# Daily Streamflow Forecasts Based on Cascade Long Short-Term Memory (LSTM) Model over the Yangtze River Basin

^{1}

^{2}

^{*}

## Abstract

**:**

## 1. Introduction

## 2. Materials and Methods

#### 2.1. Study Area and Data

^{6}km

^{2}, which accounts for about 18.8% of China’s total land area. Figure 1 shows the locations of the 49 hydrological stations used in this study. The monthly streamflow observed at these 49 stations was collected from the Yangtze River Basin Hydrological Yearbook published by the Yangtze River Conservancy Commission.

#### 2.2. The Conjunctive Surface-Subsurface Process Version 2 (CSSPv2) Model

#### 2.3. The Cascade LSTM Model

#### 2.4. Experimental Design

_{f}). Then, we use historical streamflow, soil moisture, evapotranspiration, and historical precipitation and LSTM_P to predict streamflow (LSTM_S), i.e., ([STRF, SM, ET, and P

_{f}] → STRF), to explore the capability of cascade LSTM model (Figure 2) in streamflow forecast. Note that the above experiments are conducted in the forecasting periods of 1–15 days. This experiment is called “cascade LSTM”.

#### 2.5. Evaluation of Model Performance

## 3. Results

#### 3.1. Evaluation of CSSPv2 Land Model Simulation and Default LSTM Forecast

#### 3.2. Evaluation of Cascade LSTM

#### 3.2.1. Precipitation Forecast Based on LSTM_P

#### 3.2.2. Streamflow Forecast Based on LSTM_S

#### 3.3. Evaluation of Cascade LSTM with Prefect Precipitation

## 4. Discussion

## 5. Conclusions

## Author Contributions

## Funding

## Data Availability Statement

## Conflicts of Interest

## References

- Arial, P.A.; Bellouin, N.; Coppola, E.; Jones, R.G.; Krinner, G.; Marotzke, J.; Naik, V.; Palmer, M.D.; Plattner, G.-K.; Rogelj, J.; et al. Technical Summary. Climate Change 2021: The Physical Science Basis. Contribution of Working Group I to the Sixth Assessment Report of the Intergovernmental Panel on Climate Change. 2021. Available online: https://www.ipcc.ch/report/ar6/wg1/chapter/technical-summary (accessed on 29 August 2021).
- Behnke, R.; Vavrus, S.; Allstadt, A.; Albright, T.; Thogmartin, W.E.; Radeloff, V.C. Evaluation of downscaled, gridded climate data for the conterminous United States. Ecol. Appl.
**2016**, 26, 1338–1351. [Google Scholar] [CrossRef] [PubMed] - Timmermans, B.; Wehner, M.; Cooley, D.; O’Brien, T.; Krishnan, H. An evaluation of the consistency of extremes in gridded precipitation data sets. Clim. Dyn.
**2019**, 52, 6651–6670. [Google Scholar] [CrossRef][Green Version] - Alfieri, L.; Burek, P.; Dutra, E.; Krzeminski, B.; Muraro, D.; Thielen, J.; Pappenberger, F. GloFAS—Global ensemble streamflow forecasting and flood early warning. Hydrol. Earth Syst. Sci.
**2013**, 17, 1161–1175. [Google Scholar] [CrossRef][Green Version] - Coulibaly, P.; Anctil, F.; Rasmussen, P.; Bobée, B. A recurrent neural networks approach using indices of low-frequency climatic variability to forecast regional annual runoff. Hydrol. Process.
**2000**, 14, 2755–2777. [Google Scholar] [CrossRef] - Berghuijs, W.; Larsen, J.R.; Van Emmerik, T.H.M.; Woods, R.A. A Global Assessment of Runoff Sensitivity to Changes in Precipitation, Potential Evaporation, and Other Factors. Water Resour. Res.
**2017**, 53, 8475–8486. [Google Scholar] [CrossRef][Green Version] - Yuan, X. An experimental seasonal hydrological forecasting system over the Yellow River basin—Part 2: The added value from climate forecast models. Hydrol. Earth Syst. Sci.
**2016**, 20, 2453–2466. [Google Scholar] [CrossRef][Green Version] - Yuan, X.; Ji, P.; Wang, L.; Liang, X.; Yang, K.; Ye, A.; Su, Z.; Wen, J. High-Resolution Land Surface Modeling of Hydrological Changes Over the Sanjiangyuan Region in the Eastern Tibetan Plateau: 1. Model Development and Evaluation. J. Adv. Model. Earth Syst.
**2018**, 10, 2806–2828. [Google Scholar] [CrossRef] - Frame, J.; Kratzert, F.; Raney, A.; Rahman, M.; Salas, F.R.; Nearing, G.S. Post-Processing the National Water Model with Long Short-Term Memory Networks for Streamflow Predictions and Model Diagnostics. J. Am. Water Resour. Assoc.
