This manuscript introduces the newly generated global High-Quality LAI (HiQ-LAI) Product at 500 m/5 km and 8 days resolution from 2000 to 2022. This product was generated on the GEE platform using a well-validated Spatio-Temporal Information Composition Algorithm (STICA). The HiQ-LAI can be considered as a reprocessed and value-add version of the raw official MODIS LAI products. Evaluation results demonstrate a significant improvement compared to raw MODIS LAI in terms of RMSE and Bias, enhanced temporal stability, and superior continuity especially in equatorial regions where optical remote sensing typically struggle to achieve good performance. HiQ-LAI keeps the same data format and similar quality control information with MODIS LAI, which is very convenient for data users. For me, this paper is well organized and the new product should be useful to the community of Climate Data Record (CDR). This new version of global LAI has the potential to better replace the MODIS raw product (MOD15A2) for most applications and thus desires to be published. However, there are still some minor points that need to be modified to improve the paper. Please improve these issues as follows:

1. Authors are encouraged to employ precise terminology when addressing uncertainty and accuracy in the manuscript. According to GCOS/CEOS, accuracy is defined as the proximity between the product and the reference values (doi: 10.1016/j.envsci.2015.03.018).
2. As the manuscript said, most existing filtering methods could artificially remove land surface real disturbances (e.g., forest fire, land cover change). In such cases, how does HiQ-LAI perform?
3. In the abstract “However, MODIS LAI retrievals are calculated independently for each pixel and a specific day, resulting in high noise levels in the time series and limiting its applications.” Can the expression here be more precise, for example, in the regions of XXX.
4. Page 2, Lines 50-55: “The long time series MODIS LAI dataset has made significant contributions to landmark studies on "Greening the Earth" phenomena”, suggest replace ‘long time series’ to ‘long-time series’, ‘on "Greening the Earth" phenomena’ to ‘on the "Greening the Earth" phenomena’.
5. Page 4, Figure 1: suggest adding a schematic representation of the study area's location of Section 5.3 on this global map.
6. Page 7, Lines 190-195: “we utilized the GBOV LAI measurements from a total of 29 sites spanning from 2013 to 2021 as our ground reference LAI.”, add reference here.
7. Page 7, Lines 195-205: “Furthermore, we compared MODIS LAI and HiQ-LAI in 2021 using the BELMANIP V2.1 sites (445 in total).”, “Additionally, we used DIRECT V2.1 ground measurements in this research.” Please add reference here too.
8. Page11, Table 1: the decimal places are not consistent, suggest changing them and unifying them. Besides, if there is insufficient data to compute RMSE and R2 for a particular site, it is recommended to fill the table with ‘—’ rather than ‘0.00’.
9. Page13, Lines 260-265: “the R2 for other pure vegetation types exceeds 0.88, and B1 and B3 surpassed 0.95. The consistency of mixed pixels is also relatively high, as indicated by an RMSE of 0.42 and an R2 of 0.86. However, B5 exhibits a significant disparity, with an R2 value of 0.15.” please modify these ‘R2’ to ‘R²’.
10. Page15, Lines 285-290: the usage procedures for Theil–Sen’s slope (TS) method and the Mann-Kendall (MK) test are not sufficiently clear, please provide a more detailed description and relevant mathematical formulas for Theil–Sen’s slope (TS) method and the Mann-Kendall (MK) test in Section of the methodology.

The submitted manuscript presents HiQ-LAI, a reprocessed leaf area index (LAI) dataset based on the Moderate Resolution Imaging Spectroradiometer (MODIS) LAI product that attempts to improve spatiotemporal consistency. There is a need to reduce spatiotemporal noise in global LAI products, and the HiQ-LAI dataset appears to offer improvements in this respect, so there is obvious potential for it to be useful to the community. I was able to access the 5 km dataset on Zenodo (though files appeared to be in .tif format, rather than HDF as claimed on L449), as well as the 500 m collection in Google Earth Engine without issue. Although the dataset appears useful, I do have several concerns and requests for clarification.

Main concerns:

1. Too little detail is provided about the Spatio-Temporal Information Compositing Algorithm (STICA) used to derive the dataset. Whilst I appreciate that STICA is described in a previous publication (https://doi.org/10.1109/TGRS.2023.3264280), a brief overview of how it actually works is still needed. The current description is very superficial, outlining only the names of each step, rather than how each step is actually achieved. For example, it is not clear what the Multiple Quality Assessment (MQA) information is. Does it relate to the ‘FparLai_QC’ and ‘FparExtra_QC’ quality indicators? Likewise, how is the MQA information actually used to weight different observations in the ‘fusion strategy’?

