Open Access
Add to BibSonomyAbstract
The radiometric stability of satellite sensors is crucial for generating highly consistent remote sensing measurements and products. We have presented a radiometric responsivity tracking method designed especially for optical sensors without on-board calibration systems. Using a temporally stable desert site with high reflectance, the sensor responsivity was simulated using the Second Simulation of the Satellite Signal in the Solar Spectrum (6S) radiative transfer model (RTM) with information from validated MODIS atmospheric data. Next, radiometric responsivity drifting was identified using a linear regression of the time series bidirectional reflectance distribution function (BRDF) normalized coefficients. The proposed method was applied to Chinese HJ-1A/1B charge-coupled device (CCD) sensors, which have been on-orbit operations for more than 5 years without continuous assessment of their radiometric performance. Results from the Dunhuang desert site between 2008 and 2013 indicated that the CCD sensors degraded at various rates, with the most significant degradation occurring in the blue bands, ranging from 2.8% to 4.2% yr−1. The red bands were more stable, with a degradation rate of 0.7-3.1% yr−1. A cross-sensor comparison revealed the least degradation for the HJ-1A CCD1 (blue: 2.8%; green: 2.8%; red: 0.7%; and NIR: 0.9% yr−1), whereas the degradation of HJ-1B CCD1 was most pronounced (blue: 3.5%; green: 4.1%; red: 2.3%; and NIR: 3.4% yr−1). The uncertainties of the method were evaluated theoretically based on the propagation of uncertainties from all possible sources of the RT simulations. In addition, a cross comparison with matchup ground-based absolute calibration results was conducted. The comparison demonstrated that the method was useful for continuously monitoring the radiometric performance of remote sensors, such as HJ-1A/1B CCD and GaoFen (GF) series (China's latest high-definition Earth observation satellite), and indicated the potential use of the method for high-precision absolute calibration.
© 2015 Optical Society of America
1. Introduction
Continuous and consistent multi-temporal/sensor remote sensing measurements are crucial for observing land surfaces, the atmosphere and water dynamics. However the satellite measurements and products are very sensitive to uncertainties related to sensor degradation due to the aging of components in the harsh conditions of space [1]. For sensors with onboard calibration systems, such as Landsat 7 ETM + and MODIS [2, 3], degradation is tracked and adjusted to achieve high quality satellite measurements and products, such as surface reflectance and NDVI trends. However, for sensors without onboard calibration capabilities, such as SeaWiFS [4] and NOAA AVHRR instruments [5], vicarious calibration (VC) approach is required to monitor their on-orbit radiometric performance. The VC approach is based on spectrally stable calibration sites in high-reflectance desert regions [6, 7] or on the homogeneous and clear ocean [8, 9] (both regarded as pseudo-invariant targets). In addition, VC provides an independent approach for the cross calibration of different sensors [10, 11] and for identifying the radiometric stability of onboard calibration systems (i.e., for MODIS and Landsat ETM + ) [12], which guarantees that consistent products are obtained from multi-sensor or multi-temporal measurements.
In China, remote sensing earth observations are rapidly developing, with a progressively greater number of satellites sensors that provide unparalleled data at multi spectral, spatial and temporal resolutions [13]. On September 6, 2008, China successfully launched environment and disaster monitoring and forecasting satellites (HJ-1A/1B). The payloads of the HJ-1A and HJ-1B satellites included two charge-coupled device (CCD) cameras that were designed on similar principles, including their radiometric resolutions and spectral response characteristics. The CCD cameras capture four spectral bands (430–520, 520–600, 630–690, and 760–900 nm) with a scan swath of 360 km (700 km with 2 sensors combined) and record data at 8 bits [14]. The constellations of the two satellites generate multispectral images with high spatial resolution (30 m) and a short revisit interval (2 days).
The HJ-1A/1B CCD data have been widely used for environmental issues, disaster detection, and natural resource monitoring, including land-use and land-cover change analyses (LULCC) [15], net primary productivity or biomass estimations [16, 17], and air quality monitoring (haze or aerosol retrieval) [18, 19]. Although the primary purpose is for land observation, the HJ-1A/1B CCD data collected over coastal and inland waters have successfully been used to detect multiple features, such as water suspended sediment concentrations [20], green microalgae blooms [20, 21], Chl-a and CDOM [22]. In addition, motivated by the wide range of applications of the HJ-1A/1B CCD data and the nearly identical spectral coverage and spatial resolution of Landsat TM/ETM + , the potential of extending the Landsat series Earth observations with HJ-1A/1B CCD data has been studied [18–21], and the results have shown that HJ-1A/1B CCD and Landsat 5 TM imageries are comparable and complementary.
The constellation of HJ-1A/1B satellites has operated on orbit for more than 5 years and continues to serve social and scientific communities despite being designed for a two-year lifespan. However, it is important to identify the on-orbit radiometric stability of the HJ-1A/1B CCD sensors when time-series data are used and compared with measurements from other sensors. Unfortunately, HJ-1A/1B satellites do not have onboard calibration systems for tracking the optical performance of the CCD sensors throughout the mission life. As an alternative, the China Center for Resources Satellite Data and Application (CRESDA) has conducted annual calibration efforts at the Dunhuang Calibration Site using the reflectance-based method since its launch [23]. In addition, the cross-calibration technique [24, 25] was also used to investigate the radiometric performances of the HJ-1A and HJ-1B CCD sensors.
However, both the field-based and cross-calibration efforts are constrained by limited frequency of valid measurements and are inadequate for providing continuous tracking of the long-term sensor performance. For example, the field-based calibration could only be conducted once a year for China HJ-1A/1B CCD sensors due to the high cost of filed measurements. On the other hand, cross-calibration are sensitive to the selection of matchups between observation and reference sensors (well-calibrated), which require (1) Simultaneous observations of the same location with identical or similar sun-target-view geometry from the two sensors, and (2) Clear and stable atmospheric conditions to minimize uncertainties due to atmospheric variations between observation and reference [26, 27]. Previous study proven that only 13 matchups were obtained for cross-calibration between the HJ-1A/1B CCD and Landsat TM from 2009 to 2011 [28], and 13 matchups for HJ-1A/1B CCD and Terra/MODIS from 2009 to 2012 [24], which means there are only 3 or 4 valid measurements for cross calibration each year. Therefore, uncertainties from field measurements, atmospheric variation, and discrepancy in view geometries, spectral band mismatching, and discrepancy in relative spectral response (RSR) functions etc. would pronounce more errors through propagation, which is quite difficult to eliminate due to limited matchups. Consequently, comprehensive analyses with long-term optical performance monitoring are lacking for the HJ-1A/1B CCD sensors.
