Robust Estimation of Global precipitation using Neural networks (REGN)

The acronym REGN stands for Robust Estimation of Global precipitation using Neural networks. At the same time, regn ([rɛŋn]) is the swedish word for rain. The aim of the REGN project is to develop a neural-network based implementation of the GPROF algorithm.

This repository is used to collect all of the code and and results from this project.

AGU Presentation

Intermediate results from the REGN project have been presented at AGU 2020 in the presentation

H206-07 - Using Neural Networks for Bayesian Precipitation Retrievals from GPM Passive Microwave Observations

as part of the session H206 - Space-Based Precipitation Observations and Estimation: Innovations for Science and Applications I.

Slides from the presentation can be found here.

Running the code

The code required to reproduce the presented results consists of two parts:

• The regn Python package, which implements the QRNN-based GPROF retrieval
• The Jupyter notebooks contained in the notebooks/gmi and notebook/mhs folders, which contain the Python code which performs the numerical analyses.

Python dependencies

Note: Before installing any of these dependencies it is probably a good idea to create a new environment using Python venv or conda.

Our work builds on and requires a range publicly available packages, which are collected in the requirements.txt. After cloning this repository, you can install these packages using:

$ python3 -m p install -r requirements.txt

Installing the regn package

To run any of the notebooks, the regn package must be in your PYTHONPATH. The easiest way to achieve this is probably to just install the package using pip:

$ python3 -m p install -e .

The QRNN implementation

We have recently migrated our implementation of QRNNs from the typhon package to a new, separate package called quantnn. This is still relatively new and lacks extensive documentation but is what has been used within this study.

References

For background information on quantile regression neural networks (QRNNs), refer to the following article:

• Pfreundschuh, S., Eriksson, P., Duncan, D., Rydberg, B., Håkansson, N., and Thoss, A.: A neural network approach to estimating a posteriori distributions of Bayesian retrieval problems, Atmos. Meas. Tech., 11, 4627–4643, https://doi.org/10.5194/amt-11-4627-2018, 2018.

For more information on the current GPROF algorithm, please refer to the following publications:

• Kummerow, C. D., Randel, D. L., Kulie, M., Wang, N. Y., Ferraro, R., Joseph Munchak, S., & Petkovic, V. (2015). The evolution of the Goddard profiling algorithm to a fully parametric scheme. Journal of Atmospheric and Oceanic Technology, 32(12), 2265-2280.

• The GPROF version 5 ATBD