This study proposed several versions of AusENDVI and these NDVIs can be used for studying Australia's changing vegetation dynamics and carbon, and water cycles. The paper is generally organized. The new data set would be useful for the Earth system science studies. However, I still have questions about the structure of the article. Considering these and due to the following major concerns and suggestions, I would recommend it with major revision and to determine whether to accept a revised version.

Major concern:

1. I think the article needs a flowchart to show each step, which helps the reader understand the importance of the data processing process. So far I found in the section 'Data and Code Availability' that the author lists each version of AusENDVI, but in fact, I am confused about which step each version of the data is obtained through.

2. I don't think 'Quality of existing NDVIs' is the key part of the article, this part of the results could be replaced by comparing the performance of AusENDVI with other NDVIs, e.g. by adding on the performance of AusENDVI in Figure 2, and then transforming this part into the second part of the results.

3. I think the first part of the results could be to highlight the results of each step, especially 'before and after the calibration and harmonization' and 'before and after gaping fill'. Of course, these are already in the results, but they should be in the same section to highlight the results of each step of the enhancement.

4. Is it possible to find field measurements of NDVIs in Australia to provide absolute accuracies for individual NDVIs, and if so, this would be an important support for demonstrating the accuracy of AusENDVIs.

5. The discussion is too lengthy, my suggestion is that it could be broken up into subsections.

6. As with 'Trends in peak-of-season phenology', I would suggest that the authors do the same study again, using the available NDVIs, do a trend analysis of annual averages, and then compare it to the results in the literature, to sidely bolster the credibility of these data.

Review on essd-2024-89

In the manuscript titled “Enhancing Long-Term Vegetation Monitoring in Australia: A New Approach for Harmonising and Gap-Filling AVHRR and MODIS NDVI”, Burton et al. reconstructed new harmonised NDVI datasets in Australia using the GBM method. The manuscript and figures are well prepared. I appreciate the extensive work conducted in this study, like, comparing existing datasets, producing new datasets and applications. However, from my perspective, this paper may still lack sufficient novelty to warrant publication in ESSD. Below, I outline my main concerns and provide point-to-point comments.

Main concerns:

1. In the context of the existing abundance of NDVI datasets such as VIP15 NDVI, GIMMS NDVI3g and the latest PKU NDVI, authors still aim to produce new NDVI datasets, which is challenged. I encourage this work, but authors fail to show strong motivations for doing so (like, data unavailability or any issues present in existing datasets).
2. According to the results (like, figures 2 & 8), I think PKU-consolidated dataset has been produced well, and compared to PKU data, your dataset does not show any significant and necessary improvements. Therefore, I would suggest highlighting clear improvements than other existing datasets.

Other comments:

1. No ground observations (like, Flux or PhenoCam sites) to validate your data?
2. For any designed steps (e.g., gap filling), it is expected to see the comparison of results for before and after processing (can refer to the guide: https://lpdaac.usgs.gov/documents/1328/VIP_User_Guide_ATBD_V4.pdf).
3. Add a flowchart to summarize each step and processing.
4. Add some quantified results in the abstract to show the reliability/enhancement of your datasets.
5. Lines 30-35, provide spatial and temporal resolutions information for your 41-year dataset.

This manuscript by Burton et al. proposes a new long-term NDVI dataset specifically for Australia (AusENDVI) by harmonizing and gap-filling AVHRR and MODIS data. Compared to global NDVI datasets, localized AusENDVI could provide optimized NDVI observation with the aid of prior knowledge. To this end, I agree that the AusENDVI could be a promising dataset for better understanding long-term vegetation dynamics in Australia. However, the current manuscript faces many major issues and lacks essential information that shows the superiority of AusENDVI. My overall attitude is somewhere between a severely major revision and rejection. That’s dependent on how the authors respond to the following comments.

Major comments:

First, NDVI is a spectral index calculated from red and near-infrared reflectance. Discrepancies of band settings (spectral range, FWHM, etc.) between sensors could be an important driver of the NDVI difference. This is the case for the three types of sensors involved in the manuscript, i.e., Landsat TM/ETM+, AVHRR, and MODIS. However, this source of NDVI differences in band setting has been completely ignored in the evaluation of current global NDVI datasets and generation of AusENDVI. For example, the authors failed to compare the two reference datasets, Landsat TM/ETM+ and MODIS in the manuscript.