**2021**, 57, 885–905. [Google Scholar] [CrossRef] - Gauch, M.; Mai, J.; Lin, J. The proper care and feeding of CAMELS: How limited training data affects streamflow prediction. Environ. Model. Softw.
**2020**, 135, 104926. [Google Scholar] [CrossRef] - Nearing, G.; Pelissier, C.; Kratzert, F.; Klotz, D.; Gupta, H.; Frame, J.; Sampson, A. Physically Informed Machine Learning for Hydrological Modeling Under Climate Nonstationarity. Science and Technology Infusion Climate Bulletin. NOAA’s National Weather Service. In Proceedings of the 44th NOAA Annual Climate Diagnostics and Prediction Workshop, Durham, NC, USA, 22–24 October 2019; Available online: https://www.nws.noaa.gov/ost/climate/STIP/44CDPW/44cdpw-GNearing.pdf (accessed on 26 August 2020). [CrossRef]
- Hoedt, P.; Kratzert, F.; Klotz, D.; Halmich, C.; Holzleitner, M.; Nearing, G.; Hochreiter, S.; Klambauer, G. MC-LSTM: Mass-Conserving LSTM. arXiv
**2021**, arXiv:2101.05186. [Google Scholar] [CrossRef] - Liu, J.; Yuan, X.; Zeng, J.; Jiao, Y.; Li, Y.; Zhong, L.; Yao, L. Ensemble streamflow forecasting over a cascade reservoir catchment with integrated hydrometeorological modeling and machine learning. Hydrol. Earth Syst. Sci.
**2022**, 26, 265–278. [Google Scholar] [CrossRef] - Kratzert, F.; Klotz, D.; Brenner, C.; Schulz, K.; Herrnegger, M. Rainfall–runoff modelling using Long Short-Term Memory (LSTM) networks. Hydrol. Earth Syst. Sci.
**2018**, 22, 6005–6022. [Google Scholar] [CrossRef][Green Version] - Kratzert, F.; Klotz, D.; Herrnegger, M.; Sampson, A.K.; Hochreiter, S.; Nearing, G.S. Toward Improved Predictions in Ungauged Basins: Exploiting the Power of Machine Learning. Water Resour. Res.
**2019**, 55, 11344–11354. [Google Scholar] [CrossRef][Green Version] - Rahimzad, M.; Moghaddam, N.; Alireza, Z.; Hosam, S.; Jaber, D.; Mehr, A.; Kwon, H. Performance Comparison of an LSTM-based Deep Learning Model versus Conventional Machine Learning Algorithms for Streamflow Forecasting. Water Resour. Manag.
**2021**, 35, 4167–4187. [Google Scholar] [CrossRef] - Cheng, M.; Fang, F.; Kinouchi, T.; Navon, I.; Pain, C. Long lead-time daily and monthly streamflow forecasting using machine learning methods. J. Hydrol.
**2020**, 590, 125376. [Google Scholar] [CrossRef] - Zhu, S.; Luo, X.; Yuan, X.; Xu, Z. An improved long short-term memory network for streamflow forecasting in the upper Yangtze River. Stoch. Environ. Res. Risk Assess.
**2020**, 34, 1313–1329. [Google Scholar] [CrossRef] - Cho, K.; Kim, Y. Improving streamflow prediction in the WRF-Hydro model with LSTM networks. J. Hydrol.
**2022**, 605, 127297. [Google Scholar] [CrossRef] - Hu, X.; Chu, L.; Pei, J.; Liu, W.; Bian, J. Model complexity of deep learning: A survey. Knowl. Inf. Syst.
**2021**, 63, 2585–2619. [Google Scholar] [CrossRef] - Feng, D.; Fang, K.; Shen, C. Enhancing Streamflow Forecast and Extracting Insights Using Long-Short Term Memory Networks With Data Integration at Continental Scales. Water Resour. Res.
**2020**, 56, e2019WR026793. [Google Scholar] [CrossRef] - Jain, S.; Mani, S.; Jain, S.K.; Prakash, P.; Singh, V.P.; Tullos, D.; Kumar, S.; Agarwal, S.P.; Dimri, A.P. A brief review of flood forecasting techniques and their applications. Int. J. River Basin Manag.
**2018**, 16, 329–344. [Google Scholar] [CrossRef] - Granata, F.; Di Nunno, F.; de Marinis, G. Stacked machine learning algorithms and bidirectional long short-term memory networks for multi-step ahead streamflow forecasting: A comparative study. J. Hydrol.
**2022**, 613, 128431. [Google Scholar] [CrossRef] - Bai, Y.; Bezak, N.; Sapač, K.; Klun, M.; Zhang, J. Short-Term Streamflow Forecasting Using the Feature-Enhanced Regression Model. Water Resour. Manag.
**2019**, 33, 4783–4797. [Google Scholar] [CrossRef] - Ji, P.; Yuan, X.; Shi, C.; Jiang, L.; Wang, G.; Yang, K. A Long-Term Simulation of Land Surface Conditions at High Resolution over Continental China. J. Hydrometeorol.
**2023**, 24, 285–314. [Google Scholar] [CrossRef] - Shangguan, W.; Dai, Y.; Duan, Q.; Liu, B.; Yuan, H. A global soil data set for earth system modeling. J. Adv. Model. Earth Syst.
