1. Introduction
Ever since the launch of the Communication Ocean and Meteorological Satellite (COMS) in 2010, the geostationary ocean color imager (GOCI) has provided continuous images covering an area of about 2,500 × 2,500 km2 in northeast Asian region (Ryu et al., 2012). The GOCI is indeed unique and the first geostationary ocean color sensor with a 500 m spatial resolution and is providing eight hourly observations per day. Although the main function of the GOCI is ocean color observation, its coverage still contains large land area including the Korean peninsula, Japan and part of China, Mongolia, and Russia. The total land area within the predetermined GOCI coverage is about 2.4 million km2, which occupies about 38% of the scene.
The foremost advantage of geostationary satellite images is high temporal resolution, which can enable us to observe rapid land surface changes, such as phenological development of vegetation, forest fires, flood, and heavy snow. It has been rare to use geostationary satellite data for land surface monitoring. The European meteorological satellite (MSG) has been used to derive time series observations on land surfaces and shown the potential to be used for detecting and monitoring of vegetation changes (Fensholt et al., 2011; Yan et al., 2016), leaf area index (Guan et al., 2014), drought (Rulinda et al., 2010), flooding (Proud et al., 2011), and wild fires (Amraoui et al., 2010). The GOCI ata also has been used to monitor vegetation phenology (Lee and Lee, 2016; Yeom and Kim, 2015). Although the geostationary satellite data have rather coarse spatial resolution and are still new to land remote sensing, these cases have shown the potential of geostationary satellite images for land monitoring with much improved temporal resolution.
To derive quantitative information and time-series analysis on land surface features from GOCI data, a reliable and effective atmospheric correction (AC) procedure is essential. The GOCI AC algorithm over ocean surface has been already developed and several atmosphric corrected ocean products (L1B and L2) are being provided (Ahn et al., 2012). However, the GOCI AC algorithm over land surfaces is still insufficient to apply. While atmospheric correction for ocean remote sensing has been a critical part to derive any meaningful information regarding water constituents, AC for land surface has been considered less importantly, due to the qualitative nature of information content in land remote sensing. As land remote sensing community need biophysical information of terrestrial environment from different time, sensors, and locations, AC has become increasingly important (Santer et al., 1999). The AC algorithms for ocean and land surfaces are basically the same by correcting sun and viewing angle effects, gases absorption and Rayleigh and aerosol scatterings. However, AC of land and ocean surfaces uses separate algorithm of obtaning nput parameters regarding atmospheric conditions. For example, the Moderate Resolution Imaging Spectroradiometer (MODIS) data were separately processed to produce surface reflectance for land and ocean (Vermote and Saleous, 2006).
The objective of this study is to develop an operational GOCI AC method over land surfaces, which can effectively process large amount of GOCI data obtained eight times per day. GOCI image has been well calibrated and shown to have enough radiometric quality to extract biophysical information in land surfaces (Lee et al., 2012). The GOCI land AC method should be relatively simple, fast, and accurate. Accurate GOCI land surface reflectance should be key product to be used for subsequent processing, such as to produce time series vegetation index and cloud-free composite data.
2. Method
1) Atmospheric Correction Overview
Atmospheric correction of optical imagery is mainly to derive purely reflected portion of electro-magnetic energy from the at-sensor radiance (Lt) by removing the amount of radiance attenuated by atmospheric constituents (Lee and Lee, 2015a). The radiance (Lt) measured by on-board sensor can be expressed in a simple equation (Eq. 1) and atmospheric correction is the process of extracting surface reflectance (ρ) from Lt. In Eq. 1, there are unknown variables of T, Lp, and Ed and we need to know these variables to calculate ρ from Lt. Atmospheric correction begins with the conversion of ixel’s digital number (DN) value to at-sensor radiance (Lt) by applying radiometric calibration coefficients. Often the accurate radiometric calibration coefficients are the critical element for atmospheric correction. Although early land remote sensing systems had poor radiometric calibrations (Santer et al., 1999), GOCI data were well calibrated and have reliable at-sensor radiance data.
