1. Introduction
Deep convective clouds (DCCs) are extremely cold clouds that start from the planetary boundary layer and ascend to the tropical tropopause transition layer. The absorptions due to water vapor and other gases over DCCs are minimal. DCCs are bright targets and have nearly Lambertian reflectance (Doelling et al. 2004; Hu et al. 2004; Aumann et al. 2007; Fougnie and Bach 2009; Doelling et al. 2013). They can be identified using a single longwave infrared (LWIR) channel centered at ~11-μm brightness temperature (TB11) (Hu et al. 2004; Doelling et al. 2013). DCCs have been widely used as bright calibration targets for postlaunch calibration and stability monitoring of radiometers in the visible (VIS) and near-infrared (NIR) spectrums for a variety of satellite instruments in the past decade. Hu et al. (2004), in his pioneer study, demonstrated that DCCs have a constant mean albedo over the lifetime of the Clouds and the Earth’s Radiant Energy System (CERES) on board the Tropical Rainfall Measuring Mission (TRMM) satellite. Doelling et al. (2004) applied a similar DCC technique outlined by Hu et al. (2004) to calibrate the 0.65-μm VIS and 0.86-μm NIR bands of the Advanced Very High Resolution Radiometer (AVHRR) flown on the National Oceanic and Atmospheric Administration-16 and -17 (NOAA-16 and NOAA-17) satellites. Aumann et al. (2007) applied the DCC technique to the Atmospheric Infrared Sounder (AIRS) data and found that the stability uncertainty of the first 4 years of AIRS data is 0.2% decade−1. Minnis et al. (2008) improved the DCC identification technique by introducing TB11 and reflectance uniformity tests and assessed the 0.65-μm-band calibration stability of the Visible and Infrared Scanner (VIRS) on board the TRMM and the Moderate Resolution Imaging Spectroradiometer (MODIS) on board the NASA Earth Observing System (EOS) Terra and Aqua satellites. Fougnie and Bach (2009) characterized the brightness, spectral aspects, bidirectional signature, stability, and homogeneity of DCCs, and investigated the relative calibration stability of the 0.443–1.020-μm bands of the Polarization and Directionality of Earth Reflectances (POLDER) instrument using DCC spectral ratios. Chen et al. (2013) used the DCC technique to monitor the reflective solar bands’ (RSBs) long-term response of the Fengyun-3A (FY-3A) Medium Resolution Spectral Imager (MERSI).
The DCC technique is a statistics-based method. Previous studies reported that DCC reflectance is affected by various factors, including the anisotropy of top of the atmosphere (TOA) reflectance, the choices of TB11 threshold, the reflectance and TB11 spatial uniformity thresholds, and cloud optical thickness and effective radius (Hu et al. 2004; Minnis et al. 2008; Sohn et al. 2009; Doelling et al. 2013). Angular distribution models (ADMs) were developed to account for the anisotropy effect in the DCC TOA reflectance (Hu et al. 2004; Loeb et al. 2005; Doelling et al. 2013). Sohn et al. (2009) analyzed the uncertainty of DCC calibration technique due to cloud optical thickness and effective radius variations in the identified DCC pixels for the VIS bands of Terra and Aqua MODIS. Hong et al. (2008) analyzed the interannual to diurnal variations of tropical DCCs using 7 years of the Advanced Microwave Sounding Unit B measurements. Using Aqua MODIS observations, Doelling et al. (2013) investigated extensively the sensitivity of DCC reflectance to the TB11 threshold, the reflectance and TB11 spatial uniformity thresholds, and the choice of ADMs, as well as the regional, temporal, and spectral characterization of DCCs. However, previous studies mostly focused on the 0.65-μm VIS band. The ~0.86-μm NIR band provides important observations for the satellite retrieval of green vegetation, ocean color, and aerosol. Unfortunately, MODIS channel 2 (0.86 μm) saturated over DCCs. Doelling et al. (2004), Aumann et al. (2007), Fougnie and Bach (2009), and Chen et al. (2013) applied the DCC technique to the ~0.86-μm bands of AVHRR, AIRS, POLDER, and MERSI, but the sensitivity of DCC reflectance to various factors at this spectrum has not been adequately studied to date. Moreover, the spatial resolution and the size of DCC clusters may affect the DCC spatial uniformity tests because the footprint of satellite observations varies with instrument, scan angles, and satellite altitude. However, the sensitivity of DCC reflectance to pixel spatial resolution and cluster size for both the VIS and NIR bands has not been characterized so far. In addition, previous studies usually assumed that the TB11 band is accurately calibrated using onboard blackbody calibrators; the impact of TB11 radiometric calibration bias on the DCC technique was not addressed. Disagreements also exist in DCC calibration results presented by different groups. For example, Minnis et al. (2008) found that large discrepancy exists in the trends of calibration gain calculated from the mean and mode of Aqua MODIS monthly DCC normalized radiances. Doelling et al. (2013) reported that the Aqua MODIS VIS band was stable within 0.5% decade−1 according to both the mean and mode of monthly DCC reflectance. Therefore, more efforts are needed to investigate the sensitivity of DCC reflectance to reconcile the disagreement and to make the DCC calibration method more robust.