**2014**, 6, 249–263. [Google Scholar] [CrossRef] - Liang, S.; Cheng, J.; Jia, K.; Jiang, B.; Liu, Q.; Xiao, Z.; Yao, Y.; Yuan, W.; Zhang, X.; Zhao, X.; et al. The Global Land Surface Satellite (GLASS) Product Suite. Bull. Am. Meteorol. Soc.
**2021**, 102, E323–E337. [Google Scholar] [CrossRef] - Yuan, X.; Liang, X.-Z. Evaluation of a Conjunctive Surface–Subsurface Process Model (CSSP) over the Contiguous United States at Regional–Local Scales. J. Hydrometeorol.
**2011**, 12, 579–599. [Google Scholar] [CrossRef] - Hersbach, H.; Bell, B.; Berrisford, P.; Hirahara, S.; Horanyi, A.; Muñoz-Sabater, J.; Nicolas, J.; Peubey, C.; Radu, R.; Schepers, D.; et al. The ERA5 global reanalysis. Q. J. R. Meteorol. Soc.
**2020**, 146, 1999–2049. [Google Scholar] [CrossRef] - Dai, Y.; Dickinson, R.; Wang, Y. A two-big-leaf model for canopy temperature, photosynthesis, and stomatal conductance. J. Clim.
**2004**, 17, 2281–2299. [Google Scholar] [CrossRef] - Choi, H.; Kumar, P.; Liang, X.-Z. Three-dimensional volume-averaged soil moisture transport model with a scalable parameterization of subgrid topographic variability. Water Resour. Res.
**2007**, 43, W04414. [Google Scholar] [CrossRef][Green Version] - Choi, H.; Liang, X.-Z.; Kumar, P. A Conjunctive Surface–Subsurface Flow Representation for Mesoscale Land Surface Models. J. Hydrometeorol.
**2013**, 14, 1421–1442. [Google Scholar] [CrossRef] - Liang, X.; Lettenmaier, D.P.; Wood, E.F.; Burges, S.J. A simple hydrologically based model of land surface water and energy fluxes for general circulation models. J. Geophys. Res. Atmos.
**1994**, 99, 14415–14428. [Google Scholar] [CrossRef] - Liang, X.Z.; Xu, M.; Yuan, X.; Ling, T.; Choi, H.I.; Zhang, F.; Chen, L.; Liu, S.; Su, S.; Qiao, F.; et al. Regional Climate–Weather Research and Forecasting Model. Bull. Am. Meteorol. Soc.
**2012**, 93, 1363–1387. [Google Scholar] [CrossRef] - Ji, P.; Yuan, X.; Liang, X.-Z. Do Lateral Flows Matter for the Hyperresolution Land Surface Modeling? J. Geophys. Res. Atmos.
**2017**, 122, 12077–12092. [Google Scholar] [CrossRef][Green Version] - Zheng, D.; Van Der Velde, R.; Su, Z.; Wen, J.; Wang, X. Assessment of Noah land surface model with various runoff parameterizations over a Tibetan river. J. Geophys. Res. Atmos.
**2017**, 122, 1488–1504. [Google Scholar] [CrossRef] - Ji, P.; Yuan, X.; Jiao, Y.; Wang, C.; Han, S.; Shi, C. Anthropogenic Contributions to the 2018 Extreme Flooding over the Upper Yellow River Basin in China. Bull. Am. Meteorol. Soc.
**2020**, 101, S89–S94. [Google Scholar] [CrossRef][Green Version] - Zeng, J.; Yuan, X.; Ji, P.; Shi, C. Effects of meteorological forcings and land surface model on soil moisture simulation over China. J. Hydrol.
**2021**, 603, 126978. [Google Scholar] [CrossRef] - Gupta, H.; Kling, H.; Yilmaz, K.K.; Martinez, G.F. Decomposition of the mean squared error and NSE performance criteria: Implications for improving hydrological modelling. J. Hydrol.
**2009**, 377, 80–91. [Google Scholar] [CrossRef][Green Version] - Kling, H.; Fuchs, M.; Paulin, M. Runoff conditions in the upper Danube basin under an ensemble of climate change scenarios. J. Hydrol.
**2012**, 424–425, 264–277. [Google Scholar] [CrossRef] - Chen, Y.; Yuan, H. Evaluation of nine sub-daily soil moisture model products over China using high-resolution in situ observations. J. Hydrol.
**2020**, 588, 125054. [Google Scholar] [CrossRef] - Zhang, C.; Brodeur, Z.P.; Steinschneider, S.; Herman, J.D. Leveraging Spatial Patterns in Precipitation Forecasts Using Deep Learning to Support Regional Water Management. Water Resour. Res.
**2022**, 58, e2021WR031910. [Google Scholar] [CrossRef] - Sha, Y.; Ii, D.J.G.; West, G.; Stull, R. A hybrid analog-ensemble, convolutional-neural-network method for post-processing precipitation forecasts. Mon. Weather. Rev.
**2022**, 1, 1495–1515. [Google Scholar] [CrossRef] - Chen, G.; Wang, W. Short-Term Precipitation Prediction for Contiguous United States Using Deep Learning. Geophys. Res. Lett.
**2022**, 49, e2022GL097904. [Google Scholar] [CrossRef] - Deng, H.; Chen, W.; Huang, G. Deep insight into daily runoff forecasting based on a CNN-LSTM model. Nat. Hazards
**2022**, 133, 1675–1696. [Google Scholar] [CrossRef]

