This manuscript by Young et al. presents Version 3.0 of the PhenoCam dataset, a comprehensive, multi-year, and multi-biome collection of near-surface remote sensing data for vegetation phenology monitoring. The authors provide a robust description of dataset structure, methodology, and applications, along with valuable comparisons between GCC, cameraNDVI (derived using PhenoCams), and broadbandNDVI (using typical net radiation and PAR sensors). This paper builds on and extends the PhenoCam 2.0 dataset, which is already a tremendous resource for scientists across a range of scientific communities. A new aspect of PhenoCam 3.0 is that it enables the utility of cameraNDVI data and a simplified data structure. Importantly, this work addresses not only a clear need for high-frequency, ground-based phenology observations that complement satellite datasets and support model validation but provides a well-documented community resource.

One thing that was a bit unclear to me was regarding the addition of Data Record 7 (simplified products). It may help to clarify in the main text whether these simplified files include uncertainty estimates or metadata, and if not, whether that may impact scientific use. Otherwise I only have minor comments. Overall, the paper is well-written, methodologically sound, and makes a significant contribution that is likely to be widely used. It is well-suited for publication in Earth System Science Data with minor revisions.

Around line 100 – perhaps mention the issue of clouds as well. Satellites can’t see through them, but ground based data can fill these gaps.

Line 145-  The description of the simplified GCC products should briefly clarify that while uncertainty is omitted, users may refer to Data Record 5 for uncertainty quantification if needed.

Line 149 – no need to capitalize phenology here.

Line 194: Do end-users have access to the ‘exposure time’ data? Or how is this extracted? Is it a pre-set exposure, or does it automatically adjust based on irradiance conditions? Potentially some more information would be useful. ***I now see from Box 2 it’s provided in the metadata. Do all sites have all of this metadata associated with them?

Around line 220: Perhaps mention the FOV is going to be a bit different between broadbandNDVI and cameraNDVI. How so?

Figure 1: Great figure. Might be nice to also have a smaller dot in the center of each larger transparent circle so we can hone in on exactly where these data come from. Probably not needed but an idea.

Figure 3: The camera NDVI values are quite low. Especially in SH, where they are all negative. Should this be discussed? Either way, these results are impressive, and makes you wonder, do we need NDVI at all? The GCC data seem to be more dynamic and ‘sharper’. **ah yes, just as you mention around line 425.

Figure 5: Very nice, the SNR analysis is great here and gives ideas for future researchers.

Line 485 – yes, always visually inspect the data!

Note that there is a growing number of near-surface remote sensing platforms (see recent Tansley Review by Pierrat et al., 2025 https://nph.onlinelibrary.wiley.com/doi/epdf/10.1111/nph.20405 ) and it might be nice in the discussion to discuss briefly how PhenoCam can serve to compliment these data sources, or how it can be used as an example for how future networks might consider standardizing and producing user-friendly data products.

This is a great contribution, thank you for all your efforts to keep PhenoCam alive, it has had enormous benefits to the scientific community.

Young et al. provide a substantial update to the Phenocam Dataset, with V3.0 containing expanded site/biome and temporal coverage, and the addition of a new variable, cameraNDVI and simplied summary daily files that will be useful for many/most scientific applications as well as for for education and public outreach. They present a summary of the expansion to the network, explain the data records for the new variables, and then conduct an analysis of the new cameraNDVI variable. They find it is more sensitive to changes in leaf area, while Gcc is more sensitive to pigment changes. They also explore the variance in the cameraNDVI compared to both the Gcc and broadbandNDVI calculated from measurements of incident and upwelling solar radiation at Ameriflux sites. They find the cameraNDVI gives a clearer phenology signal compared to broadbandNDVI and note that it should be viewed as complementary to Gcc given the different information in the signal.

This is a very informative overview of the new dataset for current and new users of PhenoCam data. The paper was well written, the scope and objective of the paper were clearly defined. The new data and methods clearly described. The results are nicely presented. In the discussion the authors focus on the strengths and weaknesses of the cameraNDVI, which is helpful for researchers who want to use this new variable.

I have worked on this review with a student that is new to my lab – and to this topic – and they found the review paper was very clearly written and easy to understand. They were the one to work with the data for the start of their research project. They said working with the data was easy. However, they did note that accessing V3 data is less efficient due to the lack of a queryable structure. Without predefined indexing mechanisms, all 18,102 files are stored in a single directory, making it an intensive process to parse and extract relevant information. As a result, any program built to work with V3 will face increased time and space complexity, especially when scanning, filtering, or preprocessing the data for specific sites or time ranges.

The student accessed the V2 data through the PhenoCam Network’s official download page: https://phenocam.nau.edu/webcam/network/download/, which links to the API documentation. https://phenocam.nau.edu/api/ and https://phenocam.nau.edu/api/docs/. They said the PhenoCam API provides structured, programmatic access to V2 data, organized into clearly defined endpoints, including site metadata, image data, and processed summaries like daily counts and midday images. This made V2 far easier to much easier to query, navigate, and integrate into workflows compared to the more bulk-style V3 data release. I am guessing that v3 will eventually be available through the PhenoCam API?

Beyond that question, I have only minor comments that are provided below and a question about negative camerNDVI at the shrub and grass sites provided as examples. Otherwise, I think this is really excellent dataset of ground based phenology that will be useful for a wide range of research/scientific applications and for education and outreach. I look forward to seeing this paper published.

