1. Introduction
Clouds cover approximately two-thirds of the global surface. They strongly affect Earth’s radiative energy balance by reflecting incoming solar radiation back into space (the cooling effect of clouds) and emitting infrared radiation back to the surface (the warming effect of clouds) (Baker 1997). Whether clouds cool or warm the earth–atmosphere system depends on their optical properties and spatial distributions. Because of the diverse processes clouds are involved in, clouds show up in various three-dimensional forms with strong spatial and temporal variability, which makes quantifying them very difficult. Many different methods are used to observe clouds: from the surface, in situ (aircraft based), or from space, by passive or active remote sensing techniques. Each method has its advantages, but none can depict all cloud characteristics.
The most current and widely available cloud climatologies are obtained from the surface and satellites. Historically, ground-based cloud measurements have relied on human visual observations. These data provide information about different cloud types and amounts (or fractions) at given time intervals, but are expensive and somewhat subjective (Long et al. 2006b; Pfister et al. 2003). Automated cloud observations by sky-imaging systems may be a good alternative that can avoid the influence of subjective artificial influences and retrieve continuous information on clouds. Examples include hemispheric-sky imager/total-sky imager (HSI/TSI), whole-sky camera (WSI), and Nubiscope (an infrared scanner) (Feister et al. 2010; Long et al. 2006b). Retrievals of fractional cloud cover (or cloud fraction) from these devices have been validated by many studies (Boers et al. 2010; Feister et al. 2010; Slater et al. 2001). Another surrogate may be cloud fraction calculated by techniques that use surface radiation measurements, such as an automatic partial-cloud-amount detection algorithm, which measures cloud fraction using broadband longwave downward radiation (Dürr and Philipona 2004), a methodology proposed by Long et al. (2006a) for inferring fractional sky cover from broadband shortwave radiometer measurements, and a method from Min et al. (2008) for estimating cloud fraction using spectral radiation information. Abundant surface-radiation measurements can extend the spatial distribution and time scale of surface cloud-fraction datasets.
Only satellites observe clouds over the whole globe. However, satellite remote sensing of cloud fractions is known to involve uncertainties due to spatial resolution, measurement geometry, detection sensitivity, unknown background surface influences on the measured radiance, and other factors (Long et al. 2006b; Stubenrauch et al. 2013). Surface measurements usually serve as “ground truth” to validate satellite observations. However, as indicated by many studies (Henderson-Sellers and McGuffie 1990; Kassianov et al. 2005; Rossow et al. 1993; Wu et al. 2014), surface and satellite cloud observations have different fields of view (FOVs) and different definitions of cloud fraction. Satellites observe clouds in a nadir view, in which the total cloud fraction measured by satellite is the fractional of the earth’s surface covered by cloud (Hahn et al. 2001). Surface cloud observations by the naked human eye or sky-imaging devices have a hemispherical FOV, in which total cloud fraction refers to the fraction of a solid angle that is obscured by cloud in a hemispheric sky (Kassianov et al. 2005). When comparing satellite-derived cloud fractions with surface observations, this FOV mismatch must be considered.
Ground-based active remote sensing instruments and satellite active remote sensors are also used to observe cloud properties. An active sensor always has a narrow FOV. Cloud fractions of these active sensors are defined by averaging data over time intervals. These measurements do not correspond to cloud fraction defined as an instantaneous fraction of the whole sky covered by cloud. Also, using narrow-FOV data requires the validity of assumptions for certain cloud movement. Therefore, active sensors with narrow-FOV data are not included in this research. The Moderate Resolution Imaging Spectroradiometer (MODIS) is a passive multispectral imager operating on board two polar-orbiting satellites: NASA EOS Terra and Aqua. MODIS measures radiances at 36 spectral bands, including infrared and visible bands, with spatial resolutions from 0.25 to 1 km. The MODIS cloud-mask algorithm uses up to 14 bands to detect clouds (Frey et al. 2008).
The present research aims to determine the influence of mismatched FOVs when comparing satellite-based cloud fractions with surface-observed cloud fractions. MODIS cloud observations are compared with surface cloud data observed from TSI and calculated using the shortwave (SW) radiation methods proposed by Long et al. (2006a) and SW radiation methods proposed by Xie and Liu (2013) at five Surface Radiation Network (SURFRAD) sites. When large biases exist between satellite and surface observations, reasons for the discrepancy are discussed.
2. Data and methods
In this paper, 1 yr (2007 or 2009) of instantaneous MODIS-derived cloud-fraction data and three surface-radiation budget network (SURFRAD) ground-based cloud-fraction datasets are compared. Five SURFRAD sites located in different climate zones and having sufficient data were chosen (see Fig. 1). Data for 2007 were analyzed except at the site at Desert Rock in Nevada (DRA); because TSI cloud-fraction data available from SURFRAD at the DRA sites were lacking for 2007, data from 2009 were used instead.
Map of the five SURFRAD sites. The sites that are not defined in the text are located in Bondville, Illinois (BON), and at The Pennsylvania State University (PSU).
