the Creative Commons Attribution 4.0 License.
the Creative Commons Attribution 4.0 License.
| 29 Jul 2026
Development and evaluation of a crop-specific WRF-VPRM framework: Integrating dynamic phenology and high-resolution distributions for agricultural CO2 exchange in China
Abstract. Accurate quantification of CO2 fluxes in China’s agricultural ecosystems is frequently hindered by the high spatiotemporal heterogeneity of crop types and phenology, which standard modeling frameworks often oversimplify. This study develops a crop-specific WRF-VPRM framework specifically designed for China’s complex agricultural landscape by replacing the generic cropland category with three distinct modules for rice, wheat, and maize. The primary innovation lies in the integration of high-resolution daily crop distribution data and dynamic phenological information, allowing for the optimization of core VPRM parameters to reflect crop-specific photosynthetic pathways and growing seasons. Simulation results for central and eastern China reveal a national gross ecosystem exchange (GEE) and net ecosystem exchange (NEE) of 1084.7 and 779.2 TgC yr-1, respectively, with maize and rice jointly contributing over 80 % of total carbon uptake. The model’s reliability is underscored by strong correlations between simulated NEE and provincial grain yields, alongside high consistency with OCO-2 satellite XCO2 retrievals. This refined, online-coupled system provides a more granular perspective on China’s agricultural carbon budget and offers a robust modeling tool for evaluating the feedback between crop-specific carbon dynamics and regional climate change.
- Preprint
(2776 KB) - Metadata XML
- BibTeX
- EndNote
Status: open (until 23 Sep 2026)
- RC1: 'Comment on egusphere-2026-2606', Anonymous Referee #1, 25 Aug 2026 reply
-
RC2: 'Comment on egusphere-2026-2606', Sharon Gourdji, 08 Sep 2026
reply
This seems like a reasonable study of improvements to WRF-VPRM for simulating agricultural surface-atmosphere CO2 exchange over China. However, many methodological questions are left unanswered for the reader and could potentially be addressed through future modifications to the manuscript.
First, what are the main innovations of this study relative to Dayalu et al (2018)? It seems that using daily phenology and planted area is part of it, and also potentially re-calibrating the rice parameters. (However, I’m not convinced that the parameters are improved as explained below.) But it would be good to see a comparison to results from Dayalu et al to show any demonstrated improvements in crop NEE simulation and the resulting atmospheric CO2 relative to this previous version of VPRM-China.
Second, it’s not clear what resolutions are being used for the land-cover and phenological data, the meteorology (temperature and radiation) and the simulated fluxes, and any potential aggregation of fluxes for use in the transport model. It seems that at the end of the manuscript (section 3.4.2), the authors mention 9 x 9 km for the final simulated fluxes, but this could be made clear earlier in the manuscript, especially given the crop phenology dataset is at 1km. (Was this aggregated before use in the model?) It seems to me that the meteorological field resolution can be coarser than the land-use and phenology information, since these tend to vary at finer scales. Fluxes can be simulated at the finer scale and then aggregated for use with the transport model. An overall explanation of the different resolutions used in the paper would benefit the manuscript. The title also mentions “high-resolution distributions”: what do you mean by this?
Third, the comparison to yield data needs a better explanation. For one thing, the concept of the harvest index that relates biomass to yield should be explained, but also the harvesting process itself will impact net CO2 uptake and release in a relatively sudden way. Does your phenological data account for crop harvest? After harvest, the plants may continue to photosynthesize and respire but at a reduced rate. From the point of view of the atmosphere, harvested crops are stored, transported and ultimately consumed elsewhere. For comparison to atmospheric CO2, it seems relevant to take the fate of these harvested carbon pools into account.
