Scholarlink

<p>High-altitude meteorological networks in data-scarce mountainous regions frequently experience observation gaps during extreme weather events, when accurate measurements are most critical for snowmelt-runoff forecasting and water resource management. This study develops a completeness-driven sequential iterative imputation framework to reconstruct missing meteorological observations in the Euphrates-Karasu Basin, Turkey. The approach is based on gradient boosting regression and sequences stations according to data completeness, progressively propagating information through iterative learning. Unlike conventional methods that treat all stations uniformly, the proposed framework prioritizes high-correlation stations to reduce error propagation. The framework was evaluated using 11 years of data from six meteorological variables across eight high-elevation stations. Results show strong performance, with R<sup>2</sup> values ranging from 0.945 for relative humidity to 0.985 for snowfall, and average MAE reductions of 51.2% compared to non-iterative predictions. The largest improvements were observed for snowfall, while temperature showed more limited gains due to its already high baseline performance. These differences highlight the influence of variable-specific physical characteristics on imputation success. Overall, the proposed framework effectively reconstructs missing observations and captures complex spatial variability, providing a practical and robust solution for data-sparse mountainous regions where reliable meteorological information is essential for water resource 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

Completeness-driven sequential imputation for sparse meteorological networks: a gradient boosting framework for mountainous regions

  • 期刊论文

  • 发表日期：

    2026年7月28日

查看全文
  • Burak Çırağ,
  • Reşat Acar,
  • Mete Yağanoğlu,
  • Selim Şengül

摘要

High-altitude meteorological networks in data-scarce mountainous regions frequently experience observation gaps during extreme weather events, when accurate measurements are most critical for snowmelt-runoff forecasting and water resource management. This study develops a completeness-driven sequential iterative imputation framework to reconstruct missing meteorological observations in the Euphrates-Karasu Basin, Turkey. The approach is based on gradient boosting regression and sequences stations according to data completeness, progressively propagating information through iterative learning. Unlike conventional methods that treat all stations uniformly, the proposed framework prioritizes high-correlation stations to reduce error propagation. The framework was evaluated using 11 years of data from six meteorological variables across eight high-elevation stations. Results show strong performance, with R2 values ranging from 0.945 for relative humidity to 0.985 for snowfall, and average MAE reductions of 51.2% compared to non-iterative predictions. The largest improvements were observed for snowfall, while temperature showed more limited gains due to its already high baseline performance. These differences highlight the influence of variable-specific physical characteristics on imputation success. Overall, the proposed framework effectively reconstructs missing observations and captures complex spatial variability, providing a practical and robust solution for data-sparse mountainous regions where reliable meteorological information is essential for water resource management.