Scholarlink

<p>Accurate short-term streamflow forecasts are central to water operations and flood preparedness, so this work compares how precipitation inputs and hyperparameter tuning jointly affect forecast skill. Eight precipitation datasets—India Meteorological Department (IMD), Tropical Rainfall Measuring Mission (TRMM), Climate Hazards group Infrared Precipitation with Stations (CHIRPS), Climate Prediction Center MORPHing technique (CMORPH), Climate Prediction Center (CPC), Multi-Source Weighted-Ensemble Precipitation (MSWEP), Precipitation Estimation from Remotely Sensed Information Using Artificial Neural Networks-Climate Data Record (PERSIANN CDR), and Princeton Global Forcing (PGF)—are paired with five optimizers (Ant Colony Optimization (ACO), Particle Swarm Optimization (PSO), Grid Search (GS), Randomized Search (RS), and Simulated Annealing (SA)) to tune an Extreme Gradient Boosting (XGBoost) model for one-day-ahead prediction at the Wairagarh station (Godavari Basin) over 1998–2016. Tuning is performed with 5-fold cross-validation to provide a fair basis for comparing optimizer–dataset pairs. The SA–IMD combination delivers the best performance (Nash–Sutcliffe Efficiency (NSE) 0.94 training; 0.82 testing; Root Mean Square Error (RMSE) 32.8 and 53.8&#xa0;m³/s). CMORPH–SA is also strong, with NSE values of 0.88 (training) and 0.71 (testing). TRMM–SA showed good training skills but weaker testing. MSWEP–ACO and PERSIANN–SA gave moderate results, as did CHIRPS and CPC. PGF–SA had the lowest testing accuracy. SA with IMD is the most dependable, while CMORPH–SA is competitive, and others exhibit mixed generalization. The non-parametric Friedman test identified IMD as the most dependable precipitation input and SA as the most efficient optimizer, hence endorsing the superiority of the SA + IMD configuration. These results highlight the importance of selecting the appropriate optimization algorithm and precipitation dataset for improving streamflow forecasting accuracy. The strong performance of SA, especially with IMD and CMORPH, underscores its potential in hydrological predictions. Future studies should explore advanced hybrid modeling techniques to enhance prediction robustness and precision.</p>

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

Exploring the Impact of Optimization Techniques on Streamflow Prediction Using XGBoost: A Comparative Analysis with Satellite and Reanalysis Precipitation Datasets

  • 期刊论文

  • 发表日期：

    2025年12月26日

查看全文
  • Nagireddy Masthan Reddy,
  • Roshan Srivastav

摘要

Accurate short-term streamflow forecasts are central to water operations and flood preparedness, so this work compares how precipitation inputs and hyperparameter tuning jointly affect forecast skill. Eight precipitation datasets—India Meteorological Department (IMD), Tropical Rainfall Measuring Mission (TRMM), Climate Hazards group Infrared Precipitation with Stations (CHIRPS), Climate Prediction Center MORPHing technique (CMORPH), Climate Prediction Center (CPC), Multi-Source Weighted-Ensemble Precipitation (MSWEP), Precipitation Estimation from Remotely Sensed Information Using Artificial Neural Networks-Climate Data Record (PERSIANN CDR), and Princeton Global Forcing (PGF)—are paired with five optimizers (Ant Colony Optimization (ACO), Particle Swarm Optimization (PSO), Grid Search (GS), Randomized Search (RS), and Simulated Annealing (SA)) to tune an Extreme Gradient Boosting (XGBoost) model for one-day-ahead prediction at the Wairagarh station (Godavari Basin) over 1998–2016. Tuning is performed with 5-fold cross-validation to provide a fair basis for comparing optimizer–dataset pairs. The SA–IMD combination delivers the best performance (Nash–Sutcliffe Efficiency (NSE) 0.94 training; 0.82 testing; Root Mean Square Error (RMSE) 32.8 and 53.8 m³/s). CMORPH–SA is also strong, with NSE values of 0.88 (training) and 0.71 (testing). TRMM–SA showed good training skills but weaker testing. MSWEP–ACO and PERSIANN–SA gave moderate results, as did CHIRPS and CPC. PGF–SA had the lowest testing accuracy. SA with IMD is the most dependable, while CMORPH–SA is competitive, and others exhibit mixed generalization. The non-parametric Friedman test identified IMD as the most dependable precipitation input and SA as the most efficient optimizer, hence endorsing the superiority of the SA + IMD configuration. These results highlight the importance of selecting the appropriate optimization algorithm and precipitation dataset for improving streamflow forecasting accuracy. The strong performance of SA, especially with IMD and CMORPH, underscores its potential in hydrological predictions. Future studies should explore advanced hybrid modeling techniques to enhance prediction robustness and precision.