**Figure 1.**Locations and drainage areas (km

^{2}) of 49 streamflow observational stations over the Yangtze River basin in southern China.

**Figure 3.**(

**a**) Spatial distribution of KGE between observed and CSSPv2 model-simulated monthly streamflow over the Yangtze River basin. (

**b**) The boxplot of KGE for all 49 stations. The boxes represent the 25th and 75th percentiles of KGE, the line and the dot within the box are median and mean values of KGE, respectively, and the whiskers represent 10th and 90th percentiles of KGE.

**Figure 4.**Spatial distributions of KGEs for streamflow forecasts based on default LSTM model at lead times of 1–15 days.

**Figure 5.**The relationship between the watershed area (ln km

^{2}) and the default LSTM streamflow forecast skill (KGE) at the lead times of 1–15 days.

**Figure 6.**The relationship between the watershed area (ln km

^{2}) and the LSTM_P precipitation forecast skill (KGE) at the lead times of 1–15 days.

**Figure 7.**Spatial distributions of the KGE difference (∆ KGE) between the daily streamflow forecasts by the cascade LSTM model and those by the default LSTM model.

**Figure 8.**Performance of streamflow forecasts at different lead times. The left column shows the mean values of KGE, R, β, and γ for 49 stations, and the right column shows their median values.

**Figure 9.**The relationship between the difference of streamflow forecast performance (∆ KGE of the cascade LSTM model and the default LSTM model) and the watershed area (ln km

^{2}) at the lead times of 1–15 days.

**Figure 10.**Spatial distributions of the KGE difference (∆ KGE) between the daily streamflow forecasts by the cascade LSTM with prefect precipitation and those by default LSTM.

