30-m Raster Dataset Development of Key Carbon Cycle Parameters for Terrestrial Ecosystems in the Guangdong-Hong Kong-Macao Greater Bay Area (2001–2025)

YOU Jiakai^1,2,3  ZHANG Yinghui^1,2,3*  HU Zhongwen^1,2,3  GUO Shanxin⁴  
WANG Jingzhe⁵  WU Guofeng^1,2,3

1. School of Architecture and Urban Planning, MNR Key Laboratory for Geo-Environmental Monitoring of Great Bay Area, Shenzhen University, Shenzhen 518060, China;

2. Guangdong Key Laboratory of Urban Informatics, Shenzhen 518060, China;

3. Shenzhen Key Laboratory of Spatial Smart Sensing and Services, Shenzhen 518060, China;

4. Shenzhen Institute of Advanced Technology, Chinese Academy of Sciences, Shenzhen 518055, China;

5. School of Artificial Intelligence, Shenzhen Polytechnic University, Shenzhen 518055, China

Abstract: Vegetation parameters are essential indicators for characterizing regional ecological dynamics. They provide valuable information for fine-scale ecological monitoring and for elucidating the characteristics and variations of carbon sources and sinks in terrestrial ecosystems. The Normalized Difference Vegetation Index (NDVI), Enhanced Vegetation Index (EVI), Leaf Area Index (LAI), Fraction of Absorbed Photosynthetically Active Radiation (FPAR), and Gross Primary Production (GPP) collectively constitute the key parameters for characterizing the carbon cycle of terrestrial ecosystems. Existing MODIS vegetation products exhibit spatiotemporal discontinuity due to main-algorithm failures and contain mixed pixels resulting from their 500-m spatial resolution. To address these limitations, this study focuses on the Guangdong-Hong Kong-Macao Greater Bay Area. Under the constraint of land cover consistency, the mSSA algorithm was introduced to spatiotemporally reconstruct the areas affected by the main algorithm failures across the five core MODIS vegetation-parameter products. Subsequently, the annual maximum NDVI from Landsat was extracted to represent spatial texture features. Spatial weight constraints were then applied to downscale the reconstructed dataset. These procedures yielded a 30-m resolution dataset of key terrestrial ecosystem carbon cycle parameters for the region covering 2001–2025. The generated dataset was validated against standard MODIS vegetation products. The validation results indicate a high consistency between this dataset and the MODIS products. The coefficient of determination (R²) for the vegetation parameters ranges from 0.729 to 0.941, and the fitting slopes range from 0.82 to 0.96. The dataset includes the following components: (1) boundary data for the Guangdong-Hong Kong-Macao Greater Bay Area; and (2) key carbon cycle parameter data for terrestrial ecosystems within the region. The latter comprises 8-day GPP, LAI, and FPAR datasets, alongside 16-day EVI and NDVI datasets. The dataset is archived in .shp and .tif formats. It consists of 4,893 data files with a total data volume of 1.01 TB.

Keywords: MODIS; vegetation parameters; spatial downscaling; spatiotemporal reconstruction; high spatial resolution

DOI: https://doi.org/10.3974/geodp.2026.05.08

Dataset Availability Statement:

The dataset supporting this paper was published and is accessible through the Digital Journal of Global Change Data Repository at: https://doi.org/10.3974/geodb.2026.03.10.V1.

1 Introduction

Vegetation parameters are characteristic values derived from the combined calculation of multispectral remote sensing data based on the spectral properties of vegetation. They serve as a simple and effective empirical measure of land surface vegetation conditions. These parameters hold significant application value in fields such as land cover, environmental change, primary productivity analysis, and terrestrial ecosystem carbon sinks. They act as a critical link connecting the physical structure of land surface vegetation with its physiological functions^[1–3]. The Normalized Difference Vegetation Index (NDVI)^[4], Enhanced Vegetation Index (EVI)^[5], Leaf Area Index (LAI)^[6,7], Fraction of Absorbed Photosynthetically Active Radiation (FPAR)^[8], and Gross Primary Production (GPP)^[9] collectively constitute the key parameter for characterizing the carbon cycle of terrestrial ecosystems. This dataset serves as an important foundation for fine-scale ecological monitoring in highly heterogeneous regions, such as the Guangdong-Hong Kong-Macao Greater Bay Area, and for elucidating the characteristics and variations of carbon sources and sinks^[2]. To date, various vegetation parameter products have been generated using satellite remote sensing data. Among these, MODIS products are the most widely utilized^[10–12]. However, factors such as cloud and precipitation interference can cause the main retrieval algorithm to fail. When backup algorithms are subsequently activated, they introduce anomalous fluctuations and biases into the long time-series data^[13]. Such spatiotemporal discontinuity restricts the application potential of existing vegetation parameter products for fine-scale temporal dynamic monitoring^[8].

