Reconstructing the data gap between GRACE and GRACE follow-on at the basin scale using artificial neural network

Area of Science:

• Earth Science
• Climate Science
• Hydrology

Background:

• The Gravity Recovery and Climate Experiment (GRACE) and GRACE Follow-On (GRACE-FO) missions provide crucial data for monitoring terrestrial water storage (TWS) changes.
• A significant data gap exists between the GRACE and GRACE-FO missions, hindering continuous TWS change analysis and potentially introducing biases.

Purpose of the Study:

• To develop and validate a novel method for reconstructing terrestrial water storage (TWS) data during the gap between GRACE and GRACE-FO observations.
• To provide a continuous, high-quality TWS dataset for improved hydrological and climate research, particularly in data-scarce regions.

Main Methods:

• A hybrid approach combining Multichannel Singular Spectrum Analysis (MSSA) for initial interpolation and component decomposition.
• Utilizing a Back Propagation Neural Network (BPNN) to establish relationships between reconstructed components of hydroclimatic drivers and TWS data.
• Implementing a sliding window validation technique to accurately assess model performance under realistic conditions.

Main Results:

• Successfully reconstructed TWS data gaps in 28 hot areas with severe TWS changes (mean RMSE of 2.7 cm) and 26 major river basins (mean RMSE of 2.2 cm).
• The combined MSSA-BPNN method demonstrated superior performance compared to standalone MSSA and most other artificial neural network-based approaches.
• Achieved impressive reconstruction accuracy, especially considering the nominal GRACE accuracy of ~2 cm and significant TWS variations in monitored areas.

Conclusions:

• The developed MSSA-BPNN method offers an advanced solution for gap-filling, data reconstruction, and data fusion in satellite-based Earth observation.
• The reconstructed TWS data provides valuable, continuous information essential for advancing hydrological and climate change studies.
• This research specifically benefits regions with significant TWS fluctuations where continuous data was previously unavailable.