Correlation between SCIAMACHY, MODIS, and CERES reflectance measurements: Implications for CLARREO



[1] We have analyzed the correlation between measurements by three different satellite sensors (SCIAMACHY, MODIS, and CERES) on two independent space platforms (Envisat and Terra). Though the instantaneous measurements from the two satellites are not collocated due to orbit offset, the monthly mean broadband and narrowband reflectances and their anomalies from the three instruments are highly correlated when averaged over large latitude regions. The mean reflectance from MODIS in each of those large domains is nearly the same as that derived from SCIAMACHY spectrum convolved with the filter function of the corresponding MODIS channel, with all correlation coefficients higher than 0.93. The interannual variability of monthly mean reflectance is also correlated with the variations of mean cloud and surface properties in large climate zones. The reflectance variation is correlated with the cloud fraction in low and middle latitude regions with correlation coefficients higher than 0.76 and with the snow and sea ice changes in the polar regions with correlation coefficients higher than 0.6. These correlations indicate that the nadir sampling strategy as proposed for CLARREO is sufficient for climate benchmarking of the reflected solar spectrum and provide the physical foundation for climate fingerprinting. However, the results also show the relatively large differences in trends in reflectance due to different instrument degradations and inconsistent calibrations which will affect the attribution of radiative signals of long-term climate change.

1. Introduction

[2] The radiation reflected and emitted from the Earth depends on the components of the Earth-atmosphere system. The backscattering solar spectrum is particularly sensitive to the variations of clouds and surface properties. Therefore the long-term radiative trend reflects the Earth's climate change. Since climate change signals are usually small, few to several decades may be required for sound climate change detection. There is currently no single climate mission or satellite sensor that can provide well calibrated and stable data records over such a long time; most of them cannot even normally function a decade. Therefore it is necessary to concatenate and analyze data from various sources for this endeavor. A wealth of satellite observation data of spectral and broadband solar radiances is available today. Since these data are obtained from different platforms and different sensors with different calibrations, the data consistency has been an issue for applying these heterogeneous data to climate studies [e.g., Ohring et al., 2005], which require long-term stable data records.

[3] To address this problem, the U.S. National Research Council has called for the implementation of a satellite mission, Climate Absolute Radiance and Refractory Observatory (CLARREO) [National Research Council, 2007]. This mission will make highly stable and accurate measurements of shortwave and longwave radiance spectra traceable to international standards. Complementary to CLARREO, a small satellite mission called Traceable Radiometry Underpinning Terrestrial and Helio-Studies (TRUTHS) has been proposed to the European Space Agency to measure the Earth's reflected solar spectrum [Fox et al., 2011]. In addition to serving as a reference standard for intercalibration of other satellite sensors, CLARREO and TRUTHS will provide a climate benchmark with their nadir measurements. One potential application of the solar benchmark spectra is the so-called climate fingerprinting, which attributes the radiative signals to individual climate parameters based on the spectral characteristic of each component and therefore detects climate change [Hasselmann, 1997; Leroy et al., 2008]. The solar fingerprinting requires not only long-term consistent data records but also the knowledge on how radiation responds to individual climate parameter changes, both regionally and globally. To evaluate the feasibility of using CLARREO solar benchmark to fingerprint climate changes and to define the relevant instrument requirements for that, we need to synthesize and study the current data records from different satellite sensors. As an initial step to this endeavor, this study analyzes the correlations and differences among observations from the Scanning Imaging Absorption Spectrometer for Atmospheric Chartography (SCIAMACHY) [Bovensmann et al., 1999], the Moderate Resolution Imaging Spectrometer (MODIS) [Justice et al., 1998], and the Clouds and the Earth's Radiant Energy System (CERES) [Wielicki et al., 1996].