To address these issues, previous studies have commonly used time-series reconstruction algorithms or multi-source data-fusion techniques^[14,15]. However, algorithms such as HANTS, which are based on harmonic analysis, rely heavily on the strict assumption of periodic vegetation variations. They have difficulty characterizing irregular vegetation changes in highly heterogeneous regions^[14]. Furthermore, spatiotemporal data-fusion techniques proposed by some scholars^[16–18] are often limited by the low revisit frequencies of polar-orbiting satellites. This limitation makes it difficult to capture valid cloud-free reference datasets, leading to the failure of fusion models due to a lack of effective inputs^[19,20]. Current methods are generally limited by the strict phenological cycle presets of single pixels or by the bottleneck of acquiring cloud-free reference images. They fail to fully exploit the co-evolutionary characteristics among homogeneous pixels within highly heterogeneous land surfaces.

In addition, MODIS vegetation-parameter products with a 500-m spatial resolution are still affected by mixed-pixel effects when characterizing the highly fragmented land surface of the Greater Bay Area^[22–24]. The complex land-use patterns in this region results in the mixing of multiple land-use types within a single coarse-resolution pixel. This mixed-pixel effect obscures the true spatial heterogeneity of vegetation and introduces systematic biases into regional ecological assessments. Consequently, it is difficult to support fine-scale ecological monitoring requirements^[22]. Therefore, there is an urgent need to enhance the spatial representation of vegetation parameter products through spatial downscaling techniques^[22,25,26].

To this end, this study selected the Guangdong-Hong Kong-Macao Greater Bay Area as the study region. Under the constraint of land cover consistency, the mSSA algorithm was introduced^[27]. Optimal neighboring pixels were selected through spatiotemporal quality weighting to achieve high-quality spatiotemporal reconstruction for the areas affected by the main algorithm failures across the five MODIS vegetation parameter products. Second, to address the spatial mixed-pixel effect, the annual maximum NDVI from Landsat was extracted as a texture feature to represent the spatial distribution of vegetation. This feature was utilized as a spatial weight constraint to downscale the reconstructed products. Ultimately, a 30-m spatial resolution dataset of key terrestrial ecosystem carbon cycle parameters for the Greater Bay Area spanning from 2001 to 2025 was constructed. This dataset provides data support for regional ecosystem assessment and fine-scale management.

2 Metadata of the Dataset

The metadata for the 30-m raster dataset of key carbon cycle parameters for terrestrial ecosystems in the Guangdong-Hong Kong-Macao Greater Bay Area (2001–2025)^[28] are summarized in Table 1. The provided information includes the dataset name, authors, geographical region, temporal coverage, temporal resolution, spatial resolution, dataset composition, data publishing and sharing platform, and data sharing policy.

3 Methods

3.1 Data Sources

(1) Vegetation parameter data

The production of this dataset relies on three standard MODIS products with a 500-m spatial resolution: MOD17A2H^[30], MOD15A2H^[11], and MOD13Q1^[31]. Specifically, MOD17A2H provides GPP data with an 8-day temporal resolution. MOD15A2H provides LAI and FPAR data with an 8-day temporal resolution. MOD13Q1 provides EVI and NDVI data with a 16-day temporal resolution. Each product is equipped with an independent Quality Control (QC) layer. These layers identify the retrieval algorithm type of the pixels and the contamination status from clouds, aerosols, and snow/ice. They provide a fundamental basis for the subsequent screening of high-quality observations.