Fourth, the discussion of how you simulated other land-use types is lacking, but important for comparison to total CO2 concentrations. Did you use the same parameters from Dayalu et al for the other non-crop land-use types? At times, it seems like you only simulated cropland fluxes, but then did you consider the fluxes from the other land uses to be zero? In particular, the discussion on lines 342 to 346 where you talk about the relative proportion of maize, wheat and rice to the “total net CO2 absorption”, with each of their proportions summing to 100%, is confusing.
Fifth, the VPRM parameter optimization in this study is weak, as the authors themselves state. In the paper, the authors state that the Dayalu et al rice parameters were derived in the same way as for grass. This is not what Tables 1and 2 in Dayalu et al shows (e.g. l = 0.0451 for grassland vs. l = 0.0583 for rice with rice derived using the Haenan tower in South Korea with the rice cropland ecotype). Also, the authors use the same parameters for corn and winter wheat as in Dayalu et al. It’s hard to believe that there are no agricultural flux towers in East Asia that you could use for parameter optimization other than the Changling station in northeast China (where rice isn’t even the dominant crop). In fact, on lines 441-443 in the manuscript, the authors mention that there are rice flux tower measurements in the Yangtze River Delta and the Huai River basin. Why not use these for parameter optimization? Also, they could use the same Haenam station in South Korea, as in Dayalu et al, with more recent years to derive rice parameters, and/ or take advantage of flux data from agricultural flux stations in other parts of the world that are at similar latitudes, e.g. in the Corn Belt of the USA and in Europe.
Sixth, another confusing aspect of the manuscript is that the authors switch between discussing NEE, GEE and RESP in units normalized by area (e.g. gC/m2/d) vs. total exchange (e.g. TgC/ yr) over all planted area. It becomes unclear if the differences between modeled values and observations are due to planted area estimates or the actual carbon exchange rates simulated by VPRM itself. Also, it becomes difficult to compare the intensity of net uptake when comparing numbers with different planted area (e.g. maize vs. rice).
Other specific comments are as follows:
- Consider creating a supplemental material for detailed technical information.
- Try making your figures wider to use the full width of the page. Many of them are hard to see and interpret now. Also, the colorscales are frequently confusing. Consider using green for more net uptake and/ or vegetation and red for sources to the atmosphere (consistent with other studies on this topic).
- Lines 122-123: “Evergreen vegetation is assigned a constant value of 1, whereas deciduous and grassland vegetation vary seasonally” For which parameters?
- Figure 4: Use black borders around the points for measurement stations and make them bigger? Hard to see against the elevation map now. Also, would be good to label these for reference in the text, especially for the CO2 surface and TCCON stations.
- Table 2: you mention simulating only 2020 here, but later in the paper you simulate NEE for 6 years. How to explain the discrepancy?
- Section 2.4, please provide a brief explanation of what all these indicators are and what are reasonable values.
- Section 3.1.1.: what is the spatial resolution of the simulated PAR data that you are comparing? Also, in Figure 5, which area and or locations are you comparing?
- Section 3.1.2 (a): explain the difference between surface CO2 observations and column average measurements from TCCON. How do you aggregate your own simulations for comparison with xCO2 values? Also, how well do you expect your model to simulate surface CO2 given the vertical resolution of WRF? What are the differences in the observed magnitude between surface and xCO2 measurements (before comparison to simulations)? When you conclude that your WRF-VPRM does a better job with column-averages, it is likely that the magnitude of surface CO2 variability is much higher than for column averages, so comparing biases in an absolute ppm basis may not make sense.
- Section 3.1.2 (b): Lines 314-316, at what time of year are you making these comparisons and at what timescale? Also, lines 319-323: a 15 ppm difference sounds high! Are these mainly due to fossil fuel plumes? Is there a reason why CT2022 would better capture these spikes compared to WRF-VPRM? In Figure 7, it looks like CT2022 is at a coarser resolution than WRF-VPRM which would tend to dilute peaks of CO2 near the surface. Also, not clear in this figure over what time period you are aggregating fluxes.
