Data-driven surrogate modeling with satellite data assimilation : advancing basin-scale hydrology for water balance simulation | |
| Author | Aryal, Insaf |
| Call Number | AIT Thesis no.WM-25-01 |
| Subject(s) | Hydrologic models Hydrology--Mathematical models--Data processing |
| Note | A thesis submitted in partial fulfillment of the requirements for the degree of Master of Science in Water Engineering and Management |
| Publisher | Asian Institute of Technology |
| Abstract | Hydrological modeling is a requirement for understanding the dynamics of water balance, however, physical models are faced with limitations of computational inefficiency, inadequate or overly simplified representation of complex processes, and the inability to efficiently synthesize a variety of data types. While physical models have their pros, data-driven models have the benefit of improved predictive power, scalability, and the potential for real-time analysis. In this study, a Long Short-Term Memory (LSTM) model was developed to simulate the water balance, utilizing open source data from ERA-5 reanalysis, and trained using Noah-MP model output that was conducted over Thailand as a case study. The LSTM model was then transferred to other basins to evaluate its ability to be transferred while considering the different land use, topography, and climate conditions in each basin. The model's seasonal performance was assessed in order understand how sensitive it was to variability in climate. To further refine the accuracy of water balance predictions, methods of assimilating satellite datasets, including GRACE terrestrial water storage, GLEAM evapotranspiration, and SMAP surface soil moisture, were incorporated into the data driven model to improve the representation of hydrological processes. Model performance was evaluated using observations which produced impressive outputs with respect to correlation coefficients (R) and RMSE runoff, evaporation, groundwater, and soil moisture respectively. It was also shown that the LSTM model could maintain high transferability performance across a range of basins that exhibit contrasting physical characteristics. This study highlights the opportunity that exists for data-driven and satellite data assimilation methods to improve water balance modeling and transfer geospatial models for deterministically accurate water balance predictions regardless of shifts in climate and landscape. |
| Year | 2025 |
| Type | Thesis |
| School | School of Engineering and Technology |
| Department | Department of Civil and Infrastucture Engineering (DCIE) |
| Academic Program/FoS | Water Engineering and Management (WM) |
| Chairperson(s) | Natthachet Tangdamrongsub |
| Examination Committee(s) | Shrestha, Sangam;Shanmugam, Mohana Sundaram |
| Scholarship Donor(s) | AIT Scholarship |
| Degree | Thesis (M. Sc.) - Asian Institute of Technology, 2025 |