The Visible and Infrared Imaging Radiometer Suite (VIIRS) on board the Joint Polar Satellite System (JPSS)/Suomi National Polar-Orbiting Partnership (Suomi-NPP) satellite was successfully launched on 28 October 2011. VIIRS has 22 spectral bands, with 14 RSBs, seven thermal emissive bands (TEB), and 1 day–night band (DNB). It provides moderate-resolution and radiometrically accurate observations of the globe at two spatial resolutions: 375 m (I bands) and 750 m (M bands) at nadir. VIIRS is performing well based on on-orbit verification and intensive calibration and validation results (Cao et al. 2013, 2014). With VIS, NIR, and LWIR measurements available at two spatial resolutions, VIIRS provides new opportunities for investigating the radiometric sensitivity of DCC reflectance in the VIS and NIR spectrum to different factors, such as spatial resolution.
The purpose of this paper is to study the DCC radiometric sensitivity to spatial resolution, DCC cluster size, and TB11 threshold and calibration bias using VIIRS observations. We focus on the DCC radiometric characteristics for the 0.865-μm NIR bands (M7 and I2) that have nearly identical spectral response but different spatial resolution. The VIIRS 0.672-μm VIS band (M5) was also studied for comparison purpose. The sensitivities of DCC reflectance to spatial resolution and DCC cluster size have not been studied in previous studies. The impact of the TB11 threshold on DCC reflectance has been investigated previously for different instruments (Hu et al. 2004; Aumann et al. 2007; Doelling et al. 2013). Here, we further explore this issue using VIIRS observations and investigate the influence of spatial resolution together with the TB11 threshold. Moreover, the stabilities of mean and mode of monthly DCC reflectance were extensively analyzed in this study. The paper is organized as follows. VIIRS datasets used in this paper are introduced in section 2. Section 3 presents VIIRS DCC identification criteria and methods for generating different DCC datasets. Section 4 focuses on the radiometric sensitivities of DCC reflectance to various factors. A summary is given in section 5.
2. VIIRS datasets used
The DCC technique can be applied to multiple VIIRS VIS/NIR bands, including M1–M5, M7, and I1 and I2 (M6 saturated over DCCs). This study investigates the radiometric sensitivity of DCC reflectance for bands M5 (0.672 μm), M7 (0.865 μm), and I2 (0.865 μm). Band M15 (10.763 μm) provides TB11 measurements that are required to identify DCC pixels for bands M5 and M7; I5 (11.45 μm) provides TB11 measurements for band I2. Figure 1 shows the relative spectral response (RSR) functions for all bands used in this study. Table 1 summarizes their spectral, spatial, and radiometric characteristics (Cao et al. 2013, 2014). VIIRS band M5 provides observations at a similar central wavelength as the 0.65-μm VIS band of MODIS, GOES, and AVHRR. Results from band M5 will be used to compare VIIRS VIS DCC radiometric characteristics to those derived from other instruments and VIIRS NIR bands. NIR bands M7 and I2 have nearly identical RSRs but at different spatial resolutions. DCCs from M7 and I2 will be used to study DCC radiometric sensitivity to different factors at NIR spectrum. Moreover, the RSRs of M15 and I5 are centered at slightly different wavelengths and with distinct noise-equivalent differential temperature (NEdT) characteristics (see Table 1) (Cao et al. 2014). The differences will be useful for investigating DCC radiometric sensitivity to the TB11 radiometric calibration bias.
Spectral, spatial, and radiometric characteristics of VIIRS spectral bands used in this study. M5 and M7 signal-to-noise ratios (SNRs) at low gain stage and typical scene radiances are listed because DCCs are highly reflective; I2 SNR is given at typical scene radiance; NEdTs are estimated at 205-K scene temperature.
This study focuses on the DCC characteristics over a region defined as 25°S–25°N, 150°–60°W, a portion of the intertropical convergence zone (ITCZ) that is also observed by the current GOES-East and GOES-West satellites with the understanding that the results will benefit the calibration for geostationary satellites. One year (May 2013–April 2014) of VIIRS daytime bands M5, M7, M15, I2, and I5 TOA reflectance/brightness temperature, and radiance 1-min sensor data record (SDR) granules were obtained from the NOAA/NESDIS/Center for Satellite Applications and Research (STAR) Central Data Repository (SCDR), which is essentially a short-term (4 months, revolving) mirror site of the NOAA’s Comprehensive Large Array-Data Stewardship System for VIIRS data. VIIRS SDR products are generated by the operational ground processing in the JPSS Interface Data Processing Segment (IDPS). The product short names for these five bands are SVM05, SVM07, SVM15, SVI02, and SVI05, respectively.
The VIIRS RSBs are calibrated using a full-aperture solar diffuser (SD), which is illuminated once per orbit as the satellite passes from the dark side to the sunlit side of the earth in the high latitudes of the Southern Hemisphere. The space view provides the offset measurements needed for the calibration. The degradation of SD is monitored by a solar diffuser stability monitor that is a separate device with eight detectors from 0.412 to 0.926 μm. The SD is stable over most of the VIS and NIR spectra based on the postlaunch calibration/validation results. Significant SD degradations (>30%) were observed in the M7 and I2 spectra; however, the degradations leveled off at the time of this study and have limited impact on the instrument performance and products due to weekly updates of F-factor lookup table in the calibration. Based on prelaunch laboratory test results, the calibration uncertainty of RSBs for a typical scene is expected to be less than 2%. Postlaunch on-orbit verification and intensive calibration and validation results indicate that all RSB bands used in this study are performing well (Cao et al. 2013). It is worth noting that no vicarious calibration is applied to the IDPS version of VIIRS RSB SDR products for the period of this study.