Minor comments

Introduction

Lines 117-119: could also consider Yan et al. (2019)

Methods and Materials

Section 2.4:

I appreciate data records 1 to 5 have been described previously and the authors don’t want to repeat here. But as I’m reading data record 6 I’m wondering if the RGB ROI statistics are the same as in data record 3? And if so, why repeat? I guess for transparency for the camera NDVI calculation?

I think it is the RGB stats are same from looking at the user guide on the V3 ORNL DAAC website, which as an FYI I found very useful for explaining all this, so perhaps refer readers to that document as well?

It is very helpful to see an example of how a file will look.

It is very helpful to have so many stats already calculated. Are there uncertainties provided for cameraNDVI?

Data record 7 is an excellent idea. I appreciate the idea is to keep things simple, but given the purpose can be for scientific applications (in addition to education or science outreach) why not also include the same two variables for cameraNDVI?

Section 3 (Results)

In Table 1 I think it would be useful to have the number of sites as well as site-years?

I think it could be really useful to have an additional table after Table 1 that contains the number of sites and site-years for each of the Level I ecoregions of North America. Then if researchers are focused on one or two specific regions they will be able to see the increase in number of sites and site years for that?

Figure 3: Interesting time series comparisons. I’m surprised by negative cameriaNDVI for much of the time series for the grass and shrub site though as there clearly is green vegetation there? Having said that, the authors do mention later that the values are not comparable to NDVI calculated from reflectance. I’ve also read another reviewer comment that points out that even when NDVI is above zero it’s not as high as we might expect with NDVI values we’re used to seeing from satellites. Still, the negative NDVI is a little surprising. I guess I should go and read the Petach et al. (2014) and Filippa et al. (2018) papers to learn more. If I’m remembering correctly the Wingate et al. reference I mention in my next comment addresses the issue of using DNs. But I agree with the other reviewer that explainig this a bit more when Fig. 3 is presented in the results. (while also pointing to the earlier papers) might be helpful.

Lines 404-410: Wingate et al. (2015) would be a good reference here. They used PROSAIL to show RGB signals/fractions across the European phenocam network were sensitive to chorolphyll and other pigments (and to some extent LAI), while NDVI is more sensitive to LAI (Section 3.2.1 of that paper).

Figs S2 to S6 are referenced before S1. And I’m not sure the reference for Fig. S1 at Line 449 is correct? I think that should be S2?

Also line 449: Barrow reference should be Fig. S6.

Figure 5 and S7: couldn’t hurt to have SNR_Gcc / SNR_cameraNDVI (and equivalent for S7) in the x-axis label in parentheses.

Perhaps a correlation analysis across all sites could be added between the Gcc and cameraNDVI time series so we can see across the huge range of sites which have a strong correlation or not (actually same for the SNRdiff analysis) with table summarising per ecoregion/vegetation type, or a map with point size in proportion to the correlation or SNR diff so we can see which ecoregions/vegetation types tend to have a closer correspondance (more or less variance in one or the other variable)? This would complement the examples shown in Figs. 3 to 5? Do all evergreen needleleaf sites have a clearer phenology signal in Gcc as shown in Fig. 3c?

Section 4 (Discussion)

Line 475: The authors identify NDVI < -0.5 is due to IR filter issues, but it is unclear whether such values were filtered or flagged in the dataset. This should be more explicitly discussed in the Methods or Data Records sections.

Line 485: But there is also a snow flag, so is this just to have an additional verification? Does it mean the snow flag  not reliable?

I was also wondering whether there are other QC flags (for clouds for example), but I guess this is all tied up in the Type I vs Type II vs Type III datasets?

It is tempting to ask for more of a discussion about how Gcc and cameraNDVI can be used beyond what the authors have mentioned in the last sentence of the discussion/manuscript. There are studies that have correlated Gcc and GPP for example, with highly variable results. As the authors of this study mention, NDVI can be more clearly linked to LAI than Gcc. Thinking from an ecosystem modeling perspective it seems like the new cameraNDVI variable will be of greater benefit for evaluating LAI compared to using Gcc for either LAI or GPP. However, if models were to couple with a radiative transfer model then using GCC could be more directly linked to the models. I think it’s beyond the scope to discuss this – and in any case there are plenty of other applications of these data. What I would suggest is if the authors think there are “good” and (perhaps especially) “bad” applications of either Gcc or cameraNDVI based on their expert knowledge, it would be a good opportunity to provide that perspective to the community. I personally would appreciate reading that. But again, I can see the argument that that is beyond the scope of this paper.

Other things to potentially include that would be of benefit to the reader (especially point 1 for those that may be new to using PhenoCam data when reading this v3 paper):

1. A brief update to the software applications that can be used, especially with the new variables (or just a mention to see Seyednasrollah et al. (2019) if nothing has changed.
2. Seyednasrollah et al. (2019) did a comparison of transition dates between V1 and V2. This doesn’t need to be repeated here, but a mention of the fact that the results are similar could be beneficial. I assume this is the case.

Will you publish the scripts used to process the phenocam data should anyone wish to look at that processing workflow?

References

Wingate, Lisa, Jérôme Ogée, Edoardo Cremonese, Gianluca Filippa, Toshie Mizunuma, Mirco Migliavacca, Christophe Moisy et al. "Interpreting canopy development and physiology using a European phenology camera network at flux sites." Biogeosciences 12, no. 20 (2015): 5995-6015.

Yan, D., Scott, R. L., Moore, D. J. P., Biederman, J. A., & Smith, W. K. (2019). Understanding the relationship between vegetation greenness and productivity across dryland ecosystems through the integration of PhenoCam, satellite, and eddy covariance data. Remote sensing of environment, 223, 50-62.