Estimation of Terrestrial Net Primary Productivity in China from Fengyun-3D Satellite Data

基于FY-3D卫星的中国陆地净初级生产力估算研究

+ Author Affiliations + Find other works by these authors
  • Corresponding author: Xiuzhen HAN, hanxz@cma.gov.cn
  • Funds:

    Supported by the National Key Research and Development Program of China (2018YFC1506500), Natural Science Program of China (U2142212), and National Natural Science Foundation of China (41871028)

  • doi: 10.1007/s13351-022-1183-6

PDF

  • Currently, the satellite data used to estimate terrestrial net primary productivity (NPP) in China are predominantly from foreign satellites, and very few studies have based their estimates on data from China’s Fengyun satellites. Moreover, despite their importance, the influence of land cover types and the normalized difference vegetation index (NDVI) on NPP estimation has not been clarified. This study employs the Carnegie–Ames–Stanford approach (CASA) model to compute the fraction of absorbed photosynthetically active radiation and the maximum light use efficiency suitable for the main vegetation types in China in accordance with the finer resolution observation and monitoring-global land cover (FROM-GLC) classification product. Then, the NPP is estimated from the Fengyun-3D (FY-3D) data and compared with the Moderate Resolution Imaging Spectroradiometer (MODIS) NPP product. The FY-3D NPP is also validated with existing research results and historical field-measured NPP data. In addition, the effects of land cover types and the NDVI on NPP estimation are analyzed. The results show that the CASA model and the FY-3D satellite data estimate an average NPP of 441.2 g C m−2 yr−1 in 2019 for China’s terrestrial vegetation, while the total NPP is 3.19 Pg C yr−1. Compared with the MODIS NPP, the FY-3D NPP is overestimated in areas of low vegetation productivity and is underestimated in high-productivity areas. These discrepancies are largely due to the differences between the FY-3D NDVI and MODIS NDVI. Compared with historical field-measured data, the FY-3D NPP estimation results outperformed the MODIS NPP results, although the deviation between the FY-3D NPP estimate and the in-situ measurement was large and may exceed 20% at the pixel scale. The land cover types and the NDVI significantly affected the spatial distribution of NPP and accounted for NPP deviations of 17.0% and 18.1%, respectively. Additionally, the total deviation resulting from the two factors reached 29.5%. These results show that accurate NDVI products and land cover types are important prerequisites for NPP estimation.
    在优化的光能利用率CASA( Carnegie–Ames–Stanford approach)模型基础上,引入中国区域FROM-GLC( finer resolution observation and monitoring-global land cover)土地覆盖分类产品,利用1 km 分辨率FY-3D卫星资料开展了中国陆地区域净初级生产力(NPP)估算,并与国外MODIS NPP产品和历史实测数据进行了对比分析,同时开展了土地覆盖类型和归一化植被指数(NDVI)对NPP估算的影响分析。研究表明,2019年中国陆地植被NPP为3.19 PgC a−1,平均NPP为441.2 gC m−2 a−1。与MODIS NPP产品相比,FY-3D NPP在植被生产力低的地区被高估,在生产力高的地区被低估。这些差异主要是由于FY-3D NDVI和MODIS NDVI之间差异造成。与历史站点实测数据相比,FY-3D NPP估计结果优于MODIS NPP结果,土地覆盖类型和NDVI显著影响NPP的空间分布,分别占NPP偏差的17.0%和18.1%,综合偏差达到29.5%。这些结果表明,准确的NDVI产品和土地覆盖类型是NPP估算的重要前提。本文首次利用FY-3D国产气象卫星数据实现了中国陆地区域NPP的准确估算,获得了NDVI和土地覆盖产品对NPP估算的定量影响。
  • 加载中
  • Fig. 1.  Relationship between FPAR and NDVI for the six vegetation types (Biome1: cereal and herbaceous crops; Biome2: shrubs; Biome3: broadleaf crops; Biome4: savanna; Biome5: broadleaf forest; and Biome6: coniferous forest). The blue points (NDVI and FPAR data) are from the lookup table that reports the relationship between the NDVI and FPAR of the six vegetation types in the MODIS FPAR product (MOD15A2H; Knyazikhin et al., 1999).

    Fig. 2.  Spatial distribution maps of (a) FROM-GLC land cover types and (b) MODIS IGBP land cover types in China.

    Fig. 3.  (a) NPP of China in 2019 estimated based on FY-3D data and (b) the difference (ΔNPP) between the estimated FY-3D NPP and the MODIS NPP.

    Fig. 4.  Spatial distributions of (a) the difference (ΔNDVI) between the annual average FY-3D NDVI and the annual average MODIS NDVI and (b) the scatter probability density diagrams and fitted equations of the relationship between the FY-3D NDVI and MODIS NDVI in terms of the annual average and (c) January, (d) April, (e) July, and (f) October (the blue line is the 1 : 1 line, and the red dashed line is the fitting trend line).

    Fig. 5.  Comparison of the annual average FY-3D NPP estimation results with historical field-measured NPP and MODIS NPP for typical terrestrial vegetation types in China.

    Fig. 6.  (a) Comparison of the previously measured NPP, FY-3D NPP, and MODIS NPP for 19 sites; (b) the scatter plot between the sites’ measured NPP and FY-3D NPP; and (c) the scatter plot between the MODIS NPP values and the sites’ measured NPP for the 19 sites (R is the correlation coefficient, RMSE is the root mean square error, and the red line is the 1: 1 line).

    Fig. 7.  Spatial distributions of the differences (ΔNPP) between the NPP values in China estimated using different schemes and FY-3D NPP: (a) land cover types replaced by the IGBP classification product, (b) NDVI replaced by MODIS NDVI, and (c) land cover types and NDVI both replaced.

    Table 1.  Maximum light use efficiency values of typical vegetation types in the Chinese region

    Vegetation typeMaximum light use efficiency [g C (MJ)−1]
    Deciduous coniferous forest0.485
    Evergreen coniferous forest0.389
    Deciduous broadleaf forest0.692
    Evergreen broadleaf forest0.985
    Coniferous and broadleaf mixed
    forest
    0.475
    Evergreen and deciduous broadleaf
    mixed forest
    0.768
    Shrubland0.429
    Grassland0.573
    Farmland0.542
    Other0.542
    Download: Download as CSV

    Table 2.  Historically measured NPP of typical terrestrial ecosystem sites in China based on document integration (Chen et al., 2019)

    SiteLongitude (°E)Latitude (°N)Vegetation typeYear of observationMeasured NPP (g C m−2 yr−1)
    Zhonghe100.44638.975Phragmites australis2012–2014390.00
    Ailaoshan101.01724.533Evergreen broadleaf forest2009–2010976.20
    Chanzhou109.52019.508Rubber forest20101133.50
    Huitong109.75026.833Evergreen needleleaf forest2008–2009268.50
    Taoyuan111.45028.917Rice2003675.20
    Xiaolangdi112.46735.017Mixed forest2006–2010355.00
    Dinghushan112.53623.173Mixed forest2003–2008395.90
    Ningxiang112.56728.333Mixed forest2013428.80
    Yueyang112.85029.517Poplar forest2005–2007515.70
    Shouyang113.20037.750Spring maize2012–2014509.30
    Xiping113.86733.350Poplar forest2010343.40
    Qianyanzhou115.06726.733Evergreen broadleaf forest2003–2008487.50
    Weishan116.05036.650Winter wheat and corn2006–2008559.00
    Daxing116.29139.534Poplar forest2006–2009518.30
    Yucheng116.60036.950Winter wheat and corn2003–2008366.00
    Anqing117.03330.500Poplar forest2005–2007506.10
    Tianmushan119.43330.350Mixed forest2013–2014738.18
    Anji119.67430.476Bamboo forest2011–2013585.40
    Changbaishan128.40042.400Deciduous broadleaf forest2003–2008302.33
    Download: Download as CSV

    Table 3.  Comparison of NPP in China as estimated by different methods

    MethodFY-3D NPPMODIS NPPScheme 1Scheme 2Scheme 3
    Total NPP (Pg C yr−1)3.193.233.343.463.64
    Average NPP (g C m−2 yr−1)441.2448.3462.9479.4504.9
    Percentage anomaly with total FY-3D NPP (%)1.64.78.514.1
    Mean absolute error from FY-3D NPP (g C m−2 yr−1)149.074.879.7129.4
    Download: Download as CSV
  • [1]

    Bao, G., X.-P. Xin, Y.-H. Bao, et al., 2016: Optimization of maximum light use efficiency in Inner Mongolian steppe. Spectrosc. Spect. Anal., 36, 3280–3286. (in Chinese)
    [2]

    Bartholomé, E., and A. S. Belward, 2005: GLC2000: A new approach to global land cover mapping from Earth observation data. Int. J. Remote Sens., 26, 1959–1977. doi: 10.1080/01431160412331291297.
    [3]

    Bicheron, P., P. Defourny, C. Brockmann, et al., 2011: GlobCover: Products description and validation report. Foro Mundial De La Salud, 17, 285–287. Available online at https://www.researchgate.net/publication/260137807_GLOBCOVER_products_description_and_validation_report. Accessed on 20 May 2022.
    [4]

