the Creative Commons Attribution 4.0 License.
the Creative Commons Attribution 4.0 License.
| 09 Feb 2026
Finer-Resolution Long-Term Mapping of Plant Functional Types at 30-m Resolution and Corresponding Leaf Area Index for Earth System Modeling
Abstract. Land surface data describe the heterogeneity of the terrestrial surface and serve as fundamental input for earth system models. Under ongoing climate change and increasing intensity of human activities, land surface data are required to quantify the impact of land use and land cover change (LULCC) and their contribution to regional and global climate. Plant functional type (PFT) and PFT-specific leaf area index (PFT LAI) characterize land surface and vegetation canopy attributes and are essential model inputs. However, uncertainties associated with current derivation methods may limit their application in earth system models. Moreover, long-term, high-resolution datasets of PFT distributions and PFT LAI remain scarce for fine-scale simulations. To address these gaps, we derived a global 30 m PFT map (PFT30) spanning 1985–2020, updated at five-year intervals prior to 2000 and annually thereafter. By integrating multiple high-resolution remote sensing products, we minimized the assumptions typically required for PFT fraction determination. Building on PFT30, we generated a monthly 500 m PFT LAI dataset for 1985–2020 by fusing PFT30 with the reprocessed MODIS C6.1 LAI and GIMMS LAI4g products, using a remote-sensing-derived phenology scheme instead of the empirical approaches commonly adopted in land surface models. Comparisons with three other hundred-meter global PFT products show that all datasets produce broadly consistent tree fractions, while short-vegetation fractions differ substantially; PFT30 shows better agreement with site observations. Compared with empirical schemes, the new PFT LAI dataset can better distinguish short vegetation types such as grasses, shrubs and crops, because it captures realistic phenological variations directly from remote sensing. This long-term, high-resolution PFT map and the associated PFT LAI data provide a finer representation of land surface characteristics and can be applied in land surface and earth system modeling from regional to global scales.
- Preprint
(3431 KB) - Metadata XML
- BibTeX
- EndNote
Status: closed
-
RC1: 'Comment on essd-2026-21', Anonymous Referee #1, 13 Apr 2026
- AC1: 'Reply on RC1', Wanyi Lin, 26 May 2026
-
RC2: 'Comment on essd-2026-21', Anonymous Referee #2, 12 Jun 2026
---title: "Review of ESSD-2026-21"---# Overall impressionThe authors are proposing a new product of PFTs at 30m accompanied by a disaggregated LAI per PFT. These result from a combination of other products that have not been produced for this purpose. The authors then compare this with other products and validate it with ground data that is itself not meant for this purpose. Furthermore, althought the authors claim that the innovation is the high spatial resolution of 30m, the paper never illustrates the performance of the product at 30m by showing how it effectively represents different landscapes at that scale (comparing it with satellite imagery for instance). There seems to be no consideration for the real quality and fitness-for-purpose of these products, nor whether the design choices or the dependencies from different platforms do not generate artefacts that might influence the results of the proposed combined products. Regarding the LAI, it seems the authors do not realise it is derived by modelling of the radiative transfer, and that this come with some assumptions that prevent simple linear extrapolations. More precisely, LAI is known to behave very non-linearly with scale, while it is here considered it can be dealt with linearly. All these points (and others) prevent me from recommending publication in ESSD in the present form. While I agree that there would be a need for such a product, I do not think this approach is sufficiently mature and I believe it will mislead users into thinking that the intrinsic problem is resolved when it is not.# Detailed commentsL78: The authors state: "Although many grid-scale LAI products are available from remote sensing, ecosystem models, or machine learning,they cannot be directly applied in PFT-based modeling and must be converted into PFT-specific LAI". Why? why would a grid based LAI not be suitable for PFTs so long as the LAI comes from a sufficintly fine grid? This requires more elaboration.L79: What is the difference between "PFT LAI" and normal LAI over a grid in which there is only that PFT?L106: how did you define the "best quality"? what are the criteria?L206: what justifies this linear resampling? why would LAI be linearly structured within the landscape? it very likely is not in most landscapesL216: what is a hemispheric LUT? why should it be hemispheric, and not finer or adapted to climate?L225: how is SAI computed? it is not clearly mentioned. It should be made compatible with the assumptions behind the radiative transfer modelling used for the LAI derivationL235: Fig2 is only marginally useful that general patterns have not been