\(\mathrm{Lt}=\frac{\rho T v(E o \cos \theta z T z+E d)}{\pi}+\mathrm{Lp}\) (1)
where Lp = path radiance
Tv = transmittance from target to sensor
Tz = transmittance from sun to target
Eo = Extraterrestrial solar irradiance
Ed = diffuse sky irradiance
θz = solar zenith angle
Atmospheric correction is based on the calculation of physical processes of radiative transfer from sun to earth and from earth to sensor and requires several parameters regarding atmospheric conditions, sun-target-sensor geometry, and sensor’s spectral specifications to calculate T, Lp, and Ed. To calculate the physical process of energy transfer, most atmospheric correction methods use radiative transfer (RT) model. RT model simulates the radiative transfer by calculating major interactions between atmosphere particles and electromagnetic energy (gases absorption, Rayleigh scattering, and aerosol scatterings). Among several RT models, MODerate resoluion atmospheric TRANsmission (MODTRAN) (Berk et al., 1998) and Second Simulation of a Satellite Signal in the Solar Spectrum (6S) (Vermote et al., 1997) have been widely used for atmospheric correction in remote sensing. Several atmospheric correction tools based on these RT models have been developed (Gao et al., 1996; Richter and Schläpfer, 2017).
Fig. 1 shows the overall procedure of the atmospheric correction method over land surfaces for GOCI images. Initially, the ocean area was masked out from the GOCI level 1B image that is radiometrically calibrated and geometrically corrected at-sensor radiance. Atmospheric correction of GOCI imagery is based on the 6S radiative transfer model that has been also used for to produce several standard GOCI ocean products in the Korean Institute of Ocean Science and Technology (KIOST). The 6S-based atmospheric correction requires several input data regarding atmospheric conditions, sun and viewing geometry, and sensor’s spectral response specifications.
Fig. 1. Overall scheme of the atmospheric correction over land surface in GOCI imagery.
Atmospheric effects on sensor-received signal include the radiances contributed from gases absorptions, Rayleigh scattering and aerosol scattering. To correct these atmospheric effects, we need three-dimensional atmospheric data of gas molecular and aerosol at the time of the image acquisition. However, it is almost impossible to obtain such a omplete set of atmospheric data that are spatially and temporally corresponding with te satellite image. Therefore, most atmospheric correction methods have used standard atmospheric and aerosol models that defines the type and concentrations of air molecules and aerosols. Surface reflectance is obtained by correcting absorption and scatterings that are calculated using these standard atmospheric and aerosol models. Before we apply standard atmospheric and aerosol models to correct gas absorption and scattering, we tested the sensitivity of major input parameters for the outcome of atmospheric correction for the eight GOCI spectral bands. Among the several input parameters required by the 6S model, four parameters of atmospheric model, terrain elevation, aerosol model, and aerosol optical thickness (AOT) were analyzed to evaluate the sensitivity in surface reflectance.
2) Atmospheric Model and Terrain Elevation
Standard atmospheric model defines a typical distribution of atmospheric gases properties, such as temperature, pressure, and density for a certain region of the earth. Among several atmospheric models available, three models (U.S. standard, mid latitude summer, mid latitude winter) are applicable to the GOCI target area. Atmospheric gases absorption and Rayleigh scattering are basically calculated from the parameters defined by the atmospheric model. In the mid-latitude GOCI region, gases absorption and Rayleigh scattering were not much different inclear sky conditions regardless of the atmospheric model used.
Since Rayleigh scattring is influenced by the atmospheric pressure (Wang, 2005), the standard atmospheric model needs to be adjusted by terrain elevation that is directly related to the air pressure. Terrain elevation has also effects on solar irradiance and atmospheric gases composition and density. Rayleigh reflectance was calculated by varying the terrain elevation ranging from 0 to 4000m, which reflects the topography in the GOCI land area. Fig. 2 shows the variation of Rayleigh reflectance by the terrain elevation for the eight GOCI spectral bands. Rayleigh reflectance in shorter wavelength bands (blue and green spectrum) show more susceptible to terrain elevation than red and near infrared (NIR) bands.
