Institution profile

GFZ German Research Center for Geosciences

Academic institutioneurope · de
Official website
Research library3linked papers
Opportunities0open roles
Selected work

Representative Papers

Reconstructing GRACE Terrestrial Water Storage with Spatio-Temporal Graph Neural Networks: An Application to South America

Jun 22, 2026

This study addresses the limitation of GRACE satellite observations, which provide terrestrial water storage (TWS) time series only since 2002 and thus hinder long-term climate analyses. To overcome this, the authors introduce a multivariate spatiotemporal graph neural network (MTGNN) for reconstructing monthly TWS anomalies back to 1940, leveraging daily ERA5 meteorological data—precipitation, evapotranspiration, and runoff. A novel interpretable hybrid adjacency matrix is constructed by integrating geographical distance with climate-driven lagged correlations. The model achieves high reconstruction accuracy, with Pearson correlation coefficients of 0.69 at the grid-cell scale and 0.94 at the basin scale, near-zero bias, and faithful representation of the spatial patterns associated with the 2015/16 El Niño and 2020/21 La Niña events. Notably, it attains performance comparable to existing methods while using only one-half to one-tenth of their input variables.

0 citationsRead paper

Conditional updates of neural network weights for increased out of training performance

Dec 03, 2025

Neural networks often suffer significant performance degradation under distributional shifts—such as out-of-distribution generalization, spatiotemporal extrapolation, and cross-domain transfer—due to mismatches between training and deployment data distributions. To address this, we propose a conditional dynamic weight update framework. Its core innovation is a weight anomaly regression mechanism: sensitive weight change patterns induced by distribution shifts are identified via subset retraining; an interpretable regression predictor is then constructed to map input features to weight increments; finally, model parameters are conditionally extrapolated. The method integrates weight difference extraction, regression modeling, and extrapolation techniques, and is empirically validated on multi-source climate observation datasets. Across temporal, spatial, and cross-domain extrapolation tasks, it substantially improves prediction accuracy and robustness on out-of-distribution data, while preserving interpretability and practical applicability.

0 citationsRead paper
Recent publications

Latest Papers

Reconstructing GRACE Terrestrial Water Storage with Spatio-Temporal Graph Neural Networks: An Application to South America

Jun 22, 2026

This study addresses the limitation of GRACE satellite observations, which provide terrestrial water storage (TWS) time series only since 2002 and thus hinder long-term climate analyses. To overcome this, the authors introduce a multivariate spatiotemporal graph neural network (MTGNN) for reconstructing monthly TWS anomalies back to 1940, leveraging daily ERA5 meteorological data—precipitation, evapotranspiration, and runoff. A novel interpretable hybrid adjacency matrix is constructed by integrating geographical distance with climate-driven lagged correlations. The model achieves high reconstruction accuracy, with Pearson correlation coefficients of 0.69 at the grid-cell scale and 0.94 at the basin scale, near-zero bias, and faithful representation of the spatial patterns associated with the 2015/16 El Niño and 2020/21 La Niña events. Notably, it attains performance comparable to existing methods while using only one-half to one-tenth of their input variables.

0 citationsRead paper

Conditional updates of neural network weights for increased out of training performance

Dec 03, 2025

Neural networks often suffer significant performance degradation under distributional shifts—such as out-of-distribution generalization, spatiotemporal extrapolation, and cross-domain transfer—due to mismatches between training and deployment data distributions. To address this, we propose a conditional dynamic weight update framework. Its core innovation is a weight anomaly regression mechanism: sensitive weight change patterns induced by distribution shifts are identified via subset retraining; an interpretable regression predictor is then constructed to map input features to weight increments; finally, model parameters are conditionally extrapolated. The method integrates weight difference extraction, regression modeling, and extrapolation techniques, and is empirically validated on multi-source climate observation datasets. Across temporal, spatial, and cross-domain extrapolation tasks, it substantially improves prediction accuracy and robustness on out-of-distribution data, while preserving interpretability and practical applicability.

0 citationsRead paper