Predicting Changes in Spatiotemporal Groundwater Storage Through the Integration of Multi-Satellite Data and Deep Learning Models
Continuous monitoring and accurate spatiotemporal groundwater storage change predictions can help support sustainable development and efficient groundwater resources management. In this study, remote sensing-based data were used to develop two deep learning predictive models, long short term memory...
Guardado en:
| Autores principales: | , |
|---|---|
| Formato: | article |
| Lenguaje: | EN |
| Publicado: |
IEEE
2021
|
| Materias: | |
| Acceso en línea: | https://doaj.org/article/f28ca1d49bec49e38867bbdf805fa5a2 |
| Etiquetas: |
Agregar Etiqueta
Sin Etiquetas, Sea el primero en etiquetar este registro!
|
| Sumario: | Continuous monitoring and accurate spatiotemporal groundwater storage change predictions can help support sustainable development and efficient groundwater resources management. In this study, remote sensing-based data were used to develop two deep learning predictive models, long short term memory (LSTM), and convolutional neural network-LSTM (CNN-LSTM) models. The input variables were terrestrial water storage anomaly (TWSA) recorded by the Gravity Recovery and Climate Experiment (GRACE) satellites and the GRACE-Follow-On (GRACE-FO) mission, precipitation measured by the Tropical Rainfall Measuring Mission (TRMM) satellite, average temperature and soil moisture obtained from the Global Land Data Assimilation System (GLDAS), and normalized difference vegetation index (NDVI) and modified normalized difference water index (MNDWI) acquired from the Landsat 5 and 8 satellites. A comparison of groundwater storage change predictions by the two deep learning predictive models with in situ measurements of the National Groundwater Monitoring Network (NGMN) in South Korea showed that the CNN-LSTM model was more accurate [root mean square error (RMSE) = 44.92 mm/month, correlation coefficient (<inline-formula> <tex-math notation="LaTeX">$r) =0.70$ </tex-math></inline-formula>, index of agreement (IOA) = 0.81] than the LSTM model [RMSE = 47.44 mm/month, <inline-formula> <tex-math notation="LaTeX">$r =0.62$ </tex-math></inline-formula>, and IOA = 0.77]. Furthermore, a comparison of the performance of the two models for three specific regions showed that the CNN-LSTM model captured spatiotemporal variations in groundwater storage better. Additionally, a comparison of cumulative change trends in groundwater storage with the NDVI, MNDWI, and TWSA data showed that changes in land cover and total water storage affect groundwater storage. These comparison results demonstrate that the use of parameters recorded by multiple satellites as training data for deep learning models can play an important role in the model’s performance. |
|---|