Citation: Journal of Applied Meteorology and Climatology 54, 5; 10.1175/JAMC-D-14-0206.1
Map of the five SURFRAD sites. The sites that are not defined in the text are located in Bondville, Illinois (BON), and at The Pennsylvania State University (PSU).
Citation: Journal of Applied Meteorology and Climatology 54, 5; 10.1175/JAMC-D-14-0206.1
Map of the five SURFRAD sites. The sites that are not defined in the text are located in Bondville, Illinois (BON), and at The Pennsylvania State University (PSU).
Citation: Journal of Applied Meteorology and Climatology 54, 5; 10.1175/JAMC-D-14-0206.1
a. Surface data
Three cloud-fraction datasets from the SURFRAD network were used (Augustine et al. 2000, 2005): one measured by TSI, one calculated from SW information using the methodology from Long et al. (2006a, hereinafter L06), and one calculated from SW information using the methodology from Xie and Liu (2013, hereinafter XL13).
TSI is a full-color sky camera and software system. It captures images of the hemispheric sky during daylight hours and computes a cloud fraction for each image. Cloudiness is analyzed for a 160° FOV, ignoring the 10° of the sky near the horizon. The TSI software distinguishes clear sky, thin clouds, and opaque clouds based on the red/blue (R/B) ratio of each pixel. Percentage thin and opaque cloud fractions are available for each minute on SURFRAD. Because of storage constraints, sky images are provided hourly on the SURFRAD website. In this research, thin and opaque cloud fractions were summed every minute, and then the 15-min cloud fractions centered at the MODIS overpass time were averaged following a suggestion by Kassianov et al. (2005) that 15-min averages of frequently sampled hemispheric sky cover relate much better to nadir-view cloud fraction. The TSI data and images near the MODIS overpass time were carefully examined to exclude erroneous data.
The L06 method estimates cloud fraction for an effective 160° field of view, using available downwelling direct and diffuse SW radiation data. This method first screens for overcast conditions (set to a sky-cover value of 100). For the remaining data, the estimate of fractional sky cover is based on a power-law formulation, which is empirically derived from the relationship between the normalized diffuse cloud effect and the sky fraction observed with total-sky imagers. Cloud fractions calculated by L06 were produced at 3-min intervals on the SURFRAD website. Similarly, a 15-min-period average centered at the MODIS overpass time was calculated for comparison with the MODIS cloud fraction.
The XL13 method simultaneously infers the cloud albedo and cloud fraction from surface total and direct radiation information. This method calculates cloud fraction with the information of the difference between the clear-sky and all-sky direct downwelling radiative fluxes normalized by the clear-sky direct downwelling radiative fluxes (Xie and Liu 2013). The XL13 method was applied to five SURFRAD sites to calculate cloud fractions at 1-min temporal resolution, which were then averaged into 15-min values and compared with MODIS cloud-fraction retrievals.
b. Satellite data
The 1-km (at nadir) spatial resolution MODIS cloud mask is the basis of many MODIS cloud products. It gives the likelihood information for a given pixel that is obstructed by clouds in four classes: confident clear, probably clear, uncertain–probably cloudy, or cloudy. The latter two classes are labeled as cloudy when calculating cloud fractions (Platnick et al. 2003). The MODIS level-2 cloud product generates cloud fractions at 5-km resolution by calculating the proportion of cloudy pixels from 25-pixel cloud mask groupings (Menzel et al. 2008). These cloud fractions are stored in files called MOD06 (Terra) and MYD06 (Aqua).
Cloud fractions from MOD06 and MYD06 (collection 5.1) were used in this research. Cloud-fraction data within a 30-km radius around the station were averaged as in many studies comparing satellite and surface cloud products (Kotarba 2009; Li et al. 2004; Rossow et al. 1993). Data were rejected when they were near the edge of a scan because the 30-km radius circle could cross the edge of the MODIS granule. The numbers of valid samples at each site and for all five sites are shown in Table 1.
Locations and sample numbers of the five SURFRAD sites (BON, DRA, GWN, PSU, and TBL) over 1 yr.
3. Results
As shown in Table 2, there was not much difference between the average MODIS observations and the surface TSI and L06 observations. XL13 observed lower average cloud fractions when compared with the others, which was consistent with the conclusion that XL13’s method estimate had overall smaller cloud fractions relative to L06’s method, as drawn by Xie et al. (2014). For total data from the five SURFRAD sites, MODIS overestimated by 2.34% and 3.50% relative to the mean cloud fraction observed by TSI and calculated by the L06 method, respectively. Figure 2 shows that these four cloud-fraction datasets produced nearly the same frequency distribution at the five SURFRAD sites. Moreover, the frequency distribution conformed to each site’s climate regime. The XL13 method observed more clear-sky cases at the DRA site and at Table Mesa in Boulder, Colorado (TBL), which may be the reason for the underestimation by the XL13 method. There were slight discrepancies at the TBL site. It should be mentioned that there have been some problems with the R/B ratio settings that separate clear sky, thin clouds, and opaque clouds at the TBL site; TSI observes a larger cloud fraction at the TBL site.
