Assessing proxy system models of cave dripwater δ18O variability

Abstract Large uncertainties in future climate change necessitate understanding of decadal-centennial variability in past climate. Speleothems, which capture changes in the stable isotope composition (i.e., δ18O values) of rainfall, have been widely employed to constrain hydroclimate variability in the past due to their continuous growth and potential for high temporal resolution. However, the interpretation of speleothem proxies (i.e., geochemical and growth rate records) is still imperfectly constrained. One challenge involves understanding the transformation from rainfall δ18O to calcite δ18O. A recent advance to address this challenge is the development of proxy system models (PSMs), which translate climate variables to speleothem calcite δ18O values. However, the complexity and applicability of each model varies. In order to assess which speleothem PSMs are most suitable for simulating speleothem records, we evaluate four commonly applied PSMs, which span a range of complexity (one simple, two intermediate-complexity, and one complex model). The evaluation is based on a given model’s ability to simulate observed drip δ18O time series using multi-year long cave dripwater and rainfall data sets from semi-arid central Texas and tropical Borneo. The models are based on the running mean of rainfall δ18O (RM model), dripwater transit time (PRYSM model), StorAge Selection (SAS model), and combination of flow paths in karst systems (Karstolution model). We explore the parameter settings of these models and evaluate them employing the Root Mean Square Errors (RMSEs) metric. We find that for drip sites in semi-arid climates with long residence times, model complexity of at least intermediate levels is required to simulate cave dripwater δ18O variability. The RMSEs of intermediate-complexity PSMs (PRYSM, SAS, and Karstolution) are close to the range of analytical errors, while the RMSE of the simple RM model is far larger. For drip sites in wet climates with short residence times, all four models, including the RM model, are capable of simulating cave dripwater δ18O well. In practice, we assert that the use of models of intermediate complexity such as PRYSM may be required when detailed knowledge of variables such as groundwater residence time for a given cave setting is unknown. Such PSMs are useful for global-scale studies due to their broad applicability and simplicity. We also illustrate the use of comprehensive physical models, such as Karstolution, for understanding cave processes like fractionation due to epikarst evaporation. The poorer performance of the PSMs in semi-arid environments than their performance in humid environments suggests that additional in-situ process studies are needed to improve simulations of relatively dry cave systems and provide enhanced interpretations of speleothem reconstructions from these environments.