2. Related to point 1, the authors should make clear what the difference is between the validation results they are presenting here and the validation results from the previous publication (https://doi.org/10.1109/TGRS.2023.3264280), which seem quite similar (e.g. Figure 6 of the previous publication vs. Figure 2 in this paper). How is the present study building on the previous work? Was the previous study limited in time and space, whereas this one is more comprehensive? It needs to be made clear to the reader.

3. Clarification is required regarding the choice of nearest neighbour resampling to go from 500 m to 5 km, which appears unsuitable. Surely some form of aggregation (e.g. mean value downsampling) needs to be carried out first? Otherwise, a 5 km pixel is being assigned the value of the single 500 m pixel that is closest to its centroid. That single 500 m pixel could be surrounded by pixels with very different LAI values, unless you are in an environment that is homogeneous at 5 km (which I would suggest is unheard of in terrestrial landscapes). This could be a major source of error, and I would strongly suggest the authors revisit this aspect. A proper aggregation should be easy to implement in Google Earth Engine using the ‘reduceResolution’ method (https://developers.google.com/earth-engine/guides/resample).

4. The term ‘uncertainty’ is used too loosely throughout the manuscript. For example, it’s stated on L177-178 that ‘the approach employed multiple indicators to evaluate the uncertainty for each pixel (referred to as MQA hereafter)’. If the MQA information is indeed based on categorical quality indicators, then it does not represent ‘uncertainty’. According to the International Standards Organisation’s Guide to the Expression of Uncertainty in Measurement, uncertainty ‘characterizes the dispersion of the values that could reasonably be attributed to the measurand’ (https://www.iso.org/sites/JCGM/GUM/JCGM100/C045315e-html/C045315e.html?csnumber=50461). Later in the manuscript, it’s suggested that according to validation with ground data, the HiQ-LAI dataset ‘reduces uncertainties of the raw LAI retrievals’ (L310). Again, this is improper use of the term ‘uncertainty’ – here the appropriate term is ‘error’ or ‘apparent error’ (which represents the difference between the reference value and the retrieval, see https://doi.org/10.1016/j.envsci.2015.03.018 for more details).

5. In the comparison with GBOV data, it’s stated that ‘GBOV LAI measurements’ are used, but it’s unclear whether a correction for woody material was undertaken to go from PAI to LAI. Note that the values provided by GBOV represent PAI, not LAI (https://doi.org/10.1016/j.isprsjprs.2021.02.020). If no correction was carried out, the ‘GBOV LAI’ should really be relabelled ‘GBOV PAI’, and this difference should then be considered in the discussion of the results. Related to the GBOV data, it’s also not clear whether upscaled maps or individual in situ reference measurements were used – this should be clarified.

6. The current ground-based validation is rather simple, and more information on the improvement of the HiQ-LAI dataset could be provided by reporting the validation statistics over all sites, but split by LAI magnitude, season, and biome, so we could clearly see when and where the improvements are. This would help back up the existing discussion.

7. In the introduction, the need for the HiQ-LAI dataset could be presented more clearly. Whilst I agree that the MODIS LAI product can be subject to noise, the claim needs to be backed up with specific examples and references to previous studies that have demonstrated this (for example https://doi.org/10.3390/rs12061017 and https://doi.org/10.1016/j.rse.2020.111935 demonstrate noisier temporal profiles from MODIS LAI when compared to the Copernicus Global Land Service LAI product from PROBA-V). Likewise, it’s suggested that the noise in the MODIS LAI product leads to ‘limitations on its practical applications’ – again some specific examples and references are needed here.

8. It’s stated several times that ‘MODIS LAI retrievals are calculated independently for each pixel and daily’ (e.g. L67-068, L341-342), which may be true on an internal basis. However, I understood that the C6.1 MODIS LAI products available to external users (i.e. MOD15A2H, MYD15A2H and MCD15A2H) are only available as 4-day and 8-day composites. So, some clarification (and careful wording) is needed here to avoid confusion.

9. The distinction between the results and discussion seemed somewhat arbitrary, since new results were shown in the discussion. In this case, a combined results & discussion section might make more sense. It was also a bit strange to have sections starting with figures.

Other comments:

Abstract

L19: ‘reliable validation results’ -> ‘extensive validation results’?

L35: It would be better to avoid the term ‘significantly’ unless statistical significance has been tested for – ‘substantially’ would be more appropriate. There are similar occurrences throughout the manuscript.

1. Introduction

L48: The meaning of ‘irreplaceable characteristics’ is unclear.

L64 and L66: What is the ‘specified thershold’?

L98: ‘data value-added’ -> ‘value-added data’ (here and throughout)?

L102: A reference for Google Earth Engine might be needed here (https://doi.org/10.1016/j.rse.2017.06.031).