In this paper, a practical method is proposed to track the continuous orbit radiometric performance of the Chinese HJ-1A/1B CCD sensors or other similar optical sensors without on-board calibration components. The efficiency of the method was evaluated theoretically based on the propagation of uncertainties from all possible sources and by comparing with the corresponding ground-based absolute calibration results. The results for the HJ-1A/1B CCD’s optical performance tracking and potential applications in other sensors were also discussed
2. Method for tracking radiometric responsivity
The objective of this study is to track the continuous on-orbit optical performance of HJ-1A/1B CCD sensors using radiometrically stable bright targets in the Dunhuang desert site. All available HJ-1A/1B CCD collections over Dunhuang site are acquired and screened for cloud-free high quality images. The acquisition time, the sun-target-satellite geometric angles are extracted from the metafile for each selected imagery, and the averaged digital number (DN) values over Dunhuang site are calculated. Then, two critical issues are addressed before a continuous and objective radiometric trend can be established from the long-term satellite measurements. First, time series TOA radiance are required to simulate the signal for each HJ-1A/1B CCD acquisition, according to the specific sun-target-satellite geometry, and corresponding atmospheric parameters. The Second Simulation of the Satellite Signal in the Solar Spectrum (6S) model is selected for its high accuracy and reasonable computation efficiency, using a successive orders method of scattering approximation. Moreover, the 6S code provides a more flexible approach for batching of time series data with restricted number of inputs and constants [29, 30]. Second, impacts of bidirectional reflectance distribution function (BRDF) effects, due to variations in sun-target-satellite geometries, on the time series radiometric response coefficients must be mitigated. To avoid excessive field measurements of physical BRDF models, a semiempirical kernel-driven BRDF model is adopted to normalize the illumination and viewing geometry variability between multi temporal remote sensing images [31, 32]. After that, time series radiometric responsivity is fitted using the normalized coefficients for trending estimation. The flowchart is shown in Fig. 1.
2.1 Screening and processing of satellite data
The HJ-1A/1B CCD images over the Dunhuang site from September 2008 to September 2013 were selected based on rigorous criteria. Generally, clouds, precipitation and sandstorms contaminated the images and violate the assumption of site stability. The coefficient of variation (CV), which is calculated as the ratio of the standard deviation (Std) to the average (Mean), provides a direct indication of radiometric spatial uniformity. Because the pixels affected by clouds or sandstorms would significantly increase the CV values, a CV threshold of 5% was used to exclude contaminated images. Furthermore, the images that were potentially affected by sandstorms, rainfall, or snow were further screened to avoid the subsequent effects of atmospheric particles or aerosols and the moisture content of the surface sand. The newly released MODIS Collection 6 (C6) Level 1B (L1B) products with improved long-term calibration stability [33] were obtained to distinguish the days that were affected by sandstorms, rainfall, and snow. Overall, 408 cloud-free MODIS L1B images from 2009 to 2013 were selected, and the statistics obtained from the time-series TOA reflectance were used as a baseline for the radiometric performance of the site. Thresholds for the TOA reflectance of the MODIS 465 nm band were determined as the mean plus (the upper threshold) and minus (the lower threshold) 3 times of the standard deviation, which were approximately 0.179 and 0.262, respectively. Because the site and MODIS measurements were both considered stable over time, any statistical anomalies should be excluded. Therefore, this pair of thresholds was assumed time-independent for distinguishing the images affected by sandstorms, rainfall, and snow.
To constrain the BRDF effects from the varied sun-target-satellite geometry, the HJ-1A/1B CCD sensors were arranged to enlarge the field of view to ± 31° from nadir, and the changes in the solar zenith angle (SZA) were confined to between approximately 20 and 60°. Overall, 155, 97, 154, and 82 images were selected to track the radiometric trends of the HJ-1A CCD1, HJ-1A CCD2, HJ-1B CCD1, and HJ-1B CCD2 sensors, respectively. All images were geo-referenced using the nearest-neighbor approach referring to Landsat TM images (UTM WGS-84), with the RMSE within half a pixel, and then a commom area of 90*120 pixels were selected over Dunhuang site to calculate the average DN values.
2.2 Simulation of TOA radiance
For each selected HJ-1A/1B imagery over the Dunhuang site, the TOA reflectance was simulated using 6S under specified sun-target-satellite geometry, surface and atmospheric parameters by Eq. (1):
whereis the top-of-atmosphere reflectance received by the sensor(unitless), is the path reflectance due to Rayleigh and aerosol scattering, and are the downward atmospheric transmittance from the sun to the surface and the upward atmospheric transmittance from the surface to the satellite viewing direction, is the surface reflectance (assuming a Lambertian surface during the simulation), and is the diffuse reflectance of the atmosphere.The input parameters for implementing the RT simulation include the (1) surface reflectance, (2) atmospheric parameters (ozone, total water vapor, aerosol information), and (3) sun-target-satellite geometry and sensor response function.
(1) Surface reflectance
The Dunhuang calibration/validation site is selected as a temporal stable reference endorsed by the Committee on Earth Observation Satellites (CEOS) and the Working Group on Cal/Val (WGCV). The site has a typical continental arid climate with a low aerosol loading [34], and the surface is covered by cemented gravel with almost no vegetation throughout the year [35, 36]. Therefore, the site provides great opportunity for radiometric calibration and responsivity tracking. Helde et al. observed that the temporal variability at the Dunhuang site is approximately 1% in the VIS-NIR bands, which is comparable to the Saharan and Arabian sites [37]. In addition, Hu et al. showed that the multi-year CV for the Dunhuang site is approximately 3% for the 7-year MODIS surface albedo product and multiple ground-based measurement in 10 years [38]. Thus, the site has been extensively used for VC activities among various sensors, including the NOAA/AVHRR, EOS/MODIS, FY-2B, CBERS-02B/CCD, HJ-1A/1B CCD, and HJ-1A HSI sensors [23, 25, 36, 39], and for tracking instrument degradation ATSR-2 from 1995 to 2000 [6] and FY-3A MERSI from 2008 to 2011 [40]. To further evaluate the temporal stability of the surface properties of the Dunhuang site during the study period, the BRDF-adjusted MODIS reflectance product MCD43C was acquired for Dunhuang site from 2009 to 2013. As shown in Fig. 2(a), the time-series MODIS reflectance was stable, with seasonal fluctuations mainly resulting from the periodic variations of the SZA and variability of less than 2% (1 sigma) for all bands, which was consistent with the previous results of Helder and Hu [37, 38]. Overall, the Dunhuang site is capable of providing spatial uniformity and temporal stability for long-term monitoring of satellite sensor calibration.