Fig. 2. Variation of Rayleigh reflectance by terrain elevation for the eight GOCI spectral bands.
The GOCI land area includes high mountainous region over 3000 meters elevation in northwest China and Japan. To account for the effect of terrain elevation in Rayleigh scattering and gases absorption, we used the shuttle radar topographic mission (SRTM) digital elevation model (DEM) data that were resampled to the 500m spatial resolution of the GOCI image.
3) Aerosol Model and AOT
Aerosol is atmospheric particle that is larger than gas molecule and has great influence on the at-sensor radiances in visible and NIR bands. In northeast Asia, aerosol is ighly variable in both spatially and temporally by the rapid industrialization and the environmentl conditions (Kim et al., 2007). To correct aerosol scattering in the GOCI at-sensor radiances, the 6S model requires aerosol type and aerosol optical thickness (AOT). Composition of different aerosol type were defined by several aerosol models that include continental, maritime, urban, desert, and biomass burning. Each of these aerosol models define the fraction of aerosol type (dust, oceanic, water-soluble, soot) and optical properties of each aerosol type.
Surface reflectance was simulated by increasing AOT values from 0.0 to 0.8 with four aerosol models. At-sensor radiance over large rice field from GOCI image obtained in May 14, 2011 (solar zenith angle 18°, viewing zenith angle 41.6°, relative azimuth angle 8°) were used as an input value. The other input parameters were fixed. Fig. 3 shows the variation of GOCI red and NIR band reflectance by aerosol model and AOT. Although the effect of AOT is more sensitive in shorter wavelengths of blue and green bands, we only show red and NIR bands since they are most widely used bands for land applications of vegetation and soil. The GOCI red and NIR reflectance greatly vary by AOT, which indicates that AOT is highly influential and sensitive in surface reflectance. As AOT increase from 0.0 to 0.8 for continental aerosol model, the red reflectance decreases by 0.027 and the NIR reflectance increases by 0.035. TheGOCI red and NIR reflectance were very similar for three types of aerosol model except for urbn model. Although the effect of aerosol model is not negligible, it is less influential than AOT. AOT in the GOCI land area also has high temporal variability within a day and, therefore, we must consider to variability of AOT in the GOCI atmospheric correction.
Fig. 3. Variation of the GOCI red and NIR band reflectance by aerosol model and AOT.
The ideal atmospheric correction would be to use AOT data directly derived from GOCI image itself, just like in the case of ocean atmospheric correction. However, AOT retrieval over land surface from image data is somewhat difficult task because the land surface has complex spectral characteristics unlike ocean surface (Vermote and Vermeulen, 1999). Although the land AOT estimation algorithm has been developed for the GOCI land area (Choi et al., 2016) and aerosol product will be provided by KIOST, it may not be proper to be used for the atmospheric correction for daily GOCI land images. It can be alternative to use AOT data obtained by other satellite system such as MODIS aerosol product (Levy et al., 2013). However, it may not be a proper solution for high temporal GOCI images because MODIS aerosol product is only available twice per day with Terra and Aqua system. The aerosol robotic network (AERONET) provides a long-term and near real time ground measured AOT data (Holben et al., 1998).
In ths study, we designed the spatio-temporal AOT maps using AERONET data as an input parameter of the GOCI amospheric correction. To build the spatiotemporal AOT maps, we used AERONET Level 2.0 (cloud-screened and quality-assured) data collected at the 109 AERONET stations within and near the geographical coverage of the GOCI target area (Fig. 4). Total 36 AOT maps were produced for monthly × 3 diurnal window (morning, noon, and afternoon) for every year from 2011 to 2017. After downloading AERONET data, we calculated 36 average value of AOT for monthly × 3 diurnal window for each of 109 stations. Considering the spatial variation of AOT and the distance among the AERONET stations, the 36 average values of AOT at 109 sites were then interpolated to 5 km × 5 km spatial resolution to produce spatially continuous AOT maps using inverse distance weight method (Fig. 4). The 36 spatiotemporal AOT classes were found to be statistically significant (p < 0.001) by one-way analysis of variance (ANOVA).
