Article Details

Abstract

Deep Learning becomes a powerful tool for runoff prediction, but in ungauged basins, the lack of flow observation data makes model training and prediction usually require the approach of Transfer Learning. However, the target basin often does not have enough data for fine-tuning, which makes it difficult to calibrate the model parameters. Therefore, this paper proposes an ungauged basins runoff prediction method based on conditional diffusion model. The method includes a forward noising process and a reverse denoising process. The denoising model is trained in the source basin, and then the data is recovered from the noise in the target basin as the prediction result. In addition, the denoising process is guided by the conditional data including meteorological drivers and historical runoff, and the Transformer layer is introduced into the denoising model to capture the dependence of time and features. Through the cross-validation experiment on the CAMELS-US dataset, the results show that the method has superiority.

Keywords

runoff prediction; ungauged basins; Transfer Learning; conditional diffusion model; CAMELS-US

References

1. [1] NG K W，HUANG Y F，KOO C H，et al. A Review of Hybrid Deep Learning Applications for Streamflow Forecasting [J/OL].Journal of Hydrology，2023，625：130141[2024-10-05].https://doi.org/10.1016/j.jhydrol.2023.130141.
2. [2] BLÖSCHL G，BIERKENS M F P，CHAMBEL A，et al. Twenty-three Unsolved Problems in Hydrology (UPH) - a Community Perspective [J].Hydrological Sciences Journal，2019，64（10）：1141-1158.
3. [3] IMAN M，ARABNIA H R，RASHEED K. A Review of Deep Transfer Learning and Recent Advancements [J/OL].arXiv:2201.09679 [cs.LG].[2024-09-26].https://doi.org/10.48550/arXiv.2201.09679.
4. [4] LIU C F，LIU D R，MU L. Improved Transformer Model for Enhanced Monthly Streamflow Predictions of the Yangtze River [J].IEEE Access，2022，10：58240-58253.
5. [5] VASWANI A，SHAZEER N，PARMAR N，et al. Atention is All You Need [J/OL].arXiv:1706.03762 [cs.CL].[2024-09-23].https://doi.org/10.48550/arXiv.1706.03762.
6. [6] YIN H L，ZHU W，ZHANG X W，et al. Runoff predictions in New-Gauged Basins Using Two Transformer-Based Models [J/OL].Journal of Hydrology，2023，622：129684[2024-10-06].https://doi.org/10.1016/j.jhydrol.2023.129684.
7. [7] YIN H L，GUO Z L，ZHANG X W，et al. RR-Former: Rainfall-runoff Modeling Based on Transformer [J/OL].Journal of Hydrology，2022，609：127781[2024-10-11].https://doi.org/10.1016/j.jhydrol.2022.127781.
8. [8] SHEN L，KWOK J. Non-autoregressive Conditional Diffusion Models for Time Series Prediction [J/OL].arXiv:2306.05043 [cs.LG].[2024-10-13].https://doi.org/10.48550/arXiv.2306.05043.
9. [9] SOHL-DICKSTEIN J，WEISS E，MAHESWARANATHAN N，et al. Deep Unsupervised Learning using Nonequilibrium Thermodynamics [C]//International conference on machine learning.Lille：PMLR，2015：2256-2265.
10. [10] HO J，JAIN A，ABBEEL P. Denoising Diffusion Probabilistic Models [J].Advances in Neural Information Processing Systems，2020，33：6840-6851.
11. [11] YANG Y，JIN M，WEN H，et al. A Survey on Diffusion Models for Time Series and Spatio-Temporal Data [J/OL].arXiv:2006.11239 [cs.LG].[2024-09-15].https://doi.org/10.48550/arXiv.2006.11239.
12. [12] KONG Z F，PING W，HUANG J J，et al. DiffWave: A Versatile Diffusion Model for Audio Synthesis [J/OL]. arXiv:2009.09761 [eess.AS].[2024-09-19].https://doi.org/10.48550/arXiv.2009.09761.
13. [13] XIANG Z R，YAN J，DEMIR I. A Rainfall‐Runoff Model With LSTM‐Based Sequence-to-Sequence Learning [J/OL].Water Resources Research，2020，56（1）：e2019WR025326[2024-10-17].https://doi.org/10.1029/2019WR025326.

About the authors

Zhang Wenzhao (2000-), female, Han ethnicity, from Dingxi, Gansu Province, is currently pursuing a master's degree. Her research focuses on predicting daily runoff in sparse watershed data based on deep transfer learning; Corresponding author: Yan Hua (1971-), male, Han ethnicity, from Dazhou, Sichuan, professor, PhD, research direction: intelligent information systems.