Fig. 2 (a) Trends of the MODIS reflectance product, MCD43A4, over the Dunhuang site from 2009 to 2013 and (b) Field measurements of the spectral reflectance for the Dunhuang site.
The surface reflectance of the Dunhuang site in this study was obtained from ground measurements that were conducted in 1999, 2000, 2002, 2004, 2005, 2006 and 2008 using ASD FieldSpec Spectral Radiometers with a 3 nm spectral resolution from 350 to 1,000 nm [41]. In addition, the field measurements conducted on August 26, 2009 from Gao et al. [42] and the field data collected in August 2011 from Sun et al. [40] were also adopted. The site has a stable spectral reflectance spectrum that ranges from 18% to 25% in the visible and near-infrared (NIR) regions [Fig. 2(b)]. These data both showed the stability of the ground surface at the Dunhuang site [Fig. 2(b)]. Overall, 2,135 samples were collected over the site, and the mean spectral reflectance data were determined from the multi-year measurements as the input for surface reflectance in 6S model.
(2) Atmospheric parameters
Atmospheric parameters for 6S including ozone, total water vapor, aerosol AOD. Ozone data were obtained from the OMI Satellite Level 3e and daily-averaged ozone data from the NASA Space-Based Measurement of Ozone and Air Quality Website with a grid of 0.25 × 0.25° (http://ozoneaq.gsfc.nasa.gov). The MODIS Level 2 perceptible water vapor (PWV) products from Terra MODIS Collection 5 (MOD05_L2) were obtained at a spatial resolution of 1 km using the NIR algorithm during the day from the Goddard Earth Sciences Distributed Active Archive Center (DAAC) [43]. The water content was assumed stable because the time intervals between the HJ-1A/1B and Terra overpasses are within 30 min. AOD are critical for simulating the signals received by remote sensors. Because a lack of simultaneous measurements corresponded to HJ-1A/1B overpasses, the Aqua MODIS Collection 6 “Deep Blue” aerosol data were used as an alternative for the AOD estimations (http://ladsweb.nascom.nasa.gov/data/search.html) in the 6S simulation [44]. The quality of MODIS AOD are validated and improved using matched ground-based AOD data from the Dunhuang and Dunhuang_LZU AERONET sites, and details are provided in Section 4.1.
To facilitate the simulation, the 6S code is modified to include the RSRs of the HJ-1A/1B CCD sensors, and the image specified sun-target-satellite geometry is extracted in batch. The TOA radiance is then calculated using Eq. (2).
Here, is the theoretically measured radiance from the sensor under the specified sun-target-satellite geometry, is the band-dependent extraterrestrial solar irradiance, andis the Earth–Sun distance in astronomical units (AU). Then, the radiometric response coefficient (DN/W/m2/sr/μm) was obtained from the digital numbers (DNs) extracted from the homogeneous desert pixels and the simulated radiance for each HJ-1A/1B CCD acquisition (according to Eq. (3)), In addition, the reference radiance of the dark current or noise of the instrument were determined from the pre-launch calibration of the HJ-1A/1B CCD sensors provided by the CRESDA.2.3 Responsivity trends fitting
Although images were rigorously selected to suppress BRDF effects, changes in sensor and solar view geometry must be addressed prior to long-term monitoring of the radiometric performance of the HJ-1A/1B CCD sensors. Without the bidirectional surface parameters, the Lambertian surface reflectance of the Dunhuang site was used to simulate the TOA radiance, which inevitably introduced uncertainties from the surface and atmospheric BRDF due to variations in the sun-target-satellite geometry. Since the physical BRDF models require excessive field measurements to specify the surface parameters that may vary over time [31, 32], making them difficult to implement. In this study, the semiempirical kernel-driven BRDF model of Roujean et al. [45] was adopted to normalize the time-series sensor response coefficients to a common illumination and viewing geometry, to facilitate trending analysis of the radiometric response. The modeled responsivity coefficients were assumed to consist of three additive kernels that described the isotropic scattering, geometric shadowing, and volume scattering in the Roujean model, as expressed in Eq. (4):
where θs, θv, andare the sun zenith angle (SZA), view zenith angle (VZA), and relative azimuth angle, respectively; f1 and f2 are the model kernels representing the volume scattering component and the surface scattering component, which are directly calculated as functions of the solar and view geometric angles following Roujean et al. [31, 45]. Then, parameters of ,, and were determined using multiple linear regression based on the least squares method to minimize the differences between time series measured and modeled responsivity coefficients over a wide range of view geometry parameters from Eq. (5), where i refers to the days of the image. Next, the BRDF normalized coefficients were utilized to determine long-term sensor responsivity drifting at a high frequency based on the linear regression statistics of the time-series results according to Eq. (6).Here, is the number of days on orbit, and and indicate the daily drifting rate of the sensor responsivity coefficient for band . To investigate whether the sensor responsivity trend was statistically significant, the slope of the regression line was tested using Student’s t-test. The null hypothesis (H0: slope = 0) indicates that the long-term responsivity is statistically stable. The p-value was used to estimate the probability of rejecting the null hypothesis (H0), which was set at the 5%, 1% and 0.1% level. If the p-value is <0.05, then the null hypothesis (H0) is rejected; otherwise, the null hypothesis is not rejected, and no long-term drift is present in the sensor responsivity.3. Results
3.1 Comparison with matchup field-based VC coefficients
The proposed method is validated by comparing with field-based vicarious calibration coefficients for the same periods over Dunhuang site. The reflectance-based absolute radiometric calibrations of HJ-1A/1B CCD have been conducted by CRESDA in July or August annually at the Dunhuang site, during which surface reflectance and atmospheric parameters are measured simultaneously with satellite overpass. Coefficients from field-based VC provide the only available option for radiometric calibration of HJ-1A/1B CCD data at present, and are assumed as “ground truth” for method validation. Comparisons between the results obtained in this study and sensor responsivity of the ground-based calibration are presented from 2008 to 2013, and a total of 20 matchups are obtained for each band.
