the Creative Commons Attribution 4.0 License.
the Creative Commons Attribution 4.0 License.
| 14 Oct 2026
High-resolution land-cover dynamics reveal strong regional contrasts in China’s biogenic isoprene emissions, 2000–2025
Abstract. Isoprene is a reactive biogenic volatile organic compound whose atmospheric impacts depend strongly on spatio-temporal emission patterns. Because Chinese inventories often rely on coarse or static land-cover information, we quantified how annual land-cover dynamics altered isoprene emissions across China during 2000–2025. Using a Google Earth Engine implementation of the Model of Emissions of Gases and Aerosols from Nature (MEGAN), driven by Landsat and Sentinel-2 land cover, satellite leaf area index, and 0.1° ERA5-Land meteorology, emissions were calculated on native 10–30 m grids. These were aggregated to an hourly 100 m output and compared with a 0.25° coarse simulation. National emissions averaged 9.87 ± 1.53 Tg yr-1, increasing from 8.49 in 2000 to 12.35 Tg yr-1 in 2025. Following a non-significant trend (2000–2012), emissions increased by 0.23 Tg yr-1 during 2013–2025 (p < 0.05), driven by temperature (40.4 %) and land-cover change (39.6 %). Crucially, fine land-cover data yielded a national mean only 6.2 % lower than the coarse simulation, yet exposed massive regional contrasts: +101.2 % in Beijing–Tianjin–Hebei and −20.2 % in the Pearl River Delta. Regional seasonality also varied, with summer contributing 49.4 % of emissions in Southwest China versus 78.8 % in the Northeast. Therefore, national-level agreement between inventories can mask substantial regional redistributions, with major implications for biogenic volatile organic compound inputs to atmospheric chemistry models. Estimates remain sensitive to satellite product transitions, default MEGAN emission factors, land-cover-to-plant-functional-type conversion, and meteorological forcing.
- Preprint
(2394 KB) - Metadata XML
-
Supplement
(2749 KB) - BibTeX
- EndNote
Status: open (until 25 Nov 2026)
- RC1: 'Comment on egusphere-2026-5116', Bibhasvata Dasgupta, 23 Oct 2026 reply
-
RC2: 'Comment on egusphere-2026-5116', Anonymous Referee #1, 24 Oct 2026
reply
General Comments:
The manuscript of “High-resolution land-cover dynamics reveal strong regional contrasts in China’s biogenic isoprene emissions, 2000–2025” by Ran et al. is an impressive and timely study that presents a new, high-resolution (100 m), long-term (2000–2025), and hourly biogenic isoprene inventory for China (ISOP-China-100m). The authors have undertaken a significant technical effort, leveraging Google Earth Engine to integrate dynamic land cover from Landsat/Sentinel with the established MEGAN model framework. The main finding of this manuscript is that while the use of high-resolution land cover may lead to only modest changes in the national total emission budget, it causes substantial and often countervailing redistributions of emissions at the regional and sub-regional scales. This masking effect is a critical point, as it directly impacts the accuracy of regional air quality simulations, particularly for secondary pollutants like ozone and SOA. The comprehensive evaluation, including independent land cover validation, surrogate model sensitivity analysis, and comparison with satellite HCHO columns via GEOS-Chem, adds significant confidence to the study's conclusions. This high-resolution isoprene emission inventory is great importance to the atmospheric chemistry and air quality modeling community.
The manuscript is well-written, the methods are generally robust, and the results are presented clearly. My recommendation is for publication after the authors address a few minor points, which are primarily aimed at improving clarity, strengthening the discussion of key uncertainties, and refining the interpretation of the results. I look forward to seeing the revised version in due course.Major Comments:
1. The manuscript's headline feature is its 100 m resolution. While the vegetation and PFT fractions are resolved at this fine scale, the meteorological drivers (ERA5-Land) are at a much coarser 0.1° (~10 km) resolution. This is a critical nuance. The authors mention this briefly (L87-89), but this point should be made more prominent and consistently throughout the paper (e.g., in the abstract and conclusions) to avoid any misinterpretation by readers and model users. A clearer term like "a 100-m vegetation-resolved inventory" could be helpful. Users must understand that the effective resolution for short-term emission variability is dictated by the meteorology, not the land cover grid.
2. The authors correctly identify and analyze a trend break between 2012 and 2013, which coincides with the Landsat sensor transition from L5/L7 to L8. This is a crucial point that warrants a more in-depth discussion. The current analysis (segmented trend) is an appropriate response, but the interpretation could be more critical. Is it possible that a systematic bias in land cover classification between the sensor generations (e.g., in distinguishing forest types, densities, or urban green space) is contributing significantly to the accelerated trend observed post-2013? While a full de-biasing is beyond the scope of this work, a more explicit discussion of this potential artifact in the uncertainty/limitation sections (e.g., Section 3.4 or 3.5) would substantially strengthen the manuscript by acknowledging this key uncertainty in long-term trend studies based on multi-sensor satellite data.