Statistical results of comparisons between any two of the cloud fractions observed by TSI, the cloud fractions calculated by the L06 and XL13 methods, and the matched instantaneous MODIS-derived cloud fraction in a 30-km radius around the observation site. For each entry, the three values that are separated by slashes are the correlation coeffiicent, the bias (%), and the standard deviation (%).
Frequency distribution of cloud fractions derived from MODIS, TSI, and the L06 and XL13 methods for the five SURFRAD sites (BON, DRA, GWN, PSU, and TBL).
Citation: Journal of Applied Meteorology and Climatology 54, 5; 10.1175/JAMC-D-14-0206.1
Frequency distribution of cloud fractions derived from MODIS, TSI, and the L06 and XL13 methods for the five SURFRAD sites (BON, DRA, GWN, PSU, and TBL).
Citation: Journal of Applied Meteorology and Climatology 54, 5; 10.1175/JAMC-D-14-0206.1
Frequency distribution of cloud fractions derived from MODIS, TSI, and the L06 and XL13 methods for the five SURFRAD sites (BON, DRA, GWN, PSU, and TBL).
Citation: Journal of Applied Meteorology and Climatology 54, 5; 10.1175/JAMC-D-14-0206.1
However, this does not mean that MODIS observations showed good agreement with surface observations in every case. Figure 3 shows that the scatterplots between MODIS and two surface SW-derived cloud fractions (L06 and XL13) are more discrete than those between the cloud fractions calculated by L06 and XL13. The relationships between MODIS and the two surface SW-derived cloud fractions (the left two columns) are weaker than that between the two surface SW-derived cloud-fraction datasets (Table 2). The correlation coefficients between the L06- and XL13-calculated cloud fractions at the five sites were approximately 0.95, whereas the correlations between MODIS and the various surface cloud-fraction estimates ranged from 0.8 to 0.91. Correlations between any two of the three datasets for any one station were significant (p < 0.001). There is respectable evidence that MODIS over- or underestimates cloud fractions even close to 90%. Because the L06 method is empirically derived from the relationship between the normalized diffuse cloud effect and the cloud fraction from TSI, the cloud fractions from L06 and TSI are generally in good relationships. L06-calculated cloud fractions may depend on TSI-observed cloud fractions. Thus, scatterplots and correlations between TSI and other cloud fractions are not shown in Fig. 3 (and later in Fig. 5) to avoid redundancy. Two independent cloud fractions (L06 and XL13) were compared in these two figures.
Scatterplots of three cloud-fraction datasets for the five SURFRAD sites: BON, DRA, GWN, PSU, and TBL. Shown are scatterplots of (left) L06- and (center) XL13-calculated cloud fraction as a function of measured MODIS-cloud fraction, respectively, and (right) XL13-calculated cloud fraction as a function of L06-calculated cloud fraction.
Citation: Journal of Applied Meteorology and Climatology 54, 5; 10.1175/JAMC-D-14-0206.1
Scatterplots of three cloud-fraction datasets for the five SURFRAD sites: BON, DRA, GWN, PSU, and TBL. Shown are scatterplots of (left) L06- and (center) XL13-calculated cloud fraction as a function of measured MODIS-cloud fraction, respectively, and (right) XL13-calculated cloud fraction as a function of L06-calculated cloud fraction.
Citation: Journal of Applied Meteorology and Climatology 54, 5; 10.1175/JAMC-D-14-0206.1
Scatterplots of three cloud-fraction datasets for the five SURFRAD sites: BON, DRA, GWN, PSU, and TBL. Shown are scatterplots of (left) L06- and (center) XL13-calculated cloud fraction as a function of measured MODIS-cloud fraction, respectively, and (right) XL13-calculated cloud fraction as a function of L06-calculated cloud fraction.
Citation: Journal of Applied Meteorology and Climatology 54, 5; 10.1175/JAMC-D-14-0206.1
Monthly cloud-fraction averages from the four datasets at the five sites are shown in Fig. 4. XL13 has a lower monthly average overall. It seems that MODIS overestimates cloud fractions in winter. There is a known MODIS cloud-mask problem when detecting cloud over ice and snow. MODIS may observe larger cloud fractions over such areas (Kotarba 2009). The overestimation in winter is evident except at the Goodwin Creek (GWN) site in Mississippi, which is the most southerly of the five sites (latitude = 34.24°N), where snow cover is small. There is another explanation that may account for the overestimation in winter. Cirrus daytime amounts are large in North America during December–February (Warren et al. 1986). The MODIS cloud mask is sensitive to cirrus cloud, but ground-based cloud-detecting methods have some difficulty in observing cirrus cloud. The capabilities and quality of ground-based methods in detecting cirrus clouds depend on the algorithms applied as to how clear sky is detected independent of atmospheric aerosols and solar zenith angle. Therefore, TSI underestimates cloud fraction in winter.