Figure 3 shows excellent consistency between the reflectance-based absolute calibrations and the results of this study, with an overall coefficient of correlation (R) of approximately 1 for all bands. The slopes of the regression line were close to unity with very small intercepts. The matchup comparisons indicated that the average relative differences between the two groups were approximately 2.48%, 4.54%, 2.97% and 5.18% for the blue, green, red and NIR bands, respectively (Table 1). The biases were considerably smaller than the estimated uncertainties presented in section 4.1 (7.3%, 6.7%, 6.1% and 6.1% for blue, green, red and NIR bands). The results indicated that variation at the Dunhuang site was smaller than assumed; thus, the developed method can be used to continuously monitor the radiometric performance of remote sensors and shows the potential for high-precision absolute calibration.
Fig. 3 Comparison between the reflectance-based absolute calibrations and the results of this study.
Table 1. Relative differences (%) between the vicarious field calibration and the results of this study
Since field-based vicarious approaches suffer from low frequency of measurements, the proposed method is more promising for higher frequency calibration opportunities and responsivity trending analyses. The approach used in this study could meet the desired level of radiometric calibration precision (under 5%) once sufficient surface data are collected to correct the BRDF effects of the HJ-1A/1B CCD data over the Dunhuang site, as will be discussed in section 4.2.
3.2 Tracking the radiometric responsivity of HJ-1A/1B CCD
The long-term radiometric responsivity of the HJ-1A/1B CCD sensor was tracked using the images acquired between September 2008 and September 2013, which covered a 5-year period since its launch, including the two-year design life and three years beyond the design life. Periodic patterns were observed in the time-series gain of each HJ-1A/1B CCD band for all sensors, as shown in Fig. 4. The oscillation period was approximately one year. The peak surge in the gains occurred during the late summer of each year after the period of the minimum SZA and was followed by a considerable decrease in gains during the autumn and winter seasons.
Fig. 4 Radiometric responsivity trending of HJ-1A CCD1, HJ-1A CCD2, HJ-1B CCD1, HJ-1B CCD2 over the Dunhuang site before BDRF normalization.
Similar oscillation patterns were observed and documented in the MODIS and Landsat 7 ETM + time-series data [33], which were demonstrated to result from the combined effects of variations in the illumination geometry due to varying SZAs and VZAs. In addition, similar patterns were discovered in the time-series MODIS reflectance product MCD43C, as presented in Fig. 2. Because the MCD43C data were adjusted to the nadir view, the main cause of such a seasonal variation was a change in the SZAs. However, for the HJ-1A/1B CCD sensors, this variation is more likely to result from “side-view scanning” because the field-of-view is enlarged to 360 km to obtain a wide view range. Consequently, the variations of the VZA were 0-30°, while the SZAs were confined to between approximately 20 and 60°. The BRDF effect consisted of a multivariable function of the VZA, the SZA and the relative azimuth angle. The semiempirical kernel-driven BRDF model was applied to mitigate seasonal oscillations when estimating the long-term sensor response trend. The performance of the BRDF normalization for each sensor is presented in section 4.1.
The long-term trending responses of the four HJ-1A/1B CCD sensors over the Dunhuang site were analyzed using the linear regression method to determine the time-series sensor responsivity coefficients before and after the BRDF correction. Figure 5 presents the sensor response tracking results with the fitted band specific equations. Radiometric response degradations were observed at all bands at varied levels, with negative slope values for the fitted lines ranging from −10−6 to −10−4 and from −10−5 to −10−4 before and after BRDF normalization. This result indicated daily degradation following the sensor launch. The most significant degradations occurred in the blue and green bands, with the highest slope values reaching −0.0001 for the HJ-1B CCD1 in the green band. The red and NIR bands displayed relatively gradual decreases, with the lowest slope value reaching −2 × 10−6 for the HJ-1A CCD1 red band. The radiometric performances of the HJ-1A/1B 4 CCD sensors were different in terms of the degradation rates and discrepancies in the intercepts of the fitted lines. Because the slopes were small and nearly horizontal, differences in the intercepts indicated inconsistent instrument radiometric characteristics. Although these four sensors were designed with identical principles and have been shown to be complementary to each other in many studies, inconsistencies among them do exist. These differences must be considered when taking quantitative measurements of land and comparing them with different CCD sensors.
Fig. 5 Radiometric calibration trending of HJ-1A CCD1, HJ-1A CCD2, HJ-1B CCD1, HJ-1B CCD2 over the Dunhuang site after BDRF normalization.
Next, the slope of the regression line was tested using Student’s t-test to determine whether the radiometric response trend was statistically significant under the null hypothesis (H0), in which a slope equal to zero indicates that the long-term TOA reflectance is statistically stable. Table 2 summarizes the slope, yearly gain change (%), and t-test results. The degradation rates for each band of each sensor were calculated for 2009 to 2013. The data from 2008 were excluded due to the limited number of available images. The degradation rates in the blue band were the most significant, ranging from 2.8% to 4.2% per year for all sensors and resulting in a degradation of more than 10% since launch. The HJ-1B CCD2 blue band exhibited the highest degradation, with the rate of more than 15%. The red band was among the most stable bands of the four CCD sensors, with a degradation rate between 0.7% and 3.1%. Among the four CCD sensors, the HJ-1A CCD1 changed the least (2.8%, 2.8%, 0.7%, and 0.9%), and the HJ-1B CCD1 was degraded the most (3.5%, 4.2%, 2.3%, and 3.4%). The trends were highly consistent between the data before and after BRDF normalization, except that the red and NIR bands of HJ-1A CCD1 and the NIR band of HJ-1A CCD2 were significantly degraded after applying the BRDF model.
Table 2. Summary of the HJ-1A/1B CCD sensor response trends
4. Discussion
4.1 Uncertainty budget analysis
An uncertainty analysis of the degradation results was performed by accounting for the processing chain in the TOA radiance simulation, including uncertainties in surface and atmospheric characterization and the radiative transfer code. The magnitudes of the potential errors for each individual source are presented below, including the individual errors and the overall levels of uncertainty, which were estimated based on the root sum of squares of the errors from each individual source. Multiple uncertainty sources including surface reflectance, atmospheric parameters, RTF model and BRDF impacts are considered in the analysis.
The atmospheric products used in this simulation were taken from the MODIS products, which required validation and modification over the Dunhuang site, especially for the MODIS aerosol products. The MODIS Deep Blue (DB) algorithm uses blue bands to retrieve aerosol properties over bright surfaces [46]. The DB product has been validated and assessed over deserts in Northwest China and North Africa [47–49], and approximately 73% of its retrievals fall within the expected error of 30%. Although an interval of approximately 3 h occurs between the HJ-1A/1B and Aqua MODIS overpass, it is reasonable to use DB products to estimate aerosol properties when assuming that these properties remain relatively stable during the study period since the Dunhuang site is typical for its low load and relatively stable aerosol properties. Previous studies based on continuous measurements have documented that the aerosol load and variations of the Dunhuang site are small during clear weather, despite larger diurnal changes and significant seasonal variations [50, 51].