The VIIRS TEBs are calibrated using an onboard calibrator blackbody that has been carefully characterized prelaunch. The temperature of blackbody is controlled using heater elements and thermistors. Six thermistors on the blackbody, with temperature variations less than 0.03 K, are used to measure the bulk temperature. The VIIRS longwave cold focal plane temperature is very stable with negligible variations since launch. The VIIRS TEB calibration algorithm is based on the measured blackbody temperatures, emissivity, and space view. The variation of background emission from the rotating half-angle mirror and surrounding components is also taken into account by scan-angle-dependent corrections. The postlaunch validation results showed the bias between VIIRS M15 and the equivalent MODIS band is on the order of 0.1 K after accounting for the RSR differences (Cao et al. 2013). However, a recent Cross-Track Infrared Sounder (CrIS) and VIIRS comparison study showed a brightness temperature discrepancy on the order of 0.4 K at ~205-K scene temperature in band M15 (Tobin et al. 2013a). The impact of this potential VIIRS calibration bias will be discussed later in section 4a.
VIIRS observations have unique advantages for studying the sensitivity of DCC reflectance to spatial resolution due to its improved geometric performance compared to both MODIS and AVHHR. It uses a unique pixel aggregation strategy to archive a spatial resolution of ~800 m for I bands and ~1600 m for M bands at the edge of scan. Three samples are aggregated for the pixels near nadir; two samples are aggregated for the middle scan angles; no aggregation is performed for large scan angles. As a result, the spatial resolution of the VIIRS edge-of-scan pixels is only of a factor of 2 of that of the nadir pixels, which significantly limits the geometric pixel growth toward the end of the scan. Moreover, VIIRS also has a nearly square shape for pixels at all sensor view zenith angles (VZA) (Cao et al. 2014). In addition, all VIIRS bands are well coregistered band to band and 2 × 2 I-band pixels are nested to one M-band pixel, which is especially useful for studying the resolution effects of DCC radiometric sensitivity.
3. Methodology
In this paper, VIIRS DCC pixels are identified using a method similar to those described in Minnis et al. (2008) and Doelling et al. (2013). Specifically, the general criteria for identifying VIIRS DCC pixels are as follows: 1) TB11 is less than 205 K; 2) the standard deviation of TB11 of the subject pixel and its eight adjacent pixels is less than 1 K; 3) the standard deviation of VIS/NIR reflectance of the subject pixel and its eight adjacent pixels is less than 3%, relative to the mean reflectance of these nine pixels; 4) the solar zenith angle (SZA) is less than 40°; and 5) VZA is less than 35°. VIIRS uses a bow-tie deletion technique to remove duplicated pixels in the off-nadir regions to save data transmission bandwidth (Cao et al. 2014). The purpose of using a 35° VZA threshold in this study is to avoid the bow-tie effect in the VIIRS dataset, which would complicate the DCC spatial uniformity tests and cluster identification. The above-given criteria were used to construct multiple VIIRS DCC datasets for various sensitivity studies if it is not otherwise stated.
Four VIIRS DCC datasets were created to facilitate different sensitivity studies. First, one year (May 2013–April 2014) of DCC TOA reflectance and TB11 brightness temperatures for bands M5 (750 m), M7 (750 m), and I2 (375 m) over the area of interests were generated using the above-described DCC identification criteria (dataset 1). The number of DCC pixels identified for each month ranges from ~0.5 million to 2.0 million for bands M5 and M7, and from ~2.5 million to 9.5 million for band I2. January 2014 has the fewest number of DCC pixels. For each DCC pixel, the anisotropic effect in TOA reflectance for all bands was corrected using an ADM developed by Hu et al. (2004). Monthly probability distribution functions (PDFs), as well as their means and modes, were calculated for both the DCC TOA reflectance and the ADM-adjusted DCC reflectance with a 0.003 increment (in reflectance), which was determined by varying the increment from 0.002 to 0.005 (increased by 0.0005 each time) and visually examining PDFs. The smallest increment with reasonably smooth PDFs for all bands was selected.
Second, an ADM-adjusted DCC dataset (dataset 2) was generated for representative months from each season (July 2013, October 2013, January 2014, and April 2014) for all bands. The method for generating dataset 2 is the same as that for dataset 1, except with the TB11 threshold varying from 204.5 to 205.5 K. This dataset was used to characterize the sensitivity of VIIRS DCC reflectance to the TB11 calibration bias.
The third dataset (dataset 3) consists of 3-month merged seasonal DCCs at four different spatial resolutions for all bands. It was used to investigate the sensitivity of the mean and mode of DCC reflectance to the spatial resolution and the TB11 threshold, as well as their seasonal variations. Three-month merged seasonal DCCs were used to increase the sample size for more robust statistical analysis. For our experiment in this study, we downsampled, by simple averaging, VIIRS bands M5 and M7 TOA reflectance and M15 radiance products by factors of 2, 4, and 8 to generate M-band datasets at 1500-, 3000-, and 6000-m spatial resolutions. Similarly, band I2 TOA reflectance and band I5 radiance products were downsampled to generate I-band datasets at 750-, 1500-, and 3000-m spatial resolutions. The downsampled TB11 radiances were converted to brightness temperature using an equivalent blackbody temperature lookup table. DCC pixels at different spatial resolutions were identified and the anisotropic effect was corrected using the same methods as dataset 1. Then, together with dataset 1, 3-month merged seasonal DCCs at four spatial resolutions were generated for each band. Specifically, March–April 2014 and May 2013, June–August 2013, September–November 2013, and December 2013 and January–February 2014 DCCs were combined to generate seasonal subdatasets for spring, summer, fall, and winter, respectively. PDFs of DCC reflectance were calculated at different spatial resolutions and TB11 thresholds (200–205 K) using the similar method as dataset 1. It is worth noting that the PDF increments used for the three downsampled subdatasets are 0.0035, 0.004, and 0.005, corresponding to downsampled factors of 2, 4, and 8, respectively. Larger increments were used in these cases due to their smaller sample sizes.