(2) Land cover type data

The MCD12Q1 product was used as auxiliary land-cover data^[32]. It provides surface- homogeneity constraints during the spatiotemporal reconstruction process. The LC_Type3 classification scheme was adopted. 8 core vegetation types were extracted: grasslands, shrublands, broadleaf crops, savannas, evergreen broadleaf forests, deciduous broadleaf forests, evergreen needleleaf forests, and deciduous needleleaf forests. Water bodies and non-vegetated areas were excluded. This ensures that the algorithm extracts synergistic features among similar land cover types.

(3) Spatial texture feature reference data

The annual maximum NDVI data derived from Landsat imagery at a 30-m spatial resolution, were selected as the spatial texture feature reference data for the downscaling algorithm^[33]. These data characterize the optimal growth potential of vegetation on an interannual scale and its fine spatial distribution structure.

Table 1  Metadata summary of the 30-m raster dataset of key carbon cycle parameters for terrestrial ecosystems in the Guangdong-Hong Kong-Macao Greater Bay Area (2001–2025)

3.2 Algorithm

(1) Selection of main algorithm results

Based on the QC layers of the vegetation-parameter products, strict data cleaning was conducted on the original time series. Retrieval pixels affected by cloud cover or high aerosol loading, as well as pixels generated by backup empirical algorithms, were identified and removed. Only uncontaminated, high-quality valid observations generated by the main retrieval algorithm were retained; all other invalid pixels were uniformly marked as missing values. This procedure facilitated subsequent spatiotemporal reconstruction of areas where the main algorithm failed.

(2) mSSA-based spatiotemporal reconstruction algorithm considering land cover types

To address the missing spatiotemporal data caused by main algorithm failures, the mSSA algorithm was introduced for spatiotemporal continuity reconstruction. To prevent the introduction of exogenous noise from heterogeneous surfaces, strict land cover consistency constraints were applied during the reconstruction process. For any missing pixel, the algorithm searches for valid neighboring pixels with the same land cover type within its dynamic spatial window. A comprehensive spatiotemporal weight () is then calculated:

                                                                                                           (1)

wheredenotes the number of valid observations of the neighboring pixel in the time series, and denotes the Chebyshev distance from the neighboring pixel to the central pixel. p and q are hyperparameters adjusting the spatiotemporal importance. The top K pixels of the same type with the highest comprehensive weight S are selected to construct a multiple time-series matrix. The mSSA is then utilized to extract common features, predict, and fill the missing values of the central pixel. This completes the high-quality spatiotemporal reconstruction of areas where the main algorithm failed.

(3) Spatial weight-based downscaling algorithm

To address the limitation that MODIS vegetation parameter products with a 500-m spatial resolution cannot meet the requirements of fine-scale land surface monitoring, a spatial- weight-based downscaling algorithm was adopted in this study. During the spatial downscaling process, obtaining strictly time-matched, cloud-free, single-date Landsat imagery is highly challenging due to long revisit cycles and frequent cloud and precipitation cover^[34,35]. To resolve this issue, the physical rationale and basic assumption of the proposed algorithm are as follows: although the absolute spectral reflectance of vegetation exhibits significant phenological cycle fluctuations within a year, the relative spatial distribution characteristics (i.e., spatial texture features) of heterogeneous land surface vegetation within the same region remain stable over an annual cycle^[36]. Therefore, the annual maximum NDVI from Landsat for the corresponding year was utilized as a high-resolution reference image. Fine texture features at a 30-m spatial resolution were extracted and transferred into the 500-m MODIS products.

