Scholarlink

<p>Accurate prediction of wheat grain yield (WGY) is vital for crop management and food security in arid regions. This study evaluates WGY prediction in the Dehloran Plain, western Iran, by integrating multi-source data: multi-temporal Sentinel-2 imagery, proximal Vis–NIR soil spectroscopy, and topographic attributes. A Random Forest (RF) model was developed with 135 field samples, comparing two scenarios: S1, using soil spectroscopy and remote sensing indices, and S2, which also included topographic variables. Results show that S2 significantly outperforms S1 (R² = 0.78, CCC = 0.79, nRMSE = 6.25%), emphasizing the importance of terrain attributes. Variable importance and GAM partial dependence plots reveal that soil spectral components (LV2, LV3) and topographic parameters (Effective Air, Generalized Surface, VD, CNBL) are the dominant predictors, while time-series RS data contributes less. Spatial predictions indicate higher yields in northern and eastern-central areas, with slight uncertainty. Overall, the RF model demonstrates robust predictive accuracy and low uncertainty, providing a reliable, interpretable framework for precision agriculture and sustainable land management in arid environments.</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

Integrating time series remote sensing and proximal data for wheat grain yield prediction using machine learning models

  • 期刊论文

  • 发表日期：

    2026年5月20日

查看全文
  • Samireh Nazari,
  • Shuai Zhao,
  • Shamsollah Ayoubi,
  • Mahmood Rostaminia,
  • Seyed Roohollah Mousavi,
  • Artemi Cerda

摘要

Accurate prediction of wheat grain yield (WGY) is vital for crop management and food security in arid regions. This study evaluates WGY prediction in the Dehloran Plain, western Iran, by integrating multi-source data: multi-temporal Sentinel-2 imagery, proximal Vis–NIR soil spectroscopy, and topographic attributes. A Random Forest (RF) model was developed with 135 field samples, comparing two scenarios: S1, using soil spectroscopy and remote sensing indices, and S2, which also included topographic variables. Results show that S2 significantly outperforms S1 (R² = 0.78, CCC = 0.79, nRMSE = 6.25%), emphasizing the importance of terrain attributes. Variable importance and GAM partial dependence plots reveal that soil spectral components (LV2, LV3) and topographic parameters (Effective Air, Generalized Surface, VD, CNBL) are the dominant predictors, while time-series RS data contributes less. Spatial predictions indicate higher yields in northern and eastern-central areas, with slight uncertainty. Overall, the RF model demonstrates robust predictive accuracy and low uncertainty, providing a reliable, interpretable framework for precision agriculture and sustainable land management in arid environments.