Using Artificial Intelligence Techniques for Reconstructing GRACE Data Over Nile River

Abstract

Terrestrial water storage (TWS) is crucial for the worldwide hydrologic water cycle and sustainability of water. Gravimetric missions such as the Gravity Recovery and Climate Experiments (GRACE) & GRACE-Follow on (GRACE-FO) are essential for evaluating changes in TWS.This study introduces a new approach by combining remote sensing data in the form of mascon data with deep learning models (DL) such as the Long Short-Term Memory (LSTM) Model to reconstruct GRACE data in the Nile River basin (NRB) from 2002 to 2022 to study the changes in water storage with high accuracy. This research strategy depends on applying several convolutional neural network (CNN) models, including AlexNet, VGG Net, and GoogleNet, for extracting features from time series GRACE data. After that, use the optimization algorithm (DTOFGW) to get the best hyperparameters for LSTM. Finally, comparing many different optimization algorithms showed the proposed model’s superiority. Applying statistical analysis tests illustrated the significance of our proposed model, such as ANOVA and T-test. The results of the trial showed that the proposed model (DTOFGW-LSTM) outperformed the other models by an accuracy of 96.8%, sensitivity of 0.5, specificity of 99.5%, P value of 0.85, N value of 0.97, F-score of 0.63, and confidence value of 97%.

Keywords

• grace
• grace-fo
• artificial intelligence
• lstm-dtofgw
• anova
• t-test