Scholarlink

<p>The Soil and Water Assessment Tool (SWAT) is widely used for hydrological simulation; however, systematic comparisons of uncertainty analysis algorithms under consistent modeling conditions remain limited, and integrated evaluation across performance and uncertainty metrics is often lacking. This study addresses these gaps by evaluating four SWAT-CUP algorithms —SUFI-2, GLUE, PSO, and ParaSol— under identical parameter ranges, simulation settings, and evaluation criteria in the Baihe River Basin, China. In addition, a Composite Desirability (CD) index is employed to provide an integrated assessment of model performance and uncertainty characteristics. Daily data from 2006 to 2020 were used, with 2007–2016 for calibration and 2017–2020 for validation. Thirteen parameters were calibrated using 1,000 simulations per algorithm. All methods achieved similar performance during calibration (R² = 0.84, NSE = 0.83), while validation performance showed only minor differences (R² = 0.71–0.77, NSE = 0.60–0.75). Uncertainty analysis revealed clear trade-offs among algorithms. SUFI-2 achieved a balanced performance (P-factor = 0.46, R-factor = 0.38 in validation), while PSO provided higher coverage but with wider uncertainty bands (0.52 and 0.58), and ParaSol produced narrower intervals with lower coverage (0.40 and 0.36). Based on the Composite Desirability index, SUFI-2 achieved the highest overall score (0.78), followed by ParaSol (0.74), PSO (0.71), and GLUE (0.62), a relatively more balanced performance across multiple criteria. Sensitivity analysis consistently identified CN2 as the most influential parameter, reflecting the dominant role of runoff processes. The results also confirm the presence of equifinality, as different parameter sets produced comparable simulation performance. This study provides a consistent and integrative framework for comparing uncertainty algorithms and supports method selection in similar watershed applications.</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 Evaluation of Parameter Uncertainty Algorithms in SWAT-based Hydrological Modeling

  • 期刊论文

  • 发表日期：

    2026年5月29日

查看全文
  • Biying Xu,
  • Nor Faiza Abd Rahman,
  • Vin Cent Tai,
  • Mou Leong Tan,
  • Philip W. Gassman,
  • Tengku Anita Raja Hussin

摘要

The Soil and Water Assessment Tool (SWAT) is widely used for hydrological simulation; however, systematic comparisons of uncertainty analysis algorithms under consistent modeling conditions remain limited, and integrated evaluation across performance and uncertainty metrics is often lacking. This study addresses these gaps by evaluating four SWAT-CUP algorithms —SUFI-2, GLUE, PSO, and ParaSol— under identical parameter ranges, simulation settings, and evaluation criteria in the Baihe River Basin, China. In addition, a Composite Desirability (CD) index is employed to provide an integrated assessment of model performance and uncertainty characteristics. Daily data from 2006 to 2020 were used, with 2007–2016 for calibration and 2017–2020 for validation. Thirteen parameters were calibrated using 1,000 simulations per algorithm. All methods achieved similar performance during calibration (R² = 0.84, NSE = 0.83), while validation performance showed only minor differences (R² = 0.71–0.77, NSE = 0.60–0.75). Uncertainty analysis revealed clear trade-offs among algorithms. SUFI-2 achieved a balanced performance (P-factor = 0.46, R-factor = 0.38 in validation), while PSO provided higher coverage but with wider uncertainty bands (0.52 and 0.58), and ParaSol produced narrower intervals with lower coverage (0.40 and 0.36). Based on the Composite Desirability index, SUFI-2 achieved the highest overall score (0.78), followed by ParaSol (0.74), PSO (0.71), and GLUE (0.62), a relatively more balanced performance across multiple criteria. Sensitivity analysis consistently identified CN2 as the most influential parameter, reflecting the dominant role of runoff processes. The results also confirm the presence of equifinality, as different parameter sets produced comparable simulation performance. This study provides a consistent and integrative framework for comparing uncertainty algorithms and supports method selection in similar watershed applications.