1. Introduction
Studies in the past (Staylor 1990; Rao and Chen 1993; Smith et al. 1997; Xiong et al. 2001) have shown that it is not uncommon that visible–near-infrared (VNIR) bands and shortwave infrared (SWIR) bands in satellite radiometers will degrade after launch. The stability of satellite sensors is critical for providing consistent data products from multiple instruments on different satellites. Even if the sensor is stable, bias can exist between instruments. Earlier studies (Xiong et al. 2010; Wu et al. 2008; Uprety and Cao 2011) have shown that it is possible to study the sensor stability using independent calibration sources, which include onboard calibrators, and vicarious sites. Onboard calibrators are excellent sources of calibration for satellite sensors as long as they are stable and well characterized. Studies in the past (Chander et al. 2004; Xiong et al. 2001) have shown that internal calibrators can also degrade over time. A number of studies in the past (Cao et al. 2005; Wu and Sun 2005; Wu et al. 2006; Heidinger et al. 2002, 2010) have shown that intercomparison of one satellite sensor with another stable and well-calibrated satellite sensor is one of the potential techniques for monitoring sensor performance and quantifying radiometric bias for relative and absolute calibration.
The primary purpose of the Suomi National Polar-Orbiting Partnership [Suomi-NPP (S-NPP)] Visible Infrared Imaging Radiometer Suite (VIIRS) is to fulfill the requirement of high-quality weather- and climate-related data. The calibration of reflective solar bands (RSB) in VIIRS is performed using an onboard calibrator that consists of a solar diffuser and a solar diffuser stability monitor to track the stability of the solar diffuser. VIIRS is currently undergoing extensive on-orbit calibration and validation in an effort to ensure that the satellite data products are consistent with other well-calibrated satellite radiometers. De Luccia et al. (2012) showed how some of the VIIRS VNIR bands started to degrade rapidly after launch and explained the root causes for the observed degradation. Thus, continuous monitoring of VIIRS radiometric performance is essential. More details on VIIRS instrument characteristics and on-orbit radiometric/geometric performance has been summarized in Cao et al. (2013). The absolute radiometric accuracy of VIIRS can be achieved by continuously monitoring, characterizing, and comparing the VIIRS instrument to other well-calibrated radiometers such as the Moderate Resolution Imaging Spectroradiometer (MODIS), whose absolute calibration uncertainty, as shown by Xiong et al. (2007), is within 2%. This study focuses on the VIIRS performance and characterization of the on-orbit radiometric bias by intercomparing VIIRS moderate-resolution radiometric bands with MODIS. VIIRS top of the atmosphere (TOA) reflectance measurements are compared with matching MODIS bands at the VNIR region over ocean and at the VNIR/SWIR region over desert. The desert and ocean sites have relatively stable and homogeneous radiometric properties. Cross comparison between VIIRS and MODIS over these sites can help calibrate both the low-gain and high-gain channels of VIIRS. The methodology used is the simultaneous nadir overpass extended (SNO-x) to the low latitude, which is explained below in more detail. Once the intersatellite bias between VIIRS and MODIS is quantified, radiometric consistency can be achieved with estimated uncertainties. The well-characterized and well-calibrated VIIRS data can then be intercalibrated with National Oceanic and Atmospheric Administration (NOAA) series instruments such as the Advanced Very High Resolution Radiometer (AVHRR) to establish radiometric consistency and maintain data continuity to generate multidecadal global climate data records.
2. Methodology
a. Radiometer intercomparisons at SNO extended to the low latitudes
Earlier studies (Cao and Heidinger 2002; Cao et al. 2005, 2008; Heidinger et al. 2002, 2010) have shown that SNO-based intersatellite comparison is a very useful technique for calibration and validation of satellite sensors. The methodology can be used to assess the radiometric consistency between satellites. When two or more satellites orbit the earth at different altitudes, SNOs periodically occur. Cao et al. (2004) explains in more detail about the prediction of simultaneous nadir overpasses among polar-orbiting satellites. The SNO technique can be used to compare multiple instruments at their orbital intersection with a small time difference between the instruments’ observation. Comparison of simultaneous measurements between two or more instruments at their orbital intersection with almost identical viewing conditions makes this approach extremely suitable for reducing the uncertainties associated with atmospheric absorption variability, and bidirectional reflectance distribution function (BRDF). Orbital intersection usually occurs at high-latitude polar region for polar-orbiting satellites. One of the limitations with high-latitude SNO-based comparison is that the instruments cannot be compared over a wide radiance dynamic range. However, in addition to the polar region, SNO between S-NPP and Aqua satellites occurs more frequently at low latitudes, although with a larger time difference between the instruments’ observations. This provides ample opportunities to compare S-NPP VIIRS measurements with Aqua MODIS at low latitudes that cover different target types. The SNOs at low latitudes can be used to compare instruments over uniform and stable targets such as ocean surface. SNOs at low latitudes can be extended further to compare over deserts, green vegetation, etc. SNO-x compares two instruments not only at orbital intersection but also at the overlapping region of the extended orbits. This study extends SNOs to compare VIIRS and MODIS at the ocean and the North African desert.
Since the spectral response functions of the two instruments’ matching bands may have differences in their spectral coverage and shape, it can result in different magnitudes of bias when compared at different sites. Hence, the spectral characteristics of the targets needed to be analyzed in order to quantify the radiometric bias in more detail. This paper analyzes spectral characteristics of the ocean and desert using hyperspectral measurements from the Earth Observing-1 (EO-1) Hyperion, the Airborne Visible and Infrared Imaging Spectrometer (AVIRIS), and a radiative transfer model such as the Moderate Resolution Atmospheric Transmission (MODTRAN). The difference in VIIRS bias relative to MODIS calculated using TOA reflectance over ocean and desert can be explained primarily using spectral differences, atmospheric absorption variability, calibration uncertainties, and residual cloud contamination. The primary reasons to limit the study to these two targets are as follows: these targets are radiometrically more stable (which helps to reduce radiometric uncertainties), are spatially more uniform (which reduces the registration errors), and offer a larger number of SNO events that can be extracted for bias analysis. In addition, the hyperspectral measurements for these targets are available from Hyperion and AVIRIS and are used to characterize the spectral characteristics and its impact on radiometric differences.
