the Creative Commons Attribution 4.0 License.
the Creative Commons Attribution 4.0 License.
| 02 Mar 2026
Extended global terrestrial evapotranspiration and gross primary production dataset from 1982 to near present
Abstract. The Penman–Monteith–Leuning (PML) model is a widely recognized diagnostic framework for estimating coupled terrestrial evapotranspiration (ET) and gross primary production (GPP). To address the critical need for high-fidelity, long-term, and near-present eco-hydrological records, we developed the PML-V2.2 dataset, spanning from 1982 to 2024. Driven by observation-constrained Multi-Source Weighted-Ensemble Precipitation (MSWEP) and Multi-Source Weather (MSWX) meteorological variables, the dataset comprises three complementary products: (1) PML-V2.2a, an 8-day 500 m MODIS-based product (2000–2024) optimized for near-present monitoring; (2) PML-V2.2b, a half-month 0.1° AVHRR-based product (1982–2020) anchoring long-term climate attribution; and (3) PML-V2.2c, a consolidated half-month 0.1° record integrating the former two for seamless 43-year continuity (1982–2024). Our methodological framework features an expanded bottom-up calibration using 208 flux sites (~1400 site-years) across various plant functional types (PFT) and a refined parameterization that explicitly distinguishes between irrigated and rainfed croplands. This distinction effectively mitigated systematic biases in agricultural regions, reducing ET and GPP estimation errors by 8.7 % and 16.2 %, respectively. Performance evaluation reveals high accuracy across PFTs (cross-validation Nash-Sutcliffe Efficiency, NSE > 0.60, absolute bias < 5 %), while top-down water-balance validation across 56 large river basins during 1982–2016 and 152 basins during 2003–2020 confirms exceptional reliability (NSE: 0.89–0.91). The MODIS-based (V2.2a) and AVHRR-based (V2.2b) products exhibit high statistical and spatial agreement during their overlapping period (NSE = 0.90 and 0.79 for annual ET and GPP anomalies), ensuring a seamless transition across satellite epochs. Based on the consolidated PML-V2.2c dataset, global terrestrial annual ET and GPP during 1982–2024 are estimated at 65.8 × 103 km3 yr⁻1 (with 58.0 % from transpiration) and 143.0 PgC yr⁻1, respectively. Long-term analysis reveals significant (p < 0.01) increasing trends in GPP (0.315 PgC yr⁻2) and ET (0.019 × 103 km3 yr⁻2) during 1982–2024, where rapid growth in GPP and water use efficiency is partially offset by CO2-induced physiological water savings. By bridging the gap between satellite epochs, PML-V2.2 provides an internally consistent long-term global dataset for hydrology, ecology, and other Earth science studies. The dataset is freely accessible, with the 500 m resolution PML-V2.2a product hosted on Google Earth Engine, and all 0.1° PML-V2.2a/b/c versions archived at the National Tibetan Plateau Data Center under https://doi.org/10.11888/Terre.tpdc.303314 (Xu et al., 2026).
- Preprint
(16760 KB) - Metadata XML
- BibTeX
- EndNote
Status: final response (author comments only)
-
RC1: 'Comment on essd-2026-94', Anonymous Referee #1, 23 Apr 2026
-
AC1: 'Reply on RC1', Zhenwu Xu, 22 May 2026
# Response to Reviewer
We sincerely thank the reviewer for the thorough evaluation and the highly positive feedback on our updated PML dataset and manuscript. Your constructive suggestions are extremely valuable for enhancing the transparency, utility, and overall readability of our work. We will fully incorporate these suggestions into the revised manuscript. Below is our detailed, point-by-point response to your comments.
# Response to General Comments
1.The meaning of “near present”.
We completely agree that clearly defining the temporal coverage is crucial for data users. To avoid any ambiguity caused by varying data latencies, we will explicitly specify the exact cutoff month. Unlike the “to present” MODIS products, our dataset is updated annually (when forcing is available) with a latency of approximately six months. Therefore, the current version is updated through December 2024. We will revise the Abstract and the Data Availability sections to clearly state this timeline. This clarification will provide users with precise expectations regarding the near-present coverage of the product.2. Flux tower details and data repository.