Dataset 4 is a four-season DCC dataset derived from dataset 3, but only at 750-m (M5 and M7) and 375-m (I2) spatial resolutions. In dataset 4, DCC clusters were identified for each VIIRS granule that contains DCC pixels to study the sensitivity of DCC reflectance and TB11 temperature to cluster size. DCC clusters were calculated using a chain-code algorithm (Freeman 1961) that operates by following the boundary around in a counterclockwise fashion, pixel by pixel, until it returns to the starting pixel. Figure 2 shows an example of VIIRS band I2 DCC clusters identified for a granule at 1959–2001 UTC 4 July 2013. The area and perimeter were calculated for each DCC cluster. The area/perimeter ratio, which provides a measure of the shape of the DCC cluster, was also computed. The seasonal DCCs were divided into equal-interval bins based on cluster area, perimeter, and area/perimeter ratio. The mean and mode of seasonal PDFs of DCC reflectance, the mean of TB11 temperatures, and the number of DCC pixels were calculated for each bin to investigate the relationships between DCC reflectance/mean TB11 temperature and cluster size, as well as their seasonality.
4. Results and discussions
a. Monthly probability distribution functions of VIIRS DCC reflectance
Figure 3 (left panel) shows monthly PDFs of DCC TOA reflectance for bands M5, M7, and I2 at 750-, 750-, and 375-m spatial resolutions, respectively. The mean and mode of monthly PDFs, as well as the number of DCC pixels for each month, are also listed. Spring and summer months generally have more DCC pixels, and winter months have fewer DCC pixels. The means and modes of monthly DCC TOA reflectance are larger than 0.94 in all cases. Moreover, the modes of DCC reflectance are greater than the means, consistent with Doelling et al.’s (2013) results for Aqua MODIS DCCs. Standard deviations of the mean and mode for all three bands are small, ranging from 0.6% to 0.7%. The total variations (maximum minus minimum) are 2.2% or less for the mean and 2.4% or less for the modes. VIIRS monthly DCC reflectance agrees well with previous studies in that DCCs are a highly reflective target and near-Lambertian diffuse reflectors in the VIS and NIR spectra (Doelling et al. 2004; Aumann et al. 2007; Fougnie and Bach 2009; Doelling et al. 2013).
Figure 3 (right panel) shows the monthly PDFs of the ADM-adjusted VIIRS DCC reflectance. After the adjustment, monthly PDFs become more consistent with each other, with standard deviations of mean and mode reduced to 0.5% and 0.2%, respectively, for all bands. The total variations are also reduced, with 1.9% or less for the mean and 0.6% for the mode. In addition, the mode of monthly DCC reflectance varies less than the mean for all bands. According to the NOAA Integrated Calibration/Validation System’s long-term monitoring results (NOAA 2014), the responsivity degradation of bands M7 and I2 is more than 30% since launch, while the responsivity degradation of band M5 is ~10%. However, the degradation effect is compensated for through weekly updates of the calibration coefficients (Cardema et al. 2012; Rausch et al. 2013); therefore, the calibrated reflectance for these bands should be stable, as they were found in validation studies (Cao et al. 2013). The means and modes of monthly PDFs confirm that the radiometric calibrations of all three bands are stable for highly reflective scenes during the study period despite the optical throughput degradation. Compared to the DCC TOA reflectance, the ADM-adjusted DCC reflectance has smaller variances. Therefore, the ADM-adjusted DCC reflectance (called DCC reflectance) was used to characterize the sensitivity of VIIRS DCC reflectance to various factors in the following subsections.
b. Sensitivity of DCC reflectance to TB11 calibration bias
There were concerns about the impact of the 0.4-K M15 cold calibration bias at 205-K scene temperature compared with CrIS on board the same satellite (Tobin et al. 2013a) on the DCC calibration technique. It is worth noting that VIIRS M15 brightness temperature is very stable over time and only the absolute calibration may differ from CrIS. If the M15 were to fluctuate, then the DCC calibration technique would be affected. The latest investigations by the VIIRS and CrIS science teams (Moeller 2014; Tobin et al. 2013b) indicated that the M15 cold bias has been reduced to 0.3 K by a CrIS calibration change. The remaining cold bias can be further reduced to near zero if needed by VIIRS calibration coefficient adjustments, but this potential change will cause a jump in M15 brightness temperature and therefore its impact on the DCC technique needs to be assessed.
