Methodological constrains of tree-ring stable isotope chronologies

Tree-ring stable isotope (TRSI) chronologies that combine information from living and relict wood have the potential to capture long-term trends that might be missing in traditional tree-ring width and maximum latewood density measurements. Our understanding of the possible effects of different methods to develop TRSI chronologies is, however, still incomplete. Here, we compare and evaluate five such methods applied to three multi-millennial-long oxygen isotope (δ 18 O) TRSI datasets from central Europe, the European Alps and Japan: (a) raw data, (b) cohort correction, (c) interactive mean correction, (d) outlier correction, and (e) series normalization. We show that the spectral properties preserved in the final TRSI chronologies not only depend on the data used, but also on the techniques applied. Method (a) is particularly prone to outliers if the sample size is low. Method (b) may create artificial steps and trends when single measurement series share similar start dates and/or when end and start dates are systematically skewed. Methods (c) and (d) yield similar results for annually resolved data, yet (d) is more suitable for temporally pooled datasets and less sensitive to potential biological age effects. Method (e) removes any low-frequency signal. Our findings demonstrate the risks and rewards of different TRSI chronology development techniques that must be carefully adapted to both, the data used and the question posed. • We compare five correction methods for Tree-Ring Stable Isotope chronologies using three multi-millennial datasets. • The main challenge is preserving low-frequency signals in diverse TRSI datasets with limited sample replication. • Each method has specific strengths and limitations in preserving different frequency signals. • The choice of method significantly impacts the final chronology. • The choice of method should be based on dataset characteristics and research goals. • The raw data is recommended, complemented by geographical analysis and outlier validation.