The location of SNO events between MODIS and VIIRS ranges from the high-latitude polar region to the low-latitude tropical region. SNOs between VIIRS and MODIS occur frequently at low latitudes, once every 2–3 days. The orbits in low-latitude SNOs are extended to perform sensor intercomparison. The SNO time difference is more than 10 min between the two sensors. Time difference between VIIRS and MODIS observations for SNO-x comparison is limited to within 16.5 min over desert, whereas the time difference over ocean ranges from 10 to 15 min. Distribution of SNO time difference over the desert used in this study is shown in Fig. 1 (bottom left). Maximum time difference between the VIIRS and MODIS observation used in the study is 16 min. During this time difference, the change in solar zenith angle is about 2°. The major impacts on sensor intercomparison are due to (i) change in solar geometry, which results in the change of BRDF; and (ii) change in clouds. The impact of cloud change has been removed by using clear-sky pixels through cloud mask products. In addition, since there are no major atmospheric absorptions in the VIIRS and matching MODIS bands used in this study, the impact of change in atmospheric variability is assumed to be constant within the given time difference limit of 16 min. The MODTRAN4.3-based analysis shows that the maximum change in reflectance over a desert is less than 0.2% for the time difference of 16 min, keeping the atmospheric parameters constant. In addition the number and size of regions of interest (ROIs) extracted for each SNO-x event are large enough that they also help to encompass the uncertainty due to the larger time difference. The two instruments can be compared at both an exact orbital intersection and extended SNO orbits over ocean. However, the radiometric comparison of VNIR measurements over the North African desert is possible only for the extended SNO orbits. This study uses less than ±40° latitude limits for SNO-x-based comparison. In addition, it is possible that the observation over ocean is impacted by sun glint for one instrument but not in the other mainly due to the time difference of more than 8 min. Sun-glint-impacted measurements are excluded from intercomparison. However, the chances of increased uncertainty due to cloud- and sun-glint-impacted pixels can always exist in the bias analysis.
(top left) MODIS and VIIRS orbits extended from the higher-latitude orbital intersection (SNO) to low latitude (North Africa) for intercomparison. (top right) VIIRS and MODIS orbits showing low-latitude SNO events (blue dots). The two instruments are compared at the exact orbital intersection and its extension to the low latitudes. (bottom left) SNO-x time difference between VIIRS and MODIS over the North African desert. Each dot represents the time difference for one SNO-x event.
Citation: Journal of Atmospheric and Oceanic Technology 30, 12; 10.1175/JTECH-D-13-00071.1
Figure 1 (top left) shows four MODIS and VIIRS orbits passing through the North African desert. There is no exact SNO event (orbital intersection) over desert during daytime to compare these two instruments. However, each orbit pair shown in the figure that passes through the desert is actually extended from high-latitude SNO events. Each extended SNO orbit of MODIS and VIIRS observes the same ground target with a small time difference of 15–16 min. Figure 1 (top left) shows four SNO-x events between VIIRS and MODIS. The time period for these SNO-x events are 17 December 2012: MODIS (1355:00–1413:00 UTC) and VIIRS (1338:59–1357:40 UTC); 22 October 2012: MODIS (1310:00–1320:00 UTC) and VIIRS (1251:12–1302:34 UTC); 22 October 2012: MODIS (1130:00–1140:00 UTC) and VIIRS (1106:39–1118:01 UTC); and 15 November 2012: MODIS (1220–1230 UTC) and VIIRS (1158:56–1210:18 UTC). These SNO-x events are periodic in nature and thus the extended SNO orbits are periodic as well and repeat every 16 days, which is same as the repeat cycle of the satellites. The observations over the desert by MODIS and VIIRS on these extended SNO orbits are compared at near nadir to quantify the radiometric bias. It is to be noted that even though VIIRS and MODIS orbits stretch from the SNO location (most of the time over ocean) all the way down to the desert, sensor intercomparison is performed only over the desert. The top-right panel in Fig. 1 shows the SNO events between VIIRS and MODIS over the ocean that occur at low latitudes. VIIRS is compared to MODIS measurements at the exact orbital intersection and at extended orbits to quantify VIIRS-observed bias. Collocation criteria use the maximum limit of a 6° scan angle difference over the desert. Thus, out of four SNO-x orbits shown in the top-left panel of Fig. 1, only two of them satisfy the angular difference criteria of less than 6° for intercomparison. Unlike over the desert, the scan angle difference over the ocean is limited to less than 2° for collocation and sensor intercomparison.
b. Instrument background
S-NPP VIIRS is a multispectral cross-track scanning sensor aboard the Joint Polar Satellite System (JPSS)/S-NPP satellite, which was launched on 28 October 2011 (Cao et al. 2013). It has 22 spectral bands that consist of 15 reflective solar VNIR/SWIR bands (RSB) and 7 thermal emissive bands (Cao et al. 2013). The spectral bands are categorized as moderate resolution, imagery, and day–night bands. The spatial resolution at nadir is approximately 750 m for the moderate bands and 375 m for the imagery bands (Cao et al. 2013). Unlike the previous instruments such as MODIS and AVHRR, the ground samples of VIIRS observation are aggregated in scan direction to limit changes of spatial resolution across the entire swath (Hutchison and Cracknell 2005; Cao et al. 2013).
MODIS is a National Aeronautics and Space Administration (NASA) instrument launched aboard the Terra (1999) and Aqua (2002) satellites. MODIS has 36 spectral bands with wavelengths ranging from 0.4 to 14.4 μm. It is a sun-synchronous, near-polar-orbiting satellite with 2 bands having a spatial resolution of 250 m; 5 bands, 500 m; and 29 bands, 1 km. A complete scan covers a swath width of 3000 km (Cao et al. 2013) for VIIRS and 2330 km for MODIS (Guenther et al. 2002). Both VIIRS and MODIS instruments are equipped with an onboard calibration system that consists of a solar diffuser along with the solar diffuser stability monitor for the calibration of reflective solar bands and a blackbody for the calibration of thermal emissive bands.
The EO-1 Hyperion is a push broom hyperspectral sensor that images the earth in 220 spectral bands (http://eo1.usgs.gov/sensors/hyperion) with wavelengths ranging from 0.4 to 2.5 μm. It has a footprint size of 30 m and a swath width of 7.5 km. The radiometric stability of the instrument is maintained using solar, lunar, and in-flight calibration sources (Folkman et al. 2001; Pearlman et al. 2003). AVIRIS is a whisk broom scanning hyperspectral sensor. It is an airborne instrument flying at an altitude ranging from 6000 to 17 500 m. It has 224 spectral bands ranging from 0.4 to 2.5 μm. The spatial resolution varies between 20 and 10 m depending on the altitude (Bossler et al. 2010).
c. Data processing
This study computes the radiometric bias of VIIRS moderate bands with matching MODIS bands using extended SNO over the ocean surface and African desert. Spectral coverage information for VIIRS and MODIS matching bands are given in Table 1. For VIIRS bands M-4, M-5, and M-7, MODIS matching bands used for comparison over desert (MODIS land bands 4, 1, and 2) are different from ocean (MODIS ocean bands 12, 13, and 16). A time series of VIIRS bias relative to MODIS is established using low-latitude extended SNO to quantify the VIIRS radiometric performance. Since the observed bias is target dependent (primarily due to the spectral differences of targets), the impact is quantified through hyperspectral measurements and a radiative transfer model.
VIIRS and MODIS matching bands information.
The processing steps for computing observed bias are explained below.
