摘要
准确估算植被生物物理变量，包括吸收光合有效辐射分量（fAPAR）、植被覆盖度（FVC）和叶面积指数（LAI），对于作物监测和管理至关重要。尽管Sentinel-2提供了20 m分辨率的业务化生物物理变量产品，但在不同作物类型之间仍存在不确定性，尤其是在葡萄园等木本系统中，而且该分辨率往往不足以捕捉田块内的变异性。本研究旨在：(i) 在Mediterranean herbaceous和葡萄园作物上验证Sentinel-2简化Level-2原型（SL2P）产品；(ii) 开发并评估一个PlanetScope–Sentinel-2融合框架，以生成3-m、高频率的fAPAR、FVC和LAI地图。SL2P在所有变量上均与原位参考数据表现出高度一致性（fAPAR：R 2 = 0.88，|偏差| = 0.02，不确定度 = 0.07；FVC：R 2 = 0.80，|偏差| = 0.08，不确定度 = 0.09；LAI：R 2 = 0.86，|偏差| = 0.48，不确定度 = 0.62），其中fAPAR的绝对偏差最低。为在保持物理一致性的同时提高空间和时间细节，比较了四种降尺度方法。DMS-DT在预测性能和计算效率之间提供了有利的折中（fAPAR：R 2 = 0.73，|偏差| = 0.03，不确定度 = 0.10；FVC：R 2 = 0.67，|偏差| = 0.11，不确定度 = 0.13；LAI：R 2 = 0.75，|偏差| = 0.52，不确定度 = 0.75），残差校正进一步提升了性能。总体而言，所提出的融合框架能够实现3-m、高频率的生物物理变量监测，同时保持与Sentinel-2 SL2P产品的一致性。所得产品可为需要精细尺度且频繁植被信息的应用提供潜在有用的数据源，包括灌溉管理、产量预测和其他精准农业应用。
Accurate estimation of vegetation biophysical variables, including the fraction of absorbed photosynthetically active radiation (fAPAR), fraction of vegetation cover (FVC), and leaf area index (LAI), is essential for crop monitoring and management. Although Sentinel-2 provides operational biophysical variable products at 20 m resolution, uncertainties remain across crop types, particularly in woody systems such as vineyards, and this resolution is often insufficient to capture within-field variability. This study aimed to (i) validate the Sentinel-2 Simplified Level-2 Prototype (SL2P) product over Mediterranean herbaceous and vineyard crops; and (ii) develop and assess a PlanetScope–Sentinel-2 fusion framework to generate 3-m, high-frequency maps of fAPAR, FVC, and LAI. SL2P showed strong agreement with in-situ reference data for all variables (fAPAR: R 2 = 0.88, |bias| = 0.02, uncertainty = 0.07; FVC: R 2 = 0.80, |bias| = 0.08, uncertainty = 0.09; LAI: R 2 = 0.86, |bias| = 0.48, uncertainty = 0.62), with fAPAR showing the lowest absolute bias. To increase spatial and temporal detail while preserving physical consistency, four downscaling approaches were compared. DMS-DT provided a favorable compromise between predictive performance and computational efficiency (fAPAR: R 2 = 0.73, |bias| = 0.03, uncertainty = 0.10; FVC: R 2 = 0.67, |bias| = 0.11, uncertainty = 0.13; LAI: R 2 = 0.75, |bias| = 0.52, uncertainty = 0.75), with residual correction improving performance. Overall, the proposed fusion framework enables 3-m, high-frequency monitoring of biophysical variables while maintaining consistency with Sentinel-2 SL2P products. The resulting products could provide a potentially useful data source for applications requiring fine-scale and frequent vegetation information, including irrigation management, yield prediction, and other precision-agriculture applications.