Evaluation of NDVI Estimation Considering Atmospheric and BRDF Correction through Himawari-8/AHI
- 66 Downloads
Satellite-based vegetation indices are an essential element in understanding the Earth’s surface. In this study, we estimated the normalized difference vegetation index (NDVI) using Himawari-8/Advanced Himawari Imager (AHI) data and analyzed the sensitivity of products to atmospheric and surface correction. We used the Second Simulation of a Satellite Signal in the Solar Spectrum (6S) radiative transfer model for atmospheric correction, and kernel-based semi-empirical bidirectional reflectance distribution function (BRDF) model to remove surface anisotropic effects. From this, top-of-atmosphere, top-of-canopy, and normalized NDVIs were produced. A sensitivity analysis showed that the normalized NDVI had the lowest number of missing values compared with the others and almost no low peaks during the study period. These results were validated by Terra and Aqua/Moderate Resolution Imaging Spectroradiometer (MODIS) and Project for On-Board Autonomy/Vegetation (PROBA) NDVI product, showing the root mean square error (RMSE) and bias of 0.09 and + 0.04 (MODIS) and 0.09 and − 0.04 (PROBA), respectively. These results also satisfied the FP7 Geoland2/BioPar project-defined user requirements (threshold: 0.15; target: 0.10).
KeywordsNDVI Himawari-8/AHI Atmospheric correction BRDF Normalized reflectance
Vegetation index-based remote sensing is a standard tool of Earth system science and continues to play a central role in global change, carbon cycle, land coverage/use, and terrestrial ecology research. Among various vegetation indices that have been developed, the normalized difference vegetation index (NDVI) is used the most because it provides a good representation of vegetation characteristics over the long term. Moreover, other data products derived from remote sensing use NDVI as their primary input, including the leaf area index (Pontailler et al. 2003), global land cover maps (DeFries et al. 1999), fraction of absorbed photosynthetically active radiation (Ruimy et al. 1994), net primary production (Prince and Goward 1995), air temperature (Prihodko and Goward 1997), and land surface temperature (Jin 2004).
Many studies use three types of reflectance to calculate NDVI: top-of-atmosphere (TOA), top-of-canopy (TOC), and normalized reflectance with bidirectional reflectance distribution function (BRDF) modeling. TOA reflectance does not include atmospheric correction. In this case, atmospheric effects are removed during NDVI calculations (Holben 1986). The NDVI trend is similar, regardless of whether atmospheric correction (Chen et al. 2003). TOC reflectance includes atmospheric correction. Atmospheric elements such as water vapor, total ozone column, and aerosol optical depth (AOD) affect the accuracy of satellite-based land surface products (Nagol et al. 2009). The differences in the NDVI values occurs depending on the atmospheric correction method (Ke et al. 2015). Normalized reflectance with BRDF modeling incurs changes in the NDVI value with changes in the satellite angle and solar angle (Fensholt et al. 2006a). Additionally, the resulting NDVI products must account for the BRDF effect (Vermote et al. 2008); thus, the BRDF must be modeled according to the satellite’s characteristics (Yeom and Kim 2013).
NDVI with TOA reflectance are retrieved by several organizations, including the National Oceanic and Atmospheric Administration (NOAA)/Advanced Very High Resolution Radiometer (AVHRR) and the Meteosat Second Generation (MSG)/Spinning Enhanced Visible and Infrared Imager (SEVIRI). The Satellite Pour l’Observation de la Terre (SPOT), Project for On-Board Autonomy/Vegetation (PROBA), Terra and Aqua/Moderate Resolution Imaging Spectroradiometer (MODIS), and the Suomi National Polar-orbiting Partnership (Suomi NPP)/Visible Infrared Imaging Radiometer Suite (VIIRS) have produced NDVI with TOC reflectance. Several of these organizations have geostationary satellites equipped with new sensors, such as the Geostationary Operational Environmental Satellite-16 (GOES-16), Meteosat Third Generation (MTG), and Himawari-8/9. All of these satellites have channels capable of estimating the NDVI.
South Korea operates a geostationary satellite, the Communication, Ocean, Meteorological Satellite (COMS), at the National Meteorological Satellite Center (NMSC); however, this satellite does not measure in the red/near-infrared (NIR) range required for vegetation index calculations. The Korea Meteorological Administration (KMA) launched the Geostationary Korea Multi-Purpose Satellite - 2A (GK2A) with Advanced Meteorological Imager (AMI) sensors in December 2018. This satellite has a total of 16 channels that are capable of retrieving vegetation index information. However, the appropriate vegetation index for this satellite remains under investigation.
