A Framework to Assess the Potential Uncertainties of Three FPAR Products

The remote sensing products of the fraction of photosynthetically active radiation (FPAR) are widely used in large‐scale terrestrial vegetation condition monitoring and ecological modeling. However, the validation and evaluation of these products are still insufficient due to limited ground measurements. A framework is proposed in this study that does not rely on ground FPAR observations to evaluate the potential uncertainties of three FPAR data sets over a large scale. Two independent methods, including a statistical method named extended triple colocation (ETC) and FPAR‐based gross primary production (GPP) comparative evaluation at the site level, are integrated. The framework was applied to three widely used FPAR products, including GLASS (Global Land Surface Satellite), GEOV2 (version 2 of the COPERNICUS 1‐km product), and MODIS (Moderate Resolution Imaging Spectroradiometer), in China from 2001–2015. The root‐mean‐square error (RMSE) and correlation coefficient (R) based on the ETC method showed that at the national spatial scale, GEOV2‐FPAR had the highest precision (RMSEFPAR = 0.0367, RFPAR = 0.9012), followed by MODIS‐FPAR (RMSEFPAR = 0.0482, RFPAR = 0.8392) and GLASS‐FPAR (RMSEFPAR = 0.0536, RFPAR = 0.736). In particular, each FPAR data set was locally optimal over some regions. Using a light use efficiency model, GPP was calculated using these three FPAR data sets and other identical variables and parameters at 22 eddy flux sites. General coherent results were found in both GPP evaluation and site‐level ETC analysis of these three FPAR data sets. Therefore, we conclude that this framework is reliable and applicable in evaluating the potential uncertainties of FPAR. Plain Language Summary Traditionally, the fraction of photosynthetically active radiation (FPAR) inverted by satellites is usually validated by ground‐based measurement sites. However, these sites are few, and the time of observations is not continuous, which is not suitable for large‐scale and long‐term evaluation. In the model, gross primary productivity (GPP) is generally regarded as the final product of FPAR. A large amount of GPP measured data provides us with an opportunity to indirectly validate FPAR. In addition, an extended triple colocation method, which is a statistical method used to calculate potential uncertainties of remote sensing products, is applied to the evaluation of FPAR products, and the combination of the two methods may lead to more credible conclusions. Here,