To analyze the impact of TB11 calibration bias on DCC reflectance, a sensitivity study was conducted using DCCs from four representative months (July 2013, October 2013, January 2014, and April 2014) by varying the TB11 threshold from 204.5 to 205.5 K. Figure 4 shows band M7 monthly PDFs of DCC reflectance as a function of TB11 threshold in the four representative months. The maximum impacts of 0.1–0.5-K cold or warm TB11 calibration bias for all bands are summarized in Table 2. For 0.1–0.5-K cold or warm TB11 calibration biases, the mean of DCC reflectance changed by only 0.1%–0.3%, while the mode remains unchanged in all cases, except for band M7 in January 2014 and band I2 in October 2013. In these cases, the mode changes by 0.3% (identical with the PDF increment used). Therefore, a TB11 calibration cold or warm bias on the order of 0.5 K does not have significant impact on the DCC calibration technique, especially when the mode method is used. Moreover, the results also suggest that the mode of monthly DCC reflectance is generally less sensitive to TB11 calibration bias compared to the mean.
Maximum changes of the mean and mode of monthly DCC reflectance due to 0.1–0.5-K TB11 cold or warm calibration biases for seasonal representative months during the study period.
The monthly PDFs for bands M7 and I2 also imply that the DCC calibration technique is insensitive to small TB11 calibration biases (see Table 2). The RSRs for bands M7 and I2 are nearly identical (see Fig. 1). The mean M15 brightness temperature for band M7 DCC pixels (750-m spatial resolution) is ~0.2 K higher than the mean I5 brightness temperatures for the downsampled and matched band I2 DCC pixels, which is mostly due to spectral response function differences. Note I5 has a much broader spectral response than that of M15. Point spread function and geolocation differences may also be factors with small effects. Nevertheless, the averaged difference between M7 and downsampled and matched I2 monthly DCC reflectance is ~0.2%. Moreover, it appears that the DCC technique is insensitive to instrument noise of TB11 bands. The specification and on-orbit NEdTs for band I5 are more than 4 times larger than that for band M15 (see Table 1). However, no significant difference was found in the DCC reflectance of bands I2 and M7.
c. Sensitivity of DCC reflectance to spatial resolution and TB11 threshold
Figure 5 shows the mean and mode of DCC reflectance as functions of spatial resolution and TB11 threshold for the summer season ADM-adjusted DCCs. Similar features were observed for DCCs from other seasons. Table 3 summarizes the percent changes of the mean and mode of DCC reflectance for four seasons: 1) scenario 1: at 750-m (M5 and M7) and 375-m (I2) spatial resolutions, the changes of DCC reflectance (in percent) as the TB11 threshold increases from 200 to 205 K; 2) scenario 2: at 6000-m (M5 and M7) and 3000-m (I2) spatial resolutions, the changes of DCC reflectance as the TB11 threshold increases from 200 to 205 K; 3) scenario 3: at 205-K TB11 threshold, the changes of DCC reflectance as the spatial resolution increases by a factor of 8. Figure 6 shows the mean and mode of DCC reflectance at 205-K TB11 threshold as a function of downsampling factor in four seasons.
Changes of the mean/mode of VIIRS seasonal DCC reflectance as functions of spatial resolution and TB11 threshold at three different scenarios for May 2013–April 2014.
Several obvious features were observed in Fig. 5 and Table 3. First, the means of PDFs decrease as the TB11 threshold becomes higher at all spatial resolutions and for all bands and seasons (Fig. 5, left panel; Table 3, scenarios 1 and 2), consistent with previous studies (Hu et al. 2004; Aumann et al. 2007; Doelling et al. 2013). As the TB11 threshold increases from 200 to 205 K, the band I2 mean dropped by ~1.6% to 2.3% for the 375-m spatial resolution (scenario 1), at least 50% larger than the ~0.9%–1.5% of drops for the downsampled 3000-m dataset (scenario 2). Among datasets with different spatial resolutions, the 6000-m data derived from M5 and M7 generally have the smallest reductions in the mean (≤0.9% and ≤0.8%, respectively), which indicates that the mean of DCC reflectance are less sensitive to the TB11 threshold for DCCs with coarse spatial resolutions.
Second, the means of DCC reflectance increases as the spatial resolution increases for all bands and seasons (see Fig. 5, left panel), indicating that the DCC criteria are more likely to allow the preferential selection of DCC pixels with high reflectance at a coarser spatial resolution. At the 205-K TB11 threshold, the mean of DCC reflectance derived from band I2 increases by 3.3%–4.0% for different seasons as the spatial resolution increases by a factor of 8 (Table 3; scenario 3). Similar magnitudes of increases were also observed in band M7; the increases for band M5 are even larger (4.8%–5.1%). Moreover, the number of DCC pixels drops more than 4 times after each downsampling by a factor of 2, suggesting that the same DCC identification criteria detect less total areas of DCCs for datasets at coarser spatial resolutions. In other words, the same criteria may have different impacts on DCC identification for observations from the coarser-resolution geostationary instruments such as GOES (the nominal DCC pixel size of which is ~4 km at nadir, due to the 4-km nominal spatial resolution of the GOES TB11 band) and finer spatial resolution of polar-orbiting instruments such as VIIRS (0.375 and 0.750 km, respectively) and MODIS (0.250 and 1 km, respectively).