Second, for some reason, the temporal resolution of the AusENDVI has been missing in the Abstract and Conclusion section of the manuscript. For a long-term dataset, the temporal resolution is a critical attribute that determines how well the AusENDVI could capture the abrupt vegetation changes due to climate or anthropogenic disturbances. As far as I could find in the manuscript and the data repository, AusENDVI provides monthly data records. It could be disappointing because the temporal resolution of current global NDVI datasets such as NDVI3g and NDVIpku is half a month. This issue is related to another one in that AusENDVI uses median composites while NDVI3g, NDVIpku, and MODIS NDVI use maximum composites. Why is the median? Will that underestimate vegetation growth such as vPOS?

Third, the most impressive feature of AusENDVI is that it accounts for the dominant role of precipitation in Australia. However, the strong relationship between precipitation and NDVI has been an unproved precondition in the manuscript. The authors must demonstrate pixel-wise precipitation-NDVI relation before the relationship is used to evaluate NDVI products and generate AusENDVI. For example, in Figure 8b, the abrupt increase of NDVI in 1984 does not seem to follow the precipitation anomalies (Note the authors use the precipitation anomalies to argue the deficiency of other NDVI products). A literature review without a pixel-wise relation map is not enough.

Last, the authors failed to demonstrate the improvements of AusENDVI in critical aspects such as long-term trends of vegetation and SOS.

Some minor but still important comments:

Line 96-97. Why are SOS and EOS not included?

Line 104. When is averaging used and when is nearest-neighboring used?

Line 105. How to deal with the radiometric difference between Landsat TM and ETM+ (Berner et al., 2020; https://doi.org/10.1038/s41467-020-18479-5)?

Line 212. Why is the median rather than the maximum value?

Line 122. Please provide more information on the use of the quality assurance band.

Line 128. Simply removing data in sensor transition would not only eliminate the gradual effect of sensor degradation but also the valuable information of NDVI anomaly. Note the eruption of Pinatubo (1991) and the transition of AVHRR2 and AVHRR3 (around 2000) are not accounted for.

Line 131 & Figure A1. Explain the reason why some regions experience lower data availability. How does the data availability affect the evaluation of NDVI products and AusENDVI accuracies?

Table 1. Please provide the temporal resolution of the datasets.

Line 137. Why not use existing MODIS NDVI products (MOD13Q1, MOD13C1, etc.)? It looks like AusENDVI and NDVIpku are based on different MODIS products. Will be the difference reflected in the evaluation of NDVIpku?

Line 141. How are standardized anomalies calculated?

Line 146. More details are needed for the outperformance of GBM. For example, are all the models optimized in parameters?

Line 152-153. “…in the heavily forested regions where there was little to no agreement between NDVIMCD43A4 and NDVIAVHRR…”. How was pixel quality considered in calculating agreement?

Line 155. Why is longitude not included? Give more details on NDVIMCD43A4 summary percentiles.

Line 178. Please list the hyperparameter values used.

Line 180. In addition to absolute error, a measure of error that reflects the relative error is also needed. Such a measure is particularly important for dense vegetation.

Line 185. How are the long gaps spatially and temporally distributed, particularly for dense vegetation?

Line 191-192. What do you mean by methods in the bracket?

Line 198-199. Linear temporal interpolation may under or over-estimate values for seasonal peaks or valleys or other abrupt signals.

Line 206-207. Why is not WCF used as a feature in data harmonization but in synthesis?

Line 219. Will there be any issue related to the calculation of phenology when up-sampling from monthly to two-week intervals?

Line 238-239. How was the comparison made if there are data gaps brought by, for example, clouds? What if there are insufficient valid data between 2000 and 2013 for the calculation of CV and R?

Line 242. R2 (in the text) or R (in the figure)?

Line 256-257. Present the length of the growing season please.

Line 279-280. Solid evidence is required.

Figure 5. It would be interesting to see a similar residual NDVI map for NDVIpku.

Figure 6. Notice that the increased trend of NDVI before 2000in AVHRR-CDR disappears in AusE-clim.

Figure 7. Focus needs to be placed on vegetated, particularly densely vegetated areas. Also, in Figure 7e, is the red dot line calculated without any observation data?

Line 370. What do you mean by ‘gaps in the NDVIPKU-consolidated dataset’? Non-data or data with poor quality?

Figure 8. Note that NDVIpku is generated from a different MODIS NDVI product. A comparison between MODIS NDVI products may be beneficial.