Scholarlink

<p>The present study aimed to investigate specific vegetation indices (VI) at three research sites - one grassland and two vineyards - and evaluate the potential of grassland remote sensing (RS) data to refine Normalized Difference Vegetation Index (NDVI) values in grass-covered inter-row vineyards. The vineyards differed in soil texture, with silt loam at BCS and clay at GB, located on 8–15% slopes. Field monitoring included NDVI, Photochemical Reflectance Index (PRI), and Photosynthetically Active Radiation (PAR) data at different slope positions. We also downloaded spectral data from Sentinel-2 (S2; <i>n</i> = 124) to see how well the NDVI field and S2 data correlate. Afterward, different machine learning techniques were used to refine the accuracy of the measurements, such as linear regression (LR), random forest (RF), and XGBoost. Significant differences in VI were observed between the research sites, mainly correlating with soil chemistry. While NDVI is an indicator of overall canopy vigor, PRI was more responsive to short-term physiological changes, showing higher sensitivity under stress conditions. Ground truth and RS NDVI were well correlated (<i>r</i> = 0.68), with RF providing the best accuracy when trained with day of year and the grassland data (<i>r</i> = 0.787). Each model moderately predicted grapevine VI using basic inputs (<i>r</i> &gt; 0.61), however, all three models performed well when grassland NDVI included in the training.</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

Refining grapevine vegetation status by utilizing grassland and remote sensing data

  • 期刊论文

  • 出版模式：

    开放获取

  • 发表日期：

    2025年8月27日

查看全文
  • Tibor Zsigmond,
  • Zsófia Bakacsi,
  • Ágota Horel

摘要

The present study aimed to investigate specific vegetation indices (VI) at three research sites - one grassland and two vineyards - and evaluate the potential of grassland remote sensing (RS) data to refine Normalized Difference Vegetation Index (NDVI) values in grass-covered inter-row vineyards. The vineyards differed in soil texture, with silt loam at BCS and clay at GB, located on 8–15% slopes. Field monitoring included NDVI, Photochemical Reflectance Index (PRI), and Photosynthetically Active Radiation (PAR) data at different slope positions. We also downloaded spectral data from Sentinel-2 (S2; n = 124) to see how well the NDVI field and S2 data correlate. Afterward, different machine learning techniques were used to refine the accuracy of the measurements, such as linear regression (LR), random forest (RF), and XGBoost. Significant differences in VI were observed between the research sites, mainly correlating with soil chemistry. While NDVI is an indicator of overall canopy vigor, PRI was more responsive to short-term physiological changes, showing higher sensitivity under stress conditions. Ground truth and RS NDVI were well correlated (r = 0.68), with RF providing the best accuracy when trained with day of year and the grassland data (r = 0.787). Each model moderately predicted grapevine VI using basic inputs (r > 0.61), however, all three models performed well when grassland NDVI included in the training.