    Chen, A. P., S. L. Piao, S. Luyssaert, et al., 2011: Forest annual carbon cost: Reply. Ecology, 92, 1998–2002. doi: 10.1890/11-0785.1.
    [5]

    Chen, F., H.-B. Li, and Y.-J. Liu, 2018: Spatio–temporal differentiation and influencing factors of vegetation net primary productivity using GIS and CASA: A case study in Yuanyang County, Yunnan. Chinese J. Ecol., 37, 2148–2158. doi: 10.13292/j.1000-4890.201807.003. (in Chinese)
    [6]

    Chen, R. R., Z. M. Hu, S. G. Li, et al., 2020: Assessment of normalized difference vegetation index from different data sources in grassland of northern China. J. Geo-Inf. Sci., 22, 1910–1919. doi: 10.12082/dqxxkx.2020.190237. (in Chinese)
    [7]

    Chen, T., L. Y. Zhao, B. F. Hou, et al., 2018: Remote sensing estimation net primary productivity (NPP) of ecosystem in Wuhan City using the CASA model. J. Hubei Univ. (Nat. Sci.), 40, 315–322, doi: 10.3969/j.issn.1000-2375.2018.03.020. (in Chinese)
    [8]

    Chen, Z., G. R. Yu, X. J. Zhu, et al., 2019: A dataset of primary production, respiration and net production in Chinese typical terrestrial ecosystems based on literature integration. China Sci. Data, 4, 50–58. doi: 10.11922/csdata.2018.0024.zh. (in Chinese)
    [9]

    Chen, Z.-H., Q.-Y. Ma, J. Wang, et al., 2008: Estimation of Heihe basin net primary productivity using the CASA model. J. Nat. Resour., 23, 263–273. doi: 10.3321/j.issn:1000-3037.2008.02.011. (in Chinese)
    [10]

    Dong, D., and J. Ni, 2011: Modeling changes of net primary productivity of karst vegetation in southwestern China using the CASA model. Acta Ecol. Sinica, 31, 1855–1866. (in Chinese)
    [11]

    Du, J. Q., J. M. Shu, C. X. Zhao, et al., 2016: Comparison of GIMMS NDVI3g and GIMMS NDVIg for monitoring vegetation activity and its responses to climate changes in Xinjiang during 1982–2006. Acta Ecol. Sinica, 36, 6738–6749. doi: 10.5846/stxb201504190805. (in Chinese)
    [12]

    Field, C. B., M. J. Behrenfeld, J. T. Randerson, et al., 1998: Primary production of the biosphere: Integrating terrestrial and oceanic components. Science, 281, 237–240. doi: 10.1126/science.281.5374.237.
    [13]

    Friedl, M. A., D. K. Mciver, J. C. F. Hodges, et al., 2002: Global land cover mapping from MODIS: Algorithms and early results. Remote Sens. Environ., 83, 287–302. doi: 10.1016/S0034-4257(02)00078-0.
    [14]

    Ge, M. X., J. Zhao, B. Zhong, et al., 2017. Comparison of the vegetation indexes between FY-3/VIRR, FY-3/MERSI and EOS/MODIS data. Remote Sens. Technol. Appl., 32, 262–273. (in Chinese)
    [15]

    Gong, P., 2009: Accuracy verification of global land cover map based on global flux observatory. Prog. Nat. Sci., 19, 754–759. doi: 10.3321/j.issn:1002-008X.2009.07.008. (in Chinese)
    [16]

    Gu, J., X. Li, C.-L. Huang, et al., 2013: Simulating net primary productivity of Chinese terrestrial vegetation during 2002–2010. J. Lanzhou Univ. (Nat. Sci.), 49, 203–213. doi: 10.3969/j.issn.0455-2059.2013.02.008. (in Chinese)
    [17]

    Han, X. Z., F. Wang, and Y. Han, 2019: Fengyun-3D MERSI true color imagery developed for environmental applications. J. Meteor. Res., 33, 914–924. doi: 10.1007/s13351-019-9028-7.
    [18]

    Han, X. Z., J. Yang, S. H. Tang, et al., 2020: Vegetation products derived from Fengyun-3D medium resolution spectral imager-II. J. Meteor. Res., 34, 775–785. doi: 10.1007/s13351-020-0027-5.
    [19]

    Hansen, M. C., R. S. Defries, J. R. G. Townshend, et al., 2000: Global land cover classification at 1 km spatial resolution using a classification tree approach. Int. J. Remote Sens., 21, 1331–1364. doi: 10.1080/014311600210209.
    [20]

    Hicke, J. A., G. P. Asner, J. T. Randerson, et al., 2002: Trends in North American net primary productivity derived from satellite observations, 1982–1998. Global Biogeochem. Cy., 16, 1018. doi: 10.1029/2001GB001550.
    [21]

    Hong, C. Q., X. B. Jin, C. C. Chen, et al., 2017: Overview on estimation models of land net primary productivity integrating remote sensing data. Prog. Geogr., 36, 924–939. doi: 10.18306/dlkxjz.2017.08.002. (in Chinese)
    [22]

    Huete, A., C. Justice, and W. van Leeuwen, 1999: MODIS vegetation index (MOD 13) algorithm theoretical basis document. Available online at https://lpdaac.usgs.gov/documents/104/MOD13_ATBD.pdf. Accessed on 18 May 2022.
    [23]

    Knyazikhin, Y., J. Glassy, J. L. Privette, et al., 1999: MODIS leaf area index (LAI) and fraction of photosynthetically active radiation absorbed by vegetation (FPAR) product (MOD15) algorithm theoretical basis document. Available online at https://lpdaac.usgs.gov/documents/90/MOD15_ATBD.pdf. Accessed on 18 May 2022.
    [24]

    Li, C. C., J. Wang, L. Y. Hu, et al., 2014: A circa 2010 thirty meter resolution forest map for China. Remote Sens., 6, 5325–5343. doi: 10.3390/rs6065325.
    [25]

    Li, C. H., H. J. Cao, Y. P. Fan, et al., 2019: Remote sensing estimation and analysis of net primary productivity (NPP) based on corrected CASA model: A case study of Hexi corridor. Acta Ecol. Sinica, 39, 1616–1626. doi: 10.5846/stxb201802060319. (in Chinese)
    [26]

    Liang, S. L., X. W. Li, J. D. Wang, et al., 2019: Quantitative Remote Sensing: Ideas and Algorithms. 2nd Ed., Science Press, Beijing, 941–942. (in Chinese)
    [27]

    Liu, G., R. Sun, Z. Q. Xiao, et al., 2017: Analysis of spatial and temporal variation of net primary productivity and climate controls in China from 2001 to 2014. Acta Ecol. Sinica, 37, 4936–4945. (in Chinese)
    [28]

    Liu, J. F., L. Chen, Q. Meng, et al., 2019: Research on maximum light use efficiency based on CASA-VPM model. J. Henan Agric. Sci., 48, 157–163. doi: 10.15933/j.cnki.1004-3268.2019.12.024. (in Chinese)
    [29]

    Liu, Y. H., C. H. Ye, K. W. Wang, et al., 2008: Soil moisture prediction technique in Beijing supported by RS and GIS techniques. Trans. CSAE, 24, 155–160. doi: 10.3321/j.issn:1002-6819.2008.09.031. (in Chinese)
    [30]

    Liu, Y. H., W. J. Quan, and Y. H. Gao, 2010: Net primary production and its spatio–temporal pattern in North China. J. Nat. Resour., 25, 564–573. doi: 10.11849/zrzyxb.2010.04.005. (in Chinese)
    [31]

    Loveland, T. R., B. C. Reed, J. F. Brown, et al., 2000: Development of a global land cover characteristics database and IGBP DISCover from 1 km AVHRR data. Int. J. Remote Sens., 21, 1303–1330. doi: 10.1080/014311600210191.
    [32]

    Potter, C. S., J. T. Randerson, C. B. Field, et al., 1993: Terrestrial ecosystem production: A process model based on global satellite and surface data. Global Biogeochem. Cy., 7, 811–841. doi: 10.1029/93GB02725.
    [33]

    Prince, S. D., and S. N. Goward, 1995: Global primary production: A remote sensing approach. J. Biogeogr., 22, 815–835. doi: 10.2307/2845983.
    [34]

    Qu, J. H., J. J. Yan, and Y. T. Wang, 2019: Global clear-sky data synthesis technology based on FY-3D MERSI-II instrument. Meteor. Sci. Technol., 47, 539–545. doi: 10.19517/j.1671-6345.20180442. (in Chinese)
    [35]

    Ran, Y.-H., X. Li, and L. Lu, 2009: Accuracy evaluation of the four remote sensing based land cover products over China. J. Glaciol. Geocryol., 31, 490–500. doi: 10.3724/SP.J.1300.2012.20075. (in Chinese)
    [36]

    Running, S. W., R. Nemani, J. M. Glassy, et al., 1999: MODIS daily photosynthesis (PSN) and annual net primary production (NPP) product (MOD17) algorithm theoretical basis document. Available online at https://lpdaac.usgs.gov/documents/95/MOD17_ATBD.pdf. Accessed on 18 May 2022.
    [37]

