Scholarlink

<p>The Gandak River supports livelihoods across Bihar but rises rapidly during heavy rains, leading to damaging floods and economic losses. Effective planning and emergency response depend on reliable forecasts of river stage. Using daily data from 2003–2023 at Tribeni, water levels were predicted for 1-, 5-, and 10-day lead times with tree-based models and a lightweight residual-correction hybrid model. Among single models, Random Forest (RF) provided the strongest accuracy across all horizons (R<sup>2</sup> = 0.978, 0.942, 0.928 for 1, 5, and 10&#xa0;days, respectively), with the lowest errors. XGBoost performed comparably at 1&#xa0;day and was slightly weaker at longer leads. Residual-correction hybrids delivered the best overall results: at 1&#xa0;day, XGB-RF achieved R<sup>2</sup> = 0.981 with RMSE = 0.162; at 5&#xa0;days, RF-XGB improved performance to R<sup>2</sup> = 0.945 with RMSE = 0.282; and at 10&#xa0;days, RF-XGB reached R<sup>2</sup> = 0.932 with RMSE = 0.313. These findings indicate that RF and XGBoost are effective forecasters for the Gandak River, and that a small residual-correction step can further enhance accuracy without changing inputs.</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

Comparative Analysis of Machine Learning Models for Water Level Prediction in the Urban Water Basin

  • 期刊论文

  • 发表日期：

    2026年3月16日

查看全文
  • Rahul Prakash,
  • Joseph Tripura

摘要

The Gandak River supports livelihoods across Bihar but rises rapidly during heavy rains, leading to damaging floods and economic losses. Effective planning and emergency response depend on reliable forecasts of river stage. Using daily data from 2003–2023 at Tribeni, water levels were predicted for 1-, 5-, and 10-day lead times with tree-based models and a lightweight residual-correction hybrid model. Among single models, Random Forest (RF) provided the strongest accuracy across all horizons (R2 = 0.978, 0.942, 0.928 for 1, 5, and 10 days, respectively), with the lowest errors. XGBoost performed comparably at 1 day and was slightly weaker at longer leads. Residual-correction hybrids delivered the best overall results: at 1 day, XGB-RF achieved R2 = 0.981 with RMSE = 0.162; at 5 days, RF-XGB improved performance to R2 = 0.945 with RMSE = 0.282; and at 10 days, RF-XGB reached R2 = 0.932 with RMSE = 0.313. These findings indicate that RF and XGBoost are effective forecasters for the Gandak River, and that a small residual-correction step can further enhance accuracy without changing inputs.