Third, the means of DCC reflectance in the winter show the least magnitude of variations in terms of TB11 threshold at all spatial resolutions and for all bands, while the variations in the summer and fall are relatively larger (see Table 3). Moreover, the means of the winter season DCCs are ~0.5% higher than those from other seasons (see Fig. 6). It is worth noting that winter is the time period with the fewest DCC pixels, only about half of the frequency of the other seasons. We compared the spatial distribution of DCCs for different seasons. Identified DCC pixels mostly occur over the eastern South America land region within our area of interests during winter. During other seasons, they mostly occur over ocean, especially for summer. Over the eastern South America region, the mean DCC reflectance during winter is 0.5%–1% higher than those during spring and fall. The reflectance of summer DCCs is comparable to those of winter, but the number of DCC pixels for summer is 80% less; therefore, the statistics may not be robust. Over ocean, the mean DCC reflectance for the winter is also higher than those for other seasons. However, limited ocean DCC samples are available during winter. A larger study region is needed to compare the seasonal variations of DCC reflectance over land and ocean.
Fourth, the mode of DCC reflectance for all three bands also increases with the increase of the spatial resolution and decreases with the increase of the TB11 threshold. However, the mode is less sensitive to both factors compared to the mean. In Fig. 5 (right panel) and Fig. 6, it can be observed that the mode stays invariant in many cases as the TB11 threshold and/or spatial resolution varies. Among the three bands considered in this study, the magnitude of the decrease in the modes is less than half of the decrease in means in all cases in Table 3, scenarios 1 and 3. At the 205-K TB11 threshold, the mode increases by 1.9% or less and 1.6% or less for bands M5 and M7, respectively, as the spatial resolution varies from 750 to 6000 m and ~1.3% for bands I2 as the spatial resolution changes from 375 to 3000 m, all less than half of the 4.0%–5.1% increases in means (Table 3; scenario 3). The variations for the mean and mode are similar for M5 and M7 in scenario 2 because the effect of spatial resolution dominates. Therefore, the mode of DCC reflectance is more stable than the mean in terms of spatial solution, TB11 threshold, and seasonality, consistent with Doelling et al.’s (2013) findings that the mode of monthly MODIS 0.65-μm-band DCC reflectance is more uniform regionally compared to the mean.
Fifth, the seasonal variability of the mode is also smaller than the mean, consistent with Doelling et al. (2013). Compared to the mean, the mode in the winter season is generally consistent with those in spring, summer, and fall at all spatial resolutions (see Fig. 6). The means of DCC reflectance in winter have higher absolute values and vary less compared to those in other seasons. However, the mode values are very close to each other in the four seasons, indicating the DCC mode time series have smaller seasonal cycles than the mean time series.
Finally, the DCC reflectance for the NIR bands is generally less sensitive to the spatial resolution and TB11 threshold than the VIS band. At the 205-K TB11 threshold (Table 3, scenario 3), as the spatial resolution increases by a factor of 8, the mean and mode of bands M7 and I2 monthly DCC reflectance increase by 4.2% and 1.6% or less, respectively, both smaller than the increases in band M5 (5.1% and 1.9% or less, respectively). Similar patterns were also observed at other TB11 thresholds in most cases. As the TB11 threshold changes from 200 to 205 K (Table 3, scenarios 1 and 2), the mean of monthly M7 and I2 DCC reflectance at different spatial resolutions also generally decreases in a smaller magnitude than those for band M5.
d. Sensitivity of DCC reflectance and TB11 temperature to cluster size
Figure 7 illustrates the sensitivity of mean and mode of DCC reflectance as a function of cluster size defined either by area, perimeter, or the area/perimeter ratio for band I2. The results are stratified by season as indicated by the colored lines in Fig. 7. Also the TB11 and number of DCC pixels are shown in a similar manner with cluster size. The seasonal DCCs of bands M5 and M7 have similar patterns as band I2. Table 4 summarizes the four-season averaged correlation coefficients between the mean and mode of bands M5, M7, and I2 seasonal DCC reflectance/TB11 as a function of cluster size, as well as standard deviations of the seasonal DCC means and modes as a function of cluster size. It is worth noting that altering the number or size of cluster bins will slightly change the values of the correlation coefficient and standard deviation, but the general patterns remain similar. In addition, the statistics for smaller DCC clusters are more robust due to larger cluster samples (see the fourth column of Fig. 7 for the number of DCC pixels as a function of cluster size). Large DCC clusters occur less frequently and they are prone to be artificially divided into smaller ones by granule boundaries and the SZA/VZA thresholds used in the DCC identification criteria (see Fig. 2). Therefore, more scatters were observed in Fig. 7 as the DCC cluster size becomes larger.
May 2013–April 2014 four-season averaged 1) correlation coefficients (corr) between the mean and mode of seasonal DCC reflectance/TB11 temperature as a function of cluster size and 2) standard deviations (std dev) of DCC means and modes as a function of cluster sizes.
Our cluster analysis results indicate that all three cluster size parameters, including area, perimeter, and area/perimeter ratio, are useful for characterizing the relationships between DCC reflectance/TB11 temperature and cluster size. Two interesting features can be observed. First, the mean and mode of DCC reflectance are strongly correlated positively with cluster perimeter and area/perimeter ratio, with correlation coefficients of 0.75 or higher. In general, large DCC clusters have higher mean and mode DCC reflectance. Second, mean TB11 temperature is strongly correlated negatively with DCC cluster perimeter and area/perimeter ratio, with correlation coefficients from −0.78 to −0.95, indicating that large DCC clusters have lower mean TB11 temperatures. These two features are consistent with the physical theoretical basis. Small DCC clusters and DCC pixels at the edge of a cluster are more likely to occur at the convective anvils, which are less reflective and have high TB11 temperature. On the contrary, a higher percentage of DCC pixels is located at the convective core in large clusters, especially for those with near-circular shapes (featured by greater area/perimeter ratios). Therefore, larger DCC clusters are featured by higher mean DCC reflectance and lower TB11 temperatures. Compared with the cluster perimeter and area/perimeter ratio, the cluster area is less effective, with weaker correlations with DCC reflectance and TB11 temperature. However, its underlying cause is uncertain.