    Running, S. W., R. R. Nemani, F. A. Heinsch, et al., 2004: A continuous satellite-derived measure of global terrestrial primary production. BioScience, 54, 547–560. doi: 10.1641/0006-3568(2004)054[0547:ACSMOG]2.0.CO;2.
    [38]

    Shan, T. C., W. Zheng, J. Chen, et al., 2021: A burned area mapping method for the FY-3D MERSI based on the single-temporal L1 data and multi-temporal daily active fire products. Int. J. Remote Sens., 42, 1292–1310. doi: 10.1080/01431161.2020.1826064.
    [39]

    Strahler, A., D. Muchoney, J. Botrak, et al., 1999: MODIS land cover product algorithm theoretical basis document version 5.0 MODIS land cover and land-cover change. Available online at https://lpdaac.usgs.gov/documents/86/MCD12_ATBD.pdf. Accessed on 18 May 2022.
    [40]

    Urban, O., D. Janouš, M. Acosta, et al., 2007: Ecophysiological controls over the net ecosystem exchange of mountain spruce stand. Comparison of the response in direct vs. diffuse solar radiation. Glob. Change Biol., 13, 157–168. doi: 10.1111/j.1365-2486.2006.01265.x.
    [41]

    Wang, B.-L., J.-J. Wang, Y. Yang, et al., 2013: Algorithm improvements for two important parameters of FPAR and maximum solar energy utilization efficiency. Acta Pratacul. Sinica, 22, 220–228. doi: 10.11686/cyxb20130526. (in Chinese)
    [42]

    Wang, J., Y. Y. Zhao, C. C. Li, et al., 2015: Mapping global land cover in 2001 and 2010 with spatial–temporal consistency at 250 m resolution. ISPRS J. Photogramm. Remote Sens., 103, 38–47. doi: 10.1016/j.isprsjprs.2014.03.007.
    [43]

    Wang, J. B., J. Y. Liu, Q. Q. Shao, et al., 2009: Spatial–temporal patterns of net primary productivity for 1988–2004 based on GLOPEM-CEVSA model in the “Three-River Headwaters” region of Qinghai Province, China. J. Plant Ecol., 33, 254–269. doi: 10.3773/j.issn.1005-264x.2009.02.003. (in Chinese)
    [44]

    Wang, J. L., H. Q. Lyu, G. P. Zhang, et al., 2005: Agrometeorological Forecast. Meteorological Press, Beijing, 89–109. (in Chinese)
    [45]

    Wang, W. W., C. M. Zhang, F. Li, et al., 2020: Extracting soil moisture from Fengyun-3D medium resolution spectral imager-II imagery by using a deep belief network. J. Meteor. Res., 34, 748–759. doi: 10.1007/s13351-020-9191-x.
    [46]

    Xiao, X. M., Q. Y. Zhang, D. Hollinger, et al., 2005: Modeling gross primary production of an evergreen needleleaf forest using MODIS and climate data. Ecol. Appl., 15, 954–969. doi: 10.1890/04-0470.
    [47]

    Xu, J., H.-L. Chen, S.-S. Shang, et al., 2020: Response of net primary productivity of Tibetan Plateau vegetation to climate change based on CEVSA model. Arid Land Geogr., 43, 592–601. (in Chinese)
    [48]

    Xu, Y. Q., F. J. Xiao, and L. Yu, 2020: Review of spatio–temporal distribution of net primary productity in forest ecosystem and its responses to climate change in China. Acta Ecol. Sinica, 40, 4710–4723. (in Chinese)
    [49]

    Yan, H., S.-Q. Wang, D. Billesbach, et al., 2015: Improved global simulations of gross primary product based on a new definition of water stress factor and a separate treatment of C3 and C4 plants. Ecol. Model., 297, 42–59. doi: 10.1016/j.ecolmodel.2014.11.002.
    [50]

    Yang, C., J. J. Yu, X. F. Song, et al., 2004: Reference crop evapotranspiration calculation in short interval of mountainous area in North China. Prog. Geogr., 23, 71–80. doi: 10.11820/dlkxjz.2004.06.009. (in Chinese)
    [51]

    Yang, Y. K., P. F. Xiao, X. Z. Feng, et al., 2014: Comparison and assessment of large-scale land cover datasets in China and adjacent regions. J. Remote Sens., 18, 453–475. doi: 10.11834/jrs.20143055. (in Chinese)
    [52]

    Yin, K., Y. C. Tian, C. Yuan, et al., 2015: NPP spatial and temporal pattern of vegetation in Beijing and its factor explanation based on CASA model. Remote Sens. Land Resour., 27, 133–139. doi: 10.6046/gtzyyg.2015.01.21. (in Chinese)
    [53]

    Yuan, J. G., Z. Niu, and C. L. Wang, 2006: Vegetation NPP distribution based on MODIS data and CASA model—A case study of northern Hebei Province. Chinese Geogr. Sci., 16, 334–341. doi: 10.1007/s11769-006-0334-5.
    [54]

    Zhang, F., G.-S. Zhou, and Y.-H. Wang, 2008: Dynamics simulation of net primary productivity by a satellite data-driven casa model in Inner Mongolian typical steppe, China. J. Plant Ecol., 32, 786–797. doi: 10.3773/j.issn.1005-264x.2008.04.007. (in Chinese)
    [55]

    Zhang, M. L., W. L. Jiang, Q. G. Chen, et al., 2014: Estimation of grassland net primary production in China with improved CASA model based on the CSCS. J. Desert Res., 34, 1150–1160. (in Chinese)
    [56]

    Zhao, Z. P., X. P. Wu, G. Li, et al., 2015: Drought in southwestern China and its impact on the net primary productivity of vegetation from 2009–2011. Acta Ecol. Sinica, 35, 350–360. doi: 10.5846/stxb201304040604. (in Chinese)
    [57]

    Zhou, G.-S., and X.-S. Zhang, 1995: A natural vegetation NPP model. Acta Phytoecol. Sinica, 19, 193–200. (in Chinese)
    [58]

    Zhu, W.-Q., Y.-Z. Pan, and J.-S. Zhang, 2007: Estimation of net primary productivity of Chinese terrestrial vegetation based on remote sensing. J. Plant Ecol., 31, 413–424. doi: 10.17521/cjpe.2007.0050. (in Chinese)
  • 加载中
通讯作者: 陈斌, bchen63@163.com
  • 1. 

    沈阳化工大学材料科学与工程学院 沈阳 110142

  1. 本站搜索
  2. 百度学术搜索
  3. 万方数据库搜索
  4. CNKI搜索

Estimation of Terrestrial Net Primary Productivity in China from Fengyun-3D Satellite Data

    Corresponding author: Xiuzhen HAN, hanxz@cma.gov.cn
  • 1. Earth System Modeling and Prediction Center, China Meteorological Administration, Beijing 100081
  • 2. State Key Laboratory of Severe Weather, Chinese Academy of Meteorological Sciences, China Meteorological Administration, Beijing 100081
  • 3. National Satellite Meteorological Center, China Meteorological Administration, Beijing 100081
  • 4. School of Remote Sensing and Geomatics Engineering, Nanjing University of Information Science & Technology, Nanjing 210044
Funds: Supported by the National Key Research and Development Program of China (2018YFC1506500), Natural Science Program of China (U2142212), and National Natural Science Foundation of China (41871028)

Abstract: Currently, the satellite data used to estimate terrestrial net primary productivity (NPP) in China are predominantly from foreign satellites, and very few studies have based their estimates on data from China’s Fengyun satellites. Moreover, despite their importance, the influence of land cover types and the normalized difference vegetation index (NDVI) on NPP estimation has not been clarified. This study employs the Carnegie–Ames–Stanford approach (CASA) model to compute the fraction of absorbed photosynthetically active radiation and the maximum light use efficiency suitable for the main vegetation types in China in accordance with the finer resolution observation and monitoring-global land cover (FROM-GLC) classification product. Then, the NPP is estimated from the Fengyun-3D (FY-3D) data and compared with the Moderate Resolution Imaging Spectroradiometer (MODIS) NPP product. The FY-3D NPP is also validated with existing research results and historical field-measured NPP data. In addition, the effects of land cover types and the NDVI on NPP estimation are analyzed. The results show that the CASA model and the FY-3D satellite data estimate an average NPP of 441.2 g C m−2 yr−1 in 2019 for China’s terrestrial vegetation, while the total NPP is 3.19 Pg C yr−1. Compared with the MODIS NPP, the FY-3D NPP is overestimated in areas of low vegetation productivity and is underestimated in high-productivity areas. These discrepancies are largely due to the differences between the FY-3D NDVI and MODIS NDVI. Compared with historical field-measured data, the FY-3D NPP estimation results outperformed the MODIS NPP results, although the deviation between the FY-3D NPP estimate and the in-situ measurement was large and may exceed 20% at the pixel scale. The land cover types and the NDVI significantly affected the spatial distribution of NPP and accounted for NPP deviations of 17.0% and 18.1%, respectively. Additionally, the total deviation resulting from the two factors reached 29.5%. These results show that accurate NDVI products and land cover types are important prerequisites for NPP estimation.

