Application of the machine learning methods for GRACE data based groundwater modeling: A systematic review

Vahid Nourani, Nardin Jabbarian Paknezhad, Anne Ng, Zhang Wen, Dominika Dabrowska, Selin Üzelaltınbulat

Research output: Contribution to journalReview articlepeer-review


The Gravity Recovery and Climate Experiment (GRACE) data has brought significant advancements in groundwater (GW) analysis by providing crucial information about changes in the gravity field of Earth and water storage. GRACE data are instrumental in understanding GW dynamics, monitoring aquifer depletion, and assessing water resource management strategies for sustainable utilization. Nevertheless, comprehensive reviews focusing specifically on studies related to GRACE data are lacking. In this paper, 90 original papers were considered from 2002 to 2023, which utilized machine learning (ML) methods for Downscaling GRACE (DG) and also Modeling and Forecasting GW via GRACE data (MFGG). Papers were obtained from Scopus and Web of Science databases. A total of 78% and 22% of the analyzed papers focused on DG and MFGG, respectively. The investigation of the papers revealed that the majority employed the random forest (RF) method. Subsequently, gradient boosting (GB), deep learning, and artificial neural network (ANN) methods were the most commonly applied ML techniques, respectively. The selection of input parameters has a significant impact on the modeling performance. Of the papers examined, approximately 62.5% incorporated precipitation as an input, whereas evapotranspiration and temperature were utilized in 40% and 43.75% of the cases, respectively. The results of the analysis demonstrated the capability of ML models for DG and MFGG. After investigating different studies on the application of GRACE data in DG and MFGG, it was concluded that only a few studies considered the adaptation of GRACE data trend and manner with the observed and in situ values of the wells; however, it is important to consider their compatibility in order to achieve accurate results. Moreover, it is suggested to leverage development in ML, as well as progress in computation and ensembling, to be integrated with physics-based models and harness new information for the GRACE data and GW assessment.

Original languageEnglish
Article number101113
Pages (from-to)1-13
Number of pages13
JournalGroundwater for Sustainable Development
Publication statusPublished - May 2024

Cite this