EUVIMEDEUVIMED entdecken
Open Access · CC BY 4.0

Daily Discharge Forecasting for the Lijiang River Using Heterogeneous Ensemble Learning and Karst Hydrology-Inspired Features

Jintao Shang, Junhong Chen, Shuaipu Zhang, Linfeng Wang, Songhui Ding, Zhixin Li, Yijie Bao, Shihua Su, Shucheng Wu

Water · 20.11.2026 · 18 (22) · 2899

DOI: 10.3390/w18222899 ↗

Kurz gefasst

Fachartikel in der Zeitschrift Water, online erschienen am 20.11.2026. Verfasst von Jintao Shang und 8 weiteren Autor:innen. Die Arbeit nennt 41 Literaturangaben. Der Volltext ist frei zugänglich (Open Access, CC BY 4.0). Gefördert von: Guangxi Key R&D Program, Guangxi Natural Science Foundation, National Natural Science Foundation of China.

AbstractBibliografische AngabenZitierenVerwandte Studien

Abstract (Originalsprache)

Complex surface water–groundwater exchanges and multi-timescale storage–release processes in karst basins produce pronounced nonlinearity and seasonal variability in precipitation–discharge relationships, posing a major challenge to accurate daily discharge forecasting. This study aims to improve daily discharge forecasting in karst basins by better representing antecedent hydrological conditions and seasonal differences in precipitation–discharge relationships. To this end, a weighted heterogeneous ensemble model integrating karst hydrology-inspired features with Light Gradient Boosting Machine (LightGBM), Categorical Boosting (CatBoost), and Extremely Randomized Trees (ExtraTrees) was established. The model was developed and validated for the upper Lijiang River Basin using daily hydrometeorological records of precipitation, evapotranspiration, and discharge from 1983 to 2022. Multi-timescale hydrometeorological indicators were incorporated to characterize antecedent basin wetness conditions and to capture the nonlinear precipitation–discharge response mechanisms typical of karst systems. Seasonal model partitioning into flood and non-flood periods was adopted to better represent contrasting hydrological conditions and flood peaks. The data were chronologically split into training, validation, and test sets in a 6:2:2 ratio. Model performance was evaluated using the root mean square error (RMSE), the Nash–Sutcliffe efficiency (NSE), and the Kling–Gupta efficiency (KGE). The results showed that the proposed ensemble model reliably reproduced daily discharge variability, achieving test-period RMSE, NSE, and KGE values of 93.8 m3 s−1, 0.880, and 0.907, respectively. These results demonstrate that integrating complementary tree-based models with antecedent hydrological information provides a robust approach to karst discharge forecasting. The proposed framework can support refined flood risk management, optimized water resource allocation, and sustainable development in karst regions.

Förderung

  • Guangxi Key R&D Program
  • Guangxi Natural Science Foundation
  • National Natural Science Foundation of China
  • Guangxi Young Elite Scientist Sponsorship Program

Zitieren

APA

Shang, J., Chen, J., Zhang, S., Wang, L., Ding, S., Li, Z., Bao, Y., Su, S., & Wu, S. (2026). Daily Discharge Forecasting for the Lijiang River Using Heterogeneous Ensemble Learning and Karst Hydrology-Inspired Features. Water, 18(22), 2899. https://doi.org/10.3390/w18222899

Vancouver

Shang J, Chen J, Zhang S, Wang L, Ding S, Li Z, et al. Daily Discharge Forecasting for the Lijiang River Using Heterogeneous Ensemble Learning and Karst Hydrology-Inspired Features. Water. 2026;18(22):2899. doi:10.3390/w18222899

BibTeX
@article{Shang2026,
  author = {Shang, Jintao and Chen, Junhong and Zhang, Shuaipu and Wang, Linfeng and Ding, Songhui and Li, Zhixin and Bao, Yijie and Su, Shihua and Wu, Shucheng},
  title = {Daily Discharge Forecasting for the Lijiang River Using Heterogeneous Ensemble Learning and Karst Hydrology-Inspired Features},
  journal = {Water},
  year = {2026},
  volume = {18},
  number = {22},
  pages = {2899},
  doi = {10.3390/w18222899}
}
Kostenlos starten →