A Rainfall‐Runoff Model With LSTM‐Based Sequence‐to‐Sequence Learning

Abstract

Rainfall-runoff modeling is a complex nonlinear time series problem. While there is still room for improvement, researchers have been developing physical and machine learning models for decades to predict runoff using rainfall data sets. With the advancement of computational hardware resources and algorithms, deep learning methods such as the long short-term memory (LSTM) model and sequence-to-sequence (seq2seq) modeling have shown a good deal of promise in dealing with time series problems by considering long-term dependencies and multiple outputs. This study presents an application of a prediction model based on LSTM and the seq2seq structure to estimate hourly rainfall-runoff. Focusing on two Midwestern watersheds, namely, Clear Creek and Upper Wapsipinicon River in Iowa, these models were used to predict hourly runoff for a 24-hr period using rainfall observation, rainfall forecast, runoff observation, and empirical monthly evapotranspiration data from all stations in these two watersheds. The models were evaluated using the Nash-Sutcliffe efficiency coefficient, the correlation coefficient, statistical bias, and the normalized root-mean-square error. The results show that the LSTM-seq2seq model outperforms linear regression, Lasso regression, Ridge regression, support vector regression, Gaussian processes regression, and LSTM in all stations from these two watersheds. The LSTM-seq2seq model shows sufficient predictive power and could be used to improve forecast accuracy in short-term flood forecast applications. In addition, the seq2seq method was demonstrated to be an effective method for time series predictions in hydrology. Key Points:• An hourly runoff model was developed using the LSTM sequence-to-sequence learning method for 24-hr predictions on USGS stations • The proposed model shows better performance than traditional data-driven models and is applicable to different watersheds • The advantages and limitations of seq2seq models and how this model structure could work on the rainfall-runoff modeling is presented Supporting Information:• Supporting Information S1 Correspondence to: Z. Xiang, zhongrun-xiang@uiowa.edu Citation: Xiang, Z., Yan, J., & Demir, I. (2020). A rainfall-runoff model with LSTM-based sequence-to-sequence learning. Water Resources Research, 56, e2019WR025326. https://doi.

Distributed long-term hourly streamflow predictions using deep learning – A case study for State of Iowa

“…Experimental results of integrating stage level sensors are shown in section 3.4. Xiang et al, (2020) proposed the first multi-timestep rainfall-runoff prediction model LSTM-seq2seq using the encoder-decoder LSTM method. LSTM-seq2seq shows successful applications on several USGS streamflow gages and outperforms popular machine learning models such as SVMs and LSTM in the rainfall-runoff predictions for up to 24 hours.…”

Distributed long-term hourly streamflow predictions using deep learning – A case study for State of Iowa

“…This study aims to provide 5-days hourly prediction, which is 120 timesteps for the target forecast input and output. The source history input used in this project contains 72 hours, which is determined in previous work that can provide sufficient information for runoff predictions (Xiang et al, 2020). Thus, each GRU layer contains 192 timesteps in the proposed architecture.…”

“…Mini-batch method is used in the training process in this study. The batch size of 64 has been discovered to be effective in rainfallrunoff modeling in a previous study (Xiang et al, 2020). This model used the residual sum of squares divides the total sum of squares, which is also named the variance unexplained (FVU, Equation 5) as the loss function.…”

“…This rainfall data was used in both the source for historical and forecast input in this paper. Simple moving average with 6 hours (Xiang et al, 2020) was used to reduce the noise for a smoother rainfall data.…”

“…An LSTM model that proposed by Hu et al, (2018) can forecast the hourly runoff from 1-to 6-hour lead-time with 6 different models. Xiang et al, (2020) applied the sequence learning in rainfall-runoff modeling with an Encoder-Decoder structure and the model outperforms multiple benchmarks including linear regressions, SVMs, and LSTM for predictions of the following 24 hours on several USGS streamflow gages on two watersheds. Although 24 hours of continuous prediction is an improvement, bottleneck effect in the model may reduce the efficiency for applications longer than 24 hours (Xiang et al, 2020).…”