Comparison of Long Short Term Memory Networks and the Hydrological Model in Runoff Simulation

Abstract

Runoff modeling is one of the key challenges in the field of hydrology. Various approaches exist, ranging from physically based over conceptual to fully data driven models. In this paper, we propose a data driven approach using the state-of-the-art Long-Short-Term-Memory (LSTM) network. The proposed model was applied in the Poyang Lake Basin (PYLB) and its performance was compared with an Artificial Neural Network (ANN) and the Soil & Water Assessment Tool (SWAT). We first tested the impacts of the number of previous time step (window size) in simulation accuracy. Results showed that a window in improper large size will dramatically deteriorate the model performance. In terms of PYLB, a window size of 15 days might be appropriate for both accuracy and computational efficiency. We then trained the model with 2 different input datasets, namely, dataset with precipitation only and dataset with all available meteorological variables. Results demonstrate that although LSTM with precipitation data as the only input can achieve desirable results (where the NSE ranged from 0.60 to 0.92 for the test period), the performance can be improved simply by feeding the model with more meteorological variables (where NSE ranged from 0.74 to 0.94 for the test period). Moreover, the comparison results with the ANN and the SWAT showed that the ANN can get comparable performance with the SWAT in most cases whereas the performance of LSTM is much better. The results of this study underline the potential of the LSTM for runoff modeling especially for areas where detailed topographical data are not available.

Emulation of a Process-Based Salinity Generator for the Sacramento–San Joaquin Delta of California via Deep Learning

“…This gate configuration helps the network preserve essential information over a long time and avoid rapid error signal decay. LSTM networks have only been applied recently in the field of water resources in terms of modeling rainfall-runoff process [46][47][48][49], groundwater table [50,51], water level in channels [52], water quality [53,54], and reservoir operations [55]. Given the long-term dependencies between salinity and flow/stage, LSTM should also be suitable for salinity simulation given flow and stage inputs.…”

Automated Cloud Based Long Short-Term Memory Neural Network Based SWE Prediction

“…When dealing with long-term dependencies, traditional feed forward Artificial neural networks (ANNs) are limited (Bengio et al, 1994;Fang et al, 2017). However, the Long Short-Term Memory (LSTM) network method is well-suited for long term dependencies (Hochreiter and Schmidhuber, 1997;Fan et al, 2020). LSTM is a deep neural network (DNN) method which has been successfully applied in various fields (Sahoo et al, 2019) for especially for time sequence prediction problems.…”

“…Once this is done the model can be used to predict future values. Figure 4 shows a schematic of the one-step iteration in the LSTM training/calibration procedure (Fan et al, 2020).…”

“…These sequences are split into samples which are reshaped for the LSTM model. The size of these samples is called the window size (Fan et al, 2020) and has impact on the forecast accuracy.…”

“…recurrent neural networks (RNNs) (Kumar et al, 2004) for hydrology and SWE forecasting (Huang and Cressie, 1996;Winstral et al, 2019;Magnusson et al, 2020). More recently deep learning methods such as the long short-term memory network (LSTM) have demonstrated a significant promise in hydrological time series analysis and forecasting (Xiang et al, 2020), such as soil moisture modeling (Fang et al, 2017), monthly water-table depth predictions (Zhang et al, 2018), and daily or hourly rainfall-runoff modeling (Hu et al, 2018;Kratzert et al, 2018;Le et al, 2019;Fan et al, 2020).…”