perturbed, but this is not expected since they are based on land cover. so this, at best, should be in supplementary material. What is important is to show how patterns are plausible at fine spatial scale.L254: how are you controlling that these changes do not come from the classification error from one year to the next (i.e. during a drier year a place is classified as a desert instead of a dryland)? These are known to occur and represent a lot of "apparent" change which is in reality just an artefact on single year classifications.L257: how did you control that these are not partly affected by changes in sensor?L306: Why are such large areas of Siberia covered in "shrubs" when we clearly know this is composed of larch (Decidious needleleaf trees)? THe only map that seems to show things correctly is ESA_global. This suggests there is a major issue with the product proposed (if such a clear signal is missing completely).L330: How about instead a proper validation made with data specifically collected for the purpose of validating a mapped product, instead of some ancillary data collected for a network (FLUXNET) that has an entirely different purpose?Citation: https://doi.org/
10.5194/essd-2026-21-RC2 -
RC3: 'Comment on essd-2026-21', Anonymous Referee #3, 27 Jun 2026
The manuscript presents a global 30 m plant functional type (PFT) product and an associated PFT-specific leaf area index (PFT LAI) product for Earth system modelling. The topic is relevant, and consistent long-term information on PFT fractions and LAI would be useful for land-surface, climate, ecological and biogeochemical modelling. However, in its present form I have serious concerns about the methodological basis of the dataset, the interpretation of spatial resolution, and the strength of the validation.
My main concern is that the manuscript does not convincingly demonstrate how the 30 m information is retrieved. Several input datasets and processing steps operate at coarser spatial scales. In several cases, coarser information appears to be assigned to, interpolated to, or allocated among 30 m grid cells. The method used to derive the 30 m dataset therefore needs to be explained in more detail. It should be made clear which information is directly observed at 30 m, which information is transferred from coarser datasets, which assumptions are used in the allocation procedure, and what level of independent spatial information is actually available at the nominal 30 m grid scale.
This distinction is fundamental. A product stored on a 30 m grid is not necessarily a product with validated 30 m information content. The manuscript should therefore distinguish much more clearly between nominal grid resolution, input-data resolution, and effective spatial information content. Without this clarification, the manuscript risks
The validation does not sufficiently resolve these concerns. The validation of PFT30 is based on a limited number of flux-tower sites from Shi et al. Although this dataset is useful for site-scale benchmarking, it is not sufficient to robustly validate a global 30 m product. The number of usable sites with fractional PFT-cover information is limited, the spatial distribution is strongly uneven, and important regions and ecosystem types are underrepresented. In addition, flux-tower observations represent dynamic footprints rather than individual 30 m pixels. Such data can therefore provide a useful plausibility check, but not a strong validation of pixel-level 30 m spatial detail unless footprint effects, local heterogeneity and aggregation scale are explicitly addressed.
The number of validation samples is very limited. The manuscript uses only 53 sites with fractional PFT-cover information. After aggregation into tree cover, short vegetation and bare soil, and after considering the uneven availability of bare-soil samples, the effective sample size per validation target is even smaller. This is not sufficient to support a strong global validation claim for a product that covers all major biomes and land-cover transitions from 1985 to 2020. The underlying Shi et al. dataset is concentrated mainly in North America, Europe and Australia, with very sparse representation in Asia and Africa. This leaves large regions and important ecosystems underrepresented, including tropical Africa, large parts of Asia, drylands, mountain systems, wetlands, highly fragmented agricultural mosaics, and many regions with rapid land-use change. The validation therefore cannot demonstrate global robustness.
Moreover, the validation is also weakened by the aggregation of detailed PFT classes into broad categories such as tree cover, short vegetation and bare soil. This may obscure important errors among individual PFTs, such as shrub versus grass, C3 versus C4 grass, evergreen versus deciduous vegetation, or different tree functional types. For the PFT LAI product, comparisons with an empirical phenology scheme or with MODIS-derived LAI in a regional case study may demonstrate consistency with existing products or expected seasonal behaviour, but they do not constitute independent validation of PFT-specific LAI.