Compared to the means of DCC reflectance, the modes are more stable in terms of all three cluster size parameters. Table 4 indicated that the four-season averaged standard deviations of the mode for different cluster sizes are generally much smaller than those for the mean. The mode is also more stable in individual seasons for band I2 (see Fig. 7), as well as for bands M5 and M7 (not shown). Figure 7 shows the mean generally increases with cluster size for all clusters. The mode also increases with cluster size for small clusters. However, when the DCC cluster area is larger than ~2000 km2, the perimeter exceeds ~200 km, or the perimeter/area exceeds ~4 km, the modes of band I2 DCC reflectance tend to become saturated at a value of ~0.925, which is very close to the mode of band I2 monthly DCC reflectance showed in Fig. 3 (0.9255–0.9285). Moreover, the modes also have smaller seasonal variations. In the winter season, the means of DCC reflectance are ~0.5%–1% higher than those in other seasons (Fig. 7, first column); while the modes for different seasons are more consistent with each other (Fig. 7, second column).
The cluster analysis results further support the finding that the mode of DCC reflectance is a more reliable indicator for calibration stability for individual VIS and NIR bands than the mean. The mean of monthly DCC reflectance may be biased when large quantities of small DCC clusters are present. Our analysis found that the mean and mode of DCC reflectance for small clusters are significantly different from those for large clusters. Excluding small DCC clusters from DCC analysis may improve the reliability of the DCC calibration technique, especially for instruments with finer spatial resolutions, such as VIIRS and MODIS imaging bands.
The DCC reflectance for the NIR bands is also generally less sensitive to cluster size variations than the VIS band. Table 4 shows that the four-season averaged standard deviations of the mean and mode of DCC reflectance for different cluster sizes are much smaller for band M7 than band M5 in all cases. Though the spatial resolution of band I2 DCCs (375 m) are finer than M5 (750 m) and DCC reflectance is more stable at coarser spatial resolutions, the standard deviations for I2 is smaller than M5 in three out of six cases and comparable to M5 in other cases.
5. Summary
In this study, one year of VIIRS VIS (M5) and NIR (M7 and I2) DCCs were used to study the radiometric sensitivity of DCC reflectance to spatial resolution, TB11 threshold and calibration bias, and cluster size in four seasons. VIIRS DCC pixels were identified using the widely used DCC identification criteria. The anisotropic effects in VIIRS TOA reflectance were adjusted using Hu et al.’s (2004) ADM. Monthly PDFs of the ADM-adjusted DCC reflectance indicates that all three bands are stable during the study period despite the VIIRS rotating telescope mirror throughput degradation. The standard deviations of the mean and mode of monthly DCC reflectance are 0.5% and 0.2%, respectively, in all cases. A TB11 calibration bias on the order of 0.5 K does not have a significant impact on monthly DCC reflectance, especially when the mode method is used. Our study also indicates that the mean of DCC reflectance is a function of spatial resolution, TB11 threshold, and cluster size evident in all seasons. The mean of DCC reflectance at different spatial resolutions and cluster sizes are higher in the winter than those in other seasons. Although the mode also varies, it is more stable compared with the mean in terms of TB11 calibration bias, spatial resolution, TB11 threshold, cluster size, and season. Therefore, this study suggests that the mode is more suitable to be used as an indicator of calibration stability for individual VIS and NIR bands. Moreover, the DCC reflectance for coarser-spatial-resolution datasets is less sensitive to TB11 threshold variations. The DCC reflectance for the NIR bands is generally less sensitive to spatial resolution, TB11 threshold, and cluster size than the VIS band. Results from this study improves our understanding of the DCC radiometry and allow us to use DCC as a more reliable source for VIS and NIR calibration for instruments on both polar-orbiting and geostationary satellites such as JPSS and GOES-R.
Acknowledgments
This work is funded by the JPSS program office. The authors thank the anonymous reviewers for their valuable comments, which greatly helped improve the quality of this paper. The authors would also thank the GSICS VIR/NIR subgroup for kindly providing the Hu et al. (2004) angular distribution model. 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
Aumann, H. H., Pagano T. , and Hofstadter M. , 2007: Observations of deep convective clouds as stable reflected light standard for climate research: AIRS evaluation. Atmospheric and Environmental Remote Sensing Data Processing and Utilization III: Readiness for GEOSS, M. D. Goldberg et al., Eds., International Society for Optical Engineering (SPIE Proceedings, Vol. 6684), 668410, doi:10.1117/12.734599.
Cao, C., and Coauthors, 2013: Suomi NPP VIIRS sensor data record verification, validation, and long-term performance monitoring. J. Geophys. Res. Atmos.,118, 11 664–11 678, doi:10.1002/2013JD020418.
