Patterns of precipitation δ18O through the Iberian Peninsula: Machine Learning dynamic modeling for climate proxy calibration

Keywords: Paleoclimate proxy, Stable Isotope, Artificial Intelligence, Lake, Speleothem

Oxygen isotope ratios are the most used geochemical proxy data in paleoclimatic and paleohydrological  reconstructions, currently measured in a variety of continental carbonates, including shells, sediments and speleothems. δ¹⁸O of carbonates that form in conditions of (or close to) isotopic equilibrium does not only depend on local temperature (that controls isotopic fractionation) but also on the isotopic composition of the water from which they precipitate, which in most cases is in turn determined or controlled by the δ¹⁸O of rainfall.  For this reason, understanding patterns of temporal and spatial changes in rainfall δ¹⁸O is essential for the calibration of proxies based in carbonate δ¹⁸O. Intensive monitoring programs are currently being performed in target natural systems such as lakes and karst caves. These programs however yield local isotopic records of limited duration, hampering their regional utility.      

In this work, we present a Machine Learning based approach that takes into consideration not only the dependence of rainfall δ¹⁸O to latitude and altitude, but also to the main factors affecting winter precipitation variability within the Iberian Peninsula (IP), such as the North Atlantic Oscillation and the Western Mediterranean Oscillation. This approach is based in the available isotopic data of precipitation (IAEA Global Network of Isotopes in Precipitation – GNIP) and local series.

Our model provides predicted values of δ¹⁸O with improved accuracy and uncertainty relative to current approaches. By capturing the variability of δ¹⁸O associated with diverse atmospheric conditions, it will also contribute to refine the interpretation paleoclimatic reconstructions in this region.

 

Contribution to PID2021-122854OB-I00 and research group 910198 of the UCM.