SNO-x events and data collection
The extended low-latitude SNO events between VIIRS and MODIS were predicted and analyzed from the end of February 2012 until early February 2013. For each SNO event over ocean, VIIRS scientific data records (SDR) were collected from the NOAA Comprehensive Large Array-Data Stewardship System (CLASS, http://www.class.ncdc.noaa.gov) and MODIS level 1b (L1B) data were collected from the NASA Level 1 and Atmosphere Archive and Distribution System (LAADS, http://ladsweb.nascom.nasa.gov/). There are several collections of MODIS L1B data. This study uses MODIS L1B collection 6.0 data, which are the most recent MODIS L1B product with calibration improvements over the previous versions. The spatial resolutions of data used in the comparison are 750-m L1B VIIRS moderate bands and 1-km MODIS L1B.
Mapping VIIRS data to MODIS
The radiometric comparison is performed after mapping VIIRS imagery into MODIS latitude/longitude pixel by pixel by using fast geospatial matching algorithm (GSM) explained below.
For an exact SNO-based comparison over ocean, the orbital intersection between MODIS and VIIRS image is first found. This is achieved by calculating the smallest nadir distance between the VIIRS and MODIS images. The nadir distance threshold is set to be less than 1 km at the orbital intersection. Once the orbital intersection is found, VIIRS and MODIS image data with orbital intersection as the center are extracted. The angular difference considered in the study for collocation and sensor intercomparison is less than 2°.
For an extended-SNO-based comparison over ocean and desert, VIIRS and MODIS images corresponding to extended orbits from the orbital intersection are collected. Then, a reference pixel in the MODIS image at nadir is selected, such that the location corresponding to the reference pixel lies in the overlapping area of the VIIRS and MODIS observations. VIIRS data are then searched to find a pixel that is closest in distance to the MODIS reference pixel. Once the VIIRS reference pixel is found, a portion of the MODIS and VIIRS images are extracted, such that the reference pixels are located at the center in both extracted images. The extracted VIIRS and MODIS images have the center at the same geolocation. Intercomparison is performed within the maximum scan angle difference limit of 2° over the ocean and 6° over the desert.
For each MODIS pixel, a kernel with a size of 7 × 7 pixels is chosen in VIIRS data, such that the closest distance from the MODIS pixel is one of the VIIRS pixels in the kernel window. The closest-matching VIIRS pixel is chosen based on the smallest distance calculated from the MODIS reference pixel. The threshold for the smallest distance is 1 km. If no closest pixel (less than 1 km) is found in VIIRS kernel window, then the mapped data is filled with zero. This process is repeated for each MODIS pixel until the whole VIIRS data are mapped into MODIS data. Since the spatial resolution of VIIRS (0.75 km) is higher than MODIS (1 km), very few holes (i.e., zero values for unmatched pixels) are observed. The near-nadir region that is used in the intercomparison is mostly free of holes.
Extracting ROI
After mapping VIIRS data into MODIS, ROIs are extracted from both images. The criteria for choosing an ROI are (i) size: 9 km × 9 km, (ii) a MODIS and VIIRS cloud mask product to identify cloud-contaminated pixels, (iii) spatial uniformity less than 1% for the ocean surface and less than 2% for the desert, (iv) sensor zenith limit of ± 5° for ocean surface and ± 8° for desert, (v) upper and lower limits on reflectance to avoid saturated pixels and holes that are created after mapping the VIIRS image into MODIS, and (vi) solar zenith less than 80°. VIIRS and MODIS orbits stretch from the orbital intersection (SNO location) all the way down to desert. However, the ROIs are selected only over the desert region.
The upper limit of time difference for extended low-latitude SNO between VIIRS and MODIS is limited to within 16 min. Hence, the removal of cloud contamination and sun glints are major issues over the ocean.
After scanning through the MODIS and mapped VIIRS image, a number of ROIs are selected that match the above-mentioned criteria. These ROIs are selected from extended orbits of MODIS and VIIRS that are overlapped and collocated using the GSM mapping algorithm explained in step 2.
Constructing time series of observed bias
For one SNO-x event, there can be many collocated ROIs over the MODIS and VIIRS images. For each ROI of VIIRS and MODIS, the mean reflectance is calculated. Then, a bias is calculated for that ROI as follows: radiometric bias = (VIIRS − MODIS) × 100%/MODIS. If there are N ROIs for an SNO-x event, then there will be N bias values for that event. Finally, the mean and standard deviation of these N bias values are calculated. This gives one bias value (observed bias) for each SNO-x event along with the standard deviation. This process is repeated for all SNO-x events. Observed bias along with an error bar of ±1 standard deviation of every SNO-x event is used to construct a bias time series. Bias time series are analyzed for each of the VIIRS and MODIS matching bands.
The processing steps for computing theoretical (expected) bias are summarized below.
The hyperspectral measurements from Hyperion and AVIRIS are converted from spectral radiance to reflectance using Eq. (1). The spectral response over the given site can be convolved with the instruments’ relative spectral response (RSR) functions to account for the spectral differences and quantify the spectral and radiometric biases between the VIIRS and MODIS.
d. Computing expected and residual bias
Hyperion and AVIRIS have a spectral resolution of nearly 10 nm, which might not be sufficient to account for bias due to spectral differences in bands whose bandwidths are about 10 nm. Hence, MODTRAN-based analysis can complement the quantification of expected bias. The difference between the observed bias and the expected bias due to spectral differences gives the residual bias.
3. Results and discussion
a. Radiometric bias at the SNOs
The radiometric bias of VIIRS is analyzed by comparing its TOA reflectance measurements with MODIS over the ocean and the North African desert. Bias can exist primarily due to inaccuracy in calibration, BRDF, differences in RSR functions of matching bands used for comparison and spectral differences of targets used, residual cloud contamination, and cloud shadows. VIIRS radiometric and geometric calibrations were not stable early after launch, and S-NPP VIIRS went through a period of rapidly changing lookup tables (Cao et al. 2013). The day 26 February 2012 was marked as a golden day, ensuring the start of more stable calibration for S-NPP VIIRS. Thus, this study performs an intercomparison of VIIRS observations with MODIS using data generated after the end of February 2012. The result section follows with the analysis of VIIRS-observed radiometric bias for RSB over the ocean and desert. It is to be noted that MODIS uses ocean color bands (bands 12, 13, and 16) to compare with VIIRS (M-4, M-5, and M-7) over ocean, unlike the MODIS land bands (bands 4, 1, and 2) used to compare VIIRS (M-4, M-5, and M-7) over desert. For the remaining VIIRS bands, matching MODIS bands that are used for comparison over the ocean and desert are the same. The expected bias that explains the impact on observed radiometric bias due to RSR differences (between the matching bands of MODIS and VIIRS) and target spectral differences is investigated using hyperspectral measurements and a radiative transfer model. In addition, a stable vicarious calibration site, the Antarctica Dome C, is used for validation of the bias observed over the desert.