Fig. 4. The spatio-temporal AOT maps (a) over the GOCI land area from 2011 to 2017, produced by the AERONET data collected at the 109 ground stations (b). The dotted line in Fig. 4(b) represents the geographical coverage of the GOCI image.
4) Look-up Table Generation
Once all the input parameters were furnished, the 6S radiative transfer model calculate the surface reflectance for every pixel of GOCI images, which consumesa lot of computational time. For fast and effective atmospheric correction of GOCI images acquired eight time per day, we attempted to build the pre-calculated look-up-table (LUT). Instead of calculating the radiative transfer process for every pixel, the LUT directly converts at-sensor radiance to surface reflectance. LUT is composed of surface reflectance for every combination of pre-defined values for several input parameters (Table 1), which include solar and viewing geometry, terrain elevation, atmospheric model, AOT, and the spectral response functions of the eight GOCI bands. Table 1 lists input parameters used to generate the LUT in the 6S model. Total number of pre-calculated surface reflectance were 4,366,656 for every possible combination of input parameters. From the sensitivity analysis on input parameters needed by the 6S model, the AOT showed the most influential factor and, therefore, was divided by narrow interval. The whole GOCI land area include continental, urban, desert, and biomass burning type of aerosol and it would be ideal to apply aerosol model separately by region. However, it was very difficult to spatially separate by aerosol type. To produce continuous reflectance map, we used the continental aerosol model only. The LUT generated output were then interpolated to obtain more precise surface reflectance to account for the intermediate values within a class of input parameters.
Table 1. Pre-defined input parameters to generate the ook up table with the 6S model for the GOCI atmospheric correction
3. Validation of the Atmosperic Correction
To validate the GOCI AC method, in-situ surface reflectance was measured at the rice paddy in Dangjin and the grass land in Siheung. Field measurement sites were large and homogenous enough to cover the 500m spatial resolution of GOCI image. Spectral reflectance of rice and grass were continuously measured at the same sites from the early growing season in April to the senescent stage in October. In-situ reflectance was measured by a field spectrometer (ASD Fieldspec3) with 8 degree field of view at nadir direction. Reflectance were measured at three or four different azimuth directions and then averaged. Field measured reflectance were then integrated to produce GOCI reflectance using the spectral response function of each of eight bands. To validate the GOCI atmospheric correction, we used 1,960 hourly GOCI images obtained from March to October 2016. After applying the GOCI AC method described, the GOCI surface reflectance data were obtained. Only cloud-free pixels with confident clear quality were used to compare with the in-situ measurements. The cloud detection method was based on the simple thresholds and multi-temporal data analysis method reported by Lee and Lee (2015b). Along with the in-situ measurements, MODIS surface reflectance products (MOD09, MYD09) acquired during the same time with the GOCI observations were used (tota 490 images = 245 days × 2 images per day).
As seen in Fig. 5, the time-series GOCI reflectance shows the typical pattern of crop phenology in the rice paddy. The NIR reflectance increases and red reflectance decreases after the rice transplantation in May and reaches the highest NIR reflectance in September. The spectral reflectance of the sparsely covered grass land does not show such steep increase in NIR reflectance, but it still shows gradual increase during the growing season. Among the 1,960 GOCI hourly observations, completely cloud-free observations were only 465 for the rice paddy and 327 for the grass land. It is very rare to find that all eight GOCI hourly observations in a day are cloud-free condition in both sites. Usually four to six hourly observations were maximum even in a very clear day. In such day, the GOCI hourly surface reflectance shows diurnal variation in which the difference of hourly reflectance in a day is close to 0.1 in NIR band, which is relatively large enough in the calculation of vegetation index. The diurnal variation might be caused by the different solar zenith angle (SZA) from morning to late afternoon. Most atmospheric correction methods in land remote sensing assume that the land surface is Lambertian target that reflects the same amount of energy in all directions. Obviously, the diurnal variation of hourly reflectance (grey boxes in Fig. 5) in a day indicates that the rice paddy and grass land were not Lambertian n nature and the anisotropic effect of directional reflectance need to be corrected for the subsequent cloud-free cmpositing.
