsöhazy et al (2021) driven by the daily maximum and minimum temperature as well as precipitation from the Global Meteorological Forcing Dataset for land surface modeling (v2) (http://hydrology.princeton. edu/data.php, last access: 5 May 2021; Sheffield et al, 2006) at 0.5 • resolution (hereafter GMF) over 1950-2000 CE.…”

“…A similar criterion based on the calibration only could not be used to select the TRW records for MAIDEN. The complexity of the model and the number of calibrated parameters make it more prone to overfitting problems and biological inconsistency during the calibration (Rezsöhazy et al, 2020(Rezsöhazy et al, , 2021. As a consequence, the selection of the TRW records has to be more restrictive but may also lead to a better stability of the model.…”

“…MAIDEN includes the atmospheric CO 2 concentration among other inputs, which represents a major advantage for taking into account the potential effect of the recent exponential increase in CO 2 on tree growth (Myhre et al, 2013;Boucher et al, 2014). In Rezsöhazy et al (2021), the MAIDEN model has been successfully applied over the last century to the PAGES 2k TRW database by comparing the simulated annual quantity of carbon allocated to the stem with the observed tree-ring widths (both normalized) using high-resolution climate data as inputs of the model. The model has been calibrated over the second half of the 20th century using the Bayesian procedure with Markov chain Monte Carlo sampling elaborated in Rezsöhazy et al (2020) and successfully verified over the first half of the century at 20 % of the PAGES 2k tree-ring sites.…”

“…The goal is to evaluate if and how a complex processbased model of tree growth like MAIDEN can contribute to the improvement of large-scale DA-based reconstructions of past climate variability compared to the regression model classically used so far. Because MAIDEN has been shown to be skillful in different regions of the Southern Hemisphere (SH; Rezsöhazy et al, 2021), we focus on that region and provide climate reconstructions of near-surface air temperature, precipitation and winds in the middle to high latitudes of the SH over the past 400 years. The Southern Hemisphere also constitutes an interesting study case, as few reconstructions have been performed in this region so far (Neukom et al, 2011(Neukom et al, , 2014.…”