Kriging and PINNs: Creating regular grids from non-Gaussian paleoclimatic variables

Seminar by Prof. Fabrice Lambert from Universidad Católica de Chile

26 September 2023
KST 10:30 – 11:30

The Seminar is being held in Room 1010 (Jasmin) – Integrated mechanical engineering building. Click here for the campus map.

 Earth System Models of high (ESMs) and Intermediate Complexity (EMICs) are very useful tools to test hypotheses and perform sensitivity analyses. However, they require regular input fields as boundary conditions or forcing variables. In the paleoclimatic setting, knowledge of the spatial and temporal variability of such variables comes from paleoclimatic archives, such as marine sediment cores, ice cores, speleothems, etc. Paleoclimatic archives are retrieved in specific locations where conditions and logistics permit it, resulting in unevenly distributed data points across the globe. To use empirical information in climate models, one has to interpolate the measured data to a global regular grid used as an input field for the simulation. A wide range of interpolation methods is commonly used, but most of them require a reasonably normal (Gaussian) distribution of the variable of interest. Temperature-related proxy variables usually comply with this condition, but many others do not. For example, precipitation and mineral dust are non-normally distributed both in time and in space. In this talk, I will present two approaches developed by my research group to create regular grids from non-normally distributed dust deposition fluxes measured in unevenly distributed paleoclimatic archives.