1) VIIRS-observed radiometric bias over ocean
Bias is calculated using TOA reflectance of VIIRS and MODIS over the ocean. The location of sensor intercomparison varies based on the location of occurrence of SNO events. Figure 2 shows the bias time series for VIIRS VNIR bands starting from late February 2012. Instruments are compared at the orbital intersection as well as the extended orbits with overlapped VIIRS and MODIS observations. After applying the strict filtering criteria for selecting valid ROI (as explained in the data processing section), it was found that more than 75% of low-latitude SNO events were excluded from analysis. In addition, after applying the filtering criteria for cloud screening and sun glint over the ocean, most of the SNO events that exist during April 2012 were excluded. This created a data gap in the bias time series that can be observed in the figure. The figure shows bias for most of the bands. Bias time series indicates that there is a drop in bias for some bands during April 2012. The drop in bias is considered to be due to a one-time update of solar diffuser stability monitor (SDSM) screen data, implemented in the beginning of April 2012. The update caused a drop in VIIRS SDR reflectance for M-1 through M-4 by more than 4%. There are not enough data points to confirm the exact bias trend during April. However, the time series becomes more characterizable after the start of May.
Radiometric bias time series of VIIRS bands M-1 through M-7 over the ocean. Each vertical bar represents ±1 standard deviation of bias for corresponding SNO event.
Citation: Journal of Atmospheric and Oceanic Technology 30, 12; 10.1175/JTECH-D-13-00071.1
VIIRS bias time series suggest that band M-1 has an increasing trend from the end of June 2012. However, the rest of the bands indicate a more stable bias time series. Band M-2 shows the most stable bias time series from early May to early November. However, it indicates an increasing trend thereafter. Bands M-3 and M-4 suggest a slightly increasing trend until early November and a sudden drop then after. A small drop in bias during early November can also be seen for channels M-5, M-6, and M-7. There are very few bias data points from November onward. This is because most of the VIIRS and MODIS simultaneous observations are either cloud or sun glint impacted. TOA reflectance measured above the ocean for bands M-5 through M-7 is below 5% and has fewer bias data points compared to bands M-1 to M-4. In addition, the variability in bias calculated for the majority of SNO events is larger for these bands, which are shown by the vertical bar of magnitude ±1 standard deviation for the data in Fig. 2. The bias time series for these bands also show larger variability over time. For bands M-6 and M-7, the typical reflectance measured over the ocean is less than 2%. Bias variability for each SNO event increases with the decrease in signal strength due to the decrease in the signal-to-noise ratio. VIIRS bands M-6 and M-7 over ocean have the noisiest bias time series, whereas VIIRS’s first three blue bands (M-1 to M-3) show the lowest variability in bias for each SNO event due to higher signal strength.
Bias time series are characterized by using linear fits for data from day 117 (start of May) because of (i) a major update in calibration (SDSM screen data) during April, (ii) the presence of a data gap of nearly one month in the bias time series during April, and (iii) the presence of only few bias points during the March–April period. Since bias trends are not constant, linear fits are used to characterize the time-dependent nature. Table 2 shows the linear fitting parameters of VIIRS radiometric bias time series for bands M-1 to M-7. The uncertainty in bias is less than 1% for bands M-1 through M-4. However, the uncertainty is greater than 1% for bands M-5 through M-7. In spite of the larger variability, bands M-6 and M-7 suggest a nearly constant trend in bias time series. Bias trends from early May to the end of December 2012 suggest that band M-1 has the largest change in bias of more than 4%, whereas bands M-3, M-4, M-6, and M-7 have a smaller change of less than 1.5% (Table 2). Figure 3 shows the plots of VIIRS versus MODIS reflectance measurements after applying linear fit parameters (Table 2) on MODIS reflectance. The figure includes all valid ROIs selected from each SNO event collected during the entire period of analysis. The plot shows (i) a reflectance range of each band of VIIRS and MODIS used in the intercomparison and (ii) the strong correlation coefficient with reduced variability, indicating how well the bias trends are characterized (Table 4). All VIIRS bands analyzed in this study (M-1 to M-7) are dual gain except band M-6. Since the TOA radiance observed over ocean for dual-gain bands are calibrated using high gain, the bias trends shown in Fig. 2 indicates the change in high gain over time. Each band needs to be well calibrated for both the low gain and high gain.
VIIRS-observed radiometric bias over ocean.
VIIRS vs MODIS reflectance after bias correction over the ocean using the linear model from Table 2.
Citation: Journal of Atmospheric and Oceanic Technology 30, 12; 10.1175/JTECH-D-13-00071.1
2) VIIRS-observed bias over the North African desert
VIIRS and MODIS measurements are compared at the overlapping region of extended SNO orbits. The SNO events with an orbital intersection occur at latitudes higher than 55°. Then, the orbits are extended to low latitudes, where the nadirs between VIIRS and MODIS are within 100 km. The data are compared over the North African desert. Band M-6 detectors are saturated for high-reflectance targets and hence cannot be analyzed for bias over the desert. Since the M-8 reflectance over the desert is high (greater than 50%), its bias can be quantified over the desert and thus is added in the analysis. Besides M-8, the remaining bands analyzed are all dual gain. Because of the higher signal strength over desert, the dual-gain bands M-4, M-5, and M-7 are calibrated using low gain, whereas M-1 through M-3 are still calibrated using high gain.
VIIRS bias time series for VNIR bands over the North African desert shown in Fig. 4 suggests that bands M-1 through M-4 have a decreasing trend in bias until the end of April. These bands indicate more than a 4% decrease in bias during the 2-month period (March and April). These decreasing trends in the VIIRS response are consistent with the bias trends observed over the ocean. As explained in the previous section, the cause is considered to be due to a one-time update of SDSM screen data. VIIRS bias analysis over the desert is performed for data starting from May 2013. Unlike the SNO events over the ocean, the extended SNO over the desert has more clear-sky data after November 2012 to be analyzed. Figure 4 shows that except for bands M-1 and M-8, the remaining VIIRS bands suggest a relatively stable radiometric bias time series from early May 2012 to February 2013. The bias time series are fitted using a linear model. The model fit parameters are shown in Table 3. Bands M-3 and M-4 suggest stable bias trends with the smallest change in bias of less than 0.01% over the period of analysis. M-1 suggest the largest change in bias, ranging from −3.44% in early May 2012 to 0.8% by the early February 2013. Except for M-8, which indicates a change of 1.9%, the remaining VIIRS bands suggest less than 1.4% change in bias from May 2012 to February 2013. Uncertainty in bias for each SNO event is indicated by a vertical bar of ±1 standard deviation of ROIs. Because of the larger signal strength over desert, the uncertainty in bias for each SNO event for bands M-5 and M-7 over the desert is less than 0.4%, which is about 5 times smaller compared to >2% over the ocean.
Radiometric bias time series of VIIRS bands M-1 to M-8 over the North African desert.
Citation: Journal of Atmospheric and Oceanic Technology 30, 12; 10.1175/JTECH-D-13-00071.1
VIIRS-observed radiometric bias over the desert.