基于FY-3D卫星的中国陆地净初级生产力估算研究

在优化的光能利用率CASA( Carnegie–Ames–Stanford approach)模型基础上,引入中国区域FROM-GLC( finer resolution observation and monitoring-global land cover)土地覆盖分类产品,利用1 km 分辨率FY-3D卫星资料开展了中国陆地区域净初级生产力(NPP)估算,并与国外MODIS NPP产品和历史实测数据进行了对比分析,同时开展了土地覆盖类型和归一化植被指数(NDVI)对NPP估算的影响分析。研究表明,2019年中国陆地植被NPP为3.19 PgC a−1,平均NPP为441.2 gC m−2 a−1。与MODIS NPP产品相比,FY-3D NPP在植被生产力低的地区被高估,在生产力高的地区被低估。这些差异主要是由于FY-3D NDVI和MODIS NDVI之间差异造成。与历史站点实测数据相比,FY-3D NPP估计结果优于MODIS NPP结果,土地覆盖类型和NDVI显著影响NPP的空间分布,分别占NPP偏差的17.0%和18.1%,综合偏差达到29.5%。这些结果表明,准确的NDVI产品和土地覆盖类型是NPP估算的重要前提。本文首次利用FY-3D国产气象卫星数据实现了中国陆地区域NPP的准确估算,获得了NDVI和土地覆盖产品对NPP估算的定量影响。
    • Terrestrial net primary productivity (NPP) is an important component of the surface carbon cycle and directly reflects the productivity of biomes under natural environmental conditions and the quality of terrestrial ecosystems. NPP is also a primary factor that determines whether ecosystems are carbon sources or sinks, and it regulates ecological processes and plays an important role in global change and carbon balance. Accurate estimation of terrestrial NPP is an important prerequisite for achieving carbon neutrality in China by 2060. Over the last 20 years, NPP modeling and its applications have developed rapidly, and a series of NPP estimation models applicable to regional and global scales have emerged (Prince and Goward, 1995; Field et al., 1998; Running et al., 2004; Xiao et al., 2005; Yan et al., 2015). In particular, the light use efficiency model represented by the Carnegie–Ames–Stanford approach (CASA) has become an important direction for NPP modeling development due to its application of satellite remote sensing technology (Potter et al., 1993; Zhu et al., 2007; Chen et al., 2008; Gu et al., 2013). Due to features such as simpler computation requirements, fewer parameters, and better spatial and temporal scalability (Hong et al., 2017), the CASA light use efficiency model has been applied effectively at regional and global scales. Numerous studies have performed NPP estimation at the regional scale based on the CASA model by utilizing different remote sensing data sources, such as the NOAA Advanced Very High Resolution Radiometer (AVHRR; Hicke et al., 2002; Zhang et al., 2008; Liu et al., 2010), Systeme Probatoire d’Observation de la Terre (SPOT)/vegetation data (Chen et al., 2008), Moderate Resolution Imaging Spectroradiometer (MODIS; Yuan et al., 2006; Yin et al., 2015; Chen F. et al., 2018), Landsat (Chen T. et al., 2018), and other foreign satellite data; however, very few studies have used data from the new instruments onboard Chinese satellites to carry out regional or global NPP estimation.

      On 14 November 2017, China Meteorological Administration (CMA) launched the FY-3D satellite (fourth of the Fengyun-3 series) to an altitude of 836 km. FY-3D is configured in an afternoon orbit at an equatorial crossing time (ECT) of 1400 local time. Today, a suite of products has been developed from 10 instruments on FY-3D for weather and environmental monitoring and data assimilation. The Medium Resolution Spectral Imager II (MERSI-II) on FY-3D is one of the most important payloads, has capabilities similar to or better than those of MODIS and Visible Infrared Imaging Radiometer Suite (VIIRS), and provides more than 20 environmental parameters. MERSI-II provides observations at 25 visible and infrared bands (Han et al., 2019). Among them, 6-channel data are collected at a 250-m resolution, and bands 1–4 are primarily used for ecological and environmental applications. MERSI-II can also realize the monitoring of seasonal and interannual changes at regional and global scales, which provides both a new basis for research and development of global and regional terrestrial ecological products and an opportunity to develop effective applications (Qu et al., 2019; Han et al., 2020; Wang et al., 2020; Shan et al., 2021). However, FY-3D MERSI-II data are rarely used in the quantitative assessment of ecological remote sensing. Thus, it is necessary to evaluate the applicability of FY-3D satellite data to Chinese and even global NPP estimations.

      The CASA model is a light use efficiency model that fully considers the environmental conditions and characteristic features of vegetation. Improving and effectively applying the model parameters at a regional scale in China remains a challenging task. At present, research on the optimization of CASA model parameters is primarily focused on improving light use efficiency (Dong and Ni, 2011; Wang et al., 2013) and temperature and water stress coefficients (Zhang et al., 2014; Li et al., 2019). The normalized difference vegetation index (NDVI) and land cover types are also important parameters; however, there are few studies on their impact on NPP estimation. There are many types of NDVI and land and vegetation cover-type products based on different satellites, such as NOAA/AVHRR, SPOT/VEGETATION, MODIS, Visible-Infrared Imaging Radiometer Suite (NPP/VIIRS), and FY-3. There are obvious differences among NDVI products. For instance, the numerical values of the SPOT/VEGETATION NDVI over the Qinghai–Tibet Plateau area are 0.15 higher than those based on the Global Inventory Modeling and Mapping Studies (GIMMS) NDVI from NOAA/AVHRR and the MODIS NDVI (Chen et al., 2020). Additionally, even for the same satellite data source, there may be obvious differences among different versions of NDVI (Du et al., 2016). The land and vegetation cover-type products commonly used at global and regional scales include the International Geosphere–Biosphere Programme (IGBP) land cover dataset (Loveland et al., 2000); the MODIS IGBP land cover dataset (Friedl et al., 2002), which was developed at Boston University; the UMD land cover dataset (Hansen et al., 2000), which was developed at the University of Maryland in the United States; the global land cover 2000 (GLC2000) dataset (Bartholomé and Belward, 2005), which was developed by the European Commission’s Joint Research Centre; and the GLOBCOVER 2005 land cover dataset, which was developed by the European Space Agency (Bicheron et al., 2011). However, spatial distribution patterns among land cover datasets can differ greatly for the same ground object, and the overall consistency is comparatively low among land cover datasets (Ran et al., 2009; Yang et al., 2014); therefore, it is necessary to use the appropriate classification product in the Chinese region to better estimate NPP.

      In this context, based on the CASA model, this study uses the FY-3D MERSI-II NDVI product to estimate regional NPP in China. In the setting of CASA model parameters, this study considers the optimization of the fraction of absorbed photosynthetically active radiation, the selection of land cover products suitable for China, and the maximum light energy utilization rate suitable for the main vegetation types. Then, the MODIS NPP estimation product (MOD17A3HGF; Strahler et al., 1999), NDVI product (MOD13A1; Huete et al., 1999), and land cover classification product (MCD12Q1; Running et al., 1999) are compared to explore the applicability of the FY-3D satellite data in NPP estimation and the differences between the NDVI and land cover-type products in NPP estimates. The research questions of this study are twofold: (1) Can FY-3D data be used to effectively estimate China’s terrestrial NPP? (2) What impact do land cover types and NDVI products have on NPP estimation? The results hereby obtained will provide a reference for the CASA NPP estimation model when applied to the terrestrial area of China.

    2.   Model and methods
    • Here, the NPP is estimated by using the light use efficiency model proposed by Potter et al. (1993), known as the CASA model (Potter et al., 1993). This model is expressed as follows:

      $$ \begin{aligned}[b] & {\rm{NPP}}\left( {x,t} \right) = {\rm{APAR}}(x,t) \times \varepsilon (x,t) \\ & \,\,\,\,\,\,\,\,\,\,={\rm{PAR}}(x,t) \times {\rm{FPAR}}(x,t) \times \varepsilon (x,t) \end{aligned},$$ (1)

      where $ {\rm{APAR}}\left( {x,t} \right) $ is the photosynthetically active radiation ($ {\rm{MJ}} $) absorbed by pixel $ x $ within time period $ t $ and $ \varepsilon \left( {x,t} \right) $ is the actual light use efficiency [${\rm{g}} \,\,\, ({\rm{MJ}})^{-1}$]; and $ {\rm{FPAR}}\left( {x,t} \right) $ is the absorption fraction (dimensionless) of the photosynthetically active radiation ($ {\rm{PAR}} $) absorbed by the vegetation layer, which depends on the vegetation type and the vegetation coverage status.