- Lines 340-346: How does the agricultural uptake over this period compare to the non-agricultural uptake? Your model should be simulating both, so you can make these comparisons and put the agricultural uptake in context of the total (which is crop + non-crop).
- Figure 8: hard to see, especially considering the focus on spatial patterns. Consider putting in 3 separate plots (with some in supplemental material?)
- Line 337-338: “highlighting their role in mitigating atmospheric CO2 increases” --> Crops in particular are not a permanent sink, since they are consumed by people and animals!
- Lines 372-373: nighttime positive NEE does not imply that the ecosystem is a net source on a 24-hour basis. Please clarify.
- Figure 9 and Section 3.3.2: this discussion of diurnal profiles does not specify where you are comparing diurnal cycles or if this is an average across all planted area for a given crop.
- Section 3.3: the comparisons between NEE and yield need to be clarified. First, is this annual NEE? Do your annual NEE comparisons account for harvest and transport of harvested products? How about their consumption and respiration back to atmosphere? Also, if you are comparing on a total flux basis (i.e. TgC/yr), it becomes hard to disentangle the flux strength vs. planted area in the comparison. How about presenting comparisons on a normalized area basis as well?
- Figure 9: it looks like these values are normalized by area, but the top sub-plot is a sum of the below 3 sub-plots. Does it make sense to add them on a per square meter basis? Why not just eliminate the top sub-plot?
- Figure 10 not normalized by area and inconsistent with Figure 9. Why not pick just one unit and stick with it? Or at least explain why you are switching between units normalized by area and total flux.
- Line 468: why just even years?
- Lines 470-477: why |NEE|>0? Are you trying to filter out desert areas? Also, are these correlations between provincial NEE and yield mainly reflecting spatial variability (assuming 29 provinces by 5 years)? Please explain in text.
- Lines 495-499: Units of tonnes in crop statistics are typically called production, not yield (e.g. tons/ ha * planted area in ha).
- Figure 12: what explains discrepancies like for maize in Inner Mongolia?
- Lines 551-552: for future improvements to the respiration model in VPRM, please check out the following publication: https://agupubs.onlinelibrary.wiley.com/doi/full/10.1029/2021JG006290
- Lines 574: what land uses and NEE values are being averaged in around rice fields? Do you assume zero flux for other land-use types in the vicinity of the rice fields?
- Line 637: should be “spatiotemporal variability of crop CO2 oncentrations”, or concentrations and fluxes?
Citation: https://doi.org/10.5194/egusphere-2026-2606-RC2
Viewed
| HTML | XML | Total | BibTeX | EndNote | |
|---|---|---|---|---|---|
| 169 | 60 | 21 | 250 | 22 | 17 |
- HTML: 169
- PDF: 60
- XML: 21
- Total: 250
- BibTeX: 22
- EndNote: 17
Viewed (geographical distribution)
| Country | # | Views | % |
|---|
| Total: | 0 |
| HTML: | 0 |
| PDF: | 0 |
| XML: | 0 |
- 1
The manuscript addresses a real and well-motivated gap: the standard VPRM treats cropland as a single generic vegetation class, which is clearly inadequate for a region where C3 and C4 crops with different phenologies and management calendars coexist, often in rotation on the same land. Splitting cropland into rice, wheat and maize modules and driving them with a daily-resolved crop distribution and phenology product (ChinaCropPhen1km) is a sensible and useful development, and the online coupling to WRF makes the framework potentially valuable for regional inverse and forward studies over East Asia. The manuscript is generally well organised and the figures are clear.
However, I have three major concerns:
1.) The headline flux numbers are internally inconsistent in sign and seem implausibly large. 2.) The paper is framed as an improvement over standard WRF-VPRM, but no baseline simulation with the generic cropland class is presented. 3.) the simulated CO₂ fluxes are never evaluated against flux observations.
These are addressable, but require some additional simulation and analysis. See details below.
