An Integrated Data-driven Model to Simulate Hydrological Interventions
Abstract
Data-driven models have emerged as powerful alternatives to the conventional hydrological models as they are capable of unveiling the underpinnings of multi-faceted physical-based hydrological events in a robust and timely efficient manner while bypassing the computationally intense calibration procedures. In this work, we attempt to devise a fully automated and integrated predictive data-driven model - coupled with the conventional hydrological models- to simulate the water-land dynamics via harnessing the learning architectures offered by machine (e.g. random forests) and deep learning (e.g. residual neural networks and long short-term memory networks) literature. Once well-trained, the data-driven model would then catalyze the analyses as the hydrological simulations could be - at least partially - performed by these agile predictive models. Furthermore, we deploy active machine learning module to extract optimal data points from the pool of available training set, and therefore, accelerate the learning phase of data-driven models. In specific, we explore the behavior of runoff and streamflow, two of the major hydrological components, and assess the performance of our predictive model against the conventional models i.e. the Variable Infiltration Capacity (VIC) and Catchment-based Macro-scale Floodplain (CaMaFlood). To this end, we propose an integrated model composed of three engines - operating in tandem - i.e. VIC, CaMaFlood and a data-driven learning module with an interactive graphical user interface (GUI) to emulate variant hydrological interventions.
- Publication:
-
AGU Fall Meeting Abstracts
- Pub Date:
- December 2019
- Bibcode:
- 2019AGUFM.H43J2151G
- Keywords:
-
- 0430 Computational methods and data processing;
- BIOGEOSCIENCES;
- 1805 Computational hydrology;
- HYDROLOGY;
- 1846 Model calibration;
- HYDROLOGY;
- 1873 Uncertainty assessment;
- HYDROLOGY