    • The fraction of absorbed photosynthetically active radiation (FPAR) is generally estimated by using the NDVI or ratio vegetation index, and the maximum FPAR is not larger than 0.95 (Potter et al., 1993). To optimize FPAR estimation, the FPAR backup algorithm (found in the lookup table for the relationship between the NDVI and FPAR of each vegetation type) for the MODIS FPAR product (MOD15A2H; Knyazikhin et al., 1999) is introduced to establish a quadratic polynomial regression model according to the relationship between the NDVI and FPAR of six vegetation types (Biome1, cereal, and herbaceous crops; Biome2, shrubs; Biome3, broadleaf crops; Biome4, savanna; Biome5, broadleaf forest; and Biome6, coniferous forest). All the coefficients of determination for the regression models (R2) exceed 0.99 (Fig. 1). Then, the FPAR can be estimated by using these regression models and NDVI products combined with land cover types.

      Figure 1.  Relationship between FPAR and NDVI for the six vegetation types (Biome1: cereal and herbaceous crops; Biome2: shrubs; Biome3: broadleaf crops; Biome4: savanna; Biome5: broadleaf forest; and Biome6: coniferous forest). The blue points (NDVI and FPAR data) are from the lookup table that reports the relationship between the NDVI and FPAR of the six vegetation types in the MODIS FPAR product (MOD15A2H; Knyazikhin et al., 1999).

    • Under realistic conditions, the light use efficiency $ \varepsilon $ is affected by temperature, water content, and the maximum light use efficiency $ {\varepsilon _{\max }} $, which can be expressed as follows:

      $$ \varepsilon (x,t) = {T_{\varepsilon 1}}(x,t) \times {T_{\varepsilon 2}}(x,t) \times {W_\varepsilon }(x,t) \times {\varepsilon _{\max }} ,$$ (2)

      where the temperature stress coefficients $ {T_{\varepsilon 1}}(x,t) $ and $ {T_{\varepsilon 2}}(x,t) $ can be estimated by using the method proposed by Potter et al. (1993),

      $$ \hspace{-10pt}{T}_{\varepsilon 1}\left(x\right)=0.8+0.02{T}_{{\rm{opt}}}\left(x\right)-0.0005{\left[{T}_{{\rm{opt}}}\left(x\right)\right]}^{2} ,$$ (3)
      $$\hspace{-10pt} \begin{aligned}[b] & {T}_{\varepsilon 2}\left(x,t\right)=\\ & \frac{1.1814}{\left\{1+{{\rm{e}}}^{0.2\left[{T}_{{\rm{opt}}}\left(x\right)\;-\;10\;-\;T\left(x,t\right)\right]}\right\}\times \left\{1+{{\rm{e}}}^{\left[0.3\left(\;-\;{T}_{{\rm{opt}}}\left(x\right) \;-\; 10\;+\;T\left(x,t\right)\right.\right]}\right\}} \end{aligned}.$$ (4)

      In Eqs. (2) and (3), $ {T}_{\varepsilon 1}(x,t) $ represents the limitation on photosynthesis by the intrinsic biochemical functions of plants under low- and high-temperature conditions, and $ {T}_{\varepsilon 2}(x,t) $ represents the effect of air temperature on light use efficiency when it changes from the optimum temperature $ {T}_{{\rm{opt}}}\left(x\right) $ to a high or low temperature. $ {T}_{{\rm{opt}}}\left(x\right) $ is the average air temperature of the month when the NDVI value reached its maximum in a certain area within a year, and this temperature is deemed to be the optimum temperature for vegetation growth. When the average temperature for a certain month is less than or equal to 10°C, $ {T}_{\varepsilon 1}\left(x\right) $ is 0, and photosynthesis production is deemed to be zero. If the average temperature $ T(x,t) $ for a certain month is 10 °C higher or 13°C lower than ${T}_{\rm opt}\left(x\right)$, $ {T}_{\varepsilon 2}(x,t) $ from that month is equal to half of $ {T}_{\varepsilon 2} $ from a time period in which $ T(x,t) $ is ${T}_{\rm opt}\left(x\right),$ that is, $ {T}_{\varepsilon 2}(x,t) $ = 0.4956.

      $ {W_\varepsilon }(x,t) $ is the water stress coefficient, which reflects the influence of the effective water conditions available to the plants on light use efficiency. As the effective moisture in the environment increases, $ {W_\varepsilon } $ gradually increases. Its values range from 0.5 to 1.0 and represent extremely arid to completely humid (not subject to water stress) conditions. This parameter is expressed as follows:

      $$ {W_\varepsilon } = 0.5 + 0.5{\rm{AET}}/{\rm{PET}} ,$$ (5)

      where $ {\rm{AET}} $ is the estimated actual evapotranspiration (mm), which can be obtained from the regional actual evapotranspiration model established by Zhou and Zhang (1995). This model is expressed as follows:

      $$ {\rm{AET}} = \frac{{\left[ {P \times R_{\rm{net}} \times ({P^2} + R_{\rm{ne{t}}}^2 + P \times R_{\rm{net}})} \right]}}{{\left[ {(P + R_{\rm{net}}) \times ({P^2} + R_{\rm{ne{t}}}^2)} \right]}} ,$$ (6)

      where $ P $ is the precipitation (mm) and $ R_{\rm{{\rm{net}}}} $ is the net radiation (mm), which can be converted from MJ m−2 with energy as the unit from the Penman–Monteith equation (Yang et al., 2004); the conversion coefficient is 0.408, that is, 1-mm precipitation is equivalent to 0.408 MJ m−2 net radiation. $ {\rm PET} $ is the potential evapotranspiration (mm), which is calculated here by using the Penman–Monteith equation recommended by the Food and Agriculture Organization (FAO; Yang et al., 2004). This calculation is expressed as follows:

      $$ {\rm{PET}} = \frac{{0.408\Delta (R_{\rm{net}} - G) + \gamma \frac{{900}}{{T \;+\; 273}}{u_2}({e_{\rm{s}}} - {e_{\rm{a}}})}}{{\Delta + \gamma (1 + 0.34{u_2})}} ,$$ (7)

      where ${\rm PET}$ is the potential evapotranspiration (mm day−1), $R_{\rm net}$ is the net radiation (MJ m−2 day−1), $ G $ is the soil heat flux (MJ m−2 day−1), $ T $ is the average air temperature at a height of 2 m, $\Delta$ is the slope of the saturation vapor pressure–temperature curve (hPa K−1), $ {u_2} $ is the wind speed at a height of 2 m (m s−1), $ \gamma $ is the psychrometric constant (kPa °C−1). The variables $ {e_{\rm{s}}} $ is the saturation vapor pressure (kPa), and $ {e_{\rm{a}}} $ is the actual vapor pressure (kPa). $ {e_{\rm{s}}} $ and $ {e_{\rm{a}}} $ can be expressed as follows:

      $$ {e_{{\rm s}{\text{-}}{T_{\rm max}^{}}}^{}} = 0.6108 \; \times \; \exp \left( {\frac{{17.27 \times T_{\rm max} }}{{T_{\rm max} \; +\; 237.3}}} \right) ,$$ (8)
      $$ {e_{{\rm s}{\text{-}}{T_{\rm min}}}^{}} = 0.6108 \; \times \; \exp \left( {\frac{{17.27 \times T_{\rm min} }}{{T_{\rm min} \; +\; 237.3}}} \right) ,$$ (9)
      $$ {e_{\rm{s}}} = \frac{{\rm RH}}{{100}} \times \left( {\frac{{ {e_{{\rm s}{\text{-}}{T_{\rm max}^{}}}^{}} + {e_{{\rm s}{\text{-}}{T_{\rm min}}}^{}} }}{2}} \right) ,$$ (10)
      $$ {e_{\rm{a}}} = \frac{{\rm RH}}{{100}} \times {e_{\rm{s}}} ,$$ (11)

      where $T_{\rm max}$ is the maximum temperature (K), $T_{\rm min}$ is the minimum temperature (K), and ${\rm RH}$ is the relative humidity (%). The other parameter calculation methods can be found in the relevant literature (Wang et al., 2005; Liu et al., 2008).

    • To improve the CASA model, emphasis was placed on the determination of the maximum light use efficiency $ {\varepsilon _{\max }} $; the traditional CASA model uniformly sets $ {\varepsilon _{\max }} $ to 0.389 g C (MJ)−1 (calculated based on the mass of carbon), but there are in fact great differences in $ {\varepsilon _{\max }} $ among regions and vegetation cover types. Therefore, revisions to $ {\varepsilon _{\max }} $ were made in connection with various aspects of the study, which was particularly important for increasing the model estimation’s accuracy. References are made here to the research results for maximum light use efficiency in the Chinese region (Zhu et al., 2007; Bao et al., 2016; Liu et al., 2019), which were used to determine the values of $ {\varepsilon _{\max }} $ for the typical vegetation types in the Chinese region (Table 1).

