the Creative Commons Attribution 4.0 License.
the Creative Commons Attribution 4.0 License.
Enhancing LSTM-based streamflow prediction with a spatially distributed approach
Abstract. Deep learning (DL) algorithms have previously demonstrated their effectiveness in streamflow prediction. However, in hydrological time series modelling, the performance of existing DL methods is often bound by limited spatial information, as these data-driven models are typically trained with lumped (spatially-aggregated) input data. In this study, we propose a hybrid approach, namely the Spatially Recursive (SR) model, that integrates a lumped long short-term memory (LSTM) network with a physics-based hydrological routing simulation for enhanced streamflow prediction. The lumped LSTM was trained on the basin-averaged meteorological and hydrological variables derived from 141 gauged basins located in the Great Lakes region of North America. The SR model involves applying the trained LSTM at the subbasin scale for local streamflow predictions which are then translated to the basin outlet by the hydrological routing model. We evaluated the efficacy of the SR model on predicting streamflow at 224 gauged stations across the Great Lakes region and compared its performance to that of the standalone lumped LSTM model. The results indicate that the SR model achieved performance levels on par with the lumped LSTM in basins used for training the LSTM. Additionally, the SR model was able to predict streamflow more accurately on large basins (e.g., drainage area greater than 1000 km2), underscoring the substantial information loss associated with basin-wise feature aggregation. Furthermore, the SR model outperformed the lumped LSTM when applied to basins that were not part of the LSTM training (i.e., pseudo-ungauged basins). The implication of this study is that the lumped LSTM predictions, especially in large basins and ungauged basins, can be reliably improved by considering spatial heterogeneity at finer resolution via the SR model.
- Preprint
(2070 KB) - Metadata XML
- BibTeX
- EndNote
Status: closed
-
RC1: 'Comment on hess-2023-237', Anonymous Referee #1, 04 Dec 2023
Overall, this is a very interesting study considering how to integrate deep learning models with process-based models. The paper introduces a Spatially Recursive (SR) model based on GRIP-GL data. The model first trains a basin LSTM and then uses the trained LSTM to simulate the flow of subbasins. Finally, it obtains the ultimate basin flow through a Routing-only mode. While the aspects mentioned in the paper are not individually new approach, this combination demonstrates a certain level of innovation.
- The "Routing-only mode" is an important part of the document, and perhaps a flowchart would help readers better understand its workflow. Additionally, the document mentions that the input for the Routing-only mode is hourly data. The question is how the authors transform data from the LSTM into hourly data.
- Including a map of the study area would indeed be beneficial for readers unfamiliar with the GRIP-GL project.
- The paper mentions that initially, basin data is used to train the LSTM, which is then applied to predict streamflow in subbasins using the subbasins' input data. Since there is sufficient subbasin data available, the question is why not directly train a subbasin-specific LSTM for predicting subbasin streamflow, which could then enable the prediction of basin streamflow through Routing-only mode?
- In the discussion in section 3.1, for smaller basins, the spatial segmentation might still represent a sub-basin. In this scenario, is the structure of the Spatially Recursive (SR) model still the same as it would be for multiple sub-basins, or is the Routing-only mode not used under these conditions?
- What is the role of Figure 8? Providing different delineations of the routing network within a single basin might better help in understanding the impact of routing network delineation.
Citation: https://doi.org/10.5194/hess-2023-237-RC1 - AC1: 'Reply on RC1', Q. Yu, 03 Jan 2024
-
RC2: 'Comment on hess-2023-237', Tadd Bindas, 11 Dec 2023
Thank you for letting me review your paper. Please see my supplemental material for my full review.Â
- AC2: 'Reply on RC2', Q. Yu, 03 Jan 2024
Status: closed
-
RC1: 'Comment on hess-2023-237', Anonymous Referee #1, 04 Dec 2023
Overall, this is a very interesting study considering how to integrate deep learning models with process-based models. The paper introduces a Spatially Recursive (SR) model based on GRIP-GL data. The model first trains a basin LSTM and then uses the trained LSTM to simulate the flow of subbasins. Finally, it obtains the ultimate basin flow through a Routing-only mode. While the aspects mentioned in the paper are not individually new approach, this combination demonstrates a certain level of innovation.
- The "Routing-only mode" is an important part of the document, and perhaps a flowchart would help readers better understand its workflow. Additionally, the document mentions that the input for the Routing-only mode is hourly data. The question is how the authors transform data from the LSTM into hourly data.
- Including a map of the study area would indeed be beneficial for readers unfamiliar with the GRIP-GL project.
- The paper mentions that initially, basin data is used to train the LSTM, which is then applied to predict streamflow in subbasins using the subbasins' input data. Since there is sufficient subbasin data available, the question is why not directly train a subbasin-specific LSTM for predicting subbasin streamflow, which could then enable the prediction of basin streamflow through Routing-only mode?
- In the discussion in section 3.1, for smaller basins, the spatial segmentation might still represent a sub-basin. In this scenario, is the structure of the Spatially Recursive (SR) model still the same as it would be for multiple sub-basins, or is the Routing-only mode not used under these conditions?
- What is the role of Figure 8? Providing different delineations of the routing network within a single basin might better help in understanding the impact of routing network delineation.
Citation: https://doi.org/10.5194/hess-2023-237-RC1 - AC1: 'Reply on RC1', Q. Yu, 03 Jan 2024
-
RC2: 'Comment on hess-2023-237', Tadd Bindas, 11 Dec 2023
Thank you for letting me review your paper. Please see my supplemental material for my full review.Â
- AC2: 'Reply on RC2', Q. Yu, 03 Jan 2024
Data sets
The Canadian Surface Prediction Archive (CaSPAr) Juliane Mai, Kurt C. Kornelsen, Bryan A. Tolson, Vincent Fortin, Nicolas Gasset and Djamel Bouhemhem, David Schäfer, Michael Leahy, François Anctil, Paulin Coulibaly https://caspar-data.ca/
Viewed
HTML | XML | Total | BibTeX | EndNote | |
---|---|---|---|---|---|
603 | 232 | 28 | 863 | 11 | 13 |
- HTML: 603
- PDF: 232
- XML: 28
- Total: 863
- BibTeX: 11
- EndNote: 13
Viewed (geographical distribution)
Country | # | Views | % |
---|
Total: | 0 |
HTML: | 0 |
PDF: | 0 |
XML: | 0 |
- 1