| 30 Jun 2026
Indicator-dependent vegetation trends and interannual variability across Asia–Oceania from NDVI, vegetation optical depth, and model based GPP during 2000–2021
Abstract. Vegetation change has been widely monitored using satellite-derived vegetation indices, particularly the Normalized Difference Vegetation Index (NDVI). However, different vegetation indicators represent different aspects of ecosystem conditions and may provide contrasting views of long-term vegetation change. In this study, we compared long-term trends and interannual variability among multiple indicators including two NDVI datasets, Vegetation Optical Depth (VOD), and gross primary production (GPP) simulated by a process-based ecosystem model VISIT across the Asia–Oceania region during 2000–2021. Long-term trends were estimated using Sen’s slope, and interannual variability was evaluated using detrended and standardized annual anomalies. Both NDVI datasets showed widespread positive trends across East Asia and northern Eurasia, whereas VOD exhibited weaker and more spatially heterogeneous trends, particularly in high-latitude regions. East Asia showed relatively consistent positive trends among NDVI, VOD, and GPP, suggesting coherent increases in vegetation greenness, biomass, and ecosystem productivity. In contrast, substantial discrepancies were observed in Siberia, where strong NDVI greening was not consistently accompanied by increases in VOD or GPP. Comparisons of interannual variability also revealed regional differences in agreement among indicators, with stronger relationships among NDVI, VOD, and GPP in East Asia and some semi-arid regions than in boreal ecosystems. These results indicate that assessments of vegetation change depend strongly on the indicator used and that different indicators capture different aspects of ecosystem dynamics. Combining optical, microwave, and model-based indicators therefore provides a more complete understanding of vegetation change across the diverse environments of Asia–Oceania.
This manuscript examines how long-term trends and interannual variability differ among NDVI, VOD, and model-derived GPP across the Asia–Oceania region. The combined analysis of these temporal characteristics could provide useful insights into indicator-dependent assessments of vegetation change. However, the current analysis does not adequately distinguish ecological divergence from differences arising from temporal aggregation, product uncertainty, or model structure. I believe that substantial revisions to both the analysis and its interpretation are needed before the manuscript can be considered for publication.
Major comments:
The indicators are not compared with over equivalent temporal supports. NDVI is averaged over a fixed growing-season mask, VOD is sampled in a single month selected from the month of maximum long-term mean NDVI, and GPP is averaged over the full year. These choices can themselves generate different trends and interannual variability, especially in monsoon, boreal, and seasonally dry regions. The authors should repeat the main comparisons using matched temporal definitions, for example a common growing-season mean for all indicators, annual values for all indicators where appropriate, and several alternative VOD windows such as a three-month peak-season mean or the VOD seasonal maximum. Sensitivity to a fixed versus annually varying growing season could also be tested. Without these controls, it is difficult to interpret the reported disagreement.
Statistical analysis requires more attention. The coefficient of determination is not an adequate measure of agreement because squaring the correlation removes its sign. A strong negative relationship can therefore appear as high agreement. The maps and regional summaries should report signed correlations and significance based on an effective sample size that accounts for temporal autocorrelation. In addition, the threshold R2 < 0.3 is arbitrary and needs justification, and the large number of pixel-wise tests requires multiple-testing treatment. Similar concerns apply to the trend-agreement maps. The three categories in Figs. 3 and 6 sum to 100%, but there is no category for non-significant trends or for cases in which only one product has a significant trend. Classifying the signs of very small, non-significant Sen slopes as greening, browning, or disagreement can greatly inflate the reported percentages. The authors need to distinguish significant positive agreement, significant negative agreement, opposite significant trends, mixed-significance cases, and no significant trend, while accounting for serial correlation in the Mann-Kendall tests. Indeed, trend magnitudes from NDVI, VOD, and GPP also cannot be compared directly because they have different units and scales (Fig. 1). Standardized or relative trends would be needed for such statements.
The conclusions about modeled GPP and CO2 attribution are not adequately supported. Only one configuration of one ecosystem model is evaluated, so the results should not be generalized to model-based GPP. More information is needed on the VISIT configuration, spin-up, vegetation and land-use representation, parameterization, historical climate forcing, and so on. The sensitivity experiments are also unclear. The Methods define ALLCON as the simulation in which all climate drivers vary, whereas Fig. 7 additionally includes a CLIMATE component described as the total climate contribution. Moreover, Fig. 7 displays detrended anomalies, which cannot demonstrate that CO2 is the primary driver of a long-term trend. The authors need to present undetrended component trends, define a consistent factorial or difference-based decomposition including interaction terms, and compare the inferred contributions quantitatively. Validation against independent GPP constraints or comparison with a model ensemble would also be needed to determine whether the discrepancies are specific to VISIT or robust across GPP estimates.
The interpretation of indicator disagreement is too categorical. NDVI, VOD, and GPP measure different variables, so low correlation among them is partly expected and does not by itself demonstrate ecological decoupling, dataset error, or meaningful complementary information. The manuscript repeatedly attributes regional differences to NDVI saturation, VOD contamination by soil moisture and freeze-thaw processes, or missing processes in VISIT, but these explanations are not fully support the results.
Minor comments:
L15-18: It is up to the band selection. I think it is difficult to interpret Ku-, C-, or X-band VOD consistently. How do they represent the proxy of biomass?
L39: Please cite some references.
L44-45: Yes, but it is up to the purpose of the study. For instance, if we focus on the canopy greenness, then there is not much problem. I think “vegetation dynamics” is bit unclear here.
L15-20: The abstract interprets agreement as coherent changes in greenness, biomass, and productivity. This should be moderated because VOD is not a direct biomass measurement and GPP is not independently observed here.
L58: Here, the authors clarified VOD as a proxy of water status, but it represents a biomass in this study.
L58-60: Do NDVI, VOD, and GPP each clearly represent structural, hydrological, and functional vegetation changes, respectively? This correspondence may be overly simplistic or misleading. Please revise the sentence.
L63-64: I do not agree with the statement that only a few studies have investigated this topic. Please provide appropriate references to support this claim. More generally, some of introduction sentences are presented without citations. Relevant references should be added where needed.
L95-110: The authors select the long-term datasets which cover 1980s-present but only use 2000-2021. Could you explain it?
L103-105: What is harmonized C,X,Ku-band VOD? Also, VODCA provides its own GPP product. It is worth testing it too.
L105-106: Please explain why the VISIT model was selected for this analysis. The choice appears somewhat selective, and the criteria for model selection are unclear.
L125-127: It is not matched with study period.
L125-138: I think this data processing strategy should be carefully improved. The current method is not consistent.
L139: I think they are inconsistent.
Figure. 1: Please select a single NDVI product for the main analysis.
L219-221: Assessing the consistency of IAV or trends among different products is not the primary goal of this study. If the authors intend to assess consistency across products, the same analysis should also be conducted using multiple VOD and GPP products.
Section 3.1-3.4: Because NDVI, VOD, and GPP were processed using different temporal definitions, the reported differences may reflect processing choices rather than actual differences among the indicators.