GloMarGridding: A Python Package for Spatial Interpolation to Support Structural Uncertainty Assessment of Climate Datasets

Abstract Global surface temperature datasets are constructed through processing chains that inherently introduce structural uncertainty. This arises from choices made in both the processing of input observations and in the spatial interpolation methods employed. Because these steps are often tightly integrated, it is difficult to isolate their individual contributions to uncertainty. Here we introduce GloMarGridding, a Python package designed to support the evaluation of structural uncertainty by providing flexible tools for spatial interpolation using Gaussian Process Regression Modelling (GPRM), also known as kriging. It enables the generation of spatially complete temperature fields from grid-box average and point observations, and associated uncertainties. GloMarGridding supports three spatial covariance parameterizations: fixed isotropic variograms, ellipse-based anisotropic model and empirically-derived covariance matrices. It also allows for uncertainty propagation via error covariance matrices and conditional simulation from input ensembles. By decoupling interpolation from earlier stages of dataset development - such as homogenization, quality control, and aggregation - this framework enables independent assessment of upstream processing choices and their impacts on gridded outputs.