Skip to content

Latest commit

 

History

History
47 lines (38 loc) · 2.37 KB

README.md

File metadata and controls

47 lines (38 loc) · 2.37 KB

Near-real-time monitoring of global terrestrial water storage anomalies and hydrological droughts

Shaoxing Mo, Maike Schumacher, Albert I. J. M. van Dijk, Xiaoqing Shi, Jichun Wu, Ehsan Forootan

Overview

This repository provides a PyTorch implementation of a Bayesian Convolutional Neural Network (BCNN) designed to predict GRACE/FO Terrestrial Water Storage Anomaly (TWSA) fields during the typical 3-month latency period before GRACE/FO data becomes available.

Model Inputs

The BCNN model takes the following inputs:

  • Historical GRACE/FO TWSA from the past 12 months.
  • ERA5-Land-derived:
    • Precipitation (P)
    • Temperature (T)
    • Reanalyzed TWSA (rTWSA)
  • These ERA5-Land variables are included for both:
    • The past 12 months
    • The 3-month latency period

Illustration of inputs and outputs of the BCNN model

Dependencies

  • python 3
  • PyTorch
  • h5py
  • matplotlib
  • scipy

Datasets for training and testing

The datasets used for BCNN training and testing, derived from JPL GRACE/FO Mascon and ERA5-Land datasets, are available at Google Drive. One can download the datasets, place them it in the 'datasets' subfolder, and train the BCNN model to reproduce our results.

Installation

To use this implementation, clone the repository and execute the code:

git clone https://github.com/njujinchun/DL4TWSA.git
cd DL4TWSA
python train_SVGD.py

Citation

If you find this repo useful for your research, please consider to cite:

* Mo, S., Schumacher, M., van Dijk, AIJM., Shi, X., Wu, J., Forootan, E. (2025). Near-real-time monitoring of global terrestrial water storage anomalies and hydrological droughts. Geophysical Research Letters (in press).

Questions

Contact Shaoxing Mo ([email protected]) with questions or comments.