3. The use of GEOS-Chem to evaluate the inventory against OMI HCHO is a valuable addition. However, the choice of a very coarse 4°×5° resolution is surprising, as it averages out the very high-resolution details that are the central point of the paper. Please provide a clear justification for using this coarse resolution (e.g., was it for computational expediency, or to focus on large-scale continental patterns?). More importantly, please add a discussion of how this coarse-grid evaluation might limit the ability to validate the regional-scale redistributions that the paper highlights. While the improved spatial correlation is a positive result, its interpretation is constrained by the evaluation grid size.Specific Comments:
1. Lines 95–100 and Table 1: The main text states that both simulations use the same "ERA5-Land/LAI forcing", whereas Table 1 indicates that ISOP-China-100m uses GEE-MEGAN satellite LAI while the coarse-resolution simulation uses MODIS LAI. Are the LAI inputs used in the two simulations entirely identical? If not, please clarify the specific LAI dataset and methodology used to construct ISOP-China-100m.
2. Lines 95–100: The text states that five spectral indices were used in the Random Forest model, but only NDVI and EVI are explicitly listed. What are the other three spectral indices? Please list all five indices.
3. Lines 153–155: What specifically does "data from years outside the native range of these auxiliary products were carefully extrapolated "? Please explicitly specify which surrogate years were used for each auxiliary dataset during missing years.
4. Table 2: The title of each column is incomplete. Although two columns are provided for each dataset, "multi-year mean" and "trend" are not clearly distinguished, and some units are missing. It is suggested to explicitly label them as "Mean ± SD (Tg)" and "Trend (Gg yr-1)". Additionally, please clarify whether regional percentage differences were calculated based on multi-year averages or as the mean of annual ratios.
5. Lines 333–334: Temporal resolution needs to be standardized throughout the manuscript. The abstract, methods, and conclusion state that the simulation is hourly, whereas Sect. 3.5 mentions "daily temporal frequency".
6. Lines 320–324: In Fig. 6c, the residual term of the surrogate model contributes approximately +0.021 Tg yr-1 (9.0%) to the 2013–2025 trend, but the main text only reports five drivers. It is recommended to add a description of the residual term in the main text.
7. SI Lines 38–41: Please provide further description and details regarding the predictor variables used in the surrogate model.
8. Table S1: Units for PFT area are missing. Please specify the units in the column header.
9. Line 47: "the response of O3 to BVOCs responds" contains redundant wording. Please revise.
10. Line 209: Remove the duplicated article "a" in "with a a 26-year average".
11. Line 384–385: In the conclusions, the average national emissions from ISOP-China-100m stated as "9.87 ± 1.50 Tg" is inconsistent with the value reported in the abstract. Please verify.
12. Line 416: The OMI DOI URL is written as "DATA301" in this section, whereas it appears as "DATA3010" in the references. Please verify.Citation: https://doi.org/10.5194/egusphere-2026-5116-RC2
Viewed
| HTML | XML | Total | Supplement | BibTeX | EndNote | |
|---|---|---|---|---|---|---|
| 331 | 130 | 29 | 490 | 54 | 31 | 29 |
- HTML: 331
- PDF: 130
- XML: 29
- Total: 490
- Supplement: 54
- BibTeX: 31
- EndNote: 29
Viewed (geographical distribution)
| Country | # | Views | % |
|---|
| Total: | 0 |
| HTML: | 0 |
| PDF: | 0 |
| XML: | 0 |
- 1
In ‘High-resolution land-cover dynamics reveal strong regional contrasts in China's biogenic isoprene emissions, 2000–2025’, Ran and co-authors argue that a finer land-cover map that changes every year barely changes China's national total isoprene emissions, but it shifts a lot of emissions from one region to another. They address a genuine gap: no existing product combines 100 m vegetation-resolved output, hourly resolution and 26 continuous years, and the paired same-forcing design isolates vegetation representation far better than cross-study comparisons. My primary concern is as follows. The authors claim that finer, yearly-updated land cover reshapes where isoprene is emitted, yet emissions scale almost one-to-one with the plant-type emission factor, so an error in which forest species or plant type each pixel is assigned to translates directly into these regional differences. Satellite retrieval issues, such as gaps from clouds in the high-emitting south and the 2012–2013 Landsat 5→8 transition, could also produce apparent land-cover change. That makes this uncertainty most relevant exactly where the paper's main contribution lies. With these concerns addressed, I believe it can be a useful contribution in the field.