Overall, the current validation should be described as a limited site-scale comparison or plausibility assessment, not as a robust validation of a global 30 m PFT product or a global PFT LAI product. The conclusions should be revised accordingly.
I therefore recommend rejection in its present form, or at least major revision with substantially revised framing. The authors should explain the retrieval of the 30 m dataset in more detail, distinguish nominal grid resolution from effective spatial information content, avoid implying that all 30 m variability represents independently retrieved information, provide a much stronger uncertainty assessment, and substantially improve the validation strategy.
Minor comments
- The title and abstract should more clearly distinguish between the 30 m PFT product and the 500 m PFT LAI product. The current wording may give the impression that both products provide 30 m information.
- The uncertainty propagation from input land cover, tree cover, bare soil, climate-zone data, LAI products and allocation rules should be presented more systematically.
- Equation (1) should be explained more clearly, including whether and how the procedure preserves grid-scale LAI.
- The validation should report performance separately by biome, continent, PFT group and landscape heterogeneity, where sample size allows.
- The manuscript should include a clear “recommended use and limitations” paragraph explaining suitable and unsuitable applications, especially for local-scale use at 30 m.
- The reference to Shi et al. is incomplete. The manuscript cites Shi et al. for the flux tower site attribute dataset, but the full reference appears to be missing from the reference list. The complete reference should be added:
Shi, J., Yuan, H., Lin, W., Dong, W., Liang, H., Liu, Z., Zeng, J., Zhang, H., Wei, N., Wei, Z., Zhang, S., Liu, S., Lu, X., and Dai, Y.: A flux tower site attribute dataset intended for land surface modeling, Earth System Science Data, 17, 117–134, 2025.
Citation: https://doi.org/10.5194/essd-2026-21-RC3
Status: closed
-
RC1: 'Comment on essd-2026-21', Anonymous Referee #1, 13 Apr 2026
This manuscript presents the development of a global 30-m resolution Plant Functional Type (PFT) fractional cover dataset (PFT30) and a corresponding 500-m monthly PFT-specific Leaf Area Index (LAI) product, both spanning the period 1985–2020. The core objective is to address the scarcity of long-term, high-resolution land surface data required to quantify the impacts of land use and land cover change in Earth System Models. The authors integrate multiple high-resolution remote sensing products to minimize the assumptions typically required for PFT fraction derivation. Furthermore, the study generates a PFT-specific LAI dataset using a remote-sensing-derived phenology scheme, contrasting with the empirical approaches commonly adopted in land surface models.
The study is highly relevant to the current needs of the Earth system modeling community, offering a potential solution to the challenge of representing land surface heterogeneity. The manuscript is well structured and comprehensive, and the data products appear valuable for the community. However, the manuscript in its current form requires some revisions, primarily concerning the clarity, robustness, and thoroughness of the methodological description.
Major comments:
- The method for generating PFTs for years prior to 2000 (1985, 1990, 1995) is described too simply. The strategy of using a look-up table... based on the PFT distribution for the reference year 2000 assumes that the relationship between LCT and PFT composition was static from 1985-2000. This is a strong assumption, especially in regions undergoing active land-use change. The pre-2000 methodology needs a much more detailed explanation. The limitations of this approach must be explicitly stated in the methodology and discussed in the results/conclusions.
- About PFT LAI Allocation Algorithm. The core algorithm for deriving PFT LAI is not sufficiently justified or explained. The logic behind using grids with >80% PFT cover as "representative" to calculate Ratio_ipft is sound in principle but its implementation is unclear. Eq. (1) is not explained. Does this method preserve the grid-level LAI? (It appears it does, as ∑(Frac_ipft * PFTLAI_ipft) = GridLAI). Provide a more rigorous derivation or citation for Eq. (1). Include a statement confirming that the method is mass-conservative with respect to the input GridLAI. Meanwhile, the symbolic expressions here are not concise enough and can be optimized.