In this study, we attempted to identify a vegetation index suitable for GK2A by analyzing the sensitivity of the NDVI to atmospheric correction and BRDF effects. Atmospheric corrections were performed through Second Simulation of a Satellite Signal in the Solar Spectrum (6S) radiative transfer model (RTM) and anisotropic properties of the surface are considered with semi-empirical BRDF modeling. We then constructed three types of NDVIs based on TOA reflectance, TOC reflectance, and normalized reflectance considering BRDF effects. To compare them, we analyzed the temporal and spatial differences between NDVIs by land type.
2 Data and Method
This study was conducted using Himawari-8/Advanced Himawari Imager (AHI) data, because it is difficult to obtain GK2A/AMI data at present. The satellite, developed as a next-generation multi-functional transport satellite (MTSAT), was launched on 7 October 2014 and is currently operated by the Japan Meteorological Agency (JMA).
The Himawari-8/AHI is located at longitude 140.7°E as a geostationary satellite. The satellite has a 16 channel multispectral imager to capture visible light and infrared images of the Asia-Pacific region and can provide full disk observations every 10 min and whole target area images every 2.5 min. In this study, we used the red channel (B3; central wavelength: 0.64 μm) and the NIR channel (B4; central wavelength: 0.86 μm). NDVI was calculated using data obtained over a 1-year period from January 2017 to December 2017.
Summary of data used to estimate the NDVI
L1b products (red, NIR bands)
Total ozone column
ECMWF reanalysis data
Total precipitable water
ECMWF reanalysis data
ECMWF reanalysis data
Angular parameters (SZA,VZA, RAA)
We used MODIS NDVI and PROBA NDVI to compare the estimated NDVIs in this study. MODIS NDVI were obtained using the 16-day MODIS composite product (MOD13A2) from the Terra MODIS sensor is based on the Terra MODIS level 2 (L2G) daily surface reflectance product (MOD09 series). It provides red and NIR surface reflectance corrected for atmospheric gas, thin cirrus cloud, and aerosol effects. For data processing purposes, MODIS Vegetation Indexes (VIs) are generated in square tile units (~1200 km × 1200 km at the equator) and are mapped as sinusoidal grid projections (equal area projections). Only tiles containing land features are processed. When mosaicked, all tiles cover the Earth land mass, allowing the global MODIS-VI to be generated every 16 days (Didan et al. 2015). The data produced through this process have a spatial resolution of 1 km and a temporal resolution of 16 days (Huete et al. 1999).
The PROBA NDVI used in this study was the S10 NDVI 1-km product. PROBA-V was launched by the European Space Agency (ESA) on 7 May 2013. The S10 NDVI product of the satellite was synthesized using the maximum value composite (MVC) method over 10 days based on TOC reflectance data calculated using the simplified method for atmospheric correction (SMAC). The PROBA NDVI product can be rescaled to include quality information (status map) in the main NDVI layer. Pixels wrongly identified as “land” in the S10 NDVI status map are re-classified as “sea”. The product was displayed in a regular latitude/longitude grid with the WGS 1984 (terrestrial radius: 6378 km). The resolution of the grid was 1/336°, with the center of the pixel as the reference. The spatial resolution is 1 km, and the temporal resolution was 10 days (Wolters et al. 2014).
We estimated three type NDVIs based on TOA reflectance, TOC reflectance, and normalized reflectance. We compared the three NDVIs to determine the suitable NDVI for Himawari-8. First, we converted from Himawari-8/AHI channel DN data (Red, NIR) to TOA reflectance and TOA radiance using the calibration table. Only cloudy pixels were removed from the TOA reflectance data, as atmospheric correction could not be applied. Based on TOA reflectance data, we retrieved the TOA NDVI. The TOA NDVI was calculated at 1-h intervals to coincide with the 1-h interval of cloud data. The TOC reflectance was produced from TOA reflectance for atmospheric correction. The TOC NDVI was then estimated based on TOC reflectance. Finally, we calculated the NDVI based on normalized reflectance using the BRDF model. The NDVI-based normalized reflectance was calculated at intervals of 1 day.
To compare the data obtained using the three reflectance types, we analyzed temporal and spatial differences between the acquired NDVIs by land type. We chose an NDVI that showed consistent quality, without missing values or low peaks (sudden drops in the time series). From this, we selected the optimal NDVI representation and compared its results with those obtained from MODIS and PROBA NDVIs.
The following is a description of the atmospheric correction and BRDF used for the TOC reflectance and normalized reflectance calculations in this study.
2.2.1 Atmospheric Correction for TOC Reflectance
Range and interval of input LUT parameters based on the 6S RTM (Liang et al. 2010)
Water vapor (gm cm−2)
Total ozone column (atm cm)
0.01, 0.05, 0.1, 0.15, 0.2, 0.3, 0.4, 0.6,
0.8, 1.0, 1.5, 2
If the LUT method is applied directly to atmospheric correction, discontinuities occur in the results. This is due to the discontinuity in the atmospheric correction coefficients that coincide with LUT intervals and is most prominent in SZA and VZA data in particular. Thus, in this study, we interpolated the SZA and VZA at 0.05° intervals.