Major Comments
Line 210–220: Ran et al. describe how the national emission series shows a discontinuity between 2012 and 2013, coinciding with the Landsat 5 → 8 transition, and therefore analyse trends in two segments, but the significant 2013–2025 acceleration (+0.232 Tg yr⁻¹, p < 0.005) and its 39.6% land-cover attribution begin exactly at this data-input break, and the implication that this transition has i.e. "may be distributed among correlated components" (section 3.4) is not adequately discussed. They should demonstrate, e.g. by re-deriving the 2013–2025 emission-potential trend from a homogeneous Landsat-8-only classification, that the land-cover contribution is not a sensor-change artefact, and carry the caveat into the abstract.
Line 40–45: Authors acknowledge that field measurements in China "do not always agree with emission factors transferred from global databases" (citing Klinger et al., 2002; Geron et al., 2006; Han et al., 2026), but the inventory is built on default MEGAN2.1 emission factors (Table S2) and the sensitivity analysis in section 3.5 varies temperature, radiation, LAI, water vapour, wind and pressure, never the emission factors themselves, possibly why the 9.87 Tg yr⁻¹ national mean sits at the low end of the published 7.5–37.5 Tg yr⁻¹ range. I’d recommend that they add an emission-factor scenario sensitivity test (e.g., using the China-specific library of Han et al., 2026, already cited) or explicitly state that all conclusions concern spatial and temporal structure under a fixed EF assumption, not absolute magnitudes.
Line 95–102: Ran et al. state that the companion MEGAN2.1-MODIS-0.25° run employed the same core emission equations and ERA5-Land/LAI forcing, so that inter-run differences come only from vegetation representation, but Table 1 lists GEE-MEGAN satellite LAI with 'MOD15A2H/MCD15A3H/MYD15A2H' for the fine run and only the three MODIS products for the coarse run, suggesting the LAI inputs are not identical. They should clarify whether the LAI forcing is identical in both runs; if not, they can rerun the coarse simulation with identical LAI or decompose the inter-run difference into land-cover and LAI contributions.
Line 128–134: Ran et al. convert ERA5-Land solar radiation to PPFD with a fixed factor (PPFD = 2.25 × SRR) and say that it does not resolve spectral variations due to sky conditions, but radiation is one of the dominant sensitivity drivers (±10% → −12.88% to +14.49% in national emissions), so this unquantified clear-sky bias propagates directly into the hourly emission field and potentially into the radiation term of the trend decomposition. They should quantify the effect, e.g. with a cloud-fraction-dependent conversion or a sensitivity test on the 2.25 factor and report the resulting uncertainty alongside the other sensitivity results.
Line 185–190: Ran et al. evaluate ISOP-China-100m through paired GEOS-Chem simulations compared with OMI HCHO and present the result (correlation 0.904 vs 0.890) as "independent atmospheric evidence supporting … the relative spatial validity of the ISOP-China-100m emission allocation", but the runs are at 4° × 5° and cannot resolve the regional contrasts that are the paper's central claim (BTH +101.2%, PRD −20.2%), while the margins are modest and HCHO is an indirect constraint. They should restrict the claim to large-scale validity only or add a higher-resolution regional evaluation (e.g. a nested-grid run over BTH/PRD) that actually tests the redistributed emissions.
Minor Comments
Line 285–290: Ran et al. report annual mean emissions for the seven regions in Table 2 but these sum to 9.70 Tg yr⁻¹, which is 0.17 Tg below the 9.87 Tg yr⁻¹ national mean (section 3.1), and the companion run shows the same gap (10.33 vs 10.51 Tg) — even though the table's province list covers all 31 mainland provinces and only Taiwan is "unassigned", while in section 2.1, they define the domain as "mainland China". They should explain which areas contribute to the national total, but not to any region, or correct either the table or the domain wording so the numbers reconcile.
Line 15–25: The paper present key numbers differently in the abstract, body and conclusions: 9.87 ± 1.53 (abstract) vs 9.87 ± 1.50 Tg yr⁻¹ (conclusions); 8.49 (2000) to 12.35 Tg (2025) vs a min–max range of 8.25–13.07 Tg yr⁻¹; p < 0.05 (abstract) vs p < 0.005 (section 3.1 and 3.4) for the same 2013–2025 trend; "daily temporal frequency" in section 3.5 vs "hourly" everywhere else; and 15.93 ± 1.12 Tg yr⁻¹ quoted for Wang et al. (2021) vs 15.94 ± 1.12 Tg in the source paper. They should check every repeated number, state the temporal resolution identically (hourly, per section 2.1), and re-check quoted values against the cited literature, since the ± SD, significance level and temporal resolution are the quantities readers quote first. For an emission inventory to model paper, this is significant.