This is a highly valuable suggestion that aligns perfectly with the principles of open science. To make the data more accessible and easier to reuse compared to a static supplementary table, we will directly upload the comprehensive attributes of all 208 flux tower sites (including their locations, vegetation types, and available data periods) into a permanent repository (e.g., Zenodo). Furthermore, we will also make the site-level model performance metrics and the underlying source data for key figures, especially those illustrating cross-product uncertainties, publicly available in the same repository.3. Comparison with other global products.
We agree that contextualizing our results within the broader literature will greatly strengthen the Discussion section. We will add a dedicated paragraph to compare the PML-V2.2 global mean estimates and their long-term trends against several other mainstream global products (e.g., GLEAM, SiTHv2, etc.). This can be shown as supplementary figures. Adding this comparison will help readers better understand how our new product aligns with or differs from the current scientific consensus.4. Differences in global mean values in Figure 12.
Thank you for your careful observation regarding Figure 12. We acknowledge that while the interannual variabilities and trends match closely, there is a slight discrepancy in the mean values between the b version and the a/c versions. Based on our analysis, this difference is primarily attributed to the slight differences in spatial averages in forcing data. Specifically, this is caused by systematic differences in the mean LAI magnitude between the GIMMS LAI4g and the smoothed MODIS LAI, combined with the scaling effects of simulating at a coarser spatial resolution. We will add a brief, data-supported discussion addressing this specific discrepancy in the revised text, with a new supplementary figure to help explain that.5. Comparison with PML-V2 China.
This is an excellent point. Acknowledging the previously released regional dataset, PML-V2 China (He et al., 2022), will definitely benefit users focusing on regional studies. We will add a brief comparison between our new global dataset and the regional PML-V2 China dataset in the Discussion section. This comparison will focus on highlighting regional differences and evaluating water balance closure. This addition will effectively clarify the suitability of our global product for regional-scale research in China, especially for applications requiring data from the most recent years.6. Fair comparison in Figure A5.
You raise a very valid concern regarding the comparison of water-balance-based ET in Figure A5. Comparing products that have varying spatial coverage over deserts or water bodies can indeed introduce biases. To address this and ensure a strict and fair comparison, we will establish a unified spatial mask. This mask will explicitly account for barren desert regions and exclude evaporation from inland water bodies to standardize the evaluation area across all participating datasets. We will also expand the methodology section to provide detailed descriptions of how these specific datasets were masked and processed.# Response to Specific Comments
We sincerely appreciate your meticulous review of the text. All the specific comments and grammatical corrections will be fully adopted and revised accordingly in the manuscript.Citation: https://doi.org/10.5194/essd-2026-94-AC1
-
AC1: 'Reply on RC1', Zhenwu Xu, 22 May 2026
-
RC2: 'Comment on essd-2026-94', Oscar Manuel Baez Villanueva, 26 May 2026
The article titled “Extended global terrestrial evapotranspiration and gross primary production dataset from 1982 to near present” presents the new version of the PML dataset (PML-V2.2). The dataset comprises three complementary products: (i) PML-V2.2a, an 8-day 500 m MODIS-based dataset (2000–2024); (ii) PML-V2.2b, a half-monthly 0.01deg AVHRR-based product (1982–2020) optimised for near-present monitoring; and (iii) PML-V2.2c, a half-monthly 0.1deg record consolidating the a and b products (1982–2024). Parameter optimisation was performed through a leave-one-out cross-validation procedure for each plant functional type (PFT). The manuscript is very well written, clear, and aligned with the scope of the journal. Additionally, I believe that this dataset is of particular interest to the research community! I hope the following comments and suggestions help further strengthen the manuscript.
General coments:
It would be interesting to mention in the abstract whether the product is expected to be updated regularly (e.g., annually). This information could further help the uptake of the dataset from new users.