Major comments
1. Sign convention is inconsistent, and the NEE magnitude is not credible as stated.
Eq. (1) defines NEE = GEE + RESP with GEE negative (uptake) and RESP positive (release). With GEE = −1084.7 and RESP = +305.5 TgC yr⁻¹, NEE = −779.2 TgC yr⁻¹. Yet the abstract (l. 29–30), Sect. 3.2.1 (l. 337), and the Conclusions (l. 620) all report NEE as +779.2 TgC yr⁻¹ while describing it as uptake, and Fig. 8's caption states that positive NEE indicates release. Figs. 11 and 12 then plot NEE as a positive quantity against yield. The convention needs to be fixed and applied consistently in every equation, figure, axis label and sentence.
More importantly, the magnitude itself needs justification. A net uptake of 0.78 PgC yr⁻¹ from three crops in central and eastern China corresponds to roughly one fifth of the entire global terrestrial carbon sink. Published estimates of the Chinese cropland net carbon sink are substantially smaller (https://agupubs.onlinelibrary.wiley.com/doi/full/10.1002/2013JD021297). It’s unclear to me if this result is a finding or an artefact (see also comment below)
2. Fallow-period and post-harvest respiration appear to be absent?
Fig. 3 and Fig. 10 show crop area — and therefore all crop fluxes — dropping to zero outside each crop's growing season. If cropland grid cells contribute no respiration between growing seasons (because their area fraction is set to 0), then the reported quantity is a growing-season integral, not an annual budget, and labelling it "TgC yr⁻¹" is misleading. Bare-soil heterotrophic respiration in the fallow period is a large term in cropland annual carbon budgets and it omission could explain 1.
3. The relationship between NEE and the atmospheric carbon budget is overstated
The manuscript repeatedly moves from "croplands take up CO₂ during the growing season" to "croplands mitigate the rise of atmospheric greenhouse gas concentrations" (l. 40–43, l. 337–338, and the Conclusions). These are different claims on different timescales. Carbon fixed in grain is exported at harvest and respired within months to a few years through the food and feed chain; crop residues largely decompose on similar timescales. NEE is therefore not a measure of durable sequestration, and net biome productivity for cropland systems is typically near zero or slightly positive once harvest export is included.
The defensible framing is that croplands are strong seasonal NEE sinks with a large influence on regional atmospheric CO₂ gradients — which is exactly what a WRF-VPRM study is well suited to demonstrate — and that this is distinct from long-term sequestration. I recommend adopting that framing consistently.
4. No baseline simulation
The stated innovation is replacing the generic cropland class with crop-specific modules. Demonstrating that this matters requires a control run using standard WRF-VPRM cropland parameters over the same domain and period, and a comparison of the two against the same observations. Without it, the reader cannot tell whether the crop-specific treatment changes the regional budget by 1% or 50%, or whether it improves agreement with observations at all.
5. Simulated fluxes are never evaluated against flux observations.
Sect. 3.1 evaluates PAR and atmospheric CO₂; Sect. 3.3 correlates NEE with yield. Neither directly tests NEE. Given that flux data from Changling (Dong et al., 2022) and Gucheng (Zhou et al., 2023, a wheat–maize rotation site) are already used in the study, a direct comparison of simulated versus observed half-hourly or daily NEE at these sites — ideally with independent site-years withheld from the parameter fit — should be included. Comparing the simulated diurnal NEE cycle in Fig. 9 with a tower-observed cycle would be a natural and inexpensive addition.
6. The NEE–yield correlation is largely a correlation with planting area?
Both provincial NEE totals and provincial production totals scale with provincial crop area, so the reported correlations (r = 0.982 for wheat) are close to guaranteed regardless of model skill. To make this a meaningful test, the analysis should be repeated with area-normalised quantities — NEE per unit crop area versus yield per unit area — which removes the shared area signal. Converting NEE to an expected harvestable carbon using a harvest index would be a stronger test still.
