Application study of IFAS and LSTM models on runoff simulation and flood prediction in the Tokachi River basin

Abstract

Floods are often caused by short-term heavy rainfall. An Integrated Flood Analysis System (IFAS) model is good at runoff simulation and a Long Short-Term Memory (LSTM) model is good at learning massive data and realizing rainfall forecast. In this paper, the applicability of the IFAS model to runoff simulation in the Tokachi River basin and the LSTM model to forecast hourly rainfall was studied, and the accuracy of flood prediction was also studied by inputting the optimal rainfall data forecasted by the LSTM model into the IFAS model. The research results show that the IFAS model can accurately simulate the runoff process in the Tokachi River basin. In the calibration period and the verification period, the Nash–Sutcliffe efficiency coefficient (NSE) of all simulation results are above 0.75; the LSTM model can achieve forecast hourly rainfall with high precision, the NSE of best forecast results is 0.86; the IFAS model can achieve flood prediction with high precision by using the optimal rainfall data forecasted by the LSTM model, the NSE of simulation result is 0.81. The above conclusions show that it is of great significance to combine the hourly rainfall forecasted by the LSTM model with the IFAS model for flood prediction.

Penerapan Metode Principal Component Analysis (PCA) dan Long Short-Term Memory (LSTM) dalam Memprediksi Prediksi Curah Hujan Harian

“…Penelitian tersebut memperoleh model IFAS yang akurat dengan menggunakan simulasi limpasan di DAS Tokachi menghasilkan NSE 0,75 dan model LSTM menghasilkan NSE 0,86 untuk prakiraan curah hujan per jam. Model IFAS menghasilkan nilai NSE 0,81 dengan menggunakan data curah hujan optimal yang diramalkan oleh model LSTM [16]. Chengcheng Chen, dkk.…”

Multiple Machine Learning Methods for Runoff Prediction: Contrast and Improvement

“…(Adnan et al, 2021). With the continuous improvement of human technology and understanding of nature, distributed hydrological models have evolved over the decades to fully re ect the impact of spatio-temporal heterogeneity such as watershed topography, soil, land use/cover on the water cycle (Busico et al, 2020), making it a vital role in runoff prediction (Tang et al, 2021;Chen et al, 2021;Zhou et al, 2018). However, practical applications, distributed hydrological modelstend to experience various unsolvable problems that will affect its further development and greater contribution, including low simulation in small steps, the uncertainty in the simulation accuracy caused by data, parameters and other inputs as well as by the model's own structure, and the high requirements for various data in the target river watershed during the modeling.…”

“…The LSTM models lters information through gate structures to maintain and update the state of memory cells (Qiu et al, 2020). And it has been successfully applied in the areas of nancial market forecasting (Yan and Ouyang, 2018), oil production forecasting (Sagheer and Kotb, 2019), air pollution forecasting (Xayasouk et al, 2020), electricity load forecasting (Son et al, 2020), and rainfall forecasting (Chen et al, 2021). DeepAR proposed by Salinas et al (Huang et al, 2021).…”

“…(5) shows the residual block of the encoder, where the input is rst passed through a dilated causal convolutional layer and a batch normalization layer, repeated twice (passing through the activation function ReLU layer in the process). Then the residual operation is performed together with the original input .…”