      Vegetation typeMaximum light use efficiency [g C (MJ)−1]
      Deciduous coniferous forest0.485
      Evergreen coniferous forest0.389
      Deciduous broadleaf forest0.692
      Evergreen broadleaf forest0.985
      Coniferous and broadleaf mixed
      forest
      0.475
      Evergreen and deciduous broadleaf
      mixed forest
      0.768
      Shrubland0.429
      Grassland0.573
      Farmland0.542
      Other0.542

      Table 1.  Maximum light use efficiency values of typical vegetation types in the Chinese region

    • Table 1 shows that there are obvious differences in the $ {\varepsilon _{\max }} $ for different vegetation types, especially different forest types; therefore, it is necessary to obtain accurate land and vegetation cover types. Gong (2009) pointed out that the global regional classification accuracy of the MODIS IGBP land cover product was less than 50%, and the verification accuracy of the land cover types using the 14 flux stations in China was less than 10%. Therefore, the 250-m land cover product (Wang et al., 2015) generated for the Chinese region in 2010 by the finer resolution observation and monitoring-global land cover (FROM-GLC) classification scheme (classification accuracy of 75.2%) was selected. In addition, to improve forest classification accuracy, the Chinese forest map product, which includes six main forest types with a resolution of 30 m in 2010 (Li et al., 2014), was used to correct the above FROM-GLC classification product. Then, based on the maximum land use ratio method at different spatial scale conversions, 19 types of FROM-GLC land cover products for the Chinese region with a resolution of 1 km were obtained (Fig. 2a). In addition, the 17 types of MODIS IGBP land cover classification products (MCD12Q1; Strahler et al., 1999) with a resolution of 500 m for the Chinese region in 2019 were used for comparison (Fig. 2b). The difference in spatial distribution between the FROM-GLC and MODIS IGBP land cover classification products is obvious. Regardless of the time difference, taking the FROM-GLC classification product as the true value, the evergreen broadleaf forest, evergreen coniferous forest, deciduous broadleaf forest, deciduous coniferous forest, mixed forest, and farmland (including cropland and cropland mosaics) in the MODIS IGBP classified product have spatial consistency with the FROM-GLC classification product in proportions of 27.2%, 7.8%, 55.1%, 10.3%, 7.0%, and 56.8%, respectively, indicating that the two types of land cover products are very different.

    • Evaluating the accuracy of large-scale NPP estimates has always been a difficult task. Comparison with field measurements and comparison with estimates from other models are commonly used. As it is difficult to obtain NPP field measurements in China in 2019, we adopted two schemes to verify the FY-3D NPP estimates. (1) The FY-3D NPP estimates are compared with estimates from other models, including the MODIS NPP product and the average NPP of historical main vegetation types recorded in the literature. (2) They are compared with histori-cal NPP in-situ measurements, which are selected over 19 sites for China’s typical terrestrial ecosystems from 2003 to 2014, based on literature review (Chen et al., 2019) as summarized in Table 2. To eliminate the geometric positioning error of FY-3D NPP estimates, within the range of 3 × 3 pixels centered around each site, the average FY-3D NPP values of the pixels with the same FROM-GLC land cover type (Fig. 2a) and site vegetation type (Table 2) are extracted, and these values are then compared with the field-measured NPP at the 19 sites.

      SiteLongitude (°E)Latitude (°N)Vegetation typeYear of observationMeasured NPP (g C m−2 yr−1)
      Zhonghe100.44638.975Phragmites australis2012–2014390.00
      Ailaoshan101.01724.533Evergreen broadleaf forest2009–2010976.20
      Chanzhou109.52019.508Rubber forest20101133.50
      Huitong109.75026.833Evergreen needleleaf forest2008–2009268.50
      Taoyuan111.45028.917Rice2003675.20
      Xiaolangdi112.46735.017Mixed forest2006–2010355.00
      Dinghushan112.53623.173Mixed forest2003–2008395.90
      Ningxiang112.56728.333Mixed forest2013428.80
      Yueyang112.85029.517Poplar forest2005–2007515.70
      Shouyang113.20037.750Spring maize2012–2014509.30
      Xiping113.86733.350Poplar forest2010343.40
      Qianyanzhou115.06726.733Evergreen broadleaf forest2003–2008487.50
      Weishan116.05036.650Winter wheat and corn2006–2008559.00
      Daxing116.29139.534Poplar forest2006–2009518.30
      Yucheng116.60036.950Winter wheat and corn2003–2008366.00
      Anqing117.03330.500Poplar forest2005–2007506.10
      Tianmushan119.43330.350Mixed forest2013–2014738.18
      Anji119.67430.476Bamboo forest2011–2013585.40
      Changbaishan128.40042.400Deciduous broadleaf forest2003–2008302.33

      Table 2.  Historically measured NPP of typical terrestrial ecosystem sites in China based on document integration (Chen et al., 2019)

      Figure 2.  Spatial distribution maps of (a) FROM-GLC land cover types and (b) MODIS IGBP land cover types in China.

    • For the CASA model, land cover type determines the maximum light use efficiency and FPAR, and NDVI determines the actual light use efficiency and FPAR; therefore, differences in the NDVI of the land cover types may result in obvious differences in NPP estimates. Based on the algorithm examined in this study, three schemes for estimating NPP in China’s terrestrial area are presented for comparison with corresponding values obtained by the original algorithm to analyze the effect of differences in land cover types and NDVI on NPP estimates. In Scheme 1, the FROM-GLC land cover types are replaced by the IGBP land cover types from 2019; in Scheme 2, the FY-3D NDVI is replaced by the MODIS NDVI; and in Scheme 3, the land cover types and the NDVI are both replaced.

    3.   Data and preprocessing
    • In this study, the remote sensing NPP estimation model is used to simulate China’s terrestrial vegetation NPP with a time step of months and a spatial resolution of 0.01°; the annual total NPP was the cumulative value of the 12-month NPP estimation results.

      The relevant meteorological variables include monthly average air temperature, maximum temperature, minimum temperature, relative humidity, hours of sunshine, wind speed, and precipitation. These variables are collected from 2165 meteorological stations in China in 2019 by the National Meteorological Information Center of China, and the inverse distance weighting (IDW) method is used to interpolate and generate grid images of various meteorological variables for 12 months at equal latitude and longitude increments of 0.01° × 0.01° for the study region.

      The FY-3D satellite data are obtained from the 2019 FY-3D MERSI-II NDVI monthly composite product of the National Satellite Meteorological Center of China at a resolution of 1 km and can be downloaded for free from the Fengyun Satellite Remote Sensing Data Service website (http://satellite.nsmc.org.cn/portalsite/default.aspx). The FY-3D monthly NDVI products are synthesized based on the maximum value of three 10-day NDVI products. The FY-3D 10-day NDVI product synthesis is carried out with the following method: when the number of cloudless pixels in the synthesis period is greater than or equal to 2 days, the CV-MVC (constrained view angle–maximum value composite) method is used for NDVI synthesis; that is, the value with a smaller observation angle from the two values with the largest NDVI in the observation period is selected for synthesis. If there are no clouds for only one day, the NDVI is calculated directly with the data of that day; if there are clouds in the synthesis period, the daily NDVI is calculated, and the maximum NDVI is selected for synthesis. Before NDVI synthesis, the FY-3D cloud detection product (cloud mask) is used for cloud identification.

      In addition, the 2019 MODIS NPP product (MOD17A3HGF), the MODIS IGBP land cover product (MCD12Q1), and the MODIS NDVI 16-day composite product (MOD13A1), all at a resolution of 500 m, are acquired from the U.S. NASA website (https://earthdata.nasa.gov/). Utilizing the maximum value composite method on the MODIS NDVI 16-day composite product, the monthly MODIS NDVI product from 2019 is generated. These products are all globally framed, and they are projected, spliced, and cut to generate grid images of various elements at equal latitude and longitude increments of 0.01° × 0.01° for the Chinese region.

    4.   Validation
    • The 1-km estimates of NPP in 2019 for China and the spatial distribution of the difference (ΔNPP) between the MODIS NPP product and this algorithm are shown in Fig. 3. NPP is generally found to decrease from southeast to northwest in China (Fig. 3a), which is basically consistent with previous research results (Gu et al., 2013; Liu et al., 2017; Xu Y. Q. et al., 2020). From the perspective of spatial distribution, most areas of Taiwan Island, the valleys of southwestern Yunnan, and some hilly areas along the southeast coast have good hydrothermal conditions, and the annual average NPP values in these areas are found to be comparatively high, exceeding 1000 g C m−2 yr−1. The annual NPP in Northeast and North China is generally greater than 400 g C m−2 yr−1, and the NPP of forestland exceeds 600 g C m−2 yr−1. The annual NPP of the arid and semiarid areas in the western and northern regions, and most of the Qinghai–Tibet Plateau area, is generally less than 400 g C m−2 yr−1, and the vegetation is sparse in the western Taklamakan Desert and the northern Gurbantunggut Desert areas, with annual NPP less than 50 g C m−2 yr−1. Among the main vegetation types, the average annual NPP of the evergreen broadleaf forest is the greatest (1008.2 g C m−2 yr−1), followed by the deciduous broadleaf forest (647.7 g C m−2 yr−1) and the mixed forest (471.9 g C m−2 yr−1), and the values for farmland and grassland are 447.5 and 273.7 g C m−2 yr−1, respectively.