The authors present PML-V2.2b as a product optimised for near-present monitoring. It would therefore be valuable to extend the dataset through 2025 (the last complete year) to further demonstrate this capability. In that case, the consolidated PML-V2.2c product could span 1982–2025. This would further strengthen the statement made in L66–67 regarding the production lag affecting many biophysically consistent datasets.
It would be worth considering the use of the Kling–Gupta Efficiency (KGE) instead of the Nash–Sutcliffe Efficiency (NSE). The KGE decomposes performance into correlation, bias, and variability components, providing additional diagnostic insight and facilitating comparison across regions and climatic conditions.
Out of curiosity, is there a particular reason for using MSWEP V2.8 instead of the more recent MSWEP V3.16?
Could the authors elaborate on how rainy days were treated in the case of the data from theflux sites?
Is there a sensitivity analysis associated with the optimised parameters presented in Table 2? It would be particularly interesting to assess whether parameter sensitivity varies across PFTs.
In L218, the authors mention that monthly CO2 concentrations from NOAA were used. How were these data temporally disaggregated to the 8-day and half-monthly scales?
Were the three models optimised independently? If so, it would be interesting to summarise the best-performing parameters for each version. If not, a brief discussion on the expected impact of using different forcing datasets across the three products (particularly between versions a and b) would be useful. It appears that the optimisation was performed at the 8-day temporal scale and therefore primarily for version a; it would be good to state this explicitly in the manuscript.
I appreciate the effort made by the authors to carefully merge AVHRR and MODIS observations in order to minimise discontinuities and artificial biases in long-term trends. It would be interesting to add a few sentences about the pixel-scale bidirectional consolidation process and the reverse-scaling procedure. In addition, why was the 2001–2003 period selected for the consolidation and not a longer period?
It would be very interesting to the readers to include performance metrics for all three dataset versions. Additionally, the consistency of the consolidation could be assessed at the half-monthly scale over the overlapping period across all products.
Besides the visualisation of global patterns, it would be very interesting to add a figure showcasing time series of the three products in comparison with in situ data.
Specific coments:
L22: The authors mention that PML-V2.2c exhibits “exceptional reliability”. It would be useful to specify relative to which products or benchmarks this statement is made. A comparison against other products would be very interesting!
L50–57: Please add references for the datasets mentioned.
L50: “Temporal span” or “record length” may fit better here than “temporal depth”.
L89–90: Please include the versions and references of the datasets.
Table 1: For GLEAM4, in the “key feature” column, “evaporative stress” would be more appropriate than “water stress”. In addition, the temporal coverage should be updated to 1980–2025.
L107: If I am not mistaken, bare soil evaporation should be denoted as Es instead of Eis to remain consistent with Eq. 2 and the following explanation.
L148: Perhaps “estimates” would be more appropriate than “observations” in this context.
L382–385: It would be very interesting to compare all dataset versions over the overlapping period, including metrics such as mean annual global ET, trends, and the partitioning of evaporation into its three main components, as discussed in these lines.
Citation: https://doi.org/10.5194/essd-2026-94-RC2 -
AC2: 'Reply on RC2', Zhenwu Xu, 27 May 2026
# Response to Oscar M. Baez-Villanueva (referee #2)
We sincerely thank the reviewer for the thorough evaluation and the highly positive feedback on our updated PML dataset and manuscript. Your constructive suggestions are extremely valuable for enhancing the transparency, utility, and overall readability of our work. We will fully incorporate these suggestions into the revised manuscript. Below is our detailed, point-by-point response to your comments.
# Response to General Comments
1. Data update frequency and temporal extension.
We agree that clarifying the update schedule is important for users. We will mention the annual update cycle in the Abstract. Regarding the extension to 2025, we are currently facing a widespread challenge: the severe satellite drift of MODIS since 2022 has caused significant data degradation for 2025, particularly in tropical rainforest regions. We had identified this issue during our pre-update of PML data in early Feb. 2025, therefore data for 2025 is not released at the moment. As MODIS data alone is currently insufficient to support reliable further updates, we are transitioning to VIIRS forcing data (same spatial-temporal resolution, algorithm as MODIS). The VIIRS LAI processing is completed and without drift issues, while the Albedo data (VNP43IA3) is currently being uploaded in coordination with the GEE team.
