Abstract

The most accurate rainfall-runoff predictions are currently based on deep learning. There is a concern among hydrologists that data-driven models based on deep learning may not be reliable in extrapolation or for predicting extreme events. This study tests that hypothesis using Long Short-Term Memory networks (LSTMs) and an LSTM variant that is architecturally constrained to conserve mass. The LSTM (and the mass-conserving LSTM variant) remained relatively accurate in predicting extreme (high return-period) events compared to both a conceptual model (the Sacramento Model) and a process-based model (US National Water Model), even when extreme events were not included in the training period. Adding mass balance constraints to the data-driven model (LSTM) reduced model skill during extreme events.

Highlights

  • Deep learning (DL) provides the most accurate rainfall-runoff simulations available from the hydrological sciences community 10 (Kratzert et al, 2019b, a)

  • The Long Short-Term Memory networks (LSTMs) remained relatively accurate in predict5 ing extreme events compared to both a conceptual model and a process-based model (US National Water Model), even when extreme events were not included in the training period

  • The first test period (1989-1999) is the same period used by previous studies, which allows us to confirm that the DL-based models (LSTM and Mass-Conserving LSTM (MC-LSTM)) trained for this project perform as expected relative to prior work

Read more

Summary

Introduction

Deep learning (DL) provides the most accurate rainfall-runoff simulations available from the hydrological sciences community 10 (Kratzert et al, 2019b, a) This type of finding is not new – Todini (2007) noted more than a decade ago, in his review of the history of hydrological modeling, that “physical process-oriented modellers have no confidence in the capabilities of datadriven models’ outputs with their heavy dependence on training sets, while the more system engineering-oriented modellers claim that data-driven models produce better forecasts than complex physically-based models.”.

Objectives
Methods
Results
Conclusion
Full Text
Published version (Free)

Talk to us

Join us for a 30 min session where you can share your feedback and ask us any queries you have

Schedule a call