The precision of using the MODIS DB products in the radiative simulation was further validated and improved by using the AERONET version 2, level 2 quality-assured data for the Dunhuang and Dunhuang_LZU sites (http://aeronet.gsfc.nasa.gov/). Overall, 44 matched AERONET AOD data were selected for MODIS DB AOD validation. The results indicate a high correlation between the AERONET AOD and MODIS DB AOD, with R-values of 0.88 (all quality) and 0.92 (QA = 3 only), as shown in Fig. 6. The MODIS DB AOD values have RMSE values of 0.17 with respect to the AERONET AOD, and the slope of the fitted line reached up to 0.96, which was near unity. Figure 7(a) shows that the mean difference between the AERONET AOD and MODIS DB AOD was approximately 20%, which indicated that the uncertainty of the MODIS DB estimated AOD was approximately 0.3 ± 0.06 less than the average level. Figure 7(b) shows the histogram distribution of all available MODIS DB products at the Dunhuang site from 2009 to 2013, with more than 80% of the AOD data having values of less than 0.3 and with an average value of 0.29. Then, the uncertainty of the MODIS AOD products was propagated to the simulated TOA radiance, as shown in Fig. 7(c). Uncertainties related to the AOD were analyzed by simulating the TOA radiance under possible AOD ranges before comparing it to the average AOD level of 0.3. The impacts of the AOD uncertainty were most significant in the blue band, which shows that the average difference caused by the AOD uncertainty was approximately 4% and that the NIR band exhibited a decrease of approximately 2% as the AOD changed from 0.3 to 0.5. Thus, it is reasonable to assume that the uncertainty introduced by using the MODIS DB product was less than 5% at the Dunhuang site because only clear and high quality imagery was selected through rigorous filtering. The uncertainties due to water vapor and ozone contents were determined similarly to the AOD, and the results indicated that the water vapor and ozone contents only slightly affected (~1%) the simulated TOA radiance (Table 3).
Table 3. Uncertainty budget of the TOA simulation for HJ-1A CCD1
Considering the long-term stability of the Dunhuang site, the surface reflectance variations were less than 1% [35]. The BRDF effects were normalized in the long-term trending analysis. The error was expected to be approximately 5% when considering the range of the viewing zenith angle and the relative azimuth angle [27]. The root mean square error (RMSE) was obtained by comparing the BRDF fitted results with the original data and the yield of the RMSE at the 5% probability level, with approximately 88%, 91%, 81% and 80% of the residual error concentrated between −0.05 and 0.05 for the blue, green, red and NIR bands, respectively (Fig. 8). The uncertainty of the 6S radiative transfer code simulation was approximately 1% [52]. The estimated uncertainties ranged from 6% to 7% for the NIR to the blue spectral regions. The estimated uncertainties were greater than the results from the match-up comparisons with ground-based absolute calibrations, which showed that the proposed method is more accurate than estimated.
4.2 Advantages and future improvements
As discussed above, although the proposed method was mainly developed for tracking the radiometric performance of optical sensors, especially sensors without on-board calibration capacity, it has shown great potential for achieving high-precision absolute radiometric calibration. The atmospheric conditions are relatively stable and the uncertainties due to atmospheric variations are small over highly reflective desert sites. In addition, the Dunhuang calibration site is dominated by low precipitation and a low aerosol load, which makes it possible to characterize the atmospheric conditions from the MODIS atmospheric products at a rather low uncertainty level, especially when the accuracy of such products is improved by field measurements.
The proposed method could be an independent and alternative solution for long-term and continuous tracking of the sensor responsivity, and meanwhile provides comparable precision with field-based calibration. Second, the method provides a more objective estimation of the long-term trend of the sensor responsivity, where the results were based on the statistically analysis with a time series of remote sensing images. As presented in Fig. 4, the high seasonal variations in satellite measurements make it hard and even impossible to obtain statistical significant results of the sensor performance at the low frequency of the filed-based and cross-calibration method. Furthermore, the method is an important complement and alternative of the field-based and cross-calibration methods, as a time and cost effective approach.
With continuous monitoring of the sensor response, the method provides a great potential to detect the abrupt degradation of satellite sensors, as revealed from the Landsat 5 TM with an abrupt decrease in internal calibrator response in late 1988 [53]. One possible reason that abrupt degradation was not observed is that it is uncommon and did not occur to HJ-1A/1B CCD during the study period. Nevertheless, the method is quite promising for abrupt detection with improved approach to remove noises and residual errors in the future.
The problems that remain to be addressed mainly lie in the remaining BRDF effects in time-series observations, which are common when multi temporal satellite measurements are utilized. The efficient removal of BRDF effects requires excessive field measurements to specify the surface parameters that may vary over time [31, 32], which makes such measurements difficult to implement. Alternatively, the operational MODIS BRDF product (MCD43) model parameters can be used to describe the BRDF reflectance [54], and some studies have proposed using BRDF parameterization derived from MODIS to correct for BRDF effects in the Landsat TM imagery [55]. Therefore, future improvements in precision and calibration of the proposed method are feasible after the applicability of MODIS BRDF parameters has been evaluated and extended to the HJ-1A/1B CCD imagery or other sensors. Because significant differences exist between the HJ-1A/1B and MODIS data in terms of spectral coverage and spatial resolution, cross sensor adjustment and regional modifications of the MODIS BRDF parameters are required.
Considering the continuing mission and the need for a successor of HJ-1A/1B, the China High-Resolution Earth Observation System, which includes seven HDEOSs (High-Definition Earth Observation Satellites), was planned to launch between 2013 and 2016 to provide near-real-time (NRT) observations for disaster prevention and relief, climate change monitoring, geographical mapping, and environmental and resource surveys. The first and second satellites of the GaoFen series (“GaoFen”, which means “high resolution” in Chinese) was launched in 2013 and 2014, with similar spectral characterization to HJ-1A/1B CCD and Landsat TM/ETM + . The intended mission will drive research forward and result in further applications of remote sensing. Furthermore, these satellites will promote the design and development of future satellite sensors. However, similar problems will be encountered regarding the evaluation of the radiometric performance of the GF series optical sensors, which also lack on-orbit calibration capabilities, and the method developed in this study could be applied to future sensors.
