obsarray is an extension to xarray for defining, storing and interfacing with uncertainty and measurement error-covariance information in NetCDF files using standardised metadata. These datasets that include standardised uncertainty and error-covariance information are dubbed `digital effects tables’. The obsarray documentation is available here. There are also jupyter notebooks available in the example section.
Although storage of full error-covariance matrices for large observation datasets is not practical, they are often structured to an extent that allows for simple parameterisation. obsarray makes use of a parameterisation method for error-covariance information, first developed in the FIDUCEO project, stored as attributes to uncertainty variables. In this way the datasets can be written/read in a way that this information is preserved.
One important aim of CoMet is to abstract away the complexity of dealing with error-covariances. obsarray does this by saving all error-correlation information in standardised metadata together with the uncertainty variables (and the input quantities themselves) in one dataset. These `digital effects tables’ can then be passed to punpy, which can directly use this information, so that users typically never have to interact with it. This allows efficient and easy uncertainty propagation.