Monthly average cloud fractions for the five SURFRAD sites (BON, DRA, GWN, PSU, and TBL) over 1 yr.
Citation: Journal of Applied Meteorology and Climatology 54, 5; 10.1175/JAMC-D-14-0206.1
Monthly average cloud fractions for the five SURFRAD sites (BON, DRA, GWN, PSU, and TBL) over 1 yr.
Citation: Journal of Applied Meteorology and Climatology 54, 5; 10.1175/JAMC-D-14-0206.1
Monthly average cloud fractions for the five SURFRAD sites (BON, DRA, GWN, PSU, and TBL) over 1 yr.
Citation: Journal of Applied Meteorology and Climatology 54, 5; 10.1175/JAMC-D-14-0206.1
Figure 5 shows the correlations between any two of the MODIS (L06 and XL13) three cloud-fraction datasets in different months. Clearly, the correlation coefficient between the cloud fraction calculated by the L06 method and XL13 method is relatively high throughout the year. However, the correlation coefficient between the MODIS-derived cloud fraction and surface observations is generally low in winter and summer.
Monthly correlations between any two of the MODIS, L06, and XL13 three cloud-fraction datasets for the five SURFRAD sites (BON, DRA, GWN, PSU, and TBL). Correlations for each month and each site are significant (p < 0.005).
Citation: Journal of Applied Meteorology and Climatology 54, 5; 10.1175/JAMC-D-14-0206.1
Monthly correlations between any two of the MODIS, L06, and XL13 three cloud-fraction datasets for the five SURFRAD sites (BON, DRA, GWN, PSU, and TBL). Correlations for each month and each site are significant (p < 0.005).
Citation: Journal of Applied Meteorology and Climatology 54, 5; 10.1175/JAMC-D-14-0206.1
Monthly correlations between any two of the MODIS, L06, and XL13 three cloud-fraction datasets for the five SURFRAD sites (BON, DRA, GWN, PSU, and TBL). Correlations for each month and each site are significant (p < 0.005).
Citation: Journal of Applied Meteorology and Climatology 54, 5; 10.1175/JAMC-D-14-0206.1
4. Discussions
a. Instantaneous case illustration
The objective of this research is to determine the bias between MODIS and SURFRAD cloud fractions caused by FOV mismatch. To do this, it is necessary to reduce the surface effects as much as possible. The GWN site may have few surface effects (with few snow-cover days). Every instance when a large bias existed between MODIS and the surface-derived cloud fraction at a GWN site was carefully inspected.
At the GWN site, 71% of MODIS–TSI differences were within the range from −10% to +10% and 83% were within the range from −20% to +20% (Table 3). However, there were credible instances in which the biases exceeded 60%. Cases when MODIS observed a larger cloud fraction were more numerous than cases when TSI observed a larger cloud fraction. Further study identified several reasons that may account for nearly all the large-bias cases (|MODIS − TSI| > 30%).
Numbers of cases for 10%-wide cloud-amount ranges of MODIS–TSI at the GWN site during 2007.
The first reason comes down to a difference between the sensitivity of the measurement methods. The TSI software sets two fixed R/B ratios to separate clear-sky, thin-cloud, and opaque cloud areas. For some thin-cirrus cases, the R/B ratio of the cirrus-cloud area in the TSI image is not large enough to cause the area to be classified as cloudy, whereas MODIS is sensitive to thin cirrus, thereby generating a lower estimate of TSI. On hazy days, because of the strong forward scattering caused by aerosols and haze, TSI always overestimates the cloud fraction (Long 2010).
The second reason is due to the different observation perspectives of TSI and MODIS when partial cloud is present, usually when small convective clouds or broken clouds exist in the sky. Figure 6 shows that cloud albedos are relatively higher when large discrepancies exist between cloud fractions derived from MODIS and the L06 method. Thick clouds such as cumulus clouds usually have higher cloud albedos. This corresponds well to the second reason. For almost clear and overcast cases, little difference is normally expected between satellite and surface observations. However, for partial-cloud cases, the cloud fraction derived from surface observations is strongly affected by the vertical/horizontal cloud distribution (Kassianov et al. 2005); it tends to be overestimated when the cloud sides are included. Satellite view zenith angle (VZA) has a severe impact on the MODIS cloud fraction (Maddux et al. 2010). As the VZA becomes larger, the clear space between clouds decreases because the vertical thickness of clouds can be viewed from a satellite perspective. In addition, larger VZA entails a larger pixel size and a longer observation pathlength, which combine to increase the cloud fraction with VZA. Therefore, cases with small convective clouds may lead to large differences between MODIS and surface cloud fractions, with the degree of difference depending on the VZA. Convective clouds occur more frequently in summer, and therefore the relationship between MODIS and surface observations is relatively weaker in summer (Fig. 5). Figure 7 shows that the MODIS cloud fraction increases with the VZA relative to surface observations; on average, MODIS overestimates at VZA > 40° and underestimates at VZA < 20°. Overestimation at large VZA is easy to explain from above. The underestimation at small VZA can be explained by the example in Fig. 8. The overestimation at large VZAs is relatively greater than the underestimation at small VZAs, so the average of the cloud fraction at all VZAs may result in an overestimation of MODIS.
