关键词: OCO-2/OCO-3, 归一化植被指数, 化石能源, 清洁能源

Abstract: Objective　The Yangtze River Delta (YRD) is one of the most urbanized and industrially active regions in China, accounting for a substantial portion of China's CO2 emissions. Regional monitoring of the column-averaged dry-air mole fraction of CO2 (XCO2) is crucial for understanding its spatiotemporal variability and associated natural and anthropogenic factors. This study aimed to: (1) improve the spatial coverage of OCO-2 and OCO-3 XCO2 observations using spatiotemporal weighted Kriging method; (2) characterize the spatiotemporal variations in XCO2 in the YRD region during 2020–2023, and examine the relationship between XCO2 and the normalized difference vegetation index (NDVI); and (3) combine non-growing-season XCO2 anomalies with fossil-fuel CO2 emission inventory data and power plant data to compare the XCO2 anomaly characteristics in regions with different power-generation structures. Methods　We utilized the level-2 XCO2 inversion data of the YRD region (25.5°N–35.5°N, 114°E–124°E) from January 2020 to December 2023 from NASA's OCO-2 and OCO-3 satellites (versions 10 and 11) . To fill the pervasive missing pixels, we applied a spatio-temporal weighted Kriging (STWK) interpolation method that combines spatial covariance (exponential variogram model) and temporal autocorrelation (first-order autoregressive structure) to generate interpolated XCO2 fields with 0.1° × 0.1° resolution at 15-day, monthly, seasonal, and annual time scales. Within the 95% confidence interval of the standard normal distribution, the correlation between the interpolated XCO2 prediction value and the XCO2 filling value is 0.96. To conduct the phenological analysis, we processed monthly NDVI data calculated based on the reflectance data of MODIS Terra (MOD13C1) with the same grid resolution, and performed binary classification based on the vegetation growth season (active period) and the dormant period (non-active period) to quantify the differences of XCO2 responses under different phenological conditions. To minimize the interference caused by photosynthetic uptake, we defined XCO2 anomalies as deviations from the long term seasonal baseline during non growing seasons only. These XCO2 anomalies were then coregistered with the ODIAC fossil fuel CO2 emission inventory (2020–2023) and geolocation-based power plant data from the Global Power Plant Database, which contains plant level attributes including fuel type and installed capacity. We performed K-means clustering on the emission intensity and clean energy share (including wind, solar, hydro, and nuclear power) for each grid, with the optimal cluster number determined by the elbow method (k = 3). The YRD region were identified as three regional archetypes: fossil fuel dominant (FFD), clean energy advantaged (CEA), and cleanenergy- potential (CEP). Differences in carbon emission intensity between the three regional archetypes were evaluated to assess the influence of energy composition and clean-energy penetration. Results and Discussion　The reconstructed dataset showed a clear and statistically significant upward trend in regional mean XCO2 of the YRD with annual averages increasing from 414.22 × 10−6 in 2020 to 421.36 × 10−6 in 2023, corresponding to a growth rate of 2.38 × 10−6 a⁻¹. Spatially, a persistent east to west gradient distribution of XCO2 was observed in the YRD region. The XCO2 of coastal urban agglomerations (such as Shanghai) was higher than that of inland Anhui and northern Jiangsu, and the highest values occurred in summer and early autumn due to increased human activities and stagnant meteorological conditions. Time series analysis showed that there was a clear seasonal variation for XCO2 in the region, with an average amplitude of 4.6 × 10−6. And it reached its peak at the end of spring and its lowest point at the end of summer. Cross-correlation analysis indicated a robust lagged relationship between XCO2 and NDVI. During the vegetation growth period, the monthly average XCO2 value reaches its lowest point about 2 months after the peak of NDVI. While during the non-growing period, the monthly average XCO2 value reaches its highest point about 3 months after the lowest point of NDVI, suggesting a delayed response of atmospheric CO2 variability to vegetation activity. The longer winter lag may be associated with reduced ecosystem respiration and weakened vertical mixing. Regarding the attribution of human factors, the non-growing season anomaly dataset reduces the interference of natural carbon flux. In the CEA region where clean energy power generation accounts for 35%, the growth rate of XCO2 anomaly is significantly lower than that in the FFD region where clean energy power generation only accounts for 0.4%. Quantitative analysis shows that in the CEA region, for every 1% increase in the proportion of clean energy, the growth rate of XCO2 anomaly decreases by 2.4% compared to the FFD region and by 2.1% compared to the CEP region. The CEP regions showed intermediate responses, suggesting that partial transition toward clean energy provides tangible yet incomplete benefits. The research results show that, on the basis of a complete and balanced energy structure, the higher the proportion of clean energy power generation, the better the emission reduction effect. And these results are robust to different clustering parameters and seasonal definitions, although there are still uncertainties due to limitations in emission inventories and simplified representation of atmospheric transport processes. Conclusions　This study provides an interpolated XCO2 dataset with improved spatial coverage for the YRD. The results suggest that regional CO2 variability reflects the combined effects of anthropogenic emissions and seasonal vegetation dynamics. The observed lagged relationship between NDVI and XCO2 provides a reference for understanding seasonal variations of regional carbon cycle. By focusing on non-growing-season XCO2 anomalies, we reduced the influence of seasonal vegetation activity and compared the characteristics of XCO2 anomalies in regions with different power-generation structure. Regions with higher clean-energy generation shares generally exhibit weaker XCO2 anomaly signals, although this association may also be influenced by industrial structure, atmospheric transport, and other emission sources. From a policy perspective, the observed regional differences of XCO2 suggest that power-generation structure should be considered in regional carbon-management strategies. However, the mitigation effect of increasing the share of clean-energy generation requires further evaluation using atmospheric transport modelling and sector-specific emission data.

Key words: OCO-2/OCO-3, normalized difference vegetation index, fossil energy, clean energy