      Compared with the MODIS NPP product for 2019 (Fig. 3b), the FY-3D NPP values for the forested area in the northeast and the humid region in the south, particularly the southwestern region, are obviously low, and these areas generally have higher vegetation productivity. The FY-3D NPP tends to be higher than the MODIS values for most areas in the north and on the Qinghai–Tibet Plateau, and these areas are generally areas with lower vegetation productivity. The analysis of the spatial correlation between the FY-3D NPP and MODIS NPP shows that the mean absolute error is 149.0 g C m−2 yr−1 and the coefficient of determination R2 of the fitted regression equation is only 0.60, which demonstrates the great spatial differences between the FY-3D NPP and MODIS NPP.

      Figure 3.  (a) NPP of China in 2019 estimated based on FY-3D data and (b) the difference (ΔNPP) between the estimated FY-3D NPP and the MODIS NPP.

    • To explore this spatial difference (ΔNPP) between FY-3D NPP and MODIS NPP, the annual average, January, April, July, and October FY-3D NDVI and MODIS NDVI, which are based on monthly NDVI products from 2019, are calculated. The difference (ΔNDVI) between the annual average FY-3D NDVI and MODIS NDVI is obtained, and a spatial correlation analysis is performed for the two using the annual average and the values from January, April, July, and October (Fig. 4).

      Figure 4.  Spatial distributions of (a) the difference (ΔNDVI) between the annual average FY-3D NDVI and the annual average MODIS NDVI and (b) the scatter probability density diagrams and fitted equations of the relationship between the FY-3D NDVI and MODIS NDVI in terms of the annual average and (c) January, (d) April, (e) July, and (f) October (the blue line is the 1 : 1 line, and the red dashed line is the fitting trend line).

      Similar to the ΔNPP spatial distribution map (Fig. 3b), the areas in which the ΔNPP is positive or negative (Fig. 4a) are very consistent in terms of the spatial distribution with the areas in which the ΔNDVI is positive or negative, indicating that the difference between the FY-3D NDVI and MODIS NDVI very likely determines the spatial distribution of the NPP difference between the two. There were significant spatial correlations between the FY-3D NDVI and MODIS NDVI for the annual average and in January, April, July, and October (Figs. 4bf), and the coefficients of determination R2 all exceeded 0.80. In both the annual average and in the typical months of the four seasons, compared with the MODIS NDVI, there are situations in which the FY-3D NDVI shows relatively high values in areas with low NDVI and relatively low values in areas with high NDVI. This relationship also explains, to a considerable extent, the phenomena of overestimation in areas of low vegetation productivity and underestimation in areas of high vegetation productivity in the spatial distribution of the FY-3D NPP relative to the MODIS NPP. In addition, the spatial correlation between the annual average FY-3D NDVI and the annual average MODIS NDVI (R2 = 0.9036) is clearly greater than the spatial correlation between the FY-3D NPP and the MODIS NPP (R2 = 0.5998), indicating that in addition to the NDVI, there are other important factors that affect the spatial differences in NPP estimates.

      The root mean square error (RMSE in Fig. 4) is between 0.08 and 0.13, indicating that a considerable difference between the FY-3D NDVI and the MODIS NDVI. The mean bias errors (MBE in Fig. 4) for both the annual average and the typical months, such as January, April, July, and October, are all below zero. This result indicates that the overall FY-3 NDVI values are lower than the MODIS NDVI values. An important reason for analyzing this difference in NDVI is that the red light band and near-infrared band of FY-3D MERSI-II are 0.600–0.700 and 0.815–0.915 μm, respectively, and the bandwidths are both larger than those of the red light band (0.620–0.670 μm) and near-infrared band (0.841–0.876 μm) of Earth Observing System (EOS) /MODIS, which causes the absorption of chlorophyll in the two bands of FY-3D to be weaker than that of EOS/MODIS, especially the red light band (Ge et al., 2017). Therefore, the equivalent reflectivity of the red light band during the vegetation period is higher than that of EOS/MODIS, while the reflectivity in the near-infrared band is not greatly different. The FY-3D NDVI is smaller than that of EOS/MODIS; the more vegetation grows, the greater the difference is (Ge et al., 2017).

    • The historical NPP data field measured by Zhu et al. (2007) and the 2019 MODIS NPP product for the principal vegetation types in China were utilized here for comparative analysis, as shown in Fig. 5. The NPP values for evergreen broadleaf forest, mixed forest, and deciduous forest estimated by this algorithm were comparatively close to actual measurements (discrepancies within 5%), grassland was overestimated (18.7% higher), and other vegetation types were underestimated, with deciduous coniferous forest in particular underestimated by 23.2%. For the MODIS NPP product, however, except for the value for deciduous coniferous forest, which is closer to the field-measured value, there is a phenomenon of underestimation for farmland, evergreen broadleaf forest, and deciduous broadleaf forest (12.1% to 24.5% lower). However, the NPP of other vegetation types was obviously overestimated (28.9% to 70.9% higher). In particular, the MODIS NPP was very clearly higher in the area comprising five provinces and cities in Southwest China (Sichuan, Chongqing, Yunnan, Guangxi, and Guizhou; averaging 770.0 g C m−2 yr−1), and existing research showed that the average vegetation area in this region, estimated based on another light use efficiency model known as Global Photosynthesis Ecosystem Monitoring (GLOPEM), was 540.3 g C m−2 yr−1 (Zhao et al., 2015). In addition, another study showed that the average NPP in the southwestern karst area based on the CASA model was 402.3 g C m−2 yr−1 (Dong et al., 2011). The average estimates from the algorithm employed in this study for the five southwestern provinces and cities (561.1 g C m−2 yr−1) are closer to the aforementioned research results. In addition, the FY-3D NPP in the forested area in the eastern and southeastern parts of the Qinghai–Tibet Plateau is generally 400–1150 g C m−2 yr−1, and the NPP of the central grassland/meadow area of the plateau is generally 150–400 g C m−2 yr−1. These results are more consistent with the estimates of Xu J. et al. (2020) based on the Carbon Exchange in the Vegetation–Soil–Atmosphere (CEVSA) model (the NPP of the forested area in the eastern and southeastern plateau ranged from 600 to 1200 g C m−2 yr−1, and that of the central grassland/meadow area is 200–400 g C m−2 yr−1). However, there are situations in which the MODIS NPP is high (broad area NPP > 1200 g C m−2 yr−1) in the forested area of the Qinghai–Tibet Plateau and low in the central grassland/meadow area (mostly 100–200 g C m−2 yr−1). In general, the NPP of the typical vegetation types (except for deciduous coniferous forest) estimated by this algorithm is closer in spatial distribution to the NPP measurements made in China and by previous research than to the MODIS NPP product.

    • Compared with the 19 sites’ measured NPP (Table 2), the FY-3D NPP values for 16 sites are within the range of ±30%, while the MODIS NPP values for only 13 sites are within ±30% (Fig. 6a). The FY-3D values that are not within this range including those from the Dinghushan, Yucheng, and Changbaishan sites; these values are higher than 30%.

      Figure 5.  Comparison of the annual average FY-3D NPP estimation results with historical field-measured NPP and MODIS NPP for typical terrestrial vegetation types in China.

      Figure 6.  (a) Comparison of the previously measured NPP, FY-3D NPP, and MODIS NPP for 19 sites; (b) the scatter plot between the sites’ measured NPP and FY-3D NPP; and (c) the scatter plot between the MODIS NPP values and the sites’ measured NPP for the 19 sites (R is the correlation coefficient, RMSE is the root mean square error, and the red line is the 1: 1 line).

      The spatial correlation coefficient R between the FY-3D NPP and the site field-measured NPP is 0.858 (Fig. 6b), indicating that the two have good spatial consistency. The RMSE is 126.6 g C m−2 yr−1, which is equivalent to 24% of the NPP average value (529.2 g C m−2 yr−1) for the 19 sites in Table 2; this result indicates that the NPP deviation estimated by FY-3D based on the pixel scale might be large. The main reasons for this discrepancy are algorithmic error and some other errors, such as the differences in observation year and the spatial scale effect between the satellite and the stations, etc. However, compared with the results from the MODIS NPP products and site measurements (R = 0.594, and RMSE = 192.4 g C m−2 yr−1; Fig. 6c), the estimated FY-3D NPP is better than the MODIS NPP product.

      Examining the entire Chinese region by utilizing this algorithm, the average terrestrial vegetation NPP in China is estimated to be 441.2 g C m−2 yr−1, which is slightly below the average value for terrestrial vegetation in China according to the MODIS NPP product (448.3 g C m−2 yr−1). Overall, the terrestrial vegetation NPP in China is 3.19 Pg C yr−1 (1 P = 1015), which is 1.2% lower than the value estimated by the MODIS NPP product (3.23 Pg C yr−1) and slightly higher than the average value of 3.12 Pg C yr−1 from 1989 to 1993 in China obtained by Zhu et al. (2007). The algorithm’s result is also within the range of the values of terrestrial vegetation NPP in China from 2001 to 2014 (3.02–3.49 Pg C yr−1) as estimated by Liu et al. (2017). The results presented above show that this algorithm can effectively estimate NPP throughout China using FY-3D satellite data.

    • As mentioned in Section 2.3, NPP estimates are sensitive to land cover type and NDVI. The results from NPP analyses over China, obtained by using different sensitivity test schemes, are shown in Fig. 7 and Table 3.

