The openMWR package#
Note: This documentation is still in development and may contain errors or incomplete sections.
openMWR is a Python package for training a neural network-based retrieval for a Hatpro microwave radiometer.
It includes tools for generating training datasets, training retrieval algorithms, analyzing their performance, and applying the trained algorithms to measurements operationally. It can be used to create multilinear regression retrievals, but it is focused on more complex neural network retrievals. It also supports the training of retrieval algorithms for elevation scans, to study the boundary layer.
OpenMWR also includes the torchMWRT package, a PyTorch translation of the radiative-transfer
calculation core of PyRTlib. It produces the same results as
PyRTlib while being orders of magnitude faster. That is mainly achieved by
heavy vectorizations. It also supports GPU acceleration and automatic differentiation,
which is, however, not necessary for the classical retrieval algorithms.