2. Materials

L145: Perhaps the paragraph on BELMANIP V2.1 belongs in its own section, since it’s related to product intercomparison rather than validation with ground reference data.

L162: The CEOS WGCV LPV LAI good practice document should be referenced here (https://doi.org/10.5067/doc/ceoswgcv/lpv/lai.002).

3. Methods

L185-186: Rather than the quote, it might be clearer to just say ‘spatial and temporal autocorrelation’.

4. Results

L217: ‘R2’ -> ‘R²’ (here and throughout).

Figures 2 and 3: Could the ground reference and MODIS LAI uncertainties be shown as error bars?

Table 1: It might make more sense to report the statistics by biome, or at least show the biome on the table so the reader can see which sites belong to which biomes. Also, is there a reason why the bias and RRMSE are not shown here? What about the proportion of retrievals that were compliant with uncertainty requirements (e.g. those from GCOS - https://gcos.wmo.int/en/essential-climate-variables/lai/ecv-requirements or others).

L264-266: References needed here?

5. Discussion

L305: ‘unusually high noise’ – unusual compared to what?

L325: Rather than being about ‘the dense vegetation period of summer’, I suspect this is more related to your next point – the high LAI is mostly found in tropical regions where there is greater cloud cover and aerosol load.

L326: Whilst I know what is meant by ‘pixel quality is assessed based on the algorithm path’ (i.e. main vs. back-up algorithm), it might not be clear to readers who are less familiar with the ins and outs of MODIS LAI. It would also be helpful to explain how this is assessed (I am assuming you are assigning a higher quality to the main algorithm and a lower one to the backup algorithm, but this is not explicitly mentioned).

L336: ‘Equator’ -> ‘Equatorial’ (here and throughout)?

The paper introduces the High-Quality LAI (HiQ-LAI) dataset, which is an upgraded version of the existing MODIS LAI retrievals that typically suffer from high noise levels. The Spatio-Temporal Information Compositing Algorithm (STICA) was used to create this new dataset, which incorporates pixel quality information, spatio-temporal correlation, and original retrieval to provide more accurate results. The Time-series Stability (TSS) index showed that the area with smooth LAI time-series expanded significantly across the globe, especially in equatorial regions that are known to pose challenges for optical remote sensing. The HiQ-LAI dataset outperforms raw MODIS LAI in terms of continuity and consistency, both spatially and temporally, allowing for better land surface process simulation, climate modelling, and global change research. Overall, the manuscript is well-structured, provides valuable insights into vegetation dynamics across the globe, and has the potential to enhance the quality and impact of this field of research.

However, the advantages of the proposed method for solving the noise problem of the MODIS product are not adequately demonstrated in the paper. For instance, the authors state that the previous study "overlooked spatial correlation information" and that "genuine land surface LAI anomalies (e.g., caused by forest fires) may be artificially removed, even if the LAI profile appears smoother." The paper does not provide a clear explanation of how the proposed method addresses these issues.

The methods used in this study are straightforward, but some essential points are missing in the manuscript. For instance, a brief introduction of the STICA method, the TSS metric, and some key validation methods can further improve the clarity of the manuscript.

Minor comments:

Line 21: Not entirely true. See Yuan et al., for instance.

Yuan Hua; Dai Yongjiu; Xiao Zhiqiang; Ji Duoying and Shangguan Wei; Reprocessing the MODIS Leaf Area Index products for land surface and climate modelling, Remote Sensing of Environment, 2011, 115(5): 1171-1187.

Line 67: Please check the sentence.

Line 85: The author may need to justify how the previous study "overlooked spatial correlation information."

Line 197: Why did the authors choose the year 2021? Have they tested other years, such as 2015, when strong ocean modes occurred, and NH vegetation growth was notably affected?

Bastos Ana; Ciais Philippe; Park Taejin; Zscheischler Jakob; Yue Chao; Barichivich Jonathan; Myneni Ranga B.; Peng Shushi; Piao Shilong and Zhu Zaichun; Was the extreme Northern Hemisphere greening in 2015 predictable?, Environmental Research Letters, 2017, 12(4): 044016.

Line 240: R2 should be corrected. Please check the entire manuscript. For example, see also Line 261 and 263.

Line 386: Please rephrase the sentence.

Figure 7: Could the authors explain the presence of the unusual stripes observed in the northern high latitudes? Additionally, providing a difference map between the two would offer a more straightforward visual representation. Finally, the histogram indicates that the colors in the colorbar were not effectively used.

Figure 9: The authors should provide more details on how the "improvement percentage of RMSE and R2" was calculated.

Figure 10: Please label the LAI units for the corresponding colorbars.