5. Summary and conclusions
An effective and efficient method was presented for continuously tracking the radiometric performance of optical sensors, especially sensors without on-board calibration capabilities. Taking advantage of a temporally stable and high reflectance desert site, the time-series sensor responsivity was simulated using 6S model with information from validated atmospheric data. This method was demonstrated to be superior for sensors such as the HJ-1A/1B CCD (from China) and other similar instruments when compared with field-based calibration or cross calibration approaches. In addition, this method is independent from simultaneous field measurements of satellite overpass. The uncertainties of this method were evaluated theoretically by propagating the uncertainties from all possible sources. Comparison of the results with the matchup ground-based absolute calibration results of the HJ-1A/1B CCD indicated this method’s capacity for continuously monitoring the radiometric performance of remote sensors as well as its potential for obtaining high-precision absolute calibration.
The radiometric performances of the HJ-1A/1B CCD sensors were tracked using all available images over the Dunhuang site during 5 years on-orbit operations. Continuous monitoring results of the radiometric responsivity of the HJ-1A/1B CCD indicated that degradation occurred for all bands of all of the sensors at different rates and levels, with the most significant degradation occurring for the blue bands (ranging from 2.8% to 4.2% yr−1). However, the NIR bands had degradation rates of 0.9-3.1% yr−1. Among the 4 CCD sensors, HJ-1A CCD1 was the most stable (blue: 2.8%; green: 2.8%; red: 0.7%; and NIR: 0.9% yr−1), and HJ-1B CCD1 was degraded the fastest (blue: 3.5%; green: 4.1%; red: 2.3%; and NIR: 3.4% yr−1). The tracking results are valuable for earth observation applications when time series HJ-1A/1B CCD data are used which calls for effective correction of sensor degradation, and when data from different sensors are integrated due to inconsistency of varied radiometric responsivity.
Furthermore, future improvements that overcome surface directional reflectance should be pursued to promote the use of this method for high-precision absolute calibration. Compared to the on-board or VC procedure, the proposed method offers an independent method for evaluating the radiometric stability of optical sensors, especially sensors without on-board calibration systems. The results revealed that the radiometric status of the HJ-1A/1B CCD sensors and other Earth observation satellites, such as ZiYuan (ZY-3) and China's latest high-definition Earth observation satellites, Gaofen-1 and Gaofen-2, which are confronting similar problems due to lack of orbit calibration capabilities, and the proposed method provides opportunity for regular monitoring of the present and forthcoming missions.
Acknowledgments
This work was supported by the National Natural Science Foundation of China (Nos. 41331174, 41071261, 40906092, 40971193, 41101415, 41401388 and 41406205), the National 863 Key Project (2012AA12A304), the National Basic Research Program (973 Program) (No. 2011CB707106), the Open Research Fund of Key Laboratory of Digital Earth Science, the Institute of Remote Sensing and Digital Earth at the Chinese Academy of Sciences (No. 2013LDE004), Special Fund by Surveying & Mapping and Geoinformation Research in the Public Interest (No. 201412010), the Program for Changjiang Scholars and Innovative Research Team in University (IRT1278); the Major Science and Technology Program for Water Pollution Control and Treatment (2013ZX07105-005); the Hong Kong Research Grants Council (RGC) General Research Fund (Grant No. B-Q23G); LIESMARS Special Research Funding from the “985 Project” of Wuhan University, and the Special Funds of State Key Laboratory (for equipment). The authors would like to thank the China Center for Resources Satellite Data and Application (CRESDA) for providing the HJ-1A/B CCD data sets. The authors are grateful for the hard work for field measurements from the members during the radiometric calibration works over Dunhuang site. In addition, we would like to thank Brent. N. Holben for his efforts in establishing and maintaining the Dunhuang AERONET (AErosol RObotic NETwork) site. In addition, we extend special thanks to Prof. Chuanmin Hu (University of South Florida) for providing insight and encouragement regarding this work.
References and links
1. J. Guorui, Z. Huijie, and L. Hao, “Uncertainty propagation algorithm from the radiometric calibration to the restored earth observation radiance,” Opt. Express 22(8), 9442–9449 (2014). [CrossRef] [PubMed]
2. B. L. Markham, K. J. Thome, J. A. Barsi, E. Kaita, D. L. Helder, J. L. Barker, and P. L. Scaramuzza, “Landsat-7 ETM+ on-orbit reflective-band radiometric stability and absolute calibration,” IEEE Trans. Geosci. Rem. Sens. 42(12), 2810–2820 (2004). [CrossRef]
3. X. Xiong, A. Wu, J. A. Esposito, J. Sun, N. Che, B. Guenther, and W. L. Barnes, “Trending results of MODIS optics on-obit degradation,” Proc. SPIE 4814, 337–346 (2002). [CrossRef]