We plan to update the dataset through 2025 for the final published version (expected by July/August). Moving forward, with GEE-based MODIS/VIIRS drivers, our annual update latency in subsequent years will ideally be reduced to two to six months. We will explicitly state this timeline and our transition strategy for subsequent annual updates in the revised manuscript.
2. Use of Kling-Gupta Efficiency (KGE).
This is an excellent suggestion. Currently, we provide NSE, R, RMSE, and Bias to capture different dimensions of model performance. However, we agree that KGE is a robust composite metric that facilitates easier cross-product comparisons. We will calculate the KGE metrics for all flux sites and include them in the revised manuscript to complement our existing evaluations.
3. Choice of MSWEP version.
We will briefly clarify our choice of MSWEP V2.8 in the text. When we initiated the data preparation and long-term simulations, MSWEP V3 had not been officially released (it is currently still under review). Our preliminary investigations indicated slight differences between V2.8 and V3, and the long-term stability of V3 remained uncertain at the time. Therefore, V2.8 was selected to maintain strict data consistency across the 43-year simulation. We will consider comprehensively updating to V3 in future versions once its stability is fully evaluated by the community.
4. Treatment of rainy days at flux sites.
We will elaborate on this in the methodology section. In this study, we did not explicitly filter out rainy days from the flux site data. The primary reason is that our model runs and evaluations are conducted at an 8-day temporal scale. At this temporally aggregated scale, the instantaneous uncertainties and noise introduced by precipitation events on eddy covariance measurements are largely smoothed out. This minimizes the necessity for strict daily or hourly rainy-day filtering, while preserving the continuity of the time series.
5. Parameter sensitivity analysis.
We will add a brief discussion regarding parameter sensitivity across different PFTs. Since a formal parameter sensitivity analysis was not included in the previous 2019 RSE paper, we will incorporate a variance-based sensitivity analysis (e.g., the Sobol or Morris method) in this updated version. This will make the parameterization of the PML model more transparent and scientifically robust.
6. Temporal disaggregation of CO2 data.
We will add a clarifying sentence in Section 2 to explain this processing. Specifically, the monthly NOAA CO2 concentrations were applied as static values for the 8-day or half-monthly steps within each respective month. Since terrestrial carbon-water coupling responds primarily to the long-term CO2 trend and seasonal cycles, the absence of high-frequency (sub-monthly) atmospheric CO2 fluctuations has a negligible impact on the simulations.
7. Model optimization and forcing impacts.
We will explicitly state in the manuscript that the parameter optimization was performed exclusively at the 8-day temporal scale using high-quality site meteorological observations and MODIS data, primarily targeting version a.
We will also clarify that versions a and b share the exact same parameterization scheme; the discrepancies arise solely from the different forcing datasets. We intentionally selected “bias-corrected” forcing datasets (MSWEP + MSWX) rather than raw reanalysis data (e.g., ERA5, MERRA2) to minimize such forcing-induced biases. Combined with the GIMMS LAI4g dataset, which has undergone extensive calibration and validation using in situ observations and Landsat imagery, this strategy effectively balances bottom-up simulations with top-down constraints, which underpins the high fidelity of our water balance validation.
8. Consolidation process and overlapping period.
We will add more details describing the pixel-scale bidirectional consolidation process. A bidirectional correction was adopted because unidirectional scaling might fail to perfectly preserve both identical means and consistent long-term trends. Furthermore, the 2001–2003 period was selected because it represents a highly stable overlapping window with high-quality data from both AVHRR and MODIS sensors. Compared to using a longer baseline (e.g., 2001–2020), this shorter, focused window avoids introducing abrupt shifts or artificial artifacts at the transition boundary (1999–2001).
