Innovative Model for Assessing Global Water Storage

Water is essential for life on Earth, but its availability and distribution are subject to natural and human-induced changes. Understanding how water is stored and circulated on the planet is crucial for hydrology, climate science, sustainable water management, and hazard prediction. However, measuring global water storage is challenging due to the complexity and heterogeneity of the hydrological cycle.

Satellite observations and hydrological models

One way to estimate global water storage is to use satellite observations of the Earth’s gravity field, such as those provided by the Gravity Recovery and Climate Experiment (GRACE) mission. GRACE measures the changes in the Earth’s gravity caused by the variations in the mass of water on land, in the oceans, and in the atmosphere. By subtracting the effects of other mass changes, such as those due to tectonic movements or glacial isostatic adjustment, GRACE can provide monthly estimates of total water storage anomalies (TWSAs), which represent the deviations from the long-term mean of water storage.

However, GRACE has some limitations, such as its coarse spatial resolution (about 300 km), its temporal gaps (due to satellite maintenance or replacement), and its sensitivity to noise and errors (due to measurement uncertainties or data processing). To overcome these limitations, researchers often combine GRACE data with hydrological models, which simulate the water balance and fluxes on land based on meteorological inputs, land surface characteristics, and physical laws. Hydrological models can provide finer spatial and temporal resolution, fill in the missing data, and correct for the errors in GRACE data. However, hydrological models also have their own uncertainties, such as those due to parameterization, calibration, or input data quality.

A novel deep learning approach

In their recent publication in Nature Water , researchers Junyang Gou and Professor Benedikt Soja from ETH Zurich introduced a novel deep learning approach to integrate GRACE data with hydrological models to produce global high-resolution TWSAs. Their method uses a self-supervised data assimilation framework based on deep neural networks, which can learn from both GRACE data and hydrological models without requiring any external labels or supervision.

Their method consists of two steps: first, they train a deep neural network to learn the mapping between GRACE data and hydrological model outputs at different spatial scales; second, they use the trained network to generate high-resolution TWSAs by assimilating GRACE data with hydrological model outputs. Their method can produce TWSAs at a spatial resolution of 0.25 degrees (about 28 km) and a monthly temporal resolution from 2003 to 2019.

Remarkable accuracy and benefits

The researchers evaluated their method using various metrics and comparisons with other datasets, such as in situ measurements, alternative hydrological models, or independent satellite observations. They found that their method achieves remarkable accuracy even in smaller basins or regions where GRACE data are noisy or missing. They also demonstrated that their method can capture the seasonal and interannual variations of water storage, as well as the impacts of extreme events, such as droughts or floods.

Their method promises significant benefits across various domains that rely on accurate estimates of global water storage. For example, their method can improve the understanding of the hydrological cycle and its interactions with climate change; it can support sustainable water management and planning by providing reliable information on water availability and risks; it can enhance hazard prediction and mitigation by detecting early signs of water-related disasters; and it can facilitate scientific communication and education by providing visually appealing maps of global water storage.

Conclusion

In summary, Gou and Soja introduced a new model for measuring global water storage using a novel deep learning approach that integrates satellite observations with hydrological models. Their method achieves remarkable accuracy even in smaller basins and promises significant benefits across various domains.

Recent Blog : Nature-Based Solutions: Key to Climate Change Mitigation