First, the 30-m resolution Landsat reference image was aggregated to a 500-m resolution to generate a background field image, denoted as . Second, a spatial conversion coefficient, , was calculated using the MODIS vegetation parameters reconstructed by the mSSA algorithm () and the aggregated background field image. This coefficient quantifies the numerical correspondence between dynamic parameters and the static base map at the coarse resolution scale, reflecting the spatial texture details of the land surface:

                                                                                             (2)

where  is an infinitesimally small value added to prevent a zero denominator. To eliminate obvious block boundaries generated by the coarse-resolution grid and ensure the spatial continuity of the reconstructed results, bilinear interpolation was applied to . This ensures a smooth transition of the conversion coefficient between adjacent pixels. Finally, the 30-m reference image () was multiplied by the smoothed spatial coefficient. This yields the final 30-m spatial resolution key terrestrial ecosystem carbon cycle parameter products, which preserve fine textures and are free of boundary artifacts:

                                                                                         (3)

3.3 Technical Route

The dataset was developed in 3 main stages: data preprocessing, spatiotemporal reconstruction, and spatial downscaling (Figure 1). First, pixels contaminated by clouds, snow, and aerosols, together with retrieval results from backup algorithms, were removed using the QC information from the MODIS vegetation-parameter products. Only high-quality pixels processed by the main algorithm were retained. At the same time, land-cover types from MCD12Q1 were used as spatial constraints to exclude water bodies and non-vegetated areas. Second, for areas with main-algorithm failures and missing data, a comprehensive weight was calculated from observation quality and spatial distance. Optimal pixels within neighborhoods of the same vegetation type were selected to construct a multiple time-series matrix, which was input to the mSSA model to extract common variation features and predict missing values. This step generated a spatiotemporally seamless product at 500-m spatial resolution. Finally, annual maximum NDVI from Landsat was used as a high-resolution reference image, and the 500-m products were downscaled using the spatial texture features provided by this reference image. The final product is a spatiotemporally seamless 30-m dataset of key carbon-cycle parameters (NDVI, EVI, LAI, FPAR, and GPP) for the Greater Bay Area from 2001 to 2025.

Figure 1  Flowchart of the dataset development

3.4 Data Validation Method

To verify the accuracy of the dataset, correlation and error analyses were employed to evaluate the processing results. Specifically, the coefficient of determination (R²) and root mean square error (RMSE) were calculated to examine the correlation and evaluate the errors between the standard MODIS vegetation parameter products and the generated Greater Bay Area dataset:

                                                                                             (4)

                                                                                           (5)

where  represents the total number of samples;denotes the values of the MODIS dataset products;denotes the corresponding values of the generated 30-m spatial resolution key terrestrial ecosystem carbon cycle parameter dataset for the Greater Bay Area; and  is the mean value of the MODIS dataset products.

4 Data Results and Validation

4.1 Dataset Composition

The 30-m raster dataset of key carbon cycle parameters for terrestrial ecosystems in the Guangdong-Hong Kong-Macao Greater Bay Area (2001–2025) mainly comprises 5 core parameters covering 2001–2025: GPP, LAI, and FPAR at an 8-day temporal resolution, and EVI and NDVI at a 16-day temporal resolution. The directory 30m_XX represents the dataset for vegetation parameter XX and contains subfolders such as GBA_XX_30m_ Downscaled_2001, which represents the XX parameter dataset for 2001. Each subfolder contains data files named according to the convention GBA_XXXX_30m_YYYYMMDD. tif, where YYYY, MM, and DD represent the year, month, and day, respectively. For example, GBA_NDVI_30m_20010101.tif represents NDVI data for the period beginning on January 1, 2001 (Table 2).

Table 2  Main files of the dataset

4.2 Data Validation

Figure 2 presents pixel-by-pixel comparisons between the 30-m gridded dataset of key terrestrial ecosystem carbon-cycle parameters for the Greater Bay Area (2001–2025), aggregated to 250-m and 500-m spatial resolutions, and the original MODIS dataset for four periods. The downscaled dataset maintains a strong numerical correlation with the original data. Across all seasons, the fitted lines for the five parameters closely follow the 1:1 line.

The accuracy of each parameter exhibits certain seasonal differences. NDVI, EVI, and GPP demonstrate high stability in all seasons. Their R² values are maintained above 0.865, reaching a maximum of 0.941. The corresponding RMSE values are at extremely low levels. The RMSE values for NDVI and EVI are generally stable below 0.09. The RMSE for GPP is below 0.0054 kg C·m^–2. The slopes of the fitted lines are concentrated between 0.91 and 0.96. In contrast, the fitting accuracy of LAI and FPAR exhibits some seasonal fluctuations. In winter and spring, the R² values of both parameters remain at high levels. However, in summer and autumn, their correlations slightly weaken. The R² of LAI decreases to 0.789, and the RMSE increases to 0.480. The R² of FPAR decreases to 0.729, and the RMSE increases to 0.1017. Nevertheless, the overall regression slopes for both parameters remain between 0.82 and 0.92. These results indicate that the dataset reasonably responds

numerically to the changes in land surface heterogeneity caused by seasonal transitions while preserving the spatial distribution patterns of the original observation data.

