Scholarlink

<p>Soil temperature (<i>T</i><sub><i>S</i></sub>) plays a crucial role in hydrology, ecology, and agriculture. Precise estimation of <i>T</i><sub><i>S</i></sub> is critical across these fields, prompting substantial research efforts. While past studies have largely concentrated on average <i>T</i><sub><i>S</i></sub> modeling, this research aimed to simulate both the maximum and minimum soil temperatures for two types of soil cover: bare soil and soil under turf. Data were collected from ten sites in Illinois, USA, to test and validate the proposed methodology. The study focused on modeling soil temperatures at two depths for each of the cover types. Two modeling approaches, boosted regression tree (BRT) and multivariate adaptive regression splines (MARS), were used to simulate these target parameters. A k-fold data assignment method was implemented to test model effectiveness on both local and spatial scales. The local scale involved model application at each site, while the spatial scale employed a novel merged data approach, which combined data from different locations to estimate parameters at a particular site. Model performance was evaluated using weighted <i>RMSE</i> and Nash–Sutcliffe (<i>NS</i>) efficiency criteria. Both models effectively simulated <i>T</i><sub><i>S</i></sub> on local and spatial levels. Importantly, they allow for simulation without relying on local data and can estimate soil temperature using minimal inputs, e.g., air temperature and humidity. It should be noted, however, that the present study had a limitation of excluding soil type data from the input matrix due to data unavailability. Further studies might revise the outcomes by incorporating more soil-type-based data for simulating thermal extremes.</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

Exploring soil temperature extremes: unraveling dynamics with local and spatial machine learning models

  • 期刊论文

  • 发表日期：

    2025年8月2日

查看全文
  • Sepideh Karimi

摘要

Soil temperature (TS) plays a crucial role in hydrology, ecology, and agriculture. Precise estimation of TS is critical across these fields, prompting substantial research efforts. While past studies have largely concentrated on average TS modeling, this research aimed to simulate both the maximum and minimum soil temperatures for two types of soil cover: bare soil and soil under turf. Data were collected from ten sites in Illinois, USA, to test and validate the proposed methodology. The study focused on modeling soil temperatures at two depths for each of the cover types. Two modeling approaches, boosted regression tree (BRT) and multivariate adaptive regression splines (MARS), were used to simulate these target parameters. A k-fold data assignment method was implemented to test model effectiveness on both local and spatial scales. The local scale involved model application at each site, while the spatial scale employed a novel merged data approach, which combined data from different locations to estimate parameters at a particular site. Model performance was evaluated using weighted RMSE and Nash–Sutcliffe (NS) efficiency criteria. Both models effectively simulated TS on local and spatial levels. Importantly, they allow for simulation without relying on local data and can estimate soil temperature using minimal inputs, e.g., air temperature and humidity. It should be noted, however, that the present study had a limitation of excluding soil type data from the input matrix due to data unavailability. Further studies might revise the outcomes by incorporating more soil-type-based data for simulating thermal extremes.