Minor Comments:
- In Section 2.1.7, the manuscript states that LAI4g and MODIS LAI can be combined to form a continuous long-term record. However, it remains unclear how the consistency and continuity between the two datasets are ensured in practice, particularly around the transition period. A more detailed explanation would improve confidence in the combined dataset.
- The spatial interpolation method used to resample LAI4g to 500 m resolution should be explicitly stated. The temporal aggregation from 15-day to monthly resolution should also be clarified.
- The use of 5-year interval PFT maps may introduce uncertainties in regions with rapid land cover change, which could be briefly acknowledged.
- Section 2.2.1, Step 1: "Crop PFT fraction is seen as the vegetated area in the grid if the target grid was classified as crop LCT, therefore its fraction is assigned as the remaining fraction of bare cover." This sentence is grammatically awkward and logically ambiguous. Please rephrase for clarity.
- Section 2.2.1, Step 2: It states that the shrub/grass ratio is derived from GLC_FCS30D within 1km windows. What is done for 30m pixels if the 1km window contains no relevant LCT? A brief note on the handling of edge cases would be helpful.
- Lines 195-196: the statement "The fractional cover of non-vegetated types was assigned as either 0% or 100% based on the GLC_FCS30D product" could be clearer.
- Figure 7 subplots lack detailed captions; please specify which panel corresponds to which comparison for improved readability.
- There are several instances of minor grammatical errors that require proofreading. Examples: "the roubustness" (line 65, typo); "lifeforms ecosystem" (line 43, should be "lifeform ecosystems" or "ecosystems with mixed lifeforms"); "photosynthetic pathways (e.g. C3 and C4)" (line 140, "e.g." is unnecessary).
- The study fuses 30-m PFT data with 500-m MODIS LAI. The abstract mentions a "monthly 500 m PFT LAI dataset." The methodology for upscaling the 30-m PFT fractions to match the 500-m LAI grid should be explicitly stated.
- Terminology should be used consistently. For example, in line 371, "PFT LAI values" is used, whereas "PFT LAI" is used elsewhere in the manuscript.
- The text mentions "Reprocessed MODIS v6.1 LAI" and "MODIS C6.1 LAI". Ensure the dataset version is consistent throughout the manuscript.
Citation: https://doi.org/10.5194/essd-2026-21-RC1 - AC1: 'Reply on RC1', Wanyi Lin, 26 May 2026
-
RC2: 'Comment on essd-2026-21', Anonymous Referee #2, 12 Jun 2026
---title: "Review of ESSD-2026-21"---# Overall impressionThe authors are proposing a new product of PFTs at 30m accompanied by a disaggregated LAI per PFT. These result from a combination of other products that have not been produced for this purpose. The authors then compare this with other products and validate it with ground data that is itself not meant for this purpose. Furthermore, althought the authors claim that the innovation is the high spatial resolution of 30m, the paper never illustrates the performance of the product at 30m by showing how it effectively represents different landscapes at that scale (comparing it with satellite imagery for instance). There seems to be no consideration for the real quality and fitness-for-purpose of these products, nor whether the design choices or the dependencies from different platforms do not generate artefacts that might influence the results of the proposed combined products. Regarding the LAI, it seems the authors do not realise it is derived by modelling of the radiative transfer, and that this come with some assumptions that prevent simple linear extrapolations. More precisely, LAI is known to behave very non-linearly with scale, while it is here considered it can be dealt with linearly. All these points (and others) prevent me from recommending publication in ESSD in the present form. While I agree that there would be a need for such a product, I do not think this approach is sufficiently mature and I believe it will mislead users into thinking that the intrinsic problem is resolved when it is not.# Detailed commentsL78: The authors state: "Although many grid-scale LAI products are available from remote sensing, ecosystem models, or machine learning,they cannot be directly applied in PFT-based modeling and must be converted into PFT-specific LAI". Why? why would a grid based LAI not be suitable for PFTs so long as the LAI comes from a sufficintly fine grid? This requires more elaboration.L79: What is the difference between "PFT LAI" and normal LAI over a grid in which there is only that PFT?L106: how did you define the "best quality"? what are the criteria?L206: what justifies this linear resampling? why would LAI be linearly structured within the landscape? it very likely is not in most landscapesL216: what is a hemispheric LUT? why