Technical note: Data assimilation and autoregression for using near-real-time streamflow observations in long short-term memory networks

Ingesting near-real-time observation data is a critical component of many operational hydrological forecasting systems. In this paper, we compare two strategies for ingesting near-real-time streamflow observations into long short-term memory (LSTM) rainfall-runoff models: autoregression (a forward method) and variational data assimilation. Autoregression is both more accurate and more computationally efficient than data assimilation Autoregression is sensitive to missing data, however an appropriate (and simple) training strategy mitigates this problem. We introduce a data assimilation procedure for recurrent deep learning models that uses backpropagation to make the state updates.

Automated summary

Ingesting real-time observation data is crucial for many hydrological forecasting systems. This paper compares two strategies, autoregression and variational data assimilation, for incorporating real-time streamflow observations into LSTM rainfall-runoff models. The results show that autoregression is both more accurate and computationally efficient, but it is sensitive to missing data. However, this can be mitigated by using an appropriate training strategy in data assimilation.