Cao, C., De Luccia F. J. , Xiong X. , Wolfe R. , and Weng F. , 2014: 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.,52, 1142–1156, doi:10.1109/TGRS.2013.2247768.
Cardema, J. C., Rausch K. W. , Lei N. , Moyer D. I. , and De Luccia F. J. , 2012: Operational calibration of VIIRS reflective solar band sensor data records. Earth Observing Systems XVII, J. J. Butler, X. Xiong, and X. Gu, Eds., International Society for Optical Engineering (SPIE Proceedings, Vol. 8510), 851019, doi:10.1117/12.932221.
Chen, L., Hu X. , Xu N. , and Zhang P. , 2013: The application of deep convective clouds in the calibration and response monitoring of the reflective solar bands of FY-3A/MERSI (Medium Resolution Spectral Imager). Remote Sens., 5, 6958–6975, doi:10.3390/rs5126958.
Doelling, D. R., Minnis P. , and Nguyen L. , 2004: On the use of deep convective clouds to calibrate AVHRR data. Earth Observing Systems IX, W. L. Barnes and J. J. Butler, Eds., International Society for Optical Engineering (SPIE Proceedings, Vol. 5542), 281, doi:10.1117/12.560047.
Doelling, D. R., Morstad D. , Scarino B. R. , Bhatt R. , and Gopalan A. , 2013: The characterization of deep convective clouds as an invariant calibration target and as a visible calibration technique. IEEE Trans. Geosci. Remote Sens.,51, 1147–1159, doi:10.1109/TGRS.2012.2225066.
Fougnie, B., and Bach R. , 2009: Monitoring of radiometric sensitivity changes of space sensors using deep convective clouds: Operational application to PARASOL. IEEE Trans. Geosci. Remote Sens.,47, 851–861, doi:10.1109/TGRS.2008.2005634.
Freeman, H., 1961: On the encoding of arbitrary geometric configurations . IRE Trans. Electron. Comput.,EC-10, 260–268, doi:10.1109/TEC.1961.5219197.
Hong, G., Heygster G. , Notholt J. , and Buehler S. A. , 2008: Interannual to diurnal variations in tropical and subtropical deep convective clouds and convective overshooting from seven years of AMSU-B measurements. J. Climate, 21, 4168–4189, doi:10.1175/2008JCLI1911.1.
Hu, Y., Wielicki B. A. , Ping Y. , Stackhouse P. W. Jr., Lin B. , and Young D. F. , 2004: Application of deep convective cloud albedo observation to satellite-based study of the terrestrial atmosphere: Monitoring the stability of spaceborne measurements and assessing absorption anomaly. IEEE Trans. Geosci. Remote Sens.,42, 2594–2599, doi:10.1109/TGRS.2004.834765.
Loeb, N. G., Kato S. , Loukachine K. , and Manalo-Smith N. , 2005: Angular distribution models for top-of-atmosphere radiative flux estimation from the Clouds and the Earth’s Radiant Energy System instrument on the Terra satellite. Part I: Methodology. J. Atmos. Oceanic Technol., 22, 338–351, doi:10.1175/JTECH1712.1.
Minnis, P., Doelling D. R. , Nguyen L. , Miller W. F. , and Chakrapani V. , 2008: Assessment of the visible channel calibrations of the VIRS on TRMM and MODIS on Aqua and Terra. J. Atmos. Oceanic Technol., 25, 385–400, doi:10.1175/2007JTECHA1021.1.
Moeller, C., cited 2014: S-NPP VIIRS thermal emissive band (TEB) validation update. [Available online at http://www.star.nesdis.noaa.gov/star/documents/meetings/SNPPSDR2013/dayTwo/Moeller_VIIRS.pdf.]
NOAA, cited 2014: STAR Integrated Calibration/Validation System long-term monitoring. [Available online at http://www.star.nesdis.noaa.gov/icvs/status_NPP_VIIRS.php.]
Rausch, K., Houchin S. , Cardema J. , Moy G. , Haas E. , and De Luccia F. J. , 2013: Automated calibration of the Suomi National Polar-Orbiting Partnership (S-NPP) Visible Infrared Imaging Radiometer Suite (VIIRS) reflective solar bands. J. Geophys. Res. Atmos.,118, 13 434–13 442, doi:10.1002/2013JD020479.
Sohn, B. J., Ham S.-H. , and Yang P. , 2009: Possibility of the visible-channel calibration using deep convective clouds overshooting the TTL. J. Appl. Meteor. Climatol., 48, 2271–2283, doi:10.1175/2009JAMC2197.1.
Tobin, D. C., Revercomb H. , Knuteson R. , Taylor J. , and Deslover D. , 2013a: Cal/Val of CrIS on Suomi-NPP: Intercalibration with AIRS, IASI, and VIIRS. Annual Symp. on Future Operational Environmental Satellite Systems, Austin, TX, Amer. Meteor. Soc., P700. [Available online at https://ams.confex.com/ams/93Annual/webprogram/Paper224176.html.
Tobin, D. C., Revercomb H. , Taylor J. , Knuteson B. , DeSlover D. , Borg L. , and Martin G. , cited 2013b: CrIS radiometric calibration: Uncertainty estimates and evaluations. NOAA, 30 pp. [Available online at http://www.star.nesdis.noaa.gov/star/documents/meetings/SNPPSDR2013/dayOne/Tobin_UW_CrIS.pdf.]