should it be hemispheric, and not finer or adapted to climate?L225: how is SAI computed? it is not clearly mentioned. It should be made compatible with the assumptions behind the radiative transfer modelling used for the LAI derivationL235: Fig2 is only marginally useful that general patterns have not been perturbed, but this is not expected since they are based on land cover. so this, at best, should be in supplementary material. What is important is to show how patterns are plausible at fine spatial scale.L254: how are you controlling that these changes do not come from the classification error from one year to the next (i.e. during a drier year a place is classified as a desert instead of a dryland)? These are known to occur and represent a lot of "apparent" change which is in reality just an artefact on single year classifications.L257: how did you control that these are not partly affected by changes in sensor?L306: Why are such large areas of Siberia covered in "shrubs" when we clearly know this is composed of larch (Decidious needleleaf trees)? THe only map that seems to show things correctly is ESA_global. This suggests there is a major issue with the product proposed (if such a clear signal is missing completely).L330: How about instead a proper validation made with data specifically collected for the purpose of validating a mapped product, instead of some ancillary data collected for a network (FLUXNET) that has an entirely different purpose?Citation: https://doi.org/
10.5194/essd-2026-21-RC2 -
RC3: 'Comment on essd-2026-21', Anonymous Referee #3, 27 Jun 2026
The manuscript presents a global 30 m plant functional type (PFT) product and an associated PFT-specific leaf area index (PFT LAI) product for Earth system modelling. The topic is relevant, and consistent long-term information on PFT fractions and LAI would be useful for land-surface, climate, ecological and biogeochemical modelling. However, in its present form I have serious concerns about the methodological basis of the dataset, the interpretation of spatial resolution, and the strength of the validation.
My main concern is that the manuscript does not convincingly demonstrate how the 30 m information is retrieved. Several input datasets and processing steps operate at coarser spatial scales. In several cases, coarser information appears to be assigned to, interpolated to, or allocated among 30 m grid cells. The method used to derive the 30 m dataset therefore needs to be explained in more detail. It should be made clear which information is directly observed at 30 m, which information is transferred from coarser datasets, which assumptions are used in the allocation procedure, and what level of independent spatial information is actually available at the nominal 30 m grid scale.
This distinction is fundamental. A product stored on a 30 m grid is not necessarily a product with validated 30 m information content. The manuscript should therefore distinguish much more clearly between nominal grid resolution, input-data resolution, and effective spatial information content. Without this clarification, the manuscript risks
The validation does not sufficiently resolve these concerns. The validation of PFT30 is based on a limited number of flux-tower sites from Shi et al. Although this dataset is useful for site-scale benchmarking, it is not sufficient to robustly validate a global 30 m product. The number of usable sites with fractional PFT-cover information is limited, the spatial distribution is strongly uneven, and important regions and ecosystem types are underrepresented. In addition, flux-tower observations represent dynamic footprints rather than individual 30 m pixels. Such data can therefore provide a useful plausibility check, but not a strong validation of pixel-level 30 m spatial detail unless footprint effects, local heterogeneity and aggregation scale are explicitly addressed.
The number of validation samples is very limited. The manuscript uses only 53 sites with fractional PFT-cover information. After aggregation into tree cover, short vegetation and bare soil, and after considering the uneven availability of bare-soil samples, the effective sample size per validation target is even smaller. This is not sufficient to support a strong global validation claim for a product that covers all major biomes and land-cover transitions from 1985 to 2020. The underlying Shi et al. dataset is concentrated mainly in North America, Europe and Australia, with very sparse representation in Asia and Africa. This leaves large regions and important ecosystems underrepresented, including tropical Africa, large parts of Asia, drylands, mountain systems, wetlands, highly fragmented agricultural mosaics, and many regions with rapid land-use change. The validation therefore cannot demonstrate global robustness.
