Unlocking Earth’s past: how data assimilation is revolutionizing paleoclimate research
image: (A) The process begins with a randomly simulated climate model ensemble Xb. Using a proxy system model H, forward estimation is performed to map paleoclimate states (such as sea surface temperature, SST) onto the observational space of paleoclimate proxies (such as δ18O), yielding predicted proxy values Ye. The difference between the observed proxy values Yo and the predicted values Ye is referred to as the innovation introduced into the climate model ensemble. The ensemble Kalman filter framework then utilizes this innovation to update the climate state to Xa. The assimilation output (posterior, Xa) generally lies between the model ensemble (prior, Xb) and the observed values Yo, with reduced error. (B) The types of commonly used proxies and their sources include: δ18O, δ11B, Mg/Ca, and Δ47 from biogenic carbonates; CaCO3 content, Δ47, and δ13Corg from marine sediments; δ18O and δ2H from ice cores; archaeal biomarkers such as TEX86 , UK′37, and MBT′5Me; stomatal density index, tree ring width (TWR), and δ2H . Orange-bordered ovals indicate temperature proxies, green-bordered ovals indicate pCO2 proxies, and blue-bordered ovals indicate precipitation proxies.
Credit: ©Science China Press
Understanding Earth's climatic past is essential for predicting future scenarios. In a recent review published in Science China Earth Sciences, Mingsong Li and his colleagues from Peking University discuss recent advances in the principles and applications of paleoclimate data assimilation. This emerging approach integrates paleoclimate proxies with Earth system models to enhance the precision of climate reconstructions across geological epochs and thus offers a new framework to tackle some of the persistent challenges in the field of paleoclimatology.
Traditional methods for reconstructing Earth's ancient climates rely on proxies, such as isotope analyses from ice cores and sediments, and Earth system models. However, proxies are often sparse and unevenly distributed, while models depend heavily on parameter assumptions, which can limit their reliability. The review underscores how paleoclimate data assimilation bridges these gaps by incorporating observational data directly into model simulations, leveraging techniques like ensemble Kalman filtering and particle filtering to refine predictions and minimize uncertainties. This approach ensures that reconstructions not only reflect local conditions indicated by proxies but also align with broader, global climatic trends.
The paper discusses the wide applicability of data assimilation methods to a variety of proxies and to diverse models ranging from simple box models to fully coupled Earth system models. The review also introduces available software solutions in the framework of data assimilation, like LMR, PHYDA, cfr, DASH and DeepDA. Recent progress in data assimilation is also summarized, including the reconstruction on the states of the Earth system since the Last Glacial Maximum (LGM) and during the Paleocene–Eocene Thermal Maximum (PETM), providing a comprehensive overview of advancements in techniques and their practical applications to paleoclimate research. The authors discuss the versatility of data assimilation techniques, which are applicable across various geological periods, from the Cenozoic to the Mesozoic. They highlight the potential of these methods to improve predictions and deepen our understanding of past climate extremes, such as glacial-interglacial cycles, while identifying key areas for future research. The paper provides a useful reference for researchers exploring the complexities of paleoclimate reconstructions.
See the article:
Zhang H, Li M, Hu Y. 2025. Paleoclimate data assimilation: Principles and prospects. Science China Earth Sciences, Science China Earth Sciences, 68(2): 407–424, https://doi.org/10.1007/s11430-024-1439-y