Sudden changes in the bias time series seen in 2012 for bands such as M-1 during early July and mid-December are likely to be the indications for updates in calibration. Sudden jumps in bias can also be seen for bands M-2 and M-8 during October 2012 and for M-5 during early August 2012. Even though some of the bands are more unstable than others, the bias trends for all bands can be characterized by using simple mathematical models. The table shows that the uncertainty in bias trends is less than 1% for all VIIRS bands, suggesting that it is largest for M-1, which is 0.83%, while it is lowest for M-4, which is 0.27%. Figure 5 shows the plots of VIIRS reflectance versus MODIS reflectance after applying linear fitting parameters (Table 3) on MODIS reflectance. The plots show (i) the reflectance range of each band of VIIRS and MODIS used in the intercomparison and (ii) the strong correlation coefficient with reduced variability, indicating how well the biases are characterized (Table 4).
VIIRS vs MODIS reflectance after bias correction over the North African desert using the model from Table 3.
Citation: Journal of Atmospheric and Oceanic Technology 30, 12; 10.1175/JTECH-D-13-00071.1
Correlation coefficients and number of ROIs used in analysis.
b. Spectral characterization and expected spectral bias analysis
Expected spectral bias exists due to spectral response function differences between the instruments under comparison. When two instruments observe the same target at the same time under identical viewing geometry, there should not be any bias if both are well calibrated as long as the RSR functions of the bands are exactly the same. However, the MODIS and VIIRS matching bands have some differences in their RSR shape and spectral coverage, which result in bias even when they observe the same target at the same time under identical viewing conditions. Bias due to RSR differences of the two instruments can be simulated and quantified by using hyperspectral measurements of the target. In addition, radiative transfer models such as MODTRAN can also be used to extract the expected bias for a given site. The expected bias calculated over the desert and over the ocean is shown in Table 5. For the desert, four Hyperion observations over the North African desert were used to estimate the expected bias. Hyperion observations at different times of the year from 2005 to 2012 at different locations of the North African desert were used to evaluate the expected spectral bias. The variation in time and location of Hyperion observation can add uncertainty in the residual bias primarily due to the differences in local atmospheric variability. However, the surface radiometric and spectral properties remain nearly constant for the desert. As long as the instruments RSR do not include major atmospheric absorptions such as water vapor and the Hyperion band-to-band calibration inconsistency is insignificant, the expected bias should remain fairly constant.
Expected bias [(VIIRS − MODIS) × 100%/MODIS] over ocean and the North African desert.
For the Hyperion data, an ROI of size 3 km × 3 km is chosen. Spectral bias is calculated using every pixel in the selected Hyperion ROI. Then, a mean and standard deviation are calculated to estimate the mean spectral bias and its uncertainty due to the variation of Hyperion data. Spectral bias and its uncertainty (one standard deviation) based on the variation of Hyperion data at different locations of the North African desert are shown in Table 5 for bands M-2 to M-8. In addition, over the site Mauritania (one of the SNO-x intercomparison locations), spectral uniformity, which shows the variability of Hyperion spectra over the given ROI, is plotted in Fig. 6 (bottom). Temporal spectral variability of the desert sites has not been analyzed in this study. Assuming desert sites are invariant, the effect should be small. However, the impact needs to be quantified and investigated in the future. Table 5 shows that except for VIIRS band M-5, which has the largest spectral bias uncertainty of 0.83%, the remaining bands indicate the bias uncertainty to be less than 0.25%. This indicates that the spectral bias is very stable regardless of the location used in its estimation. This can also be observed from Eq. (1), where the effect due to the absolute magnitude of Hyperion reflectance is cancelled out during convolution. For desert, the expected bias using Hyperion ranges from 0.1% to 7.8% for bands M-1 through M-8. The largest bias exists for VIIRS band M-5, whereas the smallest bias exists for band M-3. VIIRS M-6 is saturated over desert and thus is not analyzed. In addition to Hyperion, MODTRAN4 (version 2) is also used to simulate the expected spectral bias over desert. This is because (i) the Hyperion spectral range does not cover the VIIRS M-1 band and (ii) the bias observed from the MODTRAN calculation can be compared to that from Hyperion to analyze for consistency. The atmospheric model used in the MODTRAN simulation is midlatitude summer with a rural extinction for aerosol. MODTRAN simulation–based biases agree with the bias computed using Hyperion measurements to within 1% except for the VIIRS band M-5, where the difference in bias is nearly 2%. For ocean, in addition to MODTRAN, the expected bias is calculated using AVIRIS data observed over the Gulf of Mexico. Bias for bands M-1 through M-7 ranges from −0.4% to 2.7% using AVIRIS. MODTRAN simulation–based bias computation agrees well within 1% with AVIRIS results except for bands M-4 and M-7, where the difference in bias is about 2%.
(top) RSR profiles of VIIRS and MODIS sensors along with the simulated TOA reflectance spectra for the desert (Hyperion), Dome C (Hyperion), and the ocean (MODTRAN). For M-5, M-6, and M-7, RSR functions for both MODIS land bands and MODIS ocean bands are included. The bandwidth for each band is given in Table 1. (bottom) Spectral uniformity of Hyperion over the North African desert and Dome C for ROI of size 3 km × 3 km.
Citation: Journal of Atmospheric and Oceanic Technology 30, 12; 10.1175/JTECH-D-13-00071.1
Figure 6 (top) shows the Hyperion-observed reflectance spectra over the North African desert (Mauritania) and the Antarctica Dome C site. The figure also shows that MODTRAN-simulated reflectance over the ocean. Reflectance spectrum of Dome C is included in the figure because this site is used for validating the radiometric bias, which is explained in more detail in section 3d. The figure shows that the Dome C site is spectrally flat over the VNIR bands. The reflectance spectra over the North African desert show monotonously increasing behavior for VNIR bands, whereas the reflectance spectra over the ocean suggest a decreasing trend with magnitudes less than 10% for most of the bands. However, the desert complements the Dome C site with more uniform and flat spectra at longer wavelengths. It is always desirable to have bright and stable targets with flat spectra, mainly because the site can be used to perform intercomparisons and validate the sensor performance with less registration error issues. In addition, the impact of spectral differences in quantifying bias is much reduced, thus greatly reducing the complexities and uncertainties in radiometric and spectral calibration and in validation of a sensor.