Moreover, the validation is also weakened by the aggregation of detailed PFT classes into broad categories such as tree cover, short vegetation and bare soil. This may obscure important errors among individual PFTs, such as shrub versus grass, C3 versus C4 grass, evergreen versus deciduous vegetation, or different tree functional types. For the PFT LAI product, comparisons with an empirical phenology scheme or with MODIS-derived LAI in a regional case study may demonstrate consistency with existing products or expected seasonal behaviour, but they do not constitute independent validation of PFT-specific LAI.
Overall, the current validation should be described as a limited site-scale comparison or plausibility assessment, not as a robust validation of a global 30 m PFT product or a global PFT LAI product. The conclusions should be revised accordingly.
I therefore recommend rejection in its present form, or at least major revision with substantially revised framing. The authors should explain the retrieval of the 30 m dataset in more detail, distinguish nominal grid resolution from effective spatial information content, avoid implying that all 30 m variability represents independently retrieved information, provide a much stronger uncertainty assessment, and substantially improve the validation strategy.
Minor comments
- The title and abstract should more clearly distinguish between the 30 m PFT product and the 500 m PFT LAI product. The current wording may give the impression that both products provide 30 m information.
- The uncertainty propagation from input land cover, tree cover, bare soil, climate-zone data, LAI products and allocation rules should be presented more systematically.
- Equation (1) should be explained more clearly, including whether and how the procedure preserves grid-scale LAI.
- The validation should report performance separately by biome, continent, PFT group and landscape heterogeneity, where sample size allows.
- The manuscript should include a clear “recommended use and limitations” paragraph explaining suitable and unsuitable applications, especially for local-scale use at 30 m.
- The reference to Shi et al. is incomplete. The manuscript cites Shi et al. for the flux tower site attribute dataset, but the full reference appears to be missing from the reference list. The complete reference should be added:
Shi, J., Yuan, H., Lin, W., Dong, W., Liang, H., Liu, Z., Zeng, J., Zhang, H., Wei, N., Wei, Z., Zhang, S., Liu, S., Lu, X., and Dai, Y.: A flux tower site attribute dataset intended for land surface modeling, Earth System Science Data, 17, 117–134, 2025.
Citation: https://doi.org/10.5194/essd-2026-21-RC3
Data sets
Finer-Resolution Long-Term Mapping of Plant Functional Types at 30-m and 500-m Resolution for Earth System Modeling Wanyi Lin and Hua Yuan https://doi.org/10.5281/zenodo.18197461
Finer-Resolution Long-Term Plant Functional Types-specific Leaf Area Index at 500-m Resolution for Earth System Modeling Wanyi Lin and Hua Yuan https://doi.org/10.5281/zenodo.18113489
Viewed
| HTML | XML | Total | BibTeX | EndNote | |
|---|---|---|---|---|---|
| 407 | 226 | 30 | 663 | 41 | 49 |
- HTML: 407
- PDF: 226
- XML: 30
- Total: 663
- BibTeX: 41
- EndNote: 49
Viewed (geographical distribution)
| Country | # | Views | % |
|---|
| Total: | 0 |
| HTML: | 0 |
| PDF: | 0 |
| XML: | 0 |
- 1
This manuscript presents the development of a global 30-m resolution Plant Functional Type (PFT) fractional cover dataset (PFT30) and a corresponding 500-m monthly PFT-specific Leaf Area Index (LAI) product, both spanning the period 1985–2020. The core objective is to address the scarcity of long-term, high-resolution land surface data required to quantify the impacts of land use and land cover change in Earth System Models. The authors integrate multiple high-resolution remote sensing products to minimize the assumptions typically required for PFT fraction derivation. Furthermore, the study generates a PFT-specific LAI dataset using a remote-sensing-derived phenology scheme, contrasting with the empirical approaches commonly adopted in land surface models.
The study is highly relevant to the current needs of the Earth system modeling community, offering a potential solution to the challenge of representing land surface heterogeneity. The manuscript is well structured and comprehensive, and the data products appear valuable for the community. However, the manuscript in its current form requires some revisions, primarily concerning the clarity, robustness, and thoroughness of the methodological description.
Major comments:
Minor Comments: