Scholarlink

<p>The sublimation rate is a key parameter in cold-region hydrology. The simplest calculation method is the weighing method, which determines sublimation by measuring the loss of mass from a pan containing solid water. While machine learning (ML) has been widely used to model pan evaporation (E<sub>pan</sub>), its applicability to model pan sublimation (S<sub>pan</sub>) remains uncertain due to the more complex underlying mechanisms. In this study, we evaluate the performance of six different ML models—Artificial Neural Network (ANN), support vector machines (SVM), random forests (RF), adaptive neuro-fuzzy inference systems (ANFIS), convolutional neural networks (CNN), and Gradient Boosting Decision Tree (GBDT)—for simulating S<sub>pan</sub>. We collected daily meteorological parameters, including mean temperature (T), sunshine duration (Hs), relative humidity (Rh), wind speed (U), net radiation (Rn), and S<sub>pan</sub> from the Qilian Mountains between 2014 and 2024 in winter. Our results demonstrate that all models exhibit excellent simulation accuracy, with ANFIS emerging as the best performer. In the testing period, when all meteorological parameters were considered, the MAE, RMSE, KGE, and CC were 0.19&#xa0;mm d<sup>−1</sup>, 0.27&#xa0;mm d<sup>−1</sup>, 0.84, and 0.87, respectively. This study demonstrates for the first time the effectiveness of ML in simulating S<sub>pan</sub> in cold regions. More importantly, it provides a simple and scientific method for estimating sublimation.</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

Applicability of machine learning in modelling pan sublimation in cold regions

  • 期刊论文

  • 发表日期：

    2025年10月29日

查看全文
  • Kaifeng Ma,
  • Junfeng Liu,
  • Chuntan Han,
  • Rensheng Chen

摘要

The sublimation rate is a key parameter in cold-region hydrology. The simplest calculation method is the weighing method, which determines sublimation by measuring the loss of mass from a pan containing solid water. While machine learning (ML) has been widely used to model pan evaporation (Epan), its applicability to model pan sublimation (Span) remains uncertain due to the more complex underlying mechanisms. In this study, we evaluate the performance of six different ML models—Artificial Neural Network (ANN), support vector machines (SVM), random forests (RF), adaptive neuro-fuzzy inference systems (ANFIS), convolutional neural networks (CNN), and Gradient Boosting Decision Tree (GBDT)—for simulating Span. We collected daily meteorological parameters, including mean temperature (T), sunshine duration (Hs), relative humidity (Rh), wind speed (U), net radiation (Rn), and Span from the Qilian Mountains between 2014 and 2024 in winter. Our results demonstrate that all models exhibit excellent simulation accuracy, with ANFIS emerging as the best performer. In the testing period, when all meteorological parameters were considered, the MAE, RMSE, KGE, and CC were 0.19 mm d−1, 0.27 mm d−1, 0.84, and 0.87, respectively. This study demonstrates for the first time the effectiveness of ML in simulating Span in cold regions. More importantly, it provides a simple and scientific method for estimating sublimation.