7. Parameter derivation is under-documented
Table 1 gives crop-specific λ, α, β and PAR₀ derived from "in-situ measurements at the Changling vortex flux station" (l. 143–145), but the other parameters are taken from a previous paper without more explanation. Please state explicitly which site, which years, and which observations were used for each of the three crops, how the data were partitioned, and how many points entered each fit. Given that these four parameters largely determine the headline budget, the fitting procedure deserves a clear description also inter-site comparison might be useful where possible. Also: Which temperature parameters did you use? Given that you are upscaling/extrapolating to different climate regions the temperature function might strongly suppress the GEE for some regions.
8. Sub-grid fractional cover are not described.
The manuscript correctly identifies wheat–summer maize rotation as a key feature of Chinese agriculture (l. 74–76), but if I don’t miss anything the treatment is never specified. In a 9 km grid cell containing both a wheat and a subsequent maize crop, how are the two modules combined — sequentially in time, or as coexisting fractions? Are the three crop areas allowed to overlap within a cell?
9. Pscale on crops.
Sect. 2.1 states that evergreen vegetation is assigned Pscale = 1 while "deciduous and grassland vegetation vary seasonally" (l. 122–123) — but says nothing about croplands, which is the focus of the paper.
Minor comments
* l. 234–240 and Fig. 5: PAR values are given in mmol m⁻² s⁻¹ (e.g. 1289.38). This seems three orders of magnitude too large; these should be μmol m⁻² s⁻¹? Please correct throughout the text and the figure axes.
* l. 243: Correlation coefficients of exactly 1.00 across all four seasons strongly suggest the comparison is between seasonally averaged diurnal cycles (24 points) rather than time series. If so, this should be stated, and an hourly time-series comparison should be added — averaging removes almost all of the variance the model needs to reproduce.
* l. 112–113: PAR = SWDOWN/0.505 — please give units on both sides.
* l. 401 vs. l. 427: Rice is said to grow "from May to September" but the rice growing season is later said to end in late December (DOY 361). These are inconsistent; DOY 361 is also carried into the Conclusions (l. 634).
* Table 2: Domain size is listed as 3240 × 4140 km, but a 360 × 360 (l. 182) grid at 9 km resolution gives 3240 × 3240 km. Please reconcile.
* Title and abstract: The domain is central and eastern China, and Xinjiang and Xizang are excluded from the yield analysis, yet the results are described as "national." Please qualify.
* Fig. 9a: The "Total" panel appears to be the arithmetic sum of three per-area fluxes (peak ≈ −3.0 gC m⁻² h⁻¹). Summing per-unit-area fluxes across crops is not physically meaningful unless area-weighted; please clarify or replot as an area-weighted mean.
* Fig. 9 caption: GEE is defined in the text as gross ecosystem exchange but as "gross primary production" in the caption. Please use one term consistently.
* l. 291: "In contrast, the simulation of column-averaged XCO₂ exhibits higher consistency and stability" interrupts the OCO-2 discussion and does not follow from the preceding sentence. Consider moving or deleting.
* l. 559: There is an updates parameter set (https://gmd.copernicus.org/articles/18/4713/2025/) to the one cited here which is also worth citing as it directly addresses issues in cropland fluxes from the old parameter set cited here. This also matters for a discussion on the fitting procedure (see point 7 above).
* l. 594 and l. 1030: "(Ki and Kim et al., 2021)" — the reference list gives Kim and Kim (2021)
* Using a static 1km SYNMAP product seems outdated given that 100m and even 10m dynamic products exist with improved classification algorithms (e.g. https://essd.copernicus.org/articles/13/3907/2021/, https://zenodo.org/records/3939050) and have been used for VPRM studies (e.g. https://gmd.copernicus.org/articles/18/4713/2025/). It’s fine for now but requires a little disclaimer and some perspective on improvements in the outlook.