VIIRS band M-5 and matching MODIS band 1 have the largest differences in wavelength coverage. In addition, the monotonically increasing spectral response of desert for the VNIR region results in the maximum expected bias of about 8% for band M-5 over desert. Thus, the hyperspectral-measurements-based simulated bias depends on the spectral characteristics of the target and the differences in RSR functions of the two instruments. Uncertainties in expected bias can be due to a number of factors such as uncertainty in Hyperion and AVIRIS calibration, possibility of band-specific inconsistencies in instrument calibration, lower spectral resolution of Hyperion and AVIRIS, impact of the spectral characteristics of the target due to atmospheric variability and differences in observation time and location, inaccuracy in MODTRAN input parameters, and impact due to differences in solar models used to calculate Esun, which is used in calculating the expected bias.
c. Residual bias
Residual bias is the difference in observed bias and expected bias. Since impact due to spectral differences of the targets and mismatching RSR of bands of the instruments are already accounted for, in an ideal case, the residual bias should always be close to zero regardless of the target chosen for bias analysis. The magnitudes of residual bias indicate how well a sensor is calibrated in absolute scale and how large the calibration uncertainty is. However, the residual bias over the ocean and over the desert can vary from each other due to a number of factors. Except for bands M-6 and M-8, the remaining VIIRS bands analyzed in this paper are all dual-gain bands. The observations over the ocean are calibrated using high gain compared to low gain (except for bands M-1 through M-3) over the desert. Thus, bias can exist in either low gain or high gain if the gains are not characterized accurately. The TOA reflectance over the ocean for VIIRS bands M-5 and higher is less than 5%. The intercomparison of VIIRS and MODIS happens with more than 8 min of time interval. The movement of clouds and cloud-shadow during this time interval can affect the comparisons. The cloud-contaminated pixels can add large uncertainty if not detected and excluded from analysis. The cloud mask product of VIIRS and MODIS are not always capable of identifying cloud contamination at the subpixel level. This largely increases the uncertainty in bias for SNO events of M-5 and higher bands over the ocean due to a very small signal strength of less than 5%. This can result in larger variability, which can be observed in the bias time series over the ocean.
VIIRS-observed biases over the ocean and desert are not a constant function of time. It is compelling to see how well matched the shape of bias is as a function of time, such as for M-1 over the ocean and desert. However, the magnitudes of bias do not match. The difference in the bias magnitude is primarily due to the spectral differences in RSR of the VIIRS and MODIS matching bands and the difference in spectral characteristics of the ocean and desert. If the expected bias is characterized correctly and accounted for in the observed bias, then the bias time series for M-1 should match very well both in shape and magnitudes and their differences in temporal trend should be close to zero. The residual bias trends in Fig. 7 indicate three basic trends: increasing, decreasing, and nearly constant. After accounting for uncertainty, the biases in bands M-1 through M-7 are within 2% by early February 2013 except for band M-8, which indicate increasing bias trend from nearly 1% in May 2012 to 3% by early February 2013. Figure 7 also shows the residual bias trends over the ocean. Over the ocean, after accounting for bias uncertainty, VIIRS bands M-1 through M-7 agree with MODIS within 2% by the end of December 2012.
Residual bias trends (bias models from Tables 2 and 3 after subtracting the expected bias): (left) North African deserts and (right) ocean.
Citation: Journal of Atmospheric and Oceanic Technology 30, 12; 10.1175/JTECH-D-13-00071.1
The uncertainties in bias can be caused by BRDF, sun glint over the ocean, subpixel-level cloud contamination for the ocean and desert, cloud shadow, navigational errors, and calibration uncertainties in VIIRS. In addition, uncertainties in Hyperion, AVIRIS, and the MODTRAN simulation also contribute to residual bias. Hyperion and AVIRIS have 10-nm spectral resolutions, which may not be sufficient to accurately characterize the VIIRS and MODIS expected bias for bands with a spectral coverage of about 20 nm. In addition, band-to-band inconsistency in absolute calibration of Hyperion is always possible. The Hyperion measurements are from a different location of the desert and at different times of the day, which adds uncertainties due to changes in atmospheric variability. In addition, the estimation of expected spectral bias using instrument-based Esun values (calculated using different solar models) can always contribute to uncertainty larger than 2%.
VIIRS calibration is radiance based, whereas MODIS calibration is reflectance based. Thus, if VIIRS and MODIS radiance products are compared, the uncertainty in bias added due to use of different solar models in Esun calculation has to be quantified. Since radiance, reflectance, and solar zenith angle products can be extracted from L1B data for VIIRS and MODIS, Esun values can be derived by using Eq. (2). Table 6 shows the two Esun values: one derived from L1B products and the other calculated using the Thuillier (2003) solar model. The table shows that calculated Esun values for some bands can vary by more than 2% compared to derived values for both VIIRS and MODIS. Thus, bias can vary by more than 2% due to the use of different solar models for VIIRS and MODIS matching bands, if the comparison is performed using radiance products. However, impact due to the use of different solar models for VIIRS and MODIS is insignificant if the data are compared in reflectance units because the TOA reflectance product generation does not use Esun values. This study uses a reflectance-based comparison instead of a radiance-based comparison, and hence the uncertainty due to solar irradiance model should be insignificant in observed bias. Nevertheless, the Esun values need to be corrected in the operations to ensure the integrity of VIIRS calibration.
Solar spectral irradiance (W m−2 μm−1) for VIIRS moderate channels and MODIS bands.
d. VIIRS radiometric comparison and validation at Antarctica Dome C
The Antarctica Dome C site is spatially uniform (better than 1% spatial uniformity), and radiometrically stable. It is located at an elevation of 3200 m, and the atmospheric effect is minimal. The dominant atmospheric effects are Rayleigh scattering and ozone absorption. Cao et al. (2010) and Uprety and Cao (2012) have shown that the site offers stable radiometric and spectral characteristics for the calibration and validation of VNIR radiometers. The site has been recommended as a standard for calibration of VNIR radiometers (Cao et al. 2010). Dome C site can be used to quantify and validate the radiometric bias of VIIRS. However, the major limitations of the site are such that the BRDF effect is nearly 5% seasonally, it receives sunlight for only 4 months in a year, and it cannot be viewed by geostationary instruments. Figure 6 shows that Dome C is spectrally flat for VNIR bands, thus greatly reducing the complexities in radiometric and spectral calibration/validation (cal/val) of a sensor as well as the intercalibration between multiple sensors.
In this study, the Dome C site is used for validation of VIIRS bias observed over the North African desert. Uprety and Cao (2011) have shown that in spite of receiving light for only 4 months every year, the site can be used to quantify sensor degradation with reduced uncertainties. Since the dual-gain VNIR bands of VIIRS use low gain at Dome C site due to the high signal strength over snow, the site can serve as an excellent target for validating bias over the desert for the bands that are single gain or use low gain over the desert. The magnitude of bias estimated over the desert and Dome C for bands using low gain needs to be consistent. The comparison of VIIRS and MODIS at Dome C is performed using nadir view measurements. Studies in the past have shown that Antarctica snow albedo exhibits hysteresis; that is, the albedo in the morning and the afternoon are not same for the identical solar zenith angle (McGuffie and Henderson-Sellers 1985; Yamanouchi 1983). Although VIIRS and MODIS nadir view observation at Dome C occurs on different days of the austral summer, both instruments observe the Dome C site at almost the same local time (less than a 10-min difference). Thus, the impact on reflectance comparison due to hysteresis effect is insignificant. This further reduces the uncertainty in comparison. The reflectance calculated needs to be compared at similar solar zenith angles in order to reduce the uncertainty due to BRDF.