Average cloud albedo for different cloud-fraction-discrepancy (MODIS − L06) cases at the GWN site (2001–13). Cloud albedos are estimated by the XL13 method.
Citation: Journal of Applied Meteorology and Climatology 54, 5; 10.1175/JAMC-D-14-0206.1
Average cloud albedo for different cloud-fraction-discrepancy (MODIS − L06) cases at the GWN site (2001–13). Cloud albedos are estimated by the XL13 method.
Citation: Journal of Applied Meteorology and Climatology 54, 5; 10.1175/JAMC-D-14-0206.1
Average cloud albedo for different cloud-fraction-discrepancy (MODIS − L06) cases at the GWN site (2001–13). Cloud albedos are estimated by the XL13 method.
Citation: Journal of Applied Meteorology and Climatology 54, 5; 10.1175/JAMC-D-14-0206.1
Mean cloud fractions for different MODIS sensor zenith angles at the GWN site (2001–13). Sensor zenith angle is approximated to the closest 5° value from 5° to 60°.
Citation: Journal of Applied Meteorology and Climatology 54, 5; 10.1175/JAMC-D-14-0206.1
Mean cloud fractions for different MODIS sensor zenith angles at the GWN site (2001–13). Sensor zenith angle is approximated to the closest 5° value from 5° to 60°.
Citation: Journal of Applied Meteorology and Climatology 54, 5; 10.1175/JAMC-D-14-0206.1
Mean cloud fractions for different MODIS sensor zenith angles at the GWN site (2001–13). Sensor zenith angle is approximated to the closest 5° value from 5° to 60°.
Citation: Journal of Applied Meteorology and Climatology 54, 5; 10.1175/JAMC-D-14-0206.1
An example observed at 1055 LT 26 Jul 2007. (a) MODIS true-color image for the study region. The circle represents the 30-km radius around the station. (b) MODIS cloud-mask results. Gray indicates cloud cover is not determined, white indicates cloudy, red indicates probably cloudy, aquamarine indicates probably clear, and green indicates a confident clear area. (c) MODIS granule image with a rectangle showing the study region. (d) TSI true-color image. (e) TSI cloud-identification image. TSI images are the nearest MODIS passing-time images taken from the SURFRAD website at 1100 LT. The MODIS true-color image was generated using 0.659-, 0.555-, and 0.470-μm radiances, shown in red, green, and blue, respectively. The TSI true-color and TSI cloud-identification images were downloaded from the SURFRAD website. Because TSI images are provided hourly on the website, the nearest MODIS passing-time images were chosen.
Citation: Journal of Applied Meteorology and Climatology 54, 5; 10.1175/JAMC-D-14-0206.1
An example observed at 1055 LT 26 Jul 2007. (a) MODIS true-color image for the study region. The circle represents the 30-km radius around the station. (b) MODIS cloud-mask results. Gray indicates cloud cover is not determined, white indicates cloudy, red indicates probably cloudy, aquamarine indicates probably clear, and green indicates a confident clear area. (c) MODIS granule image with a rectangle showing the study region. (d) TSI true-color image. (e) TSI cloud-identification image. TSI images are the nearest MODIS passing-time images taken from the SURFRAD website at 1100 LT. The MODIS true-color image was generated using 0.659-, 0.555-, and 0.470-μm radiances, shown in red, green, and blue, respectively. The TSI true-color and TSI cloud-identification images were downloaded from the SURFRAD website. Because TSI images are provided hourly on the website, the nearest MODIS passing-time images were chosen.
Citation: Journal of Applied Meteorology and Climatology 54, 5; 10.1175/JAMC-D-14-0206.1
An example observed at 1055 LT 26 Jul 2007. (a) MODIS true-color image for the study region. The circle represents the 30-km radius around the station. (b) MODIS cloud-mask results. Gray indicates cloud cover is not determined, white indicates cloudy, red indicates probably cloudy, aquamarine indicates probably clear, and green indicates a confident clear area. (c) MODIS granule image with a rectangle showing the study region. (d) TSI true-color image. (e) TSI cloud-identification image. TSI images are the nearest MODIS passing-time images taken from the SURFRAD website at 1100 LT. The MODIS true-color image was generated using 0.659-, 0.555-, and 0.470-μm radiances, shown in red, green, and blue, respectively. The TSI true-color and TSI cloud-identification images were downloaded from the SURFRAD website. Because TSI images are provided hourly on the website, the nearest MODIS passing-time images were chosen.