To further quantify the ability of the downscaling algorithm to restore land-surface spatial heterogeneity, independent cross-validation of the 30-m Greater Bay Area dataset was performed using 10-m Sentinel-2 imagery. Directly deriving advanced parameters such as LAI and FPAR from standard Sentinel-2 products can introduce systematic errors. Therefore, this validation focused on NDVI and EVI, which directly characterize vegetation canopy features.

Figure 3 shows the scatter density distributions and linear regression results between the Greater Bay Area 30-m spatial resolution NDVI and EVI datasets and the aggregated Sentinel-2 data. During winter (DOY 001), when vegetation is relatively sparse, the Greater Bay Area products exhibit high consistency with Sentinel-2. The R² of NDVI reaches 0.758, with an RMSE of 0.1245. The R² of EVI is 0.535, with an RMSE of 0.1168. In autumn (DOY 289), when vegetation cover in the south is high, the correlation between the two vegetation parameters is slightly lower than in winter. The R² values for NDVI and EVI are 0.690 and 0.529, respectively. Overall, a large number of pixels are clustered around the 1:1 diagonal line. The slopes of the fitted lines are maintained between 0.61 and 0.74, and the intercepts are small. This indicates that the Greater Bay Area 30-m spatial resolution dataset successfully restores spatial details and numerical distributions consistent with the Sentinel-2 sensor at the 30-m scale.

Figure 2  Scatter plots for validation between the dataset GBA_VP and the original MODIS dataset

Figure 3  Scatter plots for cross-validation between the dataset GBA_VP and Sentinel-2 observation data

4.3 Data Results Analysis

(1) Spatial distribution of key terrestrial ecosystem carbon cycle parameters in the Greater Bay Area

Figure 4 shows the multi-year mean spatial distributions of the five key terrestrial ecosystem carbon-cycle parameters in the study area for January, April, July, and October during 2001–2025. The five parameters exhibit highly consistent spatial patterns. High-value areas are mainly concentrated in peripheral mountainous and hilly regions, including Zhaoqing, Huizhou, and Jiangmen, whereas low-value areas are mainly distributed in the urban agglomeration around the Pearl River Estuary, including Guangzhou, Shenzhen, Dongguan, Foshan, Zhongshan, Hong Kong, and Macao. The 30-m data reveal the boundaries of local-scale features such as intra-urban green spaces and river networks. In terms of intra-annual variation, all five parameters show clear seasonality. Values are relatively low in January, increase markedly in April, and reach their annual maxima in July. In October, values decrease relative to July but generally remain higher than those in January and April.

Figure 4  Multi-year average spatial distribution of key terrestrial ecosystem carbon cycle parameters in the Guangdong-Hong Kong-Macao Greater Bay Area (2001–2025)

(2) Temporal variation of key terrestrial ecosystem carbon cycle parameters in the Greater Bay Area

Figure 5 illustrates the temporal variations in the five key terrestrial ecosystem carbon cycle parameters (NDVI, EVI, LAI, FPAR, and GPP) in the Greater Bay Area from 2001 to 2025. On an intra-annual scale, all five parameters exhibit highly consistent, periodic seasonal fluctuations. The time-series curves are continuous. Between 2001 and 2025, all five parameters show statistically significant linear increasing trends. Specifically, the annual growth rates for NDVI, EVI, and FPAR are 0.0050, 0.0038, and 0.0050, respectively. The annual growth rates for LAI and GPP are 0.0266 and 0.0004 kg C·m^–2, respectively.