Importantly, we will expand the discussion to acknowledge that any cross-sensor consolidation inherently introduces uncertainties—a challenge we will face again with our upcoming transition to VIIRS. However, without such harmonization, continuous long-term studies across different satellite eras would be impossible. While some existing products tend to remain vague regarding their cross-sensor transition strategies, bridging this gap transparently is precisely the core value of our consolidated version “c”, which is particularly valuable for placing recent anomalies within a long-term historical context. We are committed to ensuring the methodological soundness of this processing, providing robust long-term estimates, and openly documenting the associated uncertainties in the revised manuscript.
9. Performance metrics and time series comparisons.
These are highly constructive suggestions. We will include a new figure showcasing the time series of the PML-V2.2a production data directly compared with in situ flux data, and we plan to upload more comprehensive site-level plots to our Zenodo repository for transparency.
However, for versions b and c, we would like to clarify that performing a direct time-series comparison against flux towers is scientifically problematic. The spatial mismatch between a tower's footprint (typically ~1 km) and the coarse grid resolutions (0.01° and 0.1°) over heterogeneous land surfaces introduces massive uncertainties. This spatial footprint mismatch is also the primary reason we avoid direct horizontal site-level comparisons with other coarse-resolution global products. We will explicitly discuss this limitation in the revised manuscript.
# Response to Specific Comments
We sincerely appreciate your meticulous review of the text. All the specific comments, including dataset references, terminology adjustments, removing overstated or non-standard descriptions, and updating Table 1, will be fully adopted. We will also include brief cross-product comparisons in the Discussion section as suggested.Citation: https://doi.org/10.5194/essd-2026-94-AC2
-
AC2: 'Reply on RC2', Zhenwu Xu, 27 May 2026
-
RC3: 'Comment on essd-2026-94', Anonymous Referee #1, 27 May 2026
The author had revised this manuscript accroding to my comments and can be accepted for publication at present.
Citation: https://doi.org/10.5194/essd-2026-94-RC3 -
AC3: 'Reply on RC3', Zhenwu Xu, 27 May 2026
We sincerely thank the reviewer for the positive evaluation and for recommending our manuscript for publication. We will carefully revise the manuscript according to the changes detailed in our previous response.
Citation: https://doi.org/10.5194/essd-2026-94-AC3
-
AC3: 'Reply on RC3', Zhenwu Xu, 27 May 2026
-
RC4: 'Comment on essd-2026-94', Anonymous Referee #1, 28 May 2026
It's a good revised manuscript and can be accepted for publication.
Citation: https://doi.org/10.5194/essd-2026-94-RC4
Data sets
PML-V2.2: Global terrestrial evapotranspiration and gross primary production dataset from 1982 to near present Zhenwu Xu, Yongqiang Zhang, and Dongdong Kong https://doi.org/10.11888/Terre.tpdc.303314
Viewed
| HTML | XML | Total | BibTeX | EndNote | |
|---|---|---|---|---|---|
| 1,043 | 432 | 43 | 1,518 | 38 | 58 |
- HTML: 1,043
- PDF: 432
- XML: 43
- Total: 1,518
- BibTeX: 38
- EndNote: 58
Viewed (geographical distribution)
| Country | # | Views | % |
|---|
| Total: | 0 |
| HTML: | 0 |
| PDF: | 0 |
| XML: | 0 |
- 1
The authors present a valuable update to the Penman–Monteith–Leuning (PML) model, introducing the PML-V2.2 dataset to provide a long-term (1982–2024), near-present global record of coupled terrestrial evapotranspiration (ET) and gross primary production (GPP). The manuscript is well-structured, clearly written, and the methodology is robust. The rigorous validation against both eddy covariance data and basin-scale water balance estimates provides high confidence in the dataset's fidelity. Overall, the development of this consolidated 43-year dataset bridging the AVHRR and MODIS epochs represents a significant contribution to the community. I have only a few minor suggestions to further enhance the transparency, utility, and readability of the manuscript before publication:
General Comments
Specific comments: