A Workflow for Combining microscale Imaging Techniques in Paleoclimatology

Paleoclimatic and paleoenvironmental reconstructions rely on proxies derived from physical, chemical or biological properties of the investigated archive. In order to achieve the highest spatial (hundreds of micrometers), and thus temporal resolution (subannual), various imaging techniques, such as hyperspectral imaging, computed tomography, mass spectrometry imaging (MSI), X-ray, and µXRF can be employed. However, proxies generally respond to multiple environmental variables (e.g., the GDGT-based proxy CCat is influenced by both water column temperature and nutrient concentrations). Multi-proxy studies are necessary to obtain a comprehensive understanding of past conditions and to disentangle individual biogeochemical processes.

A major roadblock in multi-proxy studies is the alignment of data across multiple datasets since manual matching of ‘wiggles’ (1D time series) can be deceptive. With imaging data this issue can be avoided since data can be matched in 2D space. Moreover, RGB images are routinely obtained alongside each method. This provides a shared data layer between methods.

Not all imaging methods can be performed on the exact same sample slice, and MSI even requires multiple samples from the same core depth to cover multiple mass windows. Consequently, four aspects are taken into account in our proposed workflow: (I) subsamples need to be referenced back to the core; (II) datasets are obtained at different positions and can have vastly different resolutions; (III) samples may be distorted during sample preparation - so even two MSI measurements from the same sediment section at the same resolution cannot be mapped directly onto each other. And although these distortions are generally small, investigating seasonal variations requires consistency at the scale of hundreds of micrometers; (IV) after a transformation between images has been found, the data needs to be transformed (i.e., resampled). This requires interpolation, which can alter properties such as sparsity. Hence, interpolation targets as well as the interpolation methods need to be selected with care.

In this work, we present a workflow capable of semi-automatically combining image datasets from (sections of) sediment cores from any two imaging methods. Advanced methods for laminated sediments are also presented, as they are particularly suitable for fine-scale matching. With this workflow we aim to replace the tedious manual teaching point selection by providing robust image registration methods for routine multi-proxy studies on subannual scales.