2.2.2 BRDF for Normalized Reflectance
3.1 Himawari-8/AHI NDVI Spatial Distribution
We examined three NDVI types (TOA NDVI, TOC NDVI, and normalized NDVI) for the Himawari-8/AHI over a 1-year period from January to December 2017. The MVC values for the TOA and TOC NDVIs were synthesized over a 1-day period to equalize the temporal resolution of the three NDVIs.
3.2 Himawari-8/AHI NDVIs Time-Series Distribution According to Land Type
The TOA NDVI showed the lowest values for all land types, with especially low values in areas where some vegetation was present, such as in evergreen forest, mixed forest, grass and crop areas (Fig. 3a–e). Evergreen forest is of particular interest, because it covers most tropical patches of Southeast Asia. The area is healthy throughout the year; nevertheless, the average TOA NDVI was 0.6, which did not reflect the characteristics of healthy vegetation.
In contrast, the TOC NDVI showed the maximum value for most land types and was relatively high in barren and shrub areas (Fig. 3c and f), which have sparser vegetation distributions. The 6S RTM estimated high coefficient values with high SZA and VZA; as a result, the NDVI values were high. We then synthesized the NDVI using the MVC method; high estimates were obtained for the TOC NDVI compared with the other two.
The normalized NDVI showed tendencies similar to those of the TOC NDVI in mixed forest and evergreen forest areas where vegetation is abundant (Fig. 3d and e). The TOA NDVI and the normalized NDVI were similar in shrub and barren areas (Fig. 3c and f), where vegetation is sparse; here volumetric scattering rarely occurs. A pattern in which the TOA NDVI and TOC NDVI decreased sharply then increased again (low peak) was apparent for most land types. These trends were determined to be due to an error that had arisen from unremoved clouds, surface moisture caused by rain, and other factors (Kim et al. 2011). In contrast, the normalized NDVI did not significantly change over the time series, regardless of land type. Therefore, normalized NDVI, with few missing values or low-peak occurrences, provided the optimal NDVI model (Gao et al. 2002).
3.3 Comparison of Other Satellite Products
The final calculated normalized NDVI was compared with the widely used MODIS and PROBA NDVIs. We performed a normalized NDVI 16-day MVC synthesis for a comparison with the MODIS NDVI and a 10-day synthesis for a comparison with PROBA NDVI data. Both datasets were tested using homogeneous pixels. Based on a 3 × 3 array of validation pixels, the standard deviation was 0.03 or less.
Comparison results between MODIS NDVI and PROBA NDVI
NDVI is not an essential climate variable, and there are no Global Climate Observation System specifications on required accuracy. The FP7 Geoland2/BioPar project defined user requirements for NDVI in terms of acceptable differences with existing satellite-derived products. The requirements specify a threshold RMSE of 0.15 and a target RMSE of 0.10 (Swinnen and Toté 2017). Regardless of the type of validation data set, the RMSE was <0.10 in this study. Other studies related to NDVI estimation have calculated the RMSE to be between 0.10 and 0.15 (Yeom and Kim 2013; Ke et al. 2015).
4 Concluding Remarks
In this study, we analyzed the sensitivity of atmospheric correction and BRDF effect to determine an NDVI suitable for the GK2A/AMI using Himawari-8/AHI data. A TOA NDVI, TOC NDVI, and normalized NDVI were compared to determine the optimal NDVI for GK2A/AMI operations. The results of our spatiotemporal analysis revealed the normalized NDVI to have the smallest number of missing values compared with the other two NDVIs, and there were almost no low peaks during the time series. The normalized NDVI calculated in this study was compared with the widely used MODIS and PROBA NDVIs. We calculated an RMSE of ≤0.1 for the normalized NDVI, regardless of the validation data set. This result satisfies FP7 Geoland2/BioPar project-defined user requirements and demonstrates the high accuracy of this NDVI estimation approach.
However, the normalized NDVI showed missing values in some areas. In the present study, the NDVI was calculated using 1-h interval data. The problem of missing values may be resolved if cloud data become available at 10-min intervals in the future. In addition, the reanalysis data of ECMWF was used to perform the current atmospheric correction. We expect that high accuracy can be achieved using satellite data with high spatial and temporal resolution in our study area.
The normalized NDVI calculated in this study can be applied to the future GK2A for geostationary-based NDVI in Asia and Australia; here, the data have a high temporal resolution, as opposed to that available from polar orbit satellites. We expect these findings to be useful for monitoring and mitigating changes in the Earth’s surface in the near future.