Citation: Journal of Applied Meteorology and Climatology 54, 5; 10.1175/JAMC-D-14-0206.1
Figure 8 shows a case observed at the GWN site on 26 July 2007 from the Terra satellite. The satellite overpass time was 1055 LT. Figure 7a shows the MODIS true-color image for the study region. Figure 7c shows the MODIS granule image. The VZA is small with an angle of 6.35°, so the pixel size is also relatively small while the satellite resolution is relatively high. As shown in Fig. 7b, in a cloud mask image, the white area represents the cloudy area, and the red area probably represents the cloudy area. These two areas add up to the cloud fraction calculated by MODIS, which is 23.5%. Figure 7d shows the TSI true-color image observed at 1100 LT. Some cumulus clouds are visible in the image. Figure 7e shows the TSI identification image; the white area is the final estimate of the cloudy area. Some cloud edges were counted as cloudy. The cumulus clouds are low, and their horizontal distribution is not very wide, but they occupy a large fraction of solid angle. The cloud fraction calculated by TSI was 69.4%. The TSI-observed cloud fraction is sensitive to the vertical cloud distribution while MODIS-observed cloud fraction is independent of the vertical cloud structures (Kassianov et al. 2005), so in some small viewing angle cases, TSI observes a higher cloud fraction when compared with MODIS.
These two reasons accounted for many of the discrepancies. MODIS may observe a smaller or larger cloud fraction than surface observations for different cases; these discrepancies will balance out when calculating the average state.
b. Interannual variations of cloud fraction at the GWN site
The TSI-measured and L06 cloud fractions were found to be in good agreement. The cloud fraction calculated using the L06 method has a long dataset in SURFRAD. Therefore, the annual averages of diurnal Aqua and Terra MODIS cloud fractions were calculated for a 30-km radius around the GWN site. A 15-min-average cloud fraction matching the satellite passing time was calculated using the L06 method, and the annual average of daily mean L06-derived cloud fraction was then calculated.
As shown in Fig. 9, the interannual variations of cloud fraction in these five annual average datasets were similar except for the annual average calculated from MODIS on board the Aqua satellite. The Aqua-MODIS cloud fraction was always relatively higher in the years when the annual mean cloud fraction was generally lower. Figure 10 shows the monthly average cloud fraction for the five datasets. It seems that MODIS overestimates cloud fraction in summer and that the overestimation is larger from Aqua than from Terra. Because both Aqua and Terra satellites operate on the World Reference System 2 (WRS-2) grid, the VZAs of Aqua and Terra were nearly the same overall. Therefore, the overestimation by Aqua can be explained by the fact that convection is stronger in the afternoon as detected by Aqua. Because there are no large differences between the interannual variations in the annual mean of the 15-min-average SW-derived cloud fraction centered on Aqua and Terra overpass time, it can be concluded that during the afternoon when Aqua overpasses, there are more convective clouds in the sky, and when the VZA is large at the same time, the result will be a strong overestimation by Aqua MODIS.
Interannual variations of cloud fractions calculated from MODIS and by the L06 method at the GWN site from 2001 to 2013. MODIS-Terra and MODIS-Aqua values were derived from the daily instantaneous MODIS cloud fractions over the GWN site from the Terra and Aqua satellites. L06-Terra and L06-Aqua were calculated from the 15-min averages of L06-derived cloud fraction centered on the Terra and Aqua passing times. L06-Whole day was calculated from the daily average of L06-derived cloud fraction.
Citation: Journal of Applied Meteorology and Climatology 54, 5; 10.1175/JAMC-D-14-0206.1
Interannual variations of cloud fractions calculated from MODIS and by the L06 method at the GWN site from 2001 to 2013. MODIS-Terra and MODIS-Aqua values were derived from the daily instantaneous MODIS cloud fractions over the GWN site from the Terra and Aqua satellites. L06-Terra and L06-Aqua were calculated from the 15-min averages of L06-derived cloud fraction centered on the Terra and Aqua passing times. L06-Whole day was calculated from the daily average of L06-derived cloud fraction.
Citation: Journal of Applied Meteorology and Climatology 54, 5; 10.1175/JAMC-D-14-0206.1
Interannual variations of cloud fractions calculated from MODIS and by the L06 method at the GWN site from 2001 to 2013. MODIS-Terra and MODIS-Aqua values were derived from the daily instantaneous MODIS cloud fractions over the GWN site from the Terra and Aqua satellites. L06-Terra and L06-Aqua were calculated from the 15-min averages of L06-derived cloud fraction centered on the Terra and Aqua passing times. L06-Whole day was calculated from the daily average of L06-derived cloud fraction.
Citation: Journal of Applied Meteorology and Climatology 54, 5; 10.1175/JAMC-D-14-0206.1
Monthly average cloud fraction derived from MODIS and by the L06 method at the GWN site during 2003–13.
Citation: Journal of Applied Meteorology and Climatology 54, 5; 10.1175/JAMC-D-14-0206.1
Monthly average cloud fraction derived from MODIS and by the L06 method at the GWN site during 2003–13.
Citation: Journal of Applied Meteorology and Climatology 54, 5; 10.1175/JAMC-D-14-0206.1
Monthly average cloud fraction derived from MODIS and by the L06 method at the GWN site during 2003–13.