**Figure 11.**The relationship between the difference of streamflow forecast performance (∆ KGE of the cascade LSTM model with prefect precipitation and the default LSTM model ) and the watershed area (ln km

^{2}) at the lead times of 1–15 days.

Hyper-Parameter | Set Up |
---|---|

Batch | 16, 32, 64, 128, 256, 512, 1024 |

Hidden cell | 8, 16, 32, 64, 128, 256 |

Dropout rate | 0.01, 0.05, 0.1, 0.15, 0.2, 0.3 |

Learning rate | 0.001, 0.005, 0.01, 0.05, 0.1, 0.2 |

**Table 2.**Percentage (%) of stations that improved compared to the default LSTM model, the cascaded LSTM model (CLSTM), and the cascaded LSTM with the perfect precipitation (CLSTM_P) model KGE and its three components (R, β and γ).

Lead Times | KGE | R | β | γ | ||||
---|---|---|---|---|---|---|---|---|

CLSTM | CLSTM_P | CLSTM | CLSTM_P | CLSTM | CLSTM_P | CLSTM | CLSTM_P | |

1 | 0.61 | 0.70 | 0.53 | 0.86 | 0.57 | 0.65 | 0.59 | 0.65 |

2 | 0.76 | 0.86 | 0.20 | 0.82 | 0.82 | 0.86 | 0.80 | 0.78 |

3 | 0.76 | 0.88 | 0.45 | 0.88 | 0.84 | 0.92 | 0.76 | 0.84 |

4 | 0.86 | 0.88 | 0.53 | 0.90 | 0.80 | 0.88 | 0.69 | 0.80 |

5 | 0.84 | 0.88 | 0.63 | 0.92 | 0.73 | 0.78 | 0.71 | 0.86 |

6 | 0.86 | 0.90 | 0.69 | 0.90 | 0.67 | 0.88 | 0.65 | 0.88 |

7 | 0.88 | 0.94 | 0.82 | 0.94 | 0.63 | 0.84 | 0.65 | 0.88 |

8 | 0.86 | 0.92 | 0.82 | 0.96 | 0.61 | 0.86 | 0.59 | 0.90 |

9 | 0.82 | 0.92 | 0.82 | 0.94 | 0.61 | 0.86 | 0.65 | 0.86 |

10 | 0.80 | 0.92 | 0.86 | 0.94 | 0.51 | 0.86 | 0.63 | 0.86 |

11 | 0.78 | 0.94 | 0.73 | 0.96 | 0.71 | 0.94 | 0.67 | 0.88 |

12 | 0.76 | 0.90 | 0.78 | 0.94 | 0.51 | 0.86 | 0.55 | 0.90 |

13 | 0.76 | 0.90 | 0.80 | 0.96 | 0.61 | 0.84 | 0.51 | 0.92 |

14 | 0.84 | 0.92 | 0.80 | 0.96 | 0.59 | 0.86 | 0.59 | 0.94 |

15 | 0.88 | 0.94 | 0.84 | 0.94 | 0.63 | 0.88 | 0.43 | 0.88 |

Disclaimer/Publisher’s Note: The statements, opinions and data contained in all publications are solely those of the individual author(s) and contributor(s) and not of MDPI and/or the editor(s). MDPI and/or the editor(s) disclaim responsibility for any injury to people or property resulting from any ideas, methods, instructions or products referred to in the content. |

© 2023 by the authors. Licensee MDPI, Basel, Switzerland. This article is an open access article distributed under the terms and conditions of the Creative Commons Attribution (CC BY) license (https://creativecommons.org/licenses/by/4.0/).

## Share and Cite

**MDPI and ACS Style**

Li, J.; Yuan, X.
Daily Streamflow Forecasts Based on Cascade Long Short-Term Memory (LSTM) Model over the Yangtze River Basin. *Water* **2023**, *15*, 1019.
https://doi.org/10.3390/w15061019

**AMA Style**

Li J, Yuan X.
Daily Streamflow Forecasts Based on Cascade Long Short-Term Memory (LSTM) Model over the Yangtze River Basin. *Water*. 2023; 15(6):1019.
https://doi.org/10.3390/w15061019

**Chicago/Turabian Style**

Li, Jiayuan, and Xing Yuan.
2023. "Daily Streamflow Forecasts Based on Cascade Long Short-Term Memory (LSTM) Model over the Yangtze River Basin" *Water* 15, no. 6: 1019.
https://doi.org/10.3390/w15061019