|The manuscript has been improved by addressing this reviewer’s comments on tree cover definition, scale discrepancy, the accuracy of MODIS IGBP land cover, and uncertainty associated with the reference field data, as well as other reviewers’ comments on the limited sample size. The reviewer is left with two additional comments, mostly related to how the previous comments are addressed. |
The first comment is with regards to how the field data collection and the associated uncertainty are reported in the paper. The revised manuscript very briefly mentioned that the uncertainty associated with the field data collection would not significantly change the results, which may well be true. I also understand that all the details about field data collection are reported in Torello-Raventos et al., (2013). However, the key steps of reference data collection should still be summarized and disclosed in this paper. It is critical to let the readers beware that the CAI reference used in the research was not directly measured in the field, but was calculated based on allometric equations.
The second comment is with regards to the suggestion of including the Landsat VCF in additional analysis. I have to rather insist on the comment as I find the authors’ argument against the suggestion unconvincing. Most of the field plots used in the study are in the size of 1 ha (some are below 1 ha); the pixel size of MODIS VCF is 250 m x 250 m (approximately 5 times larger than the field plots); the pixel size of Landsat VCF (derived based on MODIS VCF) is 30 m x 30 m (approximately 1/10 of the field plots). The field data are between 2006 and 2009, and the Landsat VCF is available around 2005 (the exact time of the data varies from place to place). The spatial mismatch between MODIS VCF and the field plots is much larger than the temporal mismatch between Landsat VCF and the field plots unless significant land cover changes were recorded at the sites. There are only 48 sites, and so the computation of data evaluation at the site level can be easily handled by a modern computer. Acknowledging the temporal mismatch, such an evaluation can still provide useful insights on how big a role the scale mismatch between the MODIS sensor and field plots plays in the data comparison. It can also provide useful insights on whether or not the biases in MODIS VCF are inherited by the later generation of remote sensing products. As the remote sensing field is moving towards mapping at increasingly higher spatial resolutions, such an evaluation would be of great interest to the producers of land cover datasets as well as the users of the newer VCF. The original comments and the authors’ responses are attached below.
“The inherent scale discrepancy between 100 m x 100m sites and 250m x 250m pixels is nicely addressed by simulations. Comparison results between the four types of simulations are also interesting. The authors could consider including the Landsat VCF data (Sexton et al. 2013) in the analysis, which is a satellite-based product most close to MODIS VCF. With a 30m x 30m spatial resolution, Landsat VCF can be averaged to close to the site scale, and a circa-2005 Landsat VCF product is available. This might generate additional insights, and might help resolve the difference with Brandt et al. 2020 in the Sahel region.
We chose to use the MODIS VCF product specifically because it’s one of the most widely used products in climatic and vegetation modelling, and has been for many years (see examples of studies in the introduction). While the 30 m GLCF product introduced by Sexton et al., 2013 is on a scale much more easily comparable to our field data, there are only 2 maps available (for 2000 and 2005), neither of which covers the field campaign period for the TROBIT project. In ecosystems as dynamic as tropical savannas, addressing the temporal mismatch would be very challenging.
In addition, performing the analysis we carried out in this paper on the GLCF product would take approximately 70 times longer than it did for MODIS VCF, rendering this correction technique computationally unfeasible for use on this much smaller scale.”