Figure 8 shows the reflectance time series of VIIRS and MODIS VNIR bands at Dome C. Radiometric bias can be clearly observed for some bands. VIIRS M-2 and M-3 bands are not shown since their equivalent MODIS bands are single gain and saturated at Dome C. The VIIRS bias time series over the desert has shown that the temporal trends of bias are not constants. Thus, VIIRS bias at Dome C needs to be compared to that over the desert over the same time period. The time period used to analyze bias at Dome C is from late October to early February 2013. Table 7 shows the bias for VIIRS bands at Dome C. The observed bias is more than 4% for bands M-5 and M-8. The large observed bias in M-5 can be explained primarily by the differences in spectral response functions of the VIIRS and MODIS equivalent bands. The expected bias at Dome C is calculated using MODTRAN for M-1 and Hyperion for the remaining bands. After taking into account the expected bias and uncertainty, the residual bias for all bands analyzed at Dome C agree with bias over the desert within 1%. The uncertainty is smallest for M-1 (0.63%) and largest for M-8 (2.28%). The residual biases at Dome C agree well with the residual bias over the desert. VIIRS bands M-1, M-4, M-5, and M-7 have a less-than-expected 2% radiometric bias at Dome C. VIIRS band M-8 suggests more than a 3% bias (consistent with the bias over the desert) with a larger uncertainty of more than 2%. The bias analysis results at Dome C are consistent with that over the desert for channels using low gain, which validates the SNO-x results over the desert.
Comparison of S-NPP VIIRS and Aqua MODIS at Antarctica Dome C. MODIS matching channels for VIIRS bands M-2 and M-3 are saturated and hence not shown. Solar zenith angles are less than 70°. Days after January 2013 in horizontal axis are added to 366 to make the reflectance time series continuous.
Citation: Journal of Atmospheric and Oceanic Technology 30, 12; 10.1175/JTECH-D-13-00071.1
Observed and expected [(VIIRS − MODIS) × 100%/MODIS], and residual (observed − expected) bias of VIIRS at Dome C.
4. Concluding remarks
The SNO-x technique was used to quantify and characterize the radiometric bias of VIIRS using MODIS as a reference. A fast geospatial mapping algorithm was developed to map VIIRS data into the MODIS geolocation grid for intercomparison purposes. VIIRS bias relative to MODIS for moderate reflective solar bands (M-1 through M-8) has been quantified. Ocean-based SNO-x analysis served for high-gain bias analysis for all bands, whereas the intercomparison over the desert served for low-gain analysis for bands M-4 and higher. The observed bias varies for each target primarily due to spectral differences of the target and difference in RSR between the VIIRS and MODIS bands. The impact of the spectral differences in observed bias is quantified using hyperspectral measurements from EO-1 Hyperion and AVIRIS and by using a radiative transfer model. After accounting for the impact due to spectral differences and bias uncertainties, the residual bias computed over the ocean and desert shows that VIIRS VNIR bands agree with MODIS within 2% with bias uncertainty less than 1%. VIIRS SWIR band M-8 analyzed over the desert indicates a nearly 3% bias with the uncertainty of 0.4%. In addition, bias for some of the VIIRS bands was validated using MODIS and VIIRS observations at the Antarctica Dome C site. The bias magnitudes and trends can vary for some bands over the desert and ocean mainly because of the dual-gain nature of VIIRS bands, the use of different MODIS bands for comparison over the ocean and desert (for VIIRS bands M-4, M-5, and M-7), and the associated uncertainties in the estimation of expected spectral bias.
The study suggests that the VIIRS bias trends are not a constant and vary with time. This is partly because VIIRS has been undergoing frequent lookup table changes during the first year after launch, and the calibration impact due to mirror degradation has not been completely eliminated in the SDR data products. A continuous effort is needed to regularly monitor VIIRS performance to keep the calibration well within specification and to reduce the calibration uncertainties. This paper shows that SNO-x is a potential approach for continuous on-orbit calibration/validation of VIIRS. The current study focuses on VIIRS RSB radiometric bands. In the future, VIIRS imagery bands and thermal emissive bands will also be analyzed using the SNO-x technique to quantify intersatellite bias.
Acknowledgments
The authors thank the VIIRS SDR team members for their dedicated support in calibration/validation of VIIRS SDR data. The authors thank Ms. Yan Bai for her support in data collection and editorial work. This work is partially funded by the JPSS program office. The manuscript contents are solely the opinions of the authors and do not constitute a statement of policy, decision, or position on behalf of NOAA or the U.S. government.
REFERENCES
Bossler, J. D., Campbell J. B. , Mcmaster R. B. , and Rizos C. , 2010: Coordinates and coordinate systems. Manual of Geospatial Science and Technology, 2nd ed. J. D. Bossler et al., Eds., CRC Press, 9–16.
Cao, C., and Heidinger A. K. , 2002: Inter-comparison of the longwave infrared channels of MODIS and AVHRR/NOAA-16 using simultaneous nadir observations at orbit intersections. Earth Observing Systems VII, W. L. Barnes, Ed., International Society for Optical Engineering (SPIE Proceedings, Vol. 4814), 306, doi:10.1117/12.451690.
Cao, C., Weinreb M. , and Xu H. , 2004: Predicting simultaneous nadir overpasses among polar-orbiting meteorological satellites for the intersatellite calibration of radiometers. J. Atmos. Oceanic Technol., 21, 537–542.
Cao, C., Xu H. , Sullivan J. , McMillin L. , Ciren P. , and Hou Y. , 2005: Intersatellite radiance biases for the High Resolution Infrared Radiation Sounders (HIRS) on board NOAA-15, -16, and -17 from simultaneous nadir observations. J. Atmos. Oceanic Technol., 22, 381–395.
Cao, C., Xiong X. , Wu A. , and Wu X. , 2008: Assessing the consistency of AVHRR and MODIS L1B reflectance for generating fundamental climate data records. J. Geophys. Res., 113, D09114, doi:10.1029/2007JD009363.
Cao, C., Uprety S. , Xiong J. , Wu A. , Jing P. , Smith D. , Chander G. , Fox N. , and Ungar S. , 2010: Establishing the Antarctic Dome C community reference standard site towards consistent measurements from Earth observation satellites. Can. J. Remote Sens., 36, 498–513, doi:10.5589/m10-075.
Cao, C., Deluccia F. , Xiong X. , Wolfe R. , and Weng F. , 2013: Early on-orbit performance of the Visible Infrared Imaging Radiometer Suite onboard the Suomi National Polar-Orbiting Partnership (S-NPP) satellite. IEEE Trans. Geosci. Remote Sens., doi:10.1109/TGRS.2013.2247768, in press.
Chander, G., Helder D. L. , Markham B. L. , Dewald J. D. , Kaita E. , Thome K. J. , Micijevic E. , and Ruggles T. A. , 2004: Landsat-5 TM reflective-band absolute radiometric calibration. IEEE Trans. Geosci. Remote Sens.,42, 2747–2760.
