Machine-learning-based reconstruction of long-term global terrestrial water storage anomalies from observed, satellite and land-surface model data
<p>Understanding long-term terrestrial water storage (TWS) variations is vital for investigating hydrological extreme events, managing water resources and assessing climate change impacts. However, the limited data duration from the Gravity Recovery and Climate Experiment (GRACE) and its follo...
Saved in:
| Main Authors: | , , |
|---|---|
| Format: | Article |
| Language: | English |
| Published: |
Copernicus Publications
2025-06-01
|
| Series: | Earth System Science Data |
| Online Access: | https://essd.copernicus.org/articles/17/2575/2025/essd-17-2575-2025.pdf |
| Tags: |
Add Tag
No Tags, Be the first to tag this record!
|
| Summary: | <p>Understanding long-term terrestrial water storage (TWS) variations is vital for investigating hydrological extreme events, managing water resources and assessing climate change impacts. However, the limited data duration from the Gravity Recovery and Climate Experiment (GRACE) and its follow-on mission (GRACE-FO) poses challenges for comprehensive long-term analysis. In this study, we reconstruct TWS anomalies (TWSAs) for the period from January 1960 to December 2022, thereby filling data gaps between the GRACE and GRACE-FO missions and generating a complete dataset for the pre-GRACE era. The workflow involves identifying optimal predictors from land surface model (LSM) outputs, meteorological variables and climatic indices using a novel Bayesian network (BN) technique for raster-based TWSA simulations. Climate indices, like the Oceanic Niño Index and Dipole Mode Index, are selected as optimal predictors for a large number of grid cells globally, along with TWSAs from LSM outputs. The most effective machine learning (ML) algorithms among convolutional neural network (CNN), support vector regression (SVR), extra trees regressor (ETR) and stacking ensemble regression (SER) models are evaluated at each grid cell to achieve optimal reproducibility. Globally, ETR performs best for most of the grid cells; this is also noticed at the river basin scale, particularly for the Ganga–Brahmaputra–Meghna, Godavari, Krishna, Limpopo and Nile river basins. The simulated TWSAs (BNML_TWSA) outperformed the TWSAs from LSM outputs when evaluated against GRACE datasets. Improvements are particularly noted in river basins such as the Godavari, Krishna, Danube and Amazon, with median correlation coefficient, Nash–Sutcliffe efficiency, and RMSE values for all grid cells in the Godavari Basin, India, being 0.927, 0.839 and 63.7 mm, respectively. A comparison with TWSAs reconstructed in recent studies indicates that the proposed BNML_TWSA outperforms them globally as well as for all of the 11 major river basins examined. Furthermore, the uncertainty of BNML_TWSA is assessed for each grid cell in terms of the standard error. Results show smaller standard error magnitudes in grid cells in arid regions compared to other regions. The presented gridded dataset is published at <span class="uri">https://doi.org/10.6084/m9.figshare.25376695</span> <span class="cit" id="xref_paren.1">(<a href="#bib1.bibx49">Mandal et al.</a>, <a href="#bib1.bibx49">2024</a>)</span>, featuring a spatial resolution of 0.50° <span class="inline-formula">×</span> 0.50° and offering global coverage.</p> |
|---|---|
| ISSN: | 1866-3508 1866-3516 |