      Figure 7.  Spatial distributions of the differences (ΔNPP) between the NPP values in China estimated using different schemes and FY-3D NPP: (a) land cover types replaced by the IGBP classification product, (b) NDVI replaced by MODIS NDVI, and (c) land cover types and NDVI both replaced.

      MethodFY-3D NPPMODIS NPPScheme 1Scheme 2Scheme 3
      Total NPP (Pg C yr−1)3.193.233.343.463.64
      Average NPP (g C m−2 yr−1)441.2448.3462.9479.4504.9
      Percentage anomaly with total FY-3D NPP (%)1.64.78.514.1
      Mean absolute error from FY-3D NPP (g C m−2 yr−1)149.074.879.7129.4

      Table 3.  Comparison of NPP in China as estimated by different methods

      Once the original FROM-GLC land cover types were replaced with the IGBP land cover types (Fig. 7a), the mean absolute error based on the pixel scale reached 74.8 g C m−2 yr−1, accounting for 17.0% of the mean NPP (441.2 g C m−2 yr−1) and increasing the total NPP and the average NPP in the terrestrial area of China by 4.7% to 3.34 Pg C yr−1 and 462.9 g C m−2 yr−1, respectively. Once the original FY-3D NDVI product was replaced with the MODIS NDVI (Fig. 7b), the mean absolute error based on the pixel scale reached 79.7 g C m−2 yr−1, accounting for 18.1% of the mean vegetation NPP and increasing the total NPP and the average NPP in China by 8.5% to 3.46 Pg C yr−1 and 479.4 g C m−2 yr−1, respectively. After replacing both the land cover types and NDVI (Fig. 7c), the mean absolute error based on the pixel scale reached 129.4 g C m−2 yr−1, accounting for 29.5% of the mean vegetation NPP and increasing the total NPP and the average NPP in the terrestrial area of China by 14.1% to 3.64 Pg C yr−1 and 504.9 g C m−2 yr−1, respectively. These results indicate that land cover type and NDVI are important factors affecting the results of NPP estimation.

    5.   Discussion and conclusions
    • Based on the CASA light use efficiency model, the China’s terrestrial NPP at a resolution of 1 km using the FY-3D MERSI-II NDVI was estimated. Although the FY-3D NPP estimation results outperformed the MODIS NPP through comparative analysis with the measured NPP data of historical stations, there are still some uncertainties because the CASA model is influenced by radiation, temperature, humidity, FPAR, etc. These elements may vary greatly in different years, resulting in deviations in comparing the estimated NPP in 2019 with the site data in historical years. In addition, the spatial scale inconsistency between the site and remote sensing observations increases this uncertainty.

      Land cover or vegetation cover types are important basic data for estimating vegetation productivity based on remote sensing imagery. Since most of the parameterization schemes for modeling vegetation productivity from remote sensing assume that the model uses the same set of parameter values for the same vegetation type, whether the vegetation types are classified accurately that is directly related to the accuracy of the NPP simulation (Liang et al., 2019). The preceding analysis indicates that, using two different land cover-type products, the deviation in NPP estimates reaches 17.0%. Such a spatial discrepancy is quite consequential; therefore, selecting a land cover classification product that is suitable for the study area and is necessary to increase the accuracy of NPP estimation.

      The NDVI directly reflects the vegetation coverage and is an important factor influencing vegetation productivity, and its changes significantly affect the spatial distribution of NPP. The preceding analysis indicates that, using the same algorithm and different NDVI products, even if the coefficient of determination for the fitted model of spatial correlation between the two exceeds 0.90, the deviation at the pixel scale also reaches 18.1%. Therefore, when using the CASA model to estimate NPP throughout China, it is necessary to fully consider the accuracy of the NDVI product. Although the CASA model can effectively estimate NPP based on FY-3D NDVI data, compared to MODIS NDVI and due to the design of the FY-3D MERSI-II sensor, FY-3D NDVI is not as sensitive to vegetation changes as MODIS NDVI. Further improvements are needed in the design of domestic satellite sensors in the future.

      The primary additional factors that affect the accuracy of NPP estimation include the following.

      (1) The model effect: Since NPP is obtained from the gross primary productivity (GPP) minus the autotrophic respiration of plants, and theoretically, the basic principle of the CASA model is designed in connection with the GPP, the NPP simulation assumes that the ratio of the autotrophic respiration of plants to GPP is constant. However, this ratio actually changes with variations in environmental temperature and the age of trees (Chen et al., 2011). Therefore, simply relying on a constant ratio of autotrophic respiration to GPP to calculate NPP inevitably results in spatial simulation errors for the system, but an ecosystem process model can better compensate for this inadequacy of the light use efficiency model (Wang et al., 2009; Hong et al., 2017). The ecosystem process model considers respiration and soil moisture changes in addition to the dynamic processes of carbon, nitrogen, and nutrients when simulating vegetation productivity. However, there is greater uncertainty when the simulation of test points is utilized for extrapolation to the regional scale, and it is difficult for a model to fully reflect the heterogeneity of the landscape; thus, the ecological remote sensing coupling model that integrates the advantages of the process model and the CASA light use efficiency model can improve the inadequacies in the theoretical design of the CASA model, despite the higher degree of complexity. Integrating the process model with the CASA model is difficult, but this model coupling remains an important development direction for NPP estimation in the future (Hong et al., 2017).

      (2) Certain value $ {\varepsilon _{\max }} $: Assigned to the maximum light use efficiency $ {\varepsilon _{\max }} $. In the CASA model, the most important parameter is the value assigned to $ {\varepsilon _{\max }} $; its size directly determines the accuracy of the simulation of vegetation productivity (Liang et al., 2019). However, this parameter differs greatly among models. Generally, $ {\varepsilon _{\max }} $ is obtained by the inversion of observed values for GPP or NPP, and differences in the observed values and mitigation limiting factors used result in enormous differences in this parameter. In this study, $ {\varepsilon _{\max }} $ is primarily determined from the application of the CASA model in China and differed substantially from the $ {\varepsilon _{\max }} $ values of different vegetation types in the MODIS NPP product, which ranged from 0.604 to 1.259. For example, the values for the deciduous coniferous forest and the evergreen coniferous forest in MODIS were 1.086 and 0.962, respectively, which are significantly different from the respective values of 0.485 and 0.389 from the algorithm analyzed in this paper.

      (3) Uncertainty in FPAR and PAR: This study simulate FPAR from the relationship between FPAR and NDVI from the MODIS lookup table, and the difference between FY-3D NDVI and MODIS NDVI was not considered, which cause some uncertainty in NPP estimation. Although we tried to revise the FY-3D monthly NDVI products to the MODIS NDVI level for FPAR estimation using cross-calibration technology, the total NPP in China is significantly higher than the MODIS NPP product and NPP estimation in Table 3, and we do not consider the FPAR model transferability.

      In addition, the effect of direct radiation is currently primarily simulated by the incident PAR in the CASA model, but the effect of scattered radiation is ignored. However, existing research has indicated that an increase in scattered radiation can clearly increase the photosynthetic capacity of vegetation; this phenomenon occurs because scattered radiation can more easily pass through the vegetation canopy and be absorbed by the leaves of understorey plants than can direct radiation, and as the scattered radiation increases, the ratio of the radiation in the blue band to that in the red band increases. Since blue light is the main band for PAR, the ratio of PAR to the total radiation increases (Urban et al., 2007). This increase results in regional simulation errors being low for the cloudy and rainy areas in southern China, but the cloudy and rainy areas in southern China are also areas of high vegetation productivity in China; therefore, it remains necessary to develop suitable constraining equations to quantitatively describe the effects of scattered and direct radiation on light use efficiency in China.

    • Utilizing FY-3D satellite data in combination with the CASA model, the mean NPP estimation in China in 2019 was 441.2 g C m−2 yr−1, and the total NPP was 3.19 Pg C yr−1, which are close to values from the MODIS NPP product (lower by 1.2%). Compared with MODIS NPP, the FY-3D NPP was overestimated in areas of low vegetation productivity, while the NPP was underestimated in high-productivity areas; this result was largely due to the difference between the FY-3D NDVI and MODIS NDVI. In comparison with the field-measured NPP data and existing research results, the NPP estimates performed better than those of the MODIS NPP product.

      The accuracy of NPP estimated at the pixel scale is largely affected by land cover types and NDVI products. The land cover types and the NDVI significantly affected the spatial distribution of NPP and were able to account for deviations in estimates of 17.0% and 18.1%, respectively, in NPP at the spatial pixel scale, and the deviation resulting from the two factors combined was 29.5%.

      Acknowledgments. We thank the China National Satellite Meteorological Center for providing the FY-3D data, the LP-DAAC and MODIS science teams for providing free MODIS products, and Tsinghua University for providing free FROM-GLC land cover products.

    Appendix
    • The NPP data used in this article can be found online at https://pan.baidu.com/s/1WLYlz_WTZDk5n08F1JPHuA, and the extraction code is c1qq.

Reference (58)

Catalog

    /

    DownLoad:  Full-Size Img  PowerPoint
    Return
    Return