Figure 5  Temporal variations of key terrestrial ecosystem carbon cycle parameters in the Greater Bay Area (2001–2025)

(3) Spatiotemporal reconstruction and downscaling effects of key terrestrial ecosystem carbon cycle parameters in the Greater Bay Area

Figure 6 compares local spatial details of the five core vegetation parameters in the original MODIS images, the spatiotemporally reconstructed images, and the final key terrestrial ecosystem carbon-cycle parameter dataset for the Greater Bay Area. The original images contain obvious data gaps (white pixels in the figure). In the spatiotemporally reconstructed images, missing pixels are effectively filled and spatial continuity is restored. In the final Greater Bay Area dataset, the spatial resolution is improved to 30 m, recovering land-surface texture and fine boundary features of land-cover objects.

Figure 6  Comparison of local spatial reconstruction and downscaling of key terrestrial ecosystem carbon cycle parameters in the Greater Bay Area

5 Discussion and Conclusions

MODIS vegetation parameter products have played a fundamental role in global ecological and climate-change research. However, in regions with high heterogeneity and intensive human disturbance, such as the Guangdong-Hong Kong-Macao Greater Bay Area, their coarse 500-m spatial resolution and frequent cloud and precipitation contamination lead to severe mixed-pixel effects and data discontinuities. This study selected the Greater Bay Area as the research region. Under strict land-cover constraints, the mSSA spatiotemporal reconstruction algorithm was introduced. A spatial-weight-based downscaling model was constructed in synergy with the fine spatial texture features extracted from the annual maximum NDVI of Landsat. Specifically, addressing the issue of inconsistent observation times between Landsat and MODIS in multi-source spatiotemporal data fusion, the annual maximum NDVI was utilized as a relatively stable, static spatial constraint base map. By calculating spatial conversion coefficients, these stable spatial texture features were mapped onto the coarse MODIS pixels^[16]. Consequently, high-quality downscaling was achieved in the cloudy and rainy southern regions, which lack high-frequency, cloud-free Landsat imagery. Thus, a spatiotemporally continuous dataset of five key terrestrial ecosystem carbon cycle parameters (NDVI, EVI, LAI, FPAR, and GPP) with a 30-m spatial resolution, spanning from 2001 to 2025, was produced.

The pixel-by-pixel validation results using the original MODIS observation data indicate that this downscaled dataset enhances the expression capability of land surface texture details while preserving the basic pattern of large-scale spatial distribution. The overall fitting accuracy demonstrates high reliability. The R² values of NDVI, EVI, and GPP remain stable above 0.865 year-round. It should be noted that the fitting accuracy of LAI and FPAR exhibits phenology-driven seasonal fluctuations. During summer and autumn, when vegetation canopy closure reaches its peak, the correlation of both parameters experiences a certain degree of attenuation. This attenuation is constrained by the saturation effect of optical remote sensing signals and the nonlinear scattering of complex canopy structures. Their R² values decrease to 0.789 and 0.729, respectively. However, their linear regression slopes remain within a reasonable range above 0.82. This dataset mitigates the limitations of existing coarse-resolution products when applied to highly fragmented land surfaces. The five key carbon cycle parameters provided offer finer underlying data support for carbon sink estimation and dynamic ecosystem monitoring in the Greater Bay Area.

Author Contributions

You, J. K., Zhang, Y. H. and Wu, G. F. conceptualized the overall design for dataset development. You, J. K. and Guo, S. X. collected and processed the data. You, J. K. and Zhang, Y. H. designed the models and algorithms. Hu, Z. W. and Wang, J. Z. conducted the data validation. You, J. K. and Zhang, Y. H. wrote the data paper.

Conflicts of Interest

The authors declare no conflicts of interest.

References

[1]        Xue, J. R., Su, B. F. Significant remote sensing vegetation indices: a review of developments and applications [J]. Journal of Sensors, 2017, 2017(1): 1353691.

[2]        Bannari, A., Morin, D., Bonn, F., et al. A review of vegetation indices [J]. Remote Sensing Reviews, 1995, 13(1–2): 95–120.

[3]        Liu, Z. T., Wang, S. J., Fang, C. L. Spatiotemporal evolution and influencing mechanism of ecosystem service value in the Guangdong-Hong Kong-Macao Greater Bay Area [J]. Acta Geographica Sinica, 2023, 33(6): 1226–1244.