Fig. 5. Temporal profile of GOCI surface reflectance obtained from the rice paddy (a) and grass land (b). Grey boxes show the diurnal variation of hourly reflectance in a day.
Fig. 6 compares GOCI reflectance and in-situ reflectance measured at the same time with the GOCI image capture. The atmospheric corrected GOCI reflectance were very close to the field-measured reflectance and they are highly correlated with each other. The GOCI reflectance are slightly higher than the field-measured reflectance by 0.058 in NIR band and 0.013 in the red band. The discrepancy between GOCI reflectance and in-situ measurements might come from different sun-target-sensor geometry. In-situ surface reflectance was measured at nadir direction while the GOCI images were taken at 43 degree view zenith angle (VZA) for the rice paddy and grass land. Similar results were reported by Fensholt et al. (2006), in which surface reflectance tends to increase at higher VZA and the anisotropic effect becomes stronger in dense vegetation canopy.
Fig. 6. Comparison between GOCI surface reflectance and in-situ measured reflectance in red and NIR bands for (a) rice paddy in Dangjin and (b) grass land in Siheung.
Fig. 7 compares the temporal profiles of GOCI surface reflectance and MODIS reflectance that were acquired at te same time. The temporal profile of GOCI surface reflectance in the rice paddy is highly comparable to the MODIS refectance with relatively high correlation coefficients, while the correlation coefficients in the grass land is slightly lower. As seen in Fig. 7, the GOCI reflectance is slightly higher than the MODIS reflectance by 0.03~0.04 in NIR band and 0.002~0.011 in the red band. Considering that MODIS atmospheric corrections over land surfaces have been continuously improved (Vermote and Saleous, 2006) and, therefore, are known to be reliable method, the close relationship with the MODIS reflectance product indicates that the GOCI atmospheric correction method can be relatively accurate and appropriate to derive surface reflectance over land surface. The slight discrepancy of surface reflectance between GOCI and MODIS could be explained by different VZA, spectral responsivity of spectral band, and atmospheric correction algorithm itself.
Fig. 7. Comparison between the GOCI reflectance and the MODIS reflectance in red and NIR bands for (a) rice paddy in Dangjin and (b) grass land in Siheung.
4. Conclusions
Development of an operational GOCI atmospheric correction algorithm is essential for time-series analysis of biophysical variables over terrestrial ecosystems. In this study, we developed the GOCI atmospheric correction method for land surfaces, which was based on the 6S radiative transfer model. AOT was the most inluential factor for the surface reflectance over land surface among several input parameters required by the 6S model. Considerng the highly variable nature of AOT in northeast Asia, we generated the spatiotemporal AOT maps using AERONET data. Unlike in ocean surface, direct extraction AOT data over land surface from GOCI image itself has been difficult although it is ideal approach for the atmospheric correction. Once the AOT extraction method is developed, the GOCI atmospheric correction method could be improved. For fast and effective computation of atmospheric effect for every pixel of GOCI images, the pre-calculated LUT approach was very useful.
The GOCI atmospheric correction method was validated by in-situ spectral measurements and MODIS reflectance products. The GOCI reflectance derived by the proposed method was very close to in-situ measured reflectance and MODIS reflectance. GOCI reflectance was slightly higher than the field-measurements and the MODIS reflectance, which might be explained by the different viewing zenith angle (VZA). Geostationary GOCI image is acquired with VZA ranging from 25 to 60 degrees while the field spectral measurements and MODIS reflectance were obtained at nadir direction. Along with the viewing angle effect, GOCI hourly surface reflectance also showed diurnal variations due to solar angle change from morning to late afternoon. To solve such problem of solar and viewing angles, a new compositing method is necessary to reduce aniotropic effects. For land applications, GOCI reflectance should be provided in a form of cloud-free composite which select clearpixels among multiple images obtained within a few days of compositing period. In this cloud-free compositing, bidirectional reflectance distribution functions (BRDF) effects need to be normalized to reduce the anisotropy effect.
