Prediction of runoff in the upper reaches of the Hei River based on the LSTM model guided by physical mechanisms
Study region: The upper reaches of the Hei River. Study focus: To improve the accuracy, interpretability and physical consistency of LSTM models for predicting streamflow, we have constructed a physically dominant loss function in the LSTM model based on the monotonic physical mechanism of rainfall-...
Saved in:
| Main Authors: | , , , , , |
|---|---|
| Format: | Article |
| Language: | English |
| Published: |
Elsevier
2025-04-01
|
| Series: | Journal of Hydrology: Regional Studies |
| Subjects: | |
| Online Access: | http://www.sciencedirect.com/science/article/pii/S2214581825000424 |
| Tags: |
Add Tag
No Tags, Be the first to tag this record!
|
| _version_ | 1832573129679962112 |
|---|---|
| author | Huazhu Xue Chaoqiang Guo Guotao Dong Chenchen Zhang Yaokang Lian Qian Yuan |
| author_facet | Huazhu Xue Chaoqiang Guo Guotao Dong Chenchen Zhang Yaokang Lian Qian Yuan |
| author_sort | Huazhu Xue |
| collection | DOAJ |
| description | Study region: The upper reaches of the Hei River. Study focus: To improve the accuracy, interpretability and physical consistency of LSTM models for predicting streamflow, we have constructed a physically dominant loss function in the LSTM model based on the monotonic physical mechanism of rainfall-runoff in the water balance. The new model (physics-guided LSTM model, PG-LSTM) was applied to predict streamflow in the upper reaches of the Hei River, and its accuracy and physical consistency in streamflow prediction were analyzed. New hydrological insights for the region: The PG-LSTM model was successfully applied to the upper reaches of the Hei River, and the accuracy was evaluated by the Nash–Sutcliffe efficiency coefficient, root mean square error and Pearson correlation coefficient. The physical consistency of the model was evaluated by the volume error, relative error, and peak flow relative difference. The results showed that the PG-LSTM model had higher accuracy and physical consistency than the traditional LSTM model. The fitting accuracy between the measured and predicted values was 0.97, which is higher than that of the traditional LSTM model (0.89). In addition, the closer the subbasin was to the outlet of the basin, the better the effect of the PG-LSTM model. This model improvement method demonstrated high streamflow prediction accuracy and interpretability, providing a scientific basis for water resource planning and management. |
| format | Article |
| id | doaj-art-2089ea34877f42e69d2b2859c64c0681 |
| institution | Kabale University |
| issn | 2214-5818 |
| language | English |
| publishDate | 2025-04-01 |
| publisher | Elsevier |
| record_format | Article |
| series | Journal of Hydrology: Regional Studies |
| spelling | doaj-art-2089ea34877f42e69d2b2859c64c06812025-02-02T05:27:35ZengElsevierJournal of Hydrology: Regional Studies2214-58182025-04-0158102218Prediction of runoff in the upper reaches of the Hei River based on the LSTM model guided by physical mechanismsHuazhu Xue0Chaoqiang Guo1Guotao Dong2Chenchen Zhang3Yaokang Lian4Qian Yuan5School of Surveying and Land Information Engineering, Henan Polytechnic University, Jiaozuo 454000, China; Henan Province Spatial Big Data Acquisition Equipment Development and Application Engineering Technology Research Center, Jiaozuo 454000, ChinaSchool of Surveying and Land Information Engineering, Henan Polytechnic University, Jiaozuo 454000, ChinaSchool of Surveying and Land Information Engineering, Henan Polytechnic University, Jiaozuo 454000, China; Heihe Water Resources and Ecological Protection Research Center, Lanzhou 730030, China; Yellow River Source Research Institute, Lanzhou 450003, China; Corresponding author at: Heihe Water Resources and Ecological Protection Research Center, Lanzhou 730030, China.School of Surveying and Land Information Engineering, Henan Polytechnic University, Jiaozuo 454000, ChinaHeihe Water Resources and Ecological Protection Research Center, Lanzhou 730030, China; Yellow River Source Research Institute, Lanzhou 450003, ChinaSchool of Surveying and Land Information Engineering, Henan Polytechnic University, Jiaozuo 454000, ChinaStudy region: The upper reaches of the Hei River. Study focus: To improve the accuracy, interpretability and physical consistency of LSTM models for predicting streamflow, we have constructed a physically dominant loss function in the LSTM model based on the monotonic physical mechanism of rainfall-runoff in the water balance. The new model (physics-guided LSTM model, PG-LSTM) was applied to predict streamflow in the upper reaches of the Hei River, and its accuracy and physical consistency in streamflow prediction were analyzed. New hydrological insights for the region: The PG-LSTM model was successfully applied to the upper reaches of the Hei River, and the accuracy was evaluated by the Nash–Sutcliffe efficiency coefficient, root mean square error and Pearson correlation coefficient. The physical consistency of the model was evaluated by the volume error, relative error, and peak flow relative difference. The results showed that the PG-LSTM model had higher accuracy and physical consistency than the traditional LSTM model. The fitting accuracy between the measured and predicted values was 0.97, which is higher than that of the traditional LSTM model (0.89). In addition, the closer the subbasin was to the outlet of the basin, the better the effect of the PG-LSTM model. This model improvement method demonstrated high streamflow prediction accuracy and interpretability, providing a scientific basis for water resource planning and management.http://www.sciencedirect.com/science/article/pii/S2214581825000424Physics-guided deep learningLSTM modelRainfall-runoff modelPhysical consistency |
| spellingShingle | Huazhu Xue Chaoqiang Guo Guotao Dong Chenchen Zhang Yaokang Lian Qian Yuan Prediction of runoff in the upper reaches of the Hei River based on the LSTM model guided by physical mechanisms Journal of Hydrology: Regional Studies Physics-guided deep learning LSTM model Rainfall-runoff model Physical consistency |
| title | Prediction of runoff in the upper reaches of the Hei River based on the LSTM model guided by physical mechanisms |
| title_full | Prediction of runoff in the upper reaches of the Hei River based on the LSTM model guided by physical mechanisms |
| title_fullStr | Prediction of runoff in the upper reaches of the Hei River based on the LSTM model guided by physical mechanisms |
| title_full_unstemmed | Prediction of runoff in the upper reaches of the Hei River based on the LSTM model guided by physical mechanisms |
| title_short | Prediction of runoff in the upper reaches of the Hei River based on the LSTM model guided by physical mechanisms |
| title_sort | prediction of runoff in the upper reaches of the hei river based on the lstm model guided by physical mechanisms |
| topic | Physics-guided deep learning LSTM model Rainfall-runoff model Physical consistency |
| url | http://www.sciencedirect.com/science/article/pii/S2214581825000424 |
| work_keys_str_mv | AT huazhuxue predictionofrunoffintheupperreachesoftheheiriverbasedonthelstmmodelguidedbyphysicalmechanisms AT chaoqiangguo predictionofrunoffintheupperreachesoftheheiriverbasedonthelstmmodelguidedbyphysicalmechanisms AT guotaodong predictionofrunoffintheupperreachesoftheheiriverbasedonthelstmmodelguidedbyphysicalmechanisms AT chenchenzhang predictionofrunoffintheupperreachesoftheheiriverbasedonthelstmmodelguidedbyphysicalmechanisms AT yaokanglian predictionofrunoffintheupperreachesoftheheiriverbasedonthelstmmodelguidedbyphysicalmechanisms AT qianyuan predictionofrunoffintheupperreachesoftheheiriverbasedonthelstmmodelguidedbyphysicalmechanisms |