2. Correlation of Reflected Solar Radiation Between SCIAMACHY, MODIS, and CERES

[4] The SCIAMACHY spectrometer is on board the Envisat space platform. Envisat is in a Sun-synchronous orbit with inclination of 98.55° and equator crossing time of 10:00 local time [Gottwald and Bovensmann, 2011]. SCIAMACHY measures the Earth's reflected spectral radiance from 240 to 1750 nm in its spectral bands 1–6 with spectral resolution from 0.2 to 1.5 nm. It also measures the spectral solar irradiance. In this study we use SCIAMACHY spectral radiance Level-1B SCI_NL_1P data with the version 7.03 of calibration and the latest available version of correction for instrument optics degradation [Bramstedt et al., 2009]. The spectral radiance and daily solar irradiance (D0) measured by SCIAMACHY are used to derive the reflectance spectra from 250 to 1750 nm with uniform spectral sampling of 1.0 nm. Corresponding to CLARREO's benchmark measurements at nadir view, the average spectral reflectance of the two near-nadir footprints of SCIAMACHY is used. The spatial coverage of each footprint on the ground is about 30 km × 240 km at this range of viewing angles.

[5] NASA's Terra satellite is also in a sun-synchronous orbit with inclination of 98.2° and equator crossing time of 10:30 local time, so the average time difference of the overpass between the SCIAMACHY on Envisat and the MODIS/CERES package on Terra is about 30 min at the same latitude. However, this does not mean that the measurements from the two platforms are nearly coincidental nor does it mean that they are collocated; on the contrary, neither their instantaneous data nor their averages over small time/space scales are similar. In accordance with the SCIAMACHY nadir views, MODIS and CERES radiances within 10° of nadir are used in this study. MODIS measures radiances in separate spectral bands, while CERES measures broadband radiance. We use the CERES Single Scanner Footprint (SSF) data product (, which includes the shortwave radiance and the collocated MODIS data in channel 1 (645 nm) and channel 2 (858 nm) in the spectrum of SCIAMACHY. Cloud properties retrieved from MODIS measurements [Minnis et al., 2006] and snow and sea ice coverage [Maslanik and Stroeve, 1999] for each CERES footprint are also included in the SSF product.

[6] Figure 1 compares the time series of monthly mean nadir reflectances between SCIAMACHY on Envisat and MODIS/CERES on Terra in the five latitude regions and the globe. The five regions include the north polar (NP, 60°N–90°N), the south polar (SP, 60°S–90°S), the northern midlatitude (NML, 30°N–60°N), the southern midlatitude (SML, 30°S–60°S), and the tropical (TRO, 30°S–30°N). All the regional averages are obtained through two steps: time/space averaging in each 5 × 5 degree of latitude-longitude grid box first and then averaging in each of the latitude regions defined above. The mean reflectance in each region, math formula, is calculated as

display math

where math formula and math formula0 represent the monthly mean radiance and mean solar incident flux, respectively. Each row of Figure 1 is for a different region and each column is for a different spectral band. The horizontal axes are for the month number, starting from January 2003. Column 1 (left) compares the reflectance from MODIS channel 1 (red) and that from SCIAMACHY spectrum convolved with the filter function of MODIS channel 1 (available at Similarly, column 2 (middle) compares the reflectances of MODIS and SCIAMACHY in the spectral band of MODIS channel 2. Column 3 (right) compares the CERES shortwave reflectance and the spectrally integrated SCIAMACHY broadband reflectance. The black plot symbols are for SCIAMACHY, the red symbols are for MODIS or CERES, and the blue symbols are for the difference between the two (red-black). For easy comparison, the difference is plotted in the same scale as the reflectance but offset by a number indicated by the horizontal black line. Though the instantaneous observations at a common location from the two platforms are not coincidental, the comparison in Figure 1 shows that the reflectances measured from the two independent orbits have the same seasonal variability and are nearly identical when averaged to those large latitude zones of great area. The correlation coefficients between the two sets of reflectance in each region are listed in the upper right corners, and they are a measure of the strength of the linear relationship between the two variables (1.0 indicates perfect correlation). The results show that the two sets of reflectance are highly correlated with each other in all regions with all correlation coefficients higher than 0.93. The nadir reflectance instead of radiance is compared because the former is much less sensitive to solar zenith angle (SZA) change and has less spectral and seasonal variations [Jin et al., 2011]. The mean SZA difference between the two measurements is generally less than 5° in all regions, with a maximum in the tropics and a minimum in the polar regions. This SZA difference has little effect on the zonal mean reflectance at nadir. When the SCIAMACHY reflectance in a MODIS channel is calculated, both the spectral radiance and irradiance are convolved with the filter function of MODIS. Since the SCIAMACHY reflectance in a MODIS channel has the same spectral coverage as the MODIS channel, the difference in reflectance between them is resulted from their different sampling and instrument calibration. While the CERES shortwave covers nearly the entire solar spectra, SCIAMACHY does not have measurements beyond 1.75 μm, in which there is about 8% of total solar incident energy and 2–3% of reflected energy on average [Lukachin et al., 2009]. To make the two broadband reflectances comparable, the same solar constant (E0 = 1365 W/m2) applied to the calculation of CERES reflectance is also used to the SCIAMACHY broadband. Through this normalization, the reflectance difference between SCIAMACHY broadband and CERES is mainly from the fact that the SCIAMACHY band pass is less than that of CERES. Therefore the seemingly good agreement between the two (shown in the column 3 of Figure 1) in fact indicates that the absolute calibration is not consistent between the two instruments (i.e., either SCIAMACHY is too high or CERES is too low), but the absolute calibration of either is beyond the scope of this study and is discussed in several other papers [e.g., Lichtenberg et al., 2006; Priestley et al., 2010]. However, the correlation between the two broadband channels remains high in all regions, indicating the 2–3% of calibration difference is systematic.

Figure 1.

Comparison of monthly mean nadir reflectance between SCIAMACHY and MODIS/CERES in five latitude regions and globe. Each row is for a different region and each column is for a different channel (from left to right: MODIS channel 1 (645 nm), channel 2 (858 nm), and CERES shortwave). The black symbol is for SCIAMACHY, the red is for MODIS or CERES, and the blue is for their difference (red-black) plotted with offset by the horizontal line.

[7] Unlike absolute reflectance, the interannual variability of reflectance is not sensitive to systematic measurement errors such as the instrument calibration bias seen above and is thus particularly useful for our study. Figure 2 compares the deseasonalized anomalies of monthly mean nadir reflectance between SCIAMCHY, MODIS, and CERES. The deseasonalized reflectance anomaly for a month is calculated as the difference between the monthly mean and the average of the same month across all years. So the anomaly represents the interannual variability. Four reflectance anomaly time series in broadband and MODIS channel 1 are compared. The anomaly for MODIS channel 2 is very similar to channel 1 and is not shown, but it is included and presented separately in the Figure 3 below. Though much smaller than the mean reflectance, the reflectance anomalies from different spectral bands and different sensors are consistent in each region. The CERES albedo anomaly from full swath data is also shown in Figure 2. Albedo at the top of atmosphere (TOA) can be considered as the angularly weighted mean reflectance over all view angles in the upper hemisphere. Its anomaly is also consistent with those for nadir reflectance, indicating that the climate benchmark of mean nadir radiance can represent that of radiative flux which cannot be measured directly from space but converted from radiance. These facts assure us that the nadir sampling error is negligible and the nadir sampling strategy as proposed by CLARREO for climate benchmark is sound, consistent with the quantitative analysis on the nadir sampling error [Doelling et al., 2011; Jin et al., 2009]. The comparability of reflectance anomalies among different spectral bands and different sensors also demonstrates the advantage of using reflectance over radiance.

Figure 2.

Deseasonalized anomalies of monthly mean reflectance for SCIAMACHY broadband (black), CERES shortwave channel (red), MODIS channel 1 (blue), and SCIAMACHY in MODIS 1 (green), for different regions. The anomaly of full swath CERES albedo is plotted in magenta.

Figure 3.

Comparison of monthly mean reflectance anomalies between SCIAMACHY and MODIS/CERES in different regions as in Figure 1. The two lines are the linear regression of the two anomalies (black for SCIAMACHY and red for MODIS/CERES). The line slopes (in units of reflectance change per decade) are shown in the top middle.

[8] We further analyzed the anomalies separately for each pair of sensors in each band to study the correlation and difference between them in Figure 3. Same to Figure 1, each row is for a different region and each column is for a different spectral band. The numbers shown in the upper right corners are the correlation coefficient between the two sets of data in the region. The solid lines are the linear regression of data (black for SCIAMACHY and red for MODIS or CERES). The two line slopes (in unit of reflectance change per decade) in each region, together with the 95% confidence ranges, are shown in the top middle and they represent the trend of reflectance change. Although the correlation between the two instruments is high, their trends in reflectance are quite different. However, the trend difference is consistent among all regions, for example, the tread from CERES or MODIS is constantly lower than that from SCIAMACHY. Therefore most of the trend difference in each region is likely due to the different instrument degradation or calibration error instead of the different time/space sampling; the sampling error is random if it exists. It should be noted that the per decade trend change is small compared with the natural variability and the slope derived from data of one decade or shorter may not represent the real climate change trend. Compared to the trends, the 95% confidence ranges are also large. However, quantifying the climate change trends and detection times is not the subject of this study and it is studied elsewhere [e.g., Feldman et al., 2011]. Our objective is to understand the difference and inconsistence of the slopes from different data sets, as well as their influence on the correlation discussed, which have implications for the application of solar fingerprinting to CLARREO-like measurements. The standard deviation (1σ) of the deseasonalized reflectance anomaly represents the natural variability. Table 1 shows the standard deviation for different sensors in different bands and different regions. Comparing the numbers row by row, we see that all instruments in all bands show largest variability in the polar regions, then decreases in order from the north middle latitude (NML), to the south middle latitude (SML), to the tropics and is smallest for the global average. The natural variability in NML is much higher than that in SML because of larger portion of land in NML where the surface albedo is more dynamic than the ocean surface dominated in SML. Comparing the σ between the two platforms in a same band (the two columns in each box), we see that the natural variability measured from the two satellites in a same channel are similar and consistent in each region, for example, 0.0026 versus 0.0027 in MODIS channel 1 over globe. The consistent variability between the two independent orbits again indicates a small nadir sampling error for both; their different slopes but consistent slope difference across all regions indicate the different instrument degradation or calibration error. The positive reflectance slope in the south polar shown by all sensors is consistent with the reported sea ice increase there in the time period shown [Turner et al., 2009]. Similarly, the negative trend in the north polar is consistent with the Arctic ice decline in recent years [Kay et al., 2011].

Table 1. Standard Deviation (1σ) of Monthly Mean Reflectance Anomalies Measured by SCIAMACHY, MODIS, and CERES on Two Independent Platforms in Three Spectral Channels and Six Latitude Regions
RegionBroadbandChannel 1Channel 2

3. Correlation of Interannual Variability Between the Reflectance and the Earth-Atmosphere Components

[9] While it is well known that the solar reflectance varies with the Earth-atmosphere components, it is not clear yet how the averaged radiative change in large climate domains relates to the domain-averaged variation of the atmospheric, cloud, and surface properties individually. Understanding the correlation between the radiative responses and the climate parameter changes in the Earth-atmosphere system, both regionally and globally, is a prerequisite for the application of solar fingerprinting technique to climate change detection. Figure 4 shows the correlation between the monthly mean reflectance anomaly and the most relevant cloud and surface property anomalies in all the regions. All these anomalies are deseasonalized. The cloud properties are retrieved from the full swath of MODIS observation data [Minnis et al., 2006]. In the example presented in Figure 4, the black lines are the SCIAMACHY broadband reflectance anomalies. The red and green lines are the associated anomalies for cloud fraction and optical depth, respectively. The blue line shows the anomaly for snow and sea ice coverage. For visual clarity, only the parameters with correlation higher than 0.5 with the reflectance variation are plotted, but all the three correlation coefficients for cloud fraction (red), optical depth (green), and snow and sea ice coverage (blue) are listed for each region. The blue dotted lines indicate the offset or zero value of the anomalies of cloud and snow/ice properties. These parameter anomalies are also rescaled for clearer illustration; this linear scaling does not affect the correlation number. The snow and sea ice anomaly has the highest correlation with reflectance anomaly in the polar regions and therefore the variation of snow and sea ice instead of clouds is mainly responsible for the interannual reflectance change in polar. However, the cloud fraction has the highest correlation, followed by the cloud optical depth, with the mean reflectance variability in lower latitude regions and globe. The anticorrelation between reflectance and cloud fraction in the polar south is probably due to the high surface albedo over the Antarctic and the cloud shadow effects on reflectance. The results here are consistent with those derived from the solar spectral kernels [Jin et al., 2011], an independent approach.

Figure 4.

Deseasonalized monthly anomalies of reflectance (black), cloud fraction (FC, in red), optical depth (τ, in green), and snow and sea ice fraction (FS, in blue). The correlation between the reflectance variation and the cloud/surface properties are shown in the upper right corner. Only parameters having correlation with reflectance higher than 0.5 are plotted.

[10] Because the reflectance anomalies in all channels are similar (see Figure 2), the correlations between and the physical properties and the reflectance in other channels shown above are similar to that presented in Figure 4 and are not shown. We summarize the correlation coefficients in Table 2, which shows that the correlation of reflectance variability and parameter changes in each region are consistent among different bands. For example, all have similar and negative correlation between reflectance and cloud fraction in the polar south and have similar and highest positive correlation between them in all lower-latitude regions. As expected, the TOA albedo has higher correlation with cloud fraction than the nadir reflectance in all regions. The reflectance in each MODIS channel is from SCIAMACHY measurements. The MODIS measured reflectance itself is not independent of the cloud properties that are retrieved from MODIS data.

Table 2. Correlation Coefficients of Anomalies Between the Reflectance/Albedo and the Three Parameters in Different Latitude Regions
  • a

    FC is cloud fraction, τ is cloud optical depth, and FSI is snow and sea ice fraction.


4. Discussion and Conclusion

[11] When averaged over large spatial and temporal scales, the time series of nadir reflectances in a common spectral band from the two independent Sun-synchronous orbits are nearly identical and highly correlated, though their instantaneous measurements are not collocated and thus not comparable. The interannual variability of monthly mean reflectance in large latitude zones, though less than 5% of the mean reflectance, are also consistent and well correlated between the two independent observations. These facts indicate that the satellite measurements in an angular range of only several degrees of nadir could provide sufficient sampling for climate benchmarking over latitude zones of great area.

[12] On the other hand, since the trend (slope) of mean reflectance change in a large climate domain is small, the relative difference in slope from different data sets could be significant due to different instrument degradation. The influence of this calibration-related slope difference on the accuracy of solar fingerprinting ought to be evaluated and accounted in future fingerprinting applications to real observational data. If such calibration error is not corrected, the derived long-term climate trend from the data will be biased. Comparing the reflectance change slopes and the natural variability values from the three sensors (SCIAMACHY, MODIS, and CERES), the time required for the climate change signal to emerge from the natural variability of reflectance could vary from one to several decades, depending on the data sets and region. Since there is currently no satellite sensor can provide stable data record for such a long period, synthesizing a climate record from heterogeneous satellite data is necessary. This requires a common and traceable radiation standard in space for calibration of various satellite instruments for data consistency.

[13] The interannual variability of monthly mean reflectance in climate zones of great area is found to correlate well with the corresponding variation of the Earth-atmosphere components. This correlation provides the physical foundation for attributing the radiative signals of climate change to different parameters of the system. The correlation analysis shows that the reflectance variation is mainly related to the snow and sea ice changes in the polar regions and to the cloud fraction and optical depth changes in low latitudes. The correlation results indicate that it is feasible to fingerprint climate changes of cloud and snow/ice properties based on well calibrated solar benchmark spectra, such as the proposed measurement data directly from, or calibrated by, CLARREO and TRUTHS.


[14] We acknowledge the European Space Agency for providing the SCIAMACHY solar radiance data and all people who helped with SCIAMACHY data processing and the NASA Langley Atmospheric Science Data Center for the CERES/MODIS data. We appreciate the valuable comments and suggestions by Bruce Wielicki and David Young at NASA Langley Center and their support for this study.