Acknowledgement
This research was a part of the project titled ‘Research for Applications of Geostationary Ocean Color Imager,’ funded by the Ministry of Oceans and Fisheries, Korea.
References
- Ahn, J., Y. Park, J. Ryu, and B. Lee, 2012. Development of atmospheric correction algorithm for Geostationary Ocean Color Imager (GOCI), Ocean Science Journal, 47(3): 247-259. https://doi.org/10.1007/s12601-012-0026-2
- Amraoui, M., C. DaCamara, and J. Pereira, 2010. Detection and monitoring of African vegetation fires using MSG-SEVIRI imagery, Remote Sensing of Environment, 114(5): 1038-1052. https://doi.org/10.1016/j.rse.2009.12.019
- Berk, A., L. Bernstein, G. Anderson, P. Acharya, D. Robertson, J. Chetwynd, and S. Adler-Golden, 1998. MODTRAN cloud and multiple scattering upgrades with application to AVIRIS, Remote Sensing of Environment, 65(3): 367-375. https://doi.org/10.1016/S0034-4257(98)00045-5
- Choi, M., J. Kim, J. Lee, M. Kim, Y. Park, U. Jeong, W. Kim, H. Hong, B. Holben, and T.F. Eck, 2016. GOCI Yonsei Aerosol Retrieval (YAER) algorithm and validation during the DRAGON-NE Asia 2012 campaign, Atmospheric Measurement Techniques, 9(2016): 1377-1398. https://doi.org/10.5194/amt-9-1377-2016
- Fensholt, R., A. Anyamba, S. Huber, S.R. Proud, C.J. Tucker, J. Small, E. Pak, M.O. Rasmussen, I. Sandholt, and C. Shisanya, 2011. 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. https://doi.org/10.1016/j.jag.2011.05.009
- Fensholt, R., I. Sandholt, S. Stisen, and C. Tucker, 2006. Analysing NDVI for the African continent using the geostationary meteosat second generation SEVIRI sensor, Remote Sensing of Environment, 101(2): 212-229. https://doi.org/10.1016/j.rse.2005.11.013
- Gao, B., K. Heidebrecht, and A.F. Goetz, 1996. Atmosphere Removal Program (ATREM) Version 2.0 Users Guide, Center for the Study of Earth from Space/CIRES, University of Colorado, Boulder, Colorado, 26: 19.
- Guan, K., D. Medvigy, E.F. Wood, K.K. Caylor, S. Li, and S. Jeong, 2014. Deriving vegetation phenological time and trajectory information over Africa using SEVIRI daily LAI, IEEE Transactions on Geoscience and Remote Sensing, 52(2): 1113-1130. https://doi.org/10.1109/TGRS.2013.2247611
- Holben, B.N., T.F. Eck, I. Slutsker, D. Tanre, J. Buis, A. Setzer, E. Vermote, J. Reagan, Y. Kaufman, and T. Nakajima, 1998. AERONET-A federated instrument network and data archive for aerosol characterization, Remote Sensing of Environment, 66(1): 1-16. https://doi.org/10.1016/S0034-4257(98)00031-5
- Kim, S., S. Yoon, J. Kim, and S. Kim, 2007. Seasonal and monthly variations of columnar aerosol optical properties over east Asia determined from multi-year MODIS, LIDAR, and AERONET Sun/sky radiometer measurements, Atmospheric Environment, 41(8): 1634-1651. https://doi.org/10.1016/j.atmosenv.2006.10.044
- Lee, H.S. and K.S. Lee, 2015a. Atmospheric correction problems with multi-temporal high spatial resolution images from different satellite sensors, Korean Journal of Remote Sensing, 31(4): 321-330. https://doi.org/10.7780/kjrs.2015.31.4.4
- Lee, H.S. and K.S. Lee, 2015b. Development of cloud detection method with geostationary ocean color imagery for land applications, Korean Journal of Remote Sensing, 31(5): 371-384 (in Korean with English abstract). https://doi.org/10.7780/kjrs.2015.31.5.2