This work was supported by “Development of Scene Analysis & Surface Algorithms” project, funded by ETRI, which is a subproject of “Development of Geostationary Meteorological Satellite Ground Segment (NMSC-2019-01)” program funded by NMSC (National Meteorological Satellite Center) of KMA(Korea Meteorological Administration). One of authors (Noh-Hun Seong) was supported by the BK21 plus Project of the Graduate School of Earth Environmental Hazard System.
- Didan, K., Munoz, A. B., Solano, R., Huete, A.: MODIS vegetation index user’s guide (MOD13 series). Vegetation Index and Phenology Lab, The University of Arizona, 1–38. (2015). https://icdc.cen.uni-hamburg.de/fileadmin/user_upload/icdc_Dokumente/MODIS/MODIS_Collection6_VegetationIndex_UsersGuide_MOD13_V03_June2015.pdf. Accessed 5 July 2019
- Duchemin, B., Berthelot, B., Dedieu, G., Leroy, M., Maisongrande, P.: Normalisation of directional effects in 10-day global syntheses derived from VEGETATION/SPOT:: II. Validation of an operational method on actual data sets. Remote Sensing of Environment. 81(1), 101–113 (2002). https://doi.org/10.1016/S0034-4257(01)00337-6 CrossRefGoogle Scholar
- Fensholt, R., Anyamba, A., Huber, S., Proud, S. R., Tucker, C. J., Small, J., Pak. E., Rasmussen, M. O., Sandholt, I., Shisanya, C.: Analysing the advantages of high temporal resolution geostationary MSG SEVIRI data compared to Polar Operational Environmental Satellite data for land surface monitoring in Africa. International Journal of Applied Earth Observation and Geoinformation, 13(5), 721–729. (2011). https://doi.org/10.1016/j.jag.2011.05.009 CrossRefGoogle Scholar
- Huete, A., Justice, C., Leeuwen, W.; MODIS vegetation index (MOD13). Algorithm Theoretical Basic Document, 3, 213, (1999). https://modis.gsfc.nasa.gov/data/atbd/atbd_mod13.pdf. Accessed 5 July 2019
- Inness, A., Ades, M., Agustí-Panareda, A., Barré, J., Benedictow, A., Blechschmidt, A., Dominguez, J.J., Engelen, R., Eskes, H., Flemming, J., Huijnen, V., Jones, L., Kipling, Z., Massart, S., Parrington, M., Peuch, V., Razinger, M., Remy, S., Schulz, M., Suttie, M.: The CAMS reanalysis of atmospheric composition. Atmos. Chem. Phys. 19, 3515–3556 (2019). https://doi.org/10.5194/acp-19-3515-2019 CrossRefGoogle Scholar
- Liang, S., Wang, D., He, M.: GOES-R advanced baseline imager (ABI) algorithm theoretical basis document for surface albedo. University of Maryland. (2010). https://www.goes-r.gov/products/ATBDs/option2/Land_Albedo_v1.0_no_color.pdf. Accessed 5 July 2019
- Paltridge, G.W., Platt, C.M.R.: Radiative processes in meteorology and climatology. Elsevier Scientific Pub. C, Amsterdam, New York (1976)Google Scholar
- Swinnen, E., Toté, C.: Gio global land component – Lot I “Operation of the global land component” algorithm theoretical basis document for NDVI, VITO. (2017). https://land.copernicus.eu/global/sites/cgls.vito.be/files/products/GIOGL1_ATBD_NDVI1km-V2.2_I2.21.pdf. Accessed 5 July 2019
- Wolters, E., Dierckx, W., Dries, J., Swinnen, E.: PROBA-V products user manual. VITO. (2014). http://proba-v.vgt.vito.be/sites/proba-v.vgt.vito.be/files/products_user_manual.pdf. Accessed 5 July 2019
- Yeom, J.M., Kim, H.O.: Feasibility of using Geostationary Ocean colour imager (GOCI) data for land applications after atmospheric correction and bidirectional reflectance distribution function modelling. Int. J. Remote Sens. 34(20), 7329–7339 (2013). https://doi.org/10.1080/2150704X.2013.817708 CrossRefGoogle Scholar
Open Access This article is licensed under a Creative Commons Attribution 4.0 International License, which permits use, sharing, adaptation, distribution and reproduction in any medium or format, as long as you give appropriate credit to the original author(s) and the source, provide a link to the Creative Commons licence, and indicate if changes were made. The images or other third party material in this article are included in the article's Creative Commons licence, unless indicated otherwise in a credit line to the material. If material is not included in the article's Creative Commons licence and your intended use is not permitted by statutory regulation or exceeds the permitted use, you will need to obtain permission directly from the copyright holder. To view a copy of this licence, visit http://creativecommons.org/licenses/by/4.0/.