Two decades of global annual fractional vegetation cover products have been derived using daily surface reflectance and land surface temperature data from the Moderate Resolution Imaging Spectroradiometer (MODIS) instrument aboard NASA's Terra satellite. These MODIS Vegetation Continuous Fields (VCF) products are unique and a distinct advance over discrete (ordinal) land cover characterizations. VCF consists of three continuous fractional cover layers at 250 m resolution summing to 100%: tree cover, non-tree vegetation and non-vegetated surface plus surface water. This paper describes the fully-automated machine learning algorithm used to create MODIS VCF and details recent algorithm improvements. Comparisons of the latest VCF products in the MOD44B collection 6, with field-based fractional tree cover and other remote sensing fractional cover products demonstrate improved agreement: the root mean square (RMS) deviations range from 9 to 23%. A comparison of MODIS canopy cover values with those from the Global Ecosystem Dynamics Investigation (GEDI) gives an R² value of 0.53. Automation of the algorithm provides flexibility for readily porting the method to derive fractional cover products from additional instruments with similar radiometric characteristics such as the Visible Infrared Imaging Radiometer (VIIRS) payload aboard Suomi NPP and JPSS platforms, and successors.

利用来自NASA Terra卫星的中分辨率成像分光辐射计（MODIS）仪器的每日表面反射率和地表温度数据，已经获得了全球二十年的年度植被覆盖度分数产品。这些MODIS植被连续场（VCF）产品是独特的，并且在离散（常规）土地覆盖特征上有明显的进步。VCF由三个连续的覆盖层组成，分辨率为250 m，总计为100％：树木覆盖，非树木植被和非植被地表以及地表水。本文介绍了用于创建MODIS VCF的全自动机器学习算法，并详细介绍了最近的算法改进。MOD44B系列中最新VCF产品的比较6 基于现场的分数树覆盖和其他遥感分数覆盖产品显示出更好的一致性：均方根（RMS）偏差在9％到23％之间。将MODIS冠层覆盖值与全球生态系统动力学调查（GEDI）的值进行比较，得出R²值为0.53。该算法的自动化提供了灵活性，可轻松移植该方法，以从具有类似辐射特征的其他仪器（例如Suomi NPP和JPSS平台上的可见红外成像辐射计（VIIRS）有效载荷）及其后继产品中获取分数封面产品。