- Lee, H.S. and K.S. Lee, 2016. Improvement of Temporal Resolution for Land Surface Monitoring by the Geostationary Ocean Color Imager Data, Korean Journal of Remote Sensing, 32(1): 25-38. https://doi.org/10.7780/kjrs.2016.32.1.3
- Lee, K., S. Park, S. Kim, H. Lee, and J. Shin, 2012. Radiometric characteristics of Geostationary Ocean Color Imager (GOCI) for land applications, Korean Journal of Remote Sensing, 28(3): 277-285. https://doi.org/10.7780/kjrs.2012.28.3.277
- Levy, R., S. Mattoo, L. Munchak, L. Remer, A. Sayer, F. Patadia, and N. Hsu, 2013. The Collection 6 MODIS aerosol products over land and ocean, Atmospheric Measurement Techniques, 6(11): 2989. https://doi.org/10.5194/amt-6-2989-2013
- Proud, S.R., R. Fensholt, L.V. Rasmussen, and I. Sandholt, 2011. Rapid response flood detection using the MSG geostationary satellite, International Journal of Applied Earth Observation and Geoinformation, 13(4): 536-544. https://doi.org/10.1016/j.jag.2011.02.002
- Richter, R. and D. Schlapfer, 2017. Atmospheric/topographic correction for satellite imagery, https://www.rese-apps.com/pdf/atcor3_manual.pdf, Accessed on Feb. 22, 2017.
- Rulinda, C.M., W. Bijker, and A. Stein, 2010. Image mining for drought monitoring in eastern Africa using Meteosat SEVIRI data, International Journal of Applied Earth Observation and Geoinformation, 12S(2010): S63-S68. https://doi.org/10.1016/j.jag.2009.10.008
- Ryu, J., H. Han, S. Cho, Y. Park, and Y. Ahn, 2012. Overview of geostationary ocean color imager (GOCI) and GOCI data processing system (GDPS), Ocean Science Journal, 47(3): 223-233. https://doi.org/10.1007/s12601-012-0024-4
- Santer, R., V. Carrere, P. Dubuisson, and J. Roger, 1999. Atmospheric correction over land for MERIS, International Journal of Remote Sensing, 20(9): 1819-1840. https://doi.org/10.1080/014311699212506
- Vermote, E. and N.Z. Saleous, 2006. Operational Atmospheric Correction of MODIS Visible and Middle Infrared Land Surface Data in the Case of an Infinite Lambertian Target, In: Qu, J.J., Gao, W., Kafatos, M., Murphy, R.E., Salomonson, V.V. (Eds.), Earth Science Satellite Remote Sensing: Vol. 1: Science and Instruments, pp. 123-153, Springer, Berlin, Heidelberg, Germany.
- Vermote, E.F., D. Tanre, J.L. Deuze, M. Herman, and J. Morcette, 1997. Second simulation of the satellite signal in the solar spectrum, 6S: An overview, IEEE Transactions on Geoscience and Remote Sensing, 35(3): 675-686. https://doi.org/10.1109/36.581987
- Vermote, E., and A. Vermeulen, 1999. Atmospheric correction algorithm: spectral reflectances (MOD09), https://modis.gsfc.nasa.gov/data/atbd/atbd_mod08.pdf, Accessed on Feb.22, 2017.
- Wang, M., 2005. A refinement for the Rayleigh radiance computation with variation of the atmospheric pressure, International Journal of Remote Sensing, 26(24): 5651-5663. https://doi.org/10.1080/01431160500168793
- Yan, D., X. Zhang, Y. Yu, and W. Guo, 2016. A comparison of tropical rainforest phenology retrieved from geostationary (seviri) and polarorbiting (modis) sensors across the congo basin, IEEE Transactions on Geoscience and Remote Sensing, 54(8): 4867-4881. https://doi.org/10.1109/TGRS.2016.2552462
- Yeom, J. and H. Kim, 2015. Comparison of NDVIs from GOCI and MODIS data towards improved assessment of crop temporal dynamics in the case of paddy rice, Remote Sensing, 7(9): 11326-11343. https://doi.org/10.3390/rs70911326