Space Geodesy Project Report

Abstract

This research aimed to address the challenge of improving the reliability of the Water- GAP Global Hydrology Model (WGHM) using the Gravity Recovery and Climate Experiment (GRACE) satellite data as a benchmark. Our problem was defined by the need to align WGHM Total Water Storage Anomaly (TWSA) data more closely with the reliable GRACE satellite measurements to enhance the simulation quality. We developed a method employing a Gaussian Mixture Model (GMM) and a Convolutional Neural Network (CNN) to identify outliers and generate new features from the WGHM data. Furthermore, time series decomposition was applied for preprocessing, feeding both the latent variables of the CNN and coefficients of the decomposition into the GMM. Our results demonstrated an improvement in alignment with the GRACE data, quantified by the Root Mean Square Error (RMSE) and visualized through various data plots. A relative improvement was observed across numerous water basins, and the time series plots provided dynamic validation of the effects of our corrections. Despite the improvements, the study highlighted inherent limitations related to outlier detec- tion, feature representation, and model component determination. It suggests the exploration of ensemble methods, Bayesian modeling, and artificial data generation for future improvements. These enhancements in the reliability of WGHM data could expand its applicability across a broader range of environmental and climate studies.