|This is my second round of reviewing. The author’s responses confirmed my concerns, and this study or this dataset includes several important flaws which may substantially mislead the future studies in this filed. Therefore, I strongly suggest the authors seriously consider the method and the GPP dataset, and take the efforts to develop the reliable method. |
First, the authors confirmed their estimate of global GPP reaches to 200 Pg C yr-1, which is almost double of the current estimates. Although the authors argued that their estimates are close to the estimates from Welp et al (2011) (150-175 Pg C yr-1) and Koffi et al. (2012) (146 Pg C yr-1). However, the estimate in this study also is higher than these two studies about higher 30%-50%. Besides, these two studies are based on atmospheric inversion methods to indirectly estimate GPP, and ecosystem respiration may highly impact their estimates to GPP. As the Welp et al (2011) claimed “best guess of 150-175” of GPP. On contrary, MODIS and FLUXCOM used site-based GPP observations to constrain their estimates, and which provide the robust estimates of GPP compared to Welp and Koffi.
The authors validated their GPP estimates at eddy covariance towers. VODCA2GPP are comparable to tower-based GPP as Fig. 1A shown. I am wondering that there are large differences over the global estimates. The method may have significant flaw that make it impossible to apply over global scale. Therefore, I strongly suggest the authors investigate the reliability of the method before producing global GPP dataset. As I pointed out that there are several unclear items in the model algorithms, which may induce large uncertainties for GPP estimates. For example, the response #5, the authors changed the definition of mdn(VOD) from landcover to vegetation density. It is totally confused what vegetation density means? By my knowledge, there is no concept of vegetation density, instead that we say Species Density, which is obvious different with the authors’ idea. It is my largest concern the authors failed to propose the robust physiological principle for using VOD to estimate GPP at all.
In addition, the authors examined MODIS and FLUXCOM dataset against eddy covariance-based GPP. However the results in this manuscript look quite different with previous reports. Especially, FLUXCOM is data-driven dataset, which should be compared with site-based GPP. However, the authors showed the underestimated GPP by FLUXCOM, which is quite different with previous studies and also difficulty to understand.
Second, the VODCA2GPP dataset showed the low performance both over spatial and temporal scales. The authors added the validations on model performance for reproducing interannual variability of GPP (response #9). However, the performance is quite low, and mean R2 value is only 0.2 or even lower. By this low performance, I can not trust the capability of VODCA2GPP, and will not use it to conduct any further analyses. So, I still doubted why we still need VODCA2GPP dataset. The authors argued that we need other satellite data source besides optical data, but it is not a reason for accepting its low performance.