[4]        Pettorelli, N., Vik, J. O., Mysterud, A., et al. Using the satellite-derived NDVI to assess ecological responses to environmental change [J]. Trends in Ecology & Evolution, 2005, 20(9): 503–510.

[5]        Jiang, Z. Y., Huete, A. R., Didan, K., et al. Development of a two-band enhanced vegetation index without a blue band [J]. Remote Sensing of Environment, 2008, 112(10): 3833–3845.

[6]        Fang, H. L., Baret, F., Plummer, S., et al. An overview of global leaf area index (LAI): methods, products, validation, and applications [J]. Reviews of Geophysics, 2019, 57(3): 739–799.

[7]        Liu, Y., Liu, R. G., Liu, L. Y., et al. Current status and perspectives of Leaf Area Index retrieval from optical remote sensing data [J]. Journal of Geo-Information Science, 2013, 15(5): 734–743.

[8]        Zhang, Y. H., Fang, H. L., Wang, Y., et al. Variation of intra-daily instantaneous FAPAR estimated from the geostationary Himawari-8 AHI data [J]. Agricultural and Forest Meteorology, 2021, 307: 108535.

[9]        Zhang, Y. H., Hu, Z. W., Wang, J. Z., et al. Temporal upscaling of MODIS instantaneous FAPAR improves forest gross primary productivity (GPP) simulation [J]. International Journal of Applied Earth Observation and Geoinformation, 2023, 121: 103360.

[10]     Zhang, X., Friedl, M. A., Schaaf, C. B., et al. Monitoring vegetation phenology using MODIS [J]. Remote Sensing of Environment, 2003, 84(3): 471–475.

[11]     Myneni, R. B., Hoffman, S., Knyazikhin, Y., et al. Global products of vegetation leaf area and fraction absorbed PAR from year one of MODIS data [J]. Remote Sensing of Environment, 2002, 83(1–2): 214–231.

[12]     Liu, L., Zhang, Y. H., Hu, Z. W., et al. MODIS global Leaf Area Index product reprocessing dataset (2001–2021) [J]. Journal of Global Change Data & Discovery, 2023, 7(3): 242–251. https://doi.org/10. 3974/geodp.2023.03.02.

[13]     Meraner, A., Ebel, P., Zhu, X. X., et al. Cloud removal in Sentinel-2 imagery using a deep residual neural network and SAR-optical data fusion [J]. ISPRS Journal of Photogrammetry and Remote Sensing, 2020, 166: 333–346.

[14]     Shen, H. F., Li, X. H., Cheng, Q., et al. Missing information reconstruction of remote sensing data: a technical review [J]. IEEE Geoscience and Remote Sensing Magazine, 2015, 3(3): 61–85.

[15]     Zhu, X. L., Cai, F. Y., Tian, J. Q., et al. Spatiotemporal fusion of multisource remote sensing data: literature survey, taxonomy, principles, applications, and future directions [J]. Remote Sensing, 2018, 10(4): 527.

[16]     Zhu, X. L., Chen, J., Gao, F., et al. An enhanced spatial and temporal adaptive reflectance fusion model for complex heterogeneous regions [J]. Remote Sensing of Environment, 2010, 114(11): 2610–2623.

[17]     Gao, F., Masek, J., Schwaller, M., et al. On the blending of the Landsat and MODIS surface reflectance: predicting daily Landsat surface reflectance [J]. IEEE Transactions on Geoscience and Remote Sensing, 2006, 44(8): 2207–2218.

[18]     Zhu, X. L., Helmer, E. H., Gao, F., et al. A flexible spatiotemporal method for fusing satellite images with different resolutions [J]. Remote Sensing of Environment, 2016, 172: 165–177.

[19]     Shen, H. F., Wu, J. A., Cheng, Q., et al. A spatiotemporal fusion based cloud removal method for remote sensing images with land cover changes [J]. IEEE Journal of Selected Topics in Applied Earth Observations and Remote Sensing, 2019, 12(3): 862–874.