4. R. A. Barnes and E. F. Zalewski, “Reflectance-based calibration of SeaWiFS. I. Calibration coefficients,” Appl. Opt. 42(9), 1629–1647 (2003). [CrossRef] [PubMed]
5. D. Koslowsky, “Signal degradation of the AVHRR shortwave channels of NOAA 11 and NOAA 14 by daily monitoring of desert targets,” Adv. Space Res. 19(9), 1355–1358 (1997). [CrossRef]
6. D. L. Smith, C. T. Mutlow, and R. C. Nagaraja, “Calibration monitoring of the visible and near-infrared channels of the Along-Track Scanning Radiometer-2 by use of stable terrestrial sites,” Appl. Opt. 41(3), 515–523 (2002). [CrossRef] [PubMed]
7. B. Fougnie, G. Bracco, B. Lafrance, C. Ruffel, O. Hagolle, and C. Tinel, “PARASOL in-flight calibration and performance,” Appl. Opt. 46(22), 5435–5451 (2007). [CrossRef] [PubMed]
8. S. Hlaing, A. Gilerson, R. Foster, M. Wang, R. Arnone, and S. Ahmed, “Radiometric calibration of ocean color satellite sensors using AERONET-OC data,” Opt. Express 22(19), 23385–23401 (2014). [CrossRef] [PubMed]
9. F. Mélin and G. Zibordi, “Vicarious calibration of satellite ocean color sensors at two coastal sites,” Appl. Opt. 49(5), 798–810 (2010). [CrossRef] [PubMed]
10. G. M. Jiang, “Intercalibration of infrared channels of polar-orbiting IRAS/FY-3A with AIRS/Aqua data,” Opt. Express 18(4), 3358–3363 (2010). [CrossRef] [PubMed]
11. E. J. Kwiatkowska, B. A. Franz, G. Meister, C. R. McClain, and X. X. Xiong, “Cross calibration of ocean-color bands from Moderate Resolution Imaging Spectroradiometer on Terra platform,” Appl. Opt. 47(36), 6796–6810 (2008). [PubMed]
12. G. Chander, X. X. Xiong, T. Y. Choi, and A. Angal, “Monitoring on-orbit calibration stability of the Terra MODIS and Landsat 7 ETM+ sensors using pseudo-invariant test sites,” Remote Sens. Environ. 114(4), 925–939 (2010). [CrossRef]
13. W. Xu, J. Gong, and M. Wang, “Development, application, and prospects for Chinese land observation satellites,” Geo-spatial Information Science 17(2), 102–109 (2014). [CrossRef]
14. Q. Wang, C. Wu, Q. Li, and J. Li, “Chinese HJ-1A/B satellites and data characteristics,” Sci. China Earth Sci. 53(S1), 51–57 (2010). [CrossRef]
15. Z. Sun, W. Shen, B. Wei, X. Liu, W. Su, C. Zhang, and J. Yang, “Object-oriented land cover classification using HJ-1 remote sensing imagery,” Sci. China Earth Sci. 53(S1), 34–44 (2010). [CrossRef]
16. Z. Guo, H. Chi, and G. Sun, “Estimating forest aboveground biomass using HJ-1 Satellite CCD and ICESat GLAS waveform data,” Sci. China Earth Sci. 53(S1), 16–25 (2010). [CrossRef]
17. R. Zhang, R. Sun, J. Du, T. Zhang, Y. Tang, H. Xu, S. Yang, and W. Jiang, “Estimations of net primary productivity and evapotranspiration based on HJ-1A/B data in Jinggangshan city, China,” J. Mt, Sci-Engl. 10, 777–789 (2013).
18. Y. Li, Y. Xue, X. He, and J. Guang, “High-resolution aerosol remote sensing retrieval over urban areas by synergetic use of HJ-1 CCD and MODIS data,” Atmos. Environ. 46, 173–180 (2012).
19. Z. T. Wang, Q. Li, S. S. Li, L. F. Chen, C. Y. Zhou, Z. F. Wang, and L. J. Zhang, “The Monitoring of Haze from HJ-1],” Guang Pu Xue Yu Guang Pu Fen Xi 32(3), 775–780 (2012). [PubMed]
20. Z. Yu, X. Chen, B. Zhou, L. Tian, X. Yuan, and L. Feng, “Assessment of total suspended sediment concentrations in Poyang Lake using HJ-1A/1B CCD imagery,” Chin. J. Oceanology Limnol. 30(2), 295–304 (2012). [CrossRef]
21. T. Cui, J. Zhang, L. E. Sun, Y. Jia, W. Zhao, Z. Wang, and J. Meng, “Satellite monitoring of massive green macroalgae bloom (GMB): imaging ability comparison of multi-source data and drifting velocity estimation,” Int. J. Remote Sens. 33(17), 5513–5527 (2012). [CrossRef]
22. L. Chen, L. Tian, F. Qiu, and X. Chen, “Water Color Constituents Remote Sensing in Wuhan Donghu Lake Using HJ-1A/B CCD Imagery,” Geomatics and Information Science of Wuhan University 36, 1280–1283 (2011).
23. H. Jiang, Q. Qin, J. Li, S. Zhao, H. Dong, W. Yuan, and R. Cui, “Validation for the absolute radiometric calibration of the HJ-1B CCD sensors of China,” in Geoscience and Remote Sensing Symposium (IGARSS), (IEEE International, 2010), 2876–2879. [CrossRef]
24. G. Chen, Z. Chen, L. Ma, and H. Zhang, “Monitoring and assessment on radiometric stability of HJ-1A CCD using MODIS data,” in Eighth International Symposium on Multispectral Image Processing and Pattern Recognition, (International Society for Optics and Photonics, 2013), 89211I.
25. Z. Chen, B. Zhang, H. Zhang, W. Zhang, and Y. Zhang, “HJ-1A multispectral imagers radiometric performance in the first year,” in Geoscience and Remote Sensing Symposium (IGARSS), (IEEE International, 2010), 4264–4267. [CrossRef]
26. S. Sterckx, S. Livens, and S. Adriaensen, “Rayleigh, deep convective clouds, and cross-sensor desert vicarious calibration validation for the PROBA-V mission,” IEEE Trans. Geosci. Rem. Sens. 51(3), 1437–1452 (2013). [CrossRef]
27. Z. Wang, P. Xiao, X. Gu, X. Feng, X. Li, H. Gao, H. Li, J. Lin, and X. Zhang, “Uncertainty analysis of cross-calibration for HJ-1 CCD camera,” Sci. China Technol. Sci. 56(3), 713–723 (2013). [CrossRef]
28. B. Jiang, S. Liang, J. R. Townshend, and Z. M. Dodson, “Assessment of the Radiometric Performance of Chinese HJ-1 Satellite CCD Instruments,” IEEE J. STARS 6, 840–850 (2013).
