Low Counts, High Stakes: Challenges in micro- Geochemical Imaging of Paleoclimate Archives

Imaging techniques, such as micro-X-ray fluorescence (µ-XRF), are critical tools for paleoenvironmental reconstructions. Recent advances in two-dimensional (2D) mapping enable detailed spatial and temporal analyses of paleoclimate archives, providing unprecedented insights into past environmental changes. However, generating high-resolution geochemical maps often requires tradeoffs between measurement time and data quality. Limited exposure times per pixel result in low counts and frequent zero values, which increase noise and variability in the data. The compositional nature of count data introduces additional challenges, as zeros and zero replacement strategies can compromise the effectiveness of log-ratio transformations.

This study investigates the impact of low-count µ-XRF data using maps from laminated lake sediments in Lake Ammersee, Germany. The varves in this archive record seasonal changes, with Ca-enriched layers forming during summer and Ti-enriched detrital layers indicating flood events. We analyzed 2D maps from eight repetition measurements and simulated extended exposure times by stacking them. Additionally, synthetic datasets were used to simulate both low (mean count intensity < 10) and high (mean count intensity > 10,000) count rate scenarios.

Our results reveal that low-count data exhibit persistent artifacts in both native and log space. Spurious correlations and clusters of elemental ratios remain evident, even after log-ratio transformations, and are most pronounced in low-intensity datasets. These artifacts complicate the reconstruction of paleoclimate proxies and the interpretation of geochemical records. Increasing count rates by extending the measurement time may provide a short-term solution to reduce these artifacts. However, current methodologies for addressing low-count compositional data and zero-replacement strategies remain inadequate, particularly for high-resolution imaging techniques. Advancing statistical approaches is essential to develop robust frameworks for interpreting paleoclimate proxies.