[20]     Emelyanova, I. V., Mcvicar, T. R., van Niel, T. G., et al. Assessing the accuracy of blending Landsat– MODIS surface reflectances in two landscapes with contrasting spatial and temporal dynamics: a framework for algorithm selection [J]. Remote Sensing of Environment, 2013, 133: 193–209.

[21]     Xiong, C., Wu, Z., Zeng, Z. Y., et al. Spatiotemporal evolution of forest landscape pattern in Guangdong- Hong Kong-Macao Greater Bay Area based on “Spatial Morphology-Fragmentation-Aggregation” [J]. Acta Ecologica Sinica, 2023, 43: 3032–3044.

[22]     Chen, X., Wang, D. W., Chen, J., et al. The mixed pixel effect in land surface phenology: a simulation study [J]. Remote Sensing of Environment, 2018, 211: 338–344.

[23]     Tao, X., Liang, S. L., He, T., et al. Estimation of fraction of absorbed photosynthetically active radiation from multiple satellite data: model development and validation [J]. Remote Sensing of Environment, 2016, 184: 539–557.

[24]     Zheng, W. H., Hu, Z. W., Zhang, Y. H., et al. Spatiotemporal dynamic of surface water bodies using dense Landsat time-series data in Guangdong-Hong Kong-Macao Greater Bay Area from 1990 to 2020 [J]. Environmental Ecology, 2025, 7(1): 53–64.

[25]     Hu, Z. W., Chu, Y. Q., Zhang, Y. H., et al. Scale matters: How spatial resolution impacts remote sensing based urban green space mapping? [J]. International Journal of Applied Earth Observation and Geoinformation, 2024, 134: 104178.

[26]     Hsieh, P. F., Lee, L. C., Chen, N. Y. Effect of spatial resolution on classification errors of pure and mixed pixels in remote sensing [J]. IEEE Transactions on Geoscience and Remote Sensing, 2001, 39(12): 2657–2663.

[27]     Oropeza, V., Sacchi, M. Simultaneous seismic data denoising and reconstruction via multichannel singular spectrum analysis [J]. Geophysics, 2011, 76(3): V25–V32.

[28]     You, J. K., Zhang, Y. H., Hu, Z. W., et al. 30-m raster dataset of key carbon cycle parameters for terrestrial ecosystems in the Guangdong-Hong Kong-Macao Greater Bay Area (2001–2025) [J/DB/OL]. Digital Journal of Global Change Data Repository, 2026. https://doi.org/10.3974/geodb.2026.03.10.V1.

[29]     GCdataPR Editorial Office. GCdataPR data sharing policy [OL]. https://doi.org/10.3974/dp.policy.2014.05 (Updated 2017).

[30]     Running, S. W., Zhao, M. Daily GPP and annual NPP (MOD17A2/A3) products NASA Earth Observing System MODIS land algorithm [J]. MOD17 User’s Guide, 2015, 2015: 1–28.

[31]     Huete, A., Didan, K., Miura, T., et al. Overview of the radiometric and biophysical performance of the MODIS vegetation indices [J]. Remote Sensing of Environment, 2002, 83(1): 195–213.

[32]     Sulla-Menashe, D., Gray, J. M., Abercrombie, S. P., et al. Hierarchical mapping of annual global land cover 2001 to present: the MODIS Collection 6 Land Cover product [J]. Remote Sensing of Environment, 2019, 222: 183–194.

[33]     Wulder, M. A., Roy, D. P., Radeloff, V. C., et al. Fifty years of Landsat science and impacts [J]. Remote Sensing of Environment, 2022, 280: 113195.

[34]     Onojeghuo, A. O., Blackburn, G. A., Wang, Q. M., et al. Rice crop phenology mapping at high spatial and temporal resolution using downscaled MODIS time-series [J]. GIScience & Remote Sensing, 2018, 55(5): 659–677.

[35]     Li, X. Q., Peng, Q. Y., Shen, R. Q., et al. Long-term reconstructed vegetation index dataset in China from fused MODIS and Landsat data [J]. Scientific Data, 2025, 12(1): 152.

[36]     Holben, B. N. Characteristics of maximum-value composite images from temporal AVHRR data [J]. International Journal of Remote Sensing, 1986, 7(11): 1417–1434.