Medium- Long-Term Runoff Forecasting Using Interpretable Hybrid Machine Learning Model for Data-Scarce Regions

doi:10.11988/ckyyb.20240319

Abstract

Abstract:

[Objectives] This study aims to analyze the applicability of existing precipitation, temperature, and runoff data in data-scarce regions, and to develop and evaluate a deep learning hybrid model driven by multi-source information for improving the accuracy of monthly runoff forecasting. [Methods] Based on historical precipitation, temperature, and runoff sequences from the Yulongkashi River, a Convolutional Neural Network-Bidirectional Gated Recurrent Unit-Attention (CNN-BiGRU-Attention) model was developed. An Improved Particle Swarm Optimization (IPSO) algorithm was used to optimize this model, forming the IPSO-CNN-BiGRU-Attention hybrid model. The performance of this model was compared with that of the Gated Recurrent Unit (GRU) model and the ABCD water balance model. [Results] The IPSO-CNN-BiGRU-Attention model that incorporated precipitation and temperature data overall outperformed the CNN-BiGRU-Attention and GRU models, showing better agreement with the observed values. As the predication period increased, the proposed model achieved a root mean square error (RMSE) of 2.11 m³/s, a mean absolute error (MAE) of 1.32 m³/s, a mean absolute percentage error (MAPE) of 73.76%, and a Nash-Sutcliffe efficiency (NSE) coefficient of 0.94. The highest forecast accuracy was observed in the first three months. [Conclusions] The IPSO-CNN-BiGRU-Attention model effectively integrates precipitation, temperature, and runoff information, significantly enhancing the accuracy of monthly runoff forecasts in data-scarce regions. The model demonstrates robust performance across different forecast horizons, particularly suitable for short-term predictions of 1-3 months. This approach offers a practical and reliable tool for hydrological forecasting and flood control/drought management in data-scarce basins.

Key words: runoff forecasting, data-scarce regions, SHAP interpretable machine learning analysis, IPSO-CNN-BiGRU-Attention model, prediction accuracy

CLC Number:

TV121径流

YOU Yu-jun, BAI Yun-gang, LU Zhen-lin, ZHANG Jiang-hui, CAO Biao, LI Wen-zhong, YU Qi-ying. Medium- Long-Term Runoff Forecasting Using Interpretable Hybrid Machine Learning Model for Data-Scarce Regions[J]. Journal of Changjiang River Scientific Research Institute, 2025, 42(7): 52-59.

Figures/Tables 9

Fig.1 Schematic diagram of feature extraction using convolutional neural network

Fig.2 Schematic diagram of BiGRU model structure

Fig.3 Execution flowchart of hyperparameter optimization usingImproved Particle Swarm Optimization (IPSO) algorithm

Fig.4 Forecasting process of ICBA runoff model

Fig.5 Annual variations of temperature, precipitation, and runoff at the same station

Fig.6 Comparison of runoff forecasts during model validation period between the three machine learning models and ABCD model

Table 1 Statistical summary of evaluation metrics for hybrid model forecast

预见期/月	模型	率定期				验证期
预见期/月	模型	MAPE/ %	RMSE/ %	MAE/ %	NSE/ %	MAPE/ %	RMSE/ %	MAE/ %	NSE/ %
1	ICBA	35	28	37	12	32	26	32	10
1	CBA	26	20	26	9	27	21	22	9
2	ICBA	33	24	34	8	35	24	30	9
2	CBA	18	15	18	7	16	14	18	7
3	ICBA	35	22	35	8	28	22	28	7
3	CBA	16	14	17	7	14	12	19	7
4	ICBA	23	17	28	6	22	17	24	6
4	CBA	16	12	16	5	15	11	16	5
5	ICBA	18	14	20	6	18	14	18	6
5	CBA	15	12	15	4	13	10	12	5
6	ICBA	16	15	16	5	14	15	14	5
6	CBA	9	11	8	4	7	9	6	4

Fig.7 Robustness assessment results of GRU, ICBA, and CBA models

Fig.8 Contribution distribution of historical monthly runoff,temperature,and precipitation to model prediction

[1]	YANG Chen-lei, BAO Teng-fei. Dam Deformation Prediction Model Based on FCM-XGBoost [J]. Journal of Changjiang River Scientific Research Institute, 2021, 38(8): 66-71.
[2]	HE Bo, MA Jing, GAO He-yu. Predicting Urban Daily Water Supply Based on Multi-granularityFeature and XGBoost Integrated Model [J]. Journal of Changjiang River Scientific Research Institute, 2020, 37(5): 43-49.
[3]	HU An-yu, BAO Teng-fei, YANG Chen-lei, ZHANG Jing-ying. LSTM-ARIMA-based Prediction of Dam Deformation: Model and Its Application [J]. Journal of Changjiang River Scientific Research Institute, 2020, 37(10): 64-68.
[4]	DONG Dan-dan, ZU An-jun, SUN Xue-lian. Model of Dam Deformation Monitoring Based on Genetic Ant Colony Optimization and Back Propagation Improved by Markov Chain [J]. Journal of Changjiang River Scientific Research Institute, 2019, 36(7): 48-54.
[5]	BAO Li-na, TANG De-shan, HU Xiao-bo, CHU Shi-ji. Runoff Prediction Model Based on Wavelet Decomposition and Arima Error Correction: Research and Application [J]. Journal of Changjiang River Scientific Research Institute, 2018, 35(12): 18-21.
[6]	ZHOU Jian-zhong, PENG Tian. Chaotic Dynamic Characteristics and Integrated Prediction of Runoff in the Upper Reaches of Yangtze River [J]. Journal of Changjiang River Scientific Research Institute, 2018, 35(10): 1-9.
[7]	QIU Zi-feng, SHEN Jian, FU Xu-dong, LUO Hao-wei. Analysis of Tunnel Monitoring Measurement Data Based on the Optimum Weighted Combinatorial Prediction Model [J]. Journal of Changjiang River Scientific Research Institute, 2016, 33(5): 53-57.
[8]	YANG Hao-xiang,LIANG Chuan, CUI Ning-bo. Prediction of Urban Water Demand by Using Weighted Grey-Markov Chain Model [J]. Journal of Changjiang River Scientific Research Institute, 2015, 32(7): 15-21.
[9]	DAI Xing-lan. Application of SVR Ensemble Model to Annual Runoff Forecasting [J]. Journal of Changjiang River Scientific Research Institute, 2015, 32(4): 12-17.
[10]	WANG Wen-Mei, SUN Rong-Lin, GAO Yan. Modified BP Neural Network for Runoff Forecasting in the Karst Area [J]. Journal of Changjiang River Scientific Research Institute, 2012, 29(4): 11-16.
[11]	吕Qiao-Sen , XU Ying, HU Zhi-Qiang. Methods of Improving Tunnel Seismic Prediction Accuracy [J]. Journal of Changjiang River Scientific Research Institute, 2012, 29(12): 41-45.