Scholarlink

<p>In seasonally frozen mid- to high-latitude regions, interactions between surface water and groundwater driven by topography and seasonal climate create strong spatial heterogeneity and temporal fluctuations in baseflow, complicating its estimation. Most existing prediction models neglect regional heterogeneity and freeze–thaw dynamics, limiting their adaptability and accuracy in such environments. Therefore, this study investigated the Colorado River Basin (CRB) and the Mississippi River Basin (MRB), which are representative of seasonally frozen regions. By incorporating the main factors influencing baseflow in ecoregions, a prediction model was implemented with spatial heterogeneity considered among ecoregions and the dynamic impact of freeze–thaw processes. The model was validated by the conductivity mass balance (CMB) method using data from 153 sites. The results indicated that: (1) elevation, temperature, leaf area index, snow depth, and potential evaporation were the dominant factors controlling baseflow in seasonally frozen regions; (2) the Random Forest model presented high prediction accuracy across all ecoregions, with an accuracy of 0.85 during the freeze–thaw period; (3) the ecoregion-based prediction model enhanced the accuracy by 26.8% compared to the global model, and incorporating ecoregions and freeze–thaw effects further increased the accuracy by 16.5% during the freeze–thaw period. These findings provide methodological support for baseflow prediction and water management.</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

Incorporating Ecoregions and Freeze–Thaw Effects Effectively Improves the Accuracy of Machine Learning Models for Baseflow Prediction in Seasonally Frozen Regions

  • 期刊论文

  • 发表日期：

    2026年2月2日

查看全文
  • Tingting Deng,
  • Hang Lyu,
  • Rong Zhong,
  • Yunting Zou,
  • Jianan Wang,
  • Yu Li,
  • Yue Guo

摘要

In seasonally frozen mid- to high-latitude regions, interactions between surface water and groundwater driven by topography and seasonal climate create strong spatial heterogeneity and temporal fluctuations in baseflow, complicating its estimation. Most existing prediction models neglect regional heterogeneity and freeze–thaw dynamics, limiting their adaptability and accuracy in such environments. Therefore, this study investigated the Colorado River Basin (CRB) and the Mississippi River Basin (MRB), which are representative of seasonally frozen regions. By incorporating the main factors influencing baseflow in ecoregions, a prediction model was implemented with spatial heterogeneity considered among ecoregions and the dynamic impact of freeze–thaw processes. The model was validated by the conductivity mass balance (CMB) method using data from 153 sites. The results indicated that: (1) elevation, temperature, leaf area index, snow depth, and potential evaporation were the dominant factors controlling baseflow in seasonally frozen regions; (2) the Random Forest model presented high prediction accuracy across all ecoregions, with an accuracy of 0.85 during the freeze–thaw period; (3) the ecoregion-based prediction model enhanced the accuracy by 26.8% compared to the global model, and incorporating ecoregions and freeze–thaw effects further increased the accuracy by 16.5% during the freeze–thaw period. These findings provide methodological support for baseflow prediction and water management.