Citation: Journal of Applied Meteorology and Climatology 54, 5; 10.1175/JAMC-D-14-0206.1
5. Conclusions
Both satellite and surface measurements capture the main features of cloud fractions. The frequency distributions and averages of the cloud fraction derived from MODIS, TSI, and the L06 and XL13 methods were relatively consistent at the five SURFRAD sites. The XL13 method observed lower cloud fractions in general. The correlation coefficients between instantaneous MODIS and surface-derived cloud fraction were nearly 0.90 for total data, which was not as good as that between the L06-calculated and XL13-calculated cloud fraction (0.95). In a considerable number of cases, MODIS misestimates a large portion of cloud fractions relative to surface measurements.
The correlation between MODIS and surface cloud fraction is relatively low in winter and summer. In winter, the biases may be largely due to sensor sensitivity, such as the MODIS cloud-mask problem when detecting clouds over snow and ice and the TSI detecting error when thin cirrus and haze appear. In summer, the biases are more strongly determined by the different FOVs of the satellite and surface observations, especially when small convective clouds exist in the FOVs.
The different FOVs of satellite and surface clouds observations affected the consistency of these two cloud-fraction datasets. For nearly overcast and clear cases, the FOV mismatch did not have much effect, whereas for partial cloud, especially with some convective or broken clouds in the FOVs, the discrepancy between satellite and surface cloud-fraction data could be very large, and on average, the sign and magnitude of the discrepancy depended on the VZA of the satellite. So an appropriate correction is strongly recommended for the MODIS satellite-based cloud algorithm to remove the influences caused by different MODIS VZAs.
The long-term average of diurnal transitory MODIS cloud-fraction observations and the short-time average surface cloud fractions matching the MODIS overpass time agreed well with regard to the long-term trend of the whole-day average of surface cloud fraction. There are only some distinctions between long-term Aqua cloud fraction and the others. The Aqua satellite may observe more convective clouds and therefore overestimate the mean cloud fraction. Therefore, when calculating long cloud-fraction trends, the polar-orbiting satellite overpass time may have an effect, because if a satellite overpasses at a time of day that is prone to convective clouds, the result is more uncertainty between the satellite and surface cloud-fraction datasets.
Acknowledgments
This study was funded by the National Basic Research Program of China (2012CB955302) and the National Natural Science Foundation of China (41175126 and 91337111). MODIS satellite data were obtained from LAADS Web (http://ladsweb.nascom.nasa.gov/data/search.html). Surface shortwave and cloud-fraction data were downloaded from the SURFRAD ftp site (
REFERENCES
Augustine, J. A., J. J. DeLuisi, and C. N. Long, 2000: SURFRAD—A national surface radiation budget network for atmospheric research. Bull. Amer. Meteor. Soc., 81, 2341–2357, doi:10.1175/1520-0477(2000)081<2341:SANSRB>2.3.CO;2.
Augustine, J. A., G. B. Hodges, C. R. Cornwall, J. J. Michalsky, and C. I. Medina, 2005: An update on SURFRAD—The GCOS Surface Radiation budget network for the continental United States. J. Atmos. Oceanic Technol., 22, 1460–1472, doi:10.1175/JTECH1806.1.
Baker, M., 1997: Cloud microphysics and climate. Science, 276, 1072–1078, doi:10.1126/science.276.5315.1072.
Boers, R., M. J. de Haij, W. M. F. Wauben, H. K. Baltink, L. H. van Ulft, M. Savenije, and C. N. Long, 2010: Optimized fractional cloudiness determination from five ground-based remote sensing techniques. J. Geophys. Res., 115, D24116, doi:10.1029/2010JD014661.
Dürr, B., and R. Philipona, 2004: Automatic cloud amount detection by surface longwave downward radiation measurements. J. Geophys. Res., 109, D05201, doi:10.1029/2003JD004182.
Feister, U., H. Möller, T. Sattler, J. Shields, U. Görsdorf, and J. Güldner, 2010: Comparison of macroscopic cloud data from ground-based measurements using VIS/NIR and IR instruments at Lindenberg, Germany. Atmos. Res., 96, 395–407, doi:10.1016/j.atmosres.2010.01.012.
Frey, R. A., S. A. Ackerman, Y. Liu, K. I. Strabala, H. Zhang, J. R. Key, and X. Wang, 2008: Cloud detection with MODIS. Part I: Improvements in the MODIS cloud mask for collection 5. J. Atmos. Oceanic Technol., 25, 1057–1072, doi:10.1175/2008JTECHA1052.1.
Hahn, C. J., W. B. Rossow, and S. G. Warren, 2001: ISCCP cloud properties associated with standard cloud types identified in individual surface observations. J. Climate, 14, 11–28, doi:10.1175/1520-0442(2001)014<0011:ICPAWS>2.0.CO;2.
Henderson-Sellers, A., and K. McGuffie, 1990: Are cloud amounts estimated from satellite sensor and conventional surface-based observations related? Remote Sens., 11, 543–550, doi:10.1080/01431169008955038.