29. E. F. Vermote, D. Tanré, J. L. Deuze, M. Herman, and J.-J. Morcette, “Second simulation of the satellite signal in the solar spectrum, 6S: An overview,” IEEE Trans. Geosci. Rem. Sens. 35(3), 675–686 (1997). [CrossRef]
30. S. M. Vicente-Serrano, F. Pérez-Cabello, and T. Lasanta, “Assessment of radiometric correction techniques in analyzing vegetation variability and change using time series of Landsat images,” Remote Sens. Environ. 112(10), 3916–3934 (2008). [CrossRef]
31. R. Latifovic, J. Cihlar, and J. Chen, “A comparison of BRDF models for the normalization of satellite optical data to a standard Sun-target-sensor geometry,” IEEE Trans. Geosci. Rem. Sens. 41(8), 1889–1898 (2003). [CrossRef]
32. J. Cihlar, H. Ly, Z. Li, J. Chen, H. Pokrant, and F. Huang, “Multitemporal, multichannel AVHRR data sets for land biosphere studies—Artifacts and corrections,” Remote Sens. Environ. 60(1), 35–57 (1997). [CrossRef]
33. A. Angal, X. Xiong, T. Choi, G. Chander, N. Mishra, and D. L. Helder, “Impact of Terra MODIS Collection 6 on long-term trending comparisons with Landsat 7 ETM+ reflective solar bands,” Romote Sens. Lett. 4(9), 873–881 (2013). [CrossRef]
34. X. Hu, Y. Zhang, Z. Liu, G. Zhang, Y. Huang, K. Qiu, Y. Wang, L. Zhang, X. Zhu, and Z. Rong, “Optical characteristics of China Radiometric Calibration Site for Remote Sensing Satellite Sensors (CRCSRSSS),” in Second International Asia-Pacific Symposium on Remote Sensing of the Atmosphere, Environment, and Space, (International Society for Optics and Photonics, 2001), pp. 77–86. [CrossRef]
35. C. Cao, L. Ma, S. Uprety, C. Li, and L. Tang, “Spectral characterization of the Dunhuang calibration/validation site using hyperspectral measurements,” Proc. SPIE 7862, 78620J (2010). [CrossRef]
36. C. Gao, X. Jiang, X. Li, and X. Li, “The cross-calibration of CBERS-02B/CCD visible-near infrared channels with Terra/MODIS channels,” Int. J. Remote Sens. 34(9-10), 3688–3698 (2013). [CrossRef]
37. D. L. Helder, B. Basnet, and D. L. Morstad, “Optimized identification of worldwide radiometric pseudo-invariant calibration sites,” Can. J. Rem. Sens. 36(5), 527–539 (2010). [CrossRef]
38. X. Hu, J. Liu, L. Sun, Z. Rong, Y. Li, Y. Zhang, Z. Zheng, R. Wu, L. Zhang, and X. Gu, “Characterization of CRCS Dunhuang test site and vicarious calibration utilization for Fengyun (FY) series sensors,” Can. J. Rem. Sens. 36(5), 566–582 (2010). [CrossRef]
39. Z. Zhang, Y. Qiu, K. Hu, X. Zhang, and J. Li, “Radiometric calibration on orbit for FY-2B meteorological satellite's visible channels with the radiometric calibration site of Dunhuang,” J. Appl. Meteorol. 3, 001 (2004).
40. L. Sun, X. Hu, M. Guo, and N. Xu, “Multisite Calibration Tracking for FY-3A MERSI Solar Bands,” IEEE Trans. Geosci. Rem. Sens. 50(12), 4929–4942 (2012). [CrossRef]
41. Dunhuang radiometric calibration site of China and China Meteorological Administration, Spectral Data Sets for Satellite Calibration Site and Typical Earth Objects (China Meteorological, Beijing, 2008).
42. H. L. Gao, X. F. Gu, T. Yu, X. Y. Li, H. Gong, J. G. Li, and G. H. Zhu, “HJ1A/HSI radiometric calibration and spectrum response function sensitivity analysis,” Spectrosc. Spect. Aanl. 30(11), 3149–3155 (2010). [PubMed]
43. B. Gao and Y. J. Kaufman, “Water vapor retrievals using Moderate Resolution Imaging Spectroradiometer (MODIS) near-infrared channels,” J. Geophys. Res. 108(D13), 4389 (2003). [CrossRef]
44. A. M. Sayer, N. C. Hsu, C. Bettenhausen, and M. J. Jeong, “Validation and uncertainty estimates for MODIS Collection 6 “Deep Blue” aerosol data,” J. Geophys. Res. 118, 7864–7872 (2013).
45. J. L. Roujean, M. Leroy, and P. Y. Deschamps, “A bidirectional reflectance model of the Earth's surface for the correction of remote sensing data,” J. Geophys. Res. 97(D18), 20455–20468 (1992). [CrossRef]
46. N. C. Hsu, S. Tsay, M. D. King, and J. R. Herman, “Deep blue retrievals of Asian aerosol properties during ACE-Asia,” IEEE Trans. Geosci. Rem. Sens. 44(11), 3180–3195 (2006). [CrossRef]
47. Y. Xie, Y. Zhang, X. Xiong, J. J. Qu, and H. Che, “Validation of MODIS aerosol optical depth product over China using CARSNET measurements,” Atmos. Environ. 45(33), 5970–5978 (2011). [CrossRef]
48. X. Li, X. Xia, S. Wang, J. Mao, and Y. Liu, “Validation of MODIS and Deep Blue aerosol optical depth retrievals in an arid/semi-arid region of northwest China,” Particuology 10(1), 132–139 (2012). [CrossRef]
49. Y. Shi, J. Zhang, J. S. Reid, E. J. Hyer, and N. C. Hsu, “Critical evaluation of the MODIS Deep Blue aerosol optical depth product for data assimilation over North Africa,” Atmos. Meas. Tech. 5(5), 7815–7865 (2012). [CrossRef]
50. X. Hu, Y. Zhang, G. Zhang, Y. Huang, and Y. Wang, “Measurements and Study of Aerosol Optical Characteristics in China Radiometric Calibration Sites,” J. Appl. Meteorol. 03, 257–266 (2001).
51. X. A. Xia, M. X. Wang, and Y. S. Wang, “Automatic and continuous measurement of aerosol properties in Dunhuang, China,” J. Environ. Sci. (China) 16(1), 40–43 (2004). [PubMed]
52. S. Y. Kotchenova, E. F. Vermote, R. Matarrese, and F. J. Klemm Jr., “Validation of a vector version of the 6S radiative transfer code for atmospheric correction of satellite data. Part I: path radiance,” Appl. Opt. 45(26), 6762–6774 (2006). [CrossRef] [PubMed]
53. D. L. Helder, T. A. Ruggles, J. D. Dewald, and S. Madhavan, “Landsat-5 Thematic Mapper reflective-band radiometric stability,” IEEE Trans. Geosci. Rem. Sens. 42(12), 2730–2746 (2004). [CrossRef]
54. C. B. Schaaf, F. Gao, A. H. Strahler, W. Lucht, X. Li, T. Tsang, N. C. Strugnell, X. Zhang, Y. Jin, J.-P. Muller, P. Lewis, M. Barnsley, P. Hobson, M. Disney, G. Roberts, M. Dunderdale, C. Doll, R. P. d’Entremont, B. Hu, S. Liang, J. L. Privette, and D. Roy, “First operational BRDF, albedo nadir reflectance products from MODIS,” Remote Sens. Environ. 83(1-2), 135–148 (2002). [CrossRef]
55. N. Flood, “Testing the local applicability of MODIS BRDF parameters for correcting Landsat TM imagery,” Romote Sens. Lett. 4(8), 793–802 (2013). [CrossRef]
