Augmenting Machine Learning with Conceptual Model to Improve Daily Streamflow Prediction
Abstract
In recent years, the applications of machine learning (ML) techniques for streamflow prediction is increasing. However, realizing the limited interpretability skills in the applications of ML techniques calls for synergizing it with available domain knowledge through physics-based models. The use of the ML technique in modeling the less known hydrological process in the conceptual model can improve its prediction skill. Here we develop a Physics Informed Machine Learning (PIML) model using the ML technique to predict hydrological variables for which the performance of the conceptual model is poor. The architecture of the conceptual model ensures the incorporation of domain specifics in the PIML model. We apply the proposed model for daily streamflow prediction in the Burhanpur subcatchment of the Tapi river basin, India. Our initial results show that the PIML model improves the Nash-Sutcliffe Efficiency (NSE) from 0.58 to 0.60 compared to the conceptual model (SIMHYD model). The hybrid approach of leveraging the strength of the ML techniques and physics-based model can be used for improvement in flood predictions.
- Publication:
-
AGU Fall Meeting Abstracts
- Pub Date:
- December 2021
- Bibcode:
- 2021AGUFM.H35S1252B