Kassianov, E., C. N. Long, and M. Ovtchinnikov, 2005: Cloud sky cover versus cloud fraction: Whole-sky simulations and observations. J. Appl. Meteor., 44, 86–98, doi:10.1175/JAM-2184.1.
Kotarba, A. Z., 2009: A comparison of MODIS-derived cloud amount with visual surface observations. Atmos. Res., 92, 522–530, doi:10.1016/j.atmosres.2009.02.001.
Li, Z., M. Cribb, F. Chang, and A. Trishchenko, 2004: Validation of MODIS-retrieved cloud fractions using whole sky imager measurements at the three ARM sites. Proc. 14th Atmospheric Radiation Measurement (ARM) Science Team Meeting, Albuquerque, NM, Atmospheric Radiation Measurement (ARM) Program, 22–26. [Available online at https://www.arm.gov/publications/proceedings/conf14/extended_abs/li1-z.pdf.]
Long, C. N., 2010: Correcting for circumsolar and near-horizon errors in sky cover retrievals from sky images. Open Atmos. Sci. J., 4, 45–52, doi:10.2174/1874282301004010045.
Long, C. N., T. P. Ackerman, K. L. Gaustad, and J. N. S. Cole, 2006a: Estimation of fractional sky cover from broadband shortwave radiometer measurements. J. Geophys. Res., 111, D11204, doi:10.1029/2005JD006475.
Long, C. N., J. Sabburg, J. Calbó, and D. Pages, 2006b: Retrieving cloud characteristics from ground-based daytime color all-sky images. J. Atmos. Oceanic Technol., 23, 633–652, doi:10.1175/JTECH1875.1.
Maddux, B., S. Ackerman, and S. Platnick, 2010: Viewing geometry dependencies in MODIS cloud products. J. Atmos. Oceanic Technol., 27, 1519–1528, doi:10.1175/2010JTECHA1432.1.
Menzel, W. P., and Coauthors, 2008: MODIS global cloud-top pressure and amount estimation: Algorithm description and results. J. Appl. Meteor. Climatol., 47, 1175–1198, doi:10.1175/2007JAMC1705.1.
Min, Q., T. Wang, C. N. Long, and M. Duan, 2008: Estimating fractional sky cover from spectral measurements. J. Geophys. Res., 113, D20208, doi:10.1029/2008JD010278.
Pfister, G., R. McKenzie, J. Liley, A. Thomas, B. Forgan, and C. N. Long, 2003: Cloud coverage based on all-sky imaging and its impact on surface solar irradiance. J. Appl. Meteor., 42, 1421–1434, doi:10.1175/1520-0450(2003)042<1421:CCBOAI>2.0.CO;2.
Platnick, S., M. D. King, S. A. Ackerman, W. P. Menzel, B. A. Baum, J. C. Riédi, and R. A. Frey, 2003: The MODIS cloud products: Algorithms and examples from Terra. IEEE Trans. Geosci. Remote Sens., 41, 459–473, doi:10.1109/TGRS.2002.808301.
Rossow, W. B., A. W. Walker, and L. C. Garder, 1993: Comparison of ISCCP and other cloud amounts. J. Climate, 6, 2394–2418, doi:10.1175/1520-0442(1993)006<2394:COIAOC>2.0.CO;2.
Slater, D., C. Long, and T. Tooman, 2001: Total sky imager/whole sky imager cloud fraction comparison. Proc. 11th ARM Science Team Meeting, Atlanta, GA, Atmospheric Radiation Measurement (ARM) Program. [Available online at https://www.arm.gov/publications/proceedings/conf11/extended_abs/slater_dw.pdf.]
Stubenrauch, C., and Coauthors, 2013: Assessment of global cloud datasets from satellites: Project and database initiated by the GEWEX Radiation Panel. Bull. Amer. Meteor. Soc., 94, 1031–1049, doi:10.1175/BAMS-D-12-00117.1.
Warren, S. G., C. J. Hahn, J. London, R. M. Chervin, and R. L. Jenne, 1986: Global distribution of total cloud cover and cloud type amounts over land. NCAR Tech. Note NCAR/TN-273+STR, doi:10.5065/D6GH9FXB.
Wu, W., Y. Liu, M. P. Jensen, T. Toto, M. J. Foster, and C. N. Long, 2014: A comparison of multiscale variations of decade-long cloud fractions from six different platforms over the Southern Great Plains in the United States. J. Geophys. Res. Atmos., 119, 3438–3459, doi:10.1002/2013JD019813.
Xie, Y., and Y. Liu, 2013: A new approach for simultaneously retrieving cloud albedo and cloud fraction from surface-based shortwave radiation measurements. Environ. Res. Lett., 8, 044023, doi:10.1088/1748-9326/8/4/044023.
Xie, Y., Y. Liu, C. N. Long, and Q. Min, 2014: Retrievals of cloud fraction and cloud albedo from surface-based shortwave radiation measurements: A comparison of 16 year measurements. J. Geophys. Res. Atmos., 119, 8925–8940, doi:10.1002/2014JD021705.
