Machine learning downscaling of GRACE/GRACE-FO data to capture spatial-temporal drought effects on groundwater storage at a local scale under data-scarcity DOI Creative Commons

Christopher Shilengwe,

Kawawa Banda, Imasiku Nyambe

et al.

ENVIRONMENTAL SYSTEMS RESEARCH, Journal Year: 2024, Volume and Issue: 13(1)

Published: Sept. 3, 2024

The continued threat from climate change and human impacts on water resources demands high-resolution continuous hydrological data accessibility for predicting trends availability. This study proposes a novel threefold downscaling method based machine learning (ML) which integrates: normalization; interaction of hydrometeorological variables; the application time series split cross-validation that produces high spatial resolution groundwater storage anomaly (GWSA) dataset Gravity Recovery Climate Experiment (GRACE) its successor mission, GRACE Follow-On (GRACE-FO). In study, relationship between terrestrial (TWSA) other land surface variables (e.g., vegetation coverage, temperature, precipitation, in situ level data) is leveraged to downscale GWSA. predicted downscaled GWSA datasets were tested using monthly observations, results showed model satisfactorily reproduced temporal variations area, with Nash-Sutcliffe efficiency (NSE) correlation coefficient values 0.8674 (random forest) 0.7909 (XGBoost), respectively. Evapotranspiration was most influential predictor variable random forest model, whereas it rainfall XGBoost model. particular, excelled aligning closely observed patterns, as evidenced by positive correlations lower error metrics (Mean Absolute Error (MAE) 54.78 mm; R-squared (R²) 0.8674). 5 km (based decreasing trend associated variability pattern. An increase drought severity during El Niño lengthened full recovery historical trends. Furthermore, lag occurrence precipitation recharge likely controlled intensity characteristics aquifer. Projected increases could further times response droughts changing climate, resetting new tipping condition. Therefore, adaptation strategies must recognise less will be available supplement supply droughts.

Language: Английский

Advances in GRACE satellite studies on terrestrial water storage: a comprehensive review DOI Creative Commons

Jyoti Karki,

Jinming Hu, Yu Zhu

et al.

Geocarto International, Journal Year: 2025, Volume and Issue: 40(1)

Published: March 28, 2025

Language: Английский

Citations

0
A non-stationary downscaling and gap-filling approach for GRACE/GRACE-FO data under climatic and anthropogenic influences DOI Creative Commons
Seyed Mojtaba Mousavimehr, Mohammad Reza Kavianpour

Applied Water Science, Journal Year: 2025, Volume and Issue: 15(5)

Published: April 7, 2025

Language: Английский

Citations

0
Machine learning downscaling of GRACE/GRACE-FO data to capture spatial-temporal drought effects on groundwater storage at a local scale under data-scarcity DOI Creative Commons

Christopher Shilengwe,

Kawawa Banda, Imasiku Nyambe

et al.

ENVIRONMENTAL SYSTEMS RESEARCH, Journal Year: 2024, Volume and Issue: 13(1)

Published: Sept. 3, 2024

The continued threat from climate change and human impacts on water resources demands high-resolution continuous hydrological data accessibility for predicting trends availability. This study proposes a novel threefold downscaling method based machine learning (ML) which integrates: normalization; interaction of hydrometeorological variables; the application time series split cross-validation that produces high spatial resolution groundwater storage anomaly (GWSA) dataset Gravity Recovery Climate Experiment (GRACE) its successor mission, GRACE Follow-On (GRACE-FO). In study, relationship between terrestrial (TWSA) other land surface variables (e.g., vegetation coverage, temperature, precipitation, in situ level data) is leveraged to downscale GWSA. predicted downscaled GWSA datasets were tested using monthly observations, results showed model satisfactorily reproduced temporal variations area, with Nash-Sutcliffe efficiency (NSE) correlation coefficient values 0.8674 (random forest) 0.7909 (XGBoost), respectively. Evapotranspiration was most influential predictor variable random forest model, whereas it rainfall XGBoost model. particular, excelled aligning closely observed patterns, as evidenced by positive correlations lower error metrics (Mean Absolute Error (MAE) 54.78 mm; R-squared (R²) 0.8674). 5 km (based decreasing trend associated variability pattern. An increase drought severity during El Niño lengthened full recovery historical trends. Furthermore, lag occurrence precipitation recharge likely controlled intensity characteristics aquifer. Projected increases could further times response droughts changing climate, resetting new tipping condition. Therefore, adaptation strategies must recognise less will be available supplement supply droughts.

Language: Английский

Citations

2