De Luccia, F., and Coauthors, 2012: Discovery and characterization of on-orbit degradation of the Visible Infrared Imaging Radiometer Suite (VIIRS) rotating telescope assembly (RTA). Earth Observing Systems XVII, J. J. Butler, X. Xiong, and X. Gu, Eds., International Society for Optical Engineering (SPIE Proceedings, Vol. 8510), 40, doi:10.1117/12.930544.
Folkman, M. A., Pearlman J. , Liao L. B. , and Jarecke P. J. , 2001: EO-1/Hyperion hyperspectral imager design, development, characterization, and calibration. Hyperspectral Remote Sensing of the Land and Atmosphere, W. L. Smith and Y. Yasuoka, Eds., International Society for Optical Engineering (SPIE Proceedings, Vol. 4151), 40, doi:10.1117/12.417022.
Guenther, B., Xiong X. , Salomonson V. V. , Barnes W. L. , and Young J. , 2002: On-orbit performance of the Earth Observing System Moderate Resolution Imaging spectroradiometer; first year of data. Remote Sens. Environ., 83, 16–30, doi:10.1016/S0034-4257(02)00097-4.
Heidinger, A. K., Cao C. , and Sullivan J. T. , 2002: Using Moderate Resolution Imaging Spectrometer (MODIS) to calibrate advanced very high resolution radiometer reflectance channels. J. Geophys. Res., 107, 4702, doi:10.1029/2001JD002035.
Heidinger, A. K., Straka W. C. III, Molling C. C. , Sullivan J. T. , and Wu X. , 2010: Deriving an inter-sensor consistent calibration for the AVHRR solar reflectance data record. Int. J. Remote Sens., 31, 6493–6517.
Hutchison, K. D., and Cracknell A. P. , 2005: Visible Infrared Imager Radiometer Suite: A New Operational Cloud Imager. CRC Press, 256 pp.
McGuffie, K., and Henderson-Sellers A. , 1985: The diurnal hysteresis of snow albedo. J. Glaciol., 31, 188–189.
Pearlman, J. S., Barry P. S. , Segal C. C. , Shepanski J. , Beiso D. , and Carman S. L. , 2003: Hyperion, a space-based imaging spectrometer. IEEE Trans. Geosci. Remote Sens., 41, 1160–1173.
Rao, C. R. N., and Chen J. , 1993: Calibration of the visible and near-infrared channels of the Advanced Very High Resolution Radiometer (AVHRR) after launch. Recent Advances in Sensors, Radiometric Calibration, and Processing of Remotely Sensed Data, P. S. Chavez Jr. and R. A. Schowengerdt, Eds., International Society for Optical Engineering (SPIE Proceedings, Vol. 1938), 56, doi:10.1117/12.161571.
Smith, D. L., Read P. D. , and Mutlow C. T. , 1997: Calibration of the visible/near-infrared channels of the Along-Track Scanning Radiometer-2 (ATSR-2). Sensors, Systems, and Next-Generation Satellites, H. Fujisada, Ed., International Society for Optical Engineering (SPIE Proceedings, Vol. 3221), 53–62.
Staylor, W. F, 1990: Degradation rates of the AVHRR visible channel for the NOAA 6, 7, and 9 spacecraft. J. Atmos. Oceanic Technol.,7, 411–423.
Teillet, P. M., Fedosejevs G. , and Thome K. J. , 2004: Spectral band difference effects on radiometric cross-calibration between multiple satellite sensors in the Landsat solar-reflective spectral domain. Sensors, Systems, and Next-Generation Satellites VIII, R. Meynart, S. P. Neeck, and H. Shimoda, Eds., International Society for Optical Engineering (SPIE Proceedings, Vol. 5570), 307, doi:10.1117/12.562709.
Thuillier, G., Hersé M. , Labs D. , Foujols T. , Peetermans W. , Gillotay D. , Simon P. C. , and Mandel H. , 2003: The Solar Spectral Irradiance from 200 to 2400 nm as Measured by the SOLSPEC spectrometer from the Atlas and Eureca missions. Solar Physics, 214, 1–22, doi:10.1023/A:1024048429145.
Uprety, S., and Cao C. , 2011: Using the Dome C site to characterize AVHRR near-infrared channel for consistent radiometric calibration. Earth Observing Systems XVI, J. J. Butler, X. Xiong. and X. Gu, Eds., International Society for Optical Engineering (SPIE Proceedings, Vol. 8153), 81531Y, doi:10.1117/12.892481.
Uprety, S., and Cao C. , 2012: Radiometric and spectral characterization and comparison of the Antarctic Dome C and Sonoran Desert sites for the calibration and validation of visible and near-infrared radiometers. J. Appl. Remote Sens., 6, 063541, doi:10.1117/1.JRS.6.063541.
Wu, A., Cao C. , and Xiong X. , 2006: Using MODIS to track calibration stability of the AVHRR on NOAA 15-18. Remote Sensing and Modeling of Ecosystems for Sustainability III, W. Gao and S. L. Ustin, Eds., International Society for Optical Engineering (SPIE Proceedings, Vol. 6298), 629812, doi:10.1117/12.681199.
Wu, A., Xiong X. , Cao C. , and Angal A. , 2008: Monitoring MODIS calibration stability of visible and near-IR bands from observed top-of-atmosphere BRDF-normalized reflectances over Libyan desert and Antarctic surfaces. Earth Observing Systems XIII, J. J. Butler and J. Xiong, Eds., International Society for Optical Engineering (SPIE Proceedings, Vol. 7081), 708113, doi:10.1117/12.795296.
Wu, X., and Sun F. , 2005: Post-launch calibration of GOES Imager visible channel using MODIS. Earth Observing Systems X, J. J. Butler, Ed., International Society for Optical Engineering (SPIE Proceedings, Vol. 5882), 58820N, doi:10.1117/12.615401.
Xiong, X., Esposito J. A. , Sun J.-Q. , Pan C. , Guenther B. W. , and Barnes W. L. , 2001: Degradation of MODIS optics and its reflective solar bands calibration. Sensors, Systems, and Next-Generation Satellites V, H. Fujisada, J. B. Lurie, and K. Weber, Eds., International Society for Optical Engineering (SPIE Proceedings, Vol. 4540), 62, doi:10.1117/12.450646.
Xiong, X., Sun J. , Barnes W. , Salomonson V. , Esposito J. , Erives H. , and Guenther B. , 2007: Multiyear on-orbit calibration and performance of Terra MODIS reflective solar bands. IEEE Trans. Geosci. Remote Sens., 45, 879–889.
Xiong, X., Sun J. , Xie X. , Barnes W. L. , and Saloruonsond V. V. , 2010: On-orbit calibration and performance of Aqua MODIS reflective solar bands. IEEE Trans. Geosci. Remote Sens., 48, 535–546.
Yamanouchi, T., 1983: Variations of incident solar flux and snow albedo on the solar zenith angle and cloud cover, at Mizuho Station, Antarctica. J. Meteor. Soc. Japan, 61, 879–893.
