SEARCH

SEARCH BY CITATION

Keywords:

  • aerosols;
  • clouds;
  • radiation;
  • surface energy budget

Abstract

  1. Top of page
  2. Abstract
  3. 1. Introduction
  4. 2. Data Set
  5. 3. Results
  6. 4. Uncertainty in Inputs Used for Surface Downward Longwave Irradiance Computations
  7. 5. Comparison With Surface Observations
  8. 6. Discussion
  9. 7. Conclusions
  10. Appendix A:: CALIOP, CPR, and MODIS Derived Cloud Properties and Flux Computations With Them
  11. Acknowledgments
  12. References
  13. Supporting Information

[1] One year of instantaneous top-of-atmosphere (TOA) and surface shortwave and longwave irradiances are computed using cloud and aerosol properties derived from instruments on the A-Train Constellation: the Cloud-Aerosol Lidar with Orthogonal Polarization (CALIOP) on the Cloud-Aerosol Lidar and Infrared Pathfinder Satellite Observation (CALIPSO) satellite, the CloudSat Cloud Profiling Radar (CPR), and the Aqua Moderate Resolution Imaging Spectrometer (MODIS). When modeled irradiances are compared with those computed with cloud properties derived from MODIS radiances by a Clouds and the Earth's Radiant Energy System (CERES) cloud algorithm, the global and annual mean of modeled instantaneous TOA irradiances decreases by 12.5 W m−2 (5.0%) for reflected shortwave and 2.5 W m−2 (1.1%) for longwave irradiances. As a result, the global annual mean of instantaneous TOA irradiances agrees better with CERES-derived irradiances to within 0.5W m−2 (out of 237.8 W m−2) for reflected shortwave and 2.6W m−2 (out of 240.1 W m−2) for longwave irradiances. In addition, the global annual mean of instantaneous surface downward longwave irradiances increases by 3.6 W m−2 (1.0%) when CALIOP- and CPR-derived cloud properties are used. The global annual mean of instantaneous surface downward shortwave irradiances also increases by 8.6 W m−2 (1.6%), indicating that the net surface irradiance increases when CALIOP- and CPR-derived cloud properties are used. Increasing the surface downward longwave irradiance is caused by larger cloud fractions (the global annual mean by 0.11, 0.04 excluding clouds with optical thickness less than 0.3) and lower cloud base heights (the global annual mean by 1.6 km). The increase of the surface downward longwave irradiance in the Arctic exceeds 10 W m−2 (∼4%) in winter because CALIOP and CPR detect more clouds in comparison with the cloud detection by the CERES cloud algorithm during polar night. The global annual mean surface downward longwave irradiance of 345.4 W m−2 is estimated by combining the modeled instantaneous surface longwave irradiance computed with CALIOP and CPR cloud profiles with the global annual mean longwave irradiance from the CERES product (AVG), which includes the diurnal variation of the irradiance. The estimated bias error is −1.5 W m−2 and the uncertainty is 6.9 W m−2. The uncertainty is predominately caused by the near-surface temperature and column water vapor amount uncertainties.

1. Introduction

  1. Top of page
  2. Abstract
  3. 1. Introduction
  4. 2. Data Set
  5. 3. Results
  6. 4. Uncertainty in Inputs Used for Surface Downward Longwave Irradiance Computations
  7. 5. Comparison With Surface Observations
  8. 6. Discussion
  9. 7. Conclusions
  10. Appendix A:: CALIOP, CPR, and MODIS Derived Cloud Properties and Flux Computations With Them
  11. Acknowledgments
  12. References
  13. Supporting Information

[2] The radiation budget at the Earth's surface plays a critical role in the energy and water cycles of the planet. The global mean net surface irradiance is balanced by the surface latent and sensible heat fluxes and ocean heating rate [Wong et al., 2006]. In addition, the radiative net energy deposition in the atmosphere and vertical and horizontal profiles of the energy deposition determine the dynamics in the atmosphere. Understanding how the radiation at the surface, within the atmosphere and top-of-atmosphere change in response to climate forcing also requires an understanding of cloud, water vapor, and surface feedback processes. Sustained global observations of the radiation budget at different levels in the atmosphere and the associated atmospheric and surface properties are thus critical [Wielicki et al., 1995].

[3] Unfortunately, direct observations of surface irradiance are currently available only over a limited number of ground sites over land, and a handful of offshore and island locations. Therefore, a global estimate of the surface radiation budget must be determined indirectly through radiative transfer model calculations initialized using satellite-derived cloud and aerosol properties and meteorological data from assimilation models. Until the launches of the Cloud-Aerosol Lidar and Infrared Pathfinder Satellite Observation (CALIPSO) [Winker et al., 2010] and CloudSat [Stephens et al., 2008] missions, global cloud properties have only been available from passive satellite instruments. For example, Zhang et al. [2004] used data from the International Satellite Cloud Climatology Project (ISCCP[Rossow and Schiffer, 1991]) to estimate global surface irradiances. The surface irradiance estimate by Zhang et al. [2004] relied on cloud retrievals derived from operational weather satellites (geostationary and polar orbiting satellites), with limited calibration accuracy. The surface irradiance estimated in the Clouds and the Earth's Radiant Energy System (CERES) project relies on Moderate Resolution Imaging Spectrometer (MODIS) derived cloud and aerosol properties and also uses CERES-derived top-of-atmosphere (TOA) irradiance as a constraint [Charlock et al., 2006]. Despite the relative success in using such data sets, the accuracies of surface radiation estimates are constrained by the limitations in accurately retrieving all of the necessary cloud parameters needed for the radiative transfer models.

[4] As pointed out by Trenberth et al. [2009], one challenge in estimating the surface radiation budget is properly determining the cloud base height because, in passive satellite instrument retrievals, it is indirectly derived from the cloud top height. In addition, the cloud top height retrieval itself contains errors because it is usually determined by a combination of the effective cloud top temperature and vertical temperature profile. Furthermore, the cloud base height estimate used in irradiance computations relies on climatological vertical cloud profiles [Zhang et al., 2004] or an empirical relationship based on the cloud physical thickness, cloud top temperature, and optical thickness [Minnis et al., 2010, 2011]. Both climatological cloud profiles and the empirical relationship were derived from limited sets of data and therefore cannot provide an accurate cloud base for all regions and seasons.

[5] CALIPSO and CloudSat, launched in 2006 as part of NASA's A-train satellite constellation, provide detailed information on cloud and aerosol vertical profiles from tropics to the poles. When cloud vertical profiles derived from Cloud-Aerosol Lidar with Orthogonal Polarization (CALIOP [Hunt et al., 2009]) and Cloud Profiling Radar (CPR [Im et al., 2005]) data are used in surface irradiance computations, uncertainty in the surface irradiance, especially in the downward longwave irradiance, is expected to decrease, provided that the vertical cloud profile taken over their ground track represents a global mean cloud profile (i.e., the sampling error does not exceed modeling errors).

[6] In this paper, we investigate how CALIOP- and CPR-derived cloud vertical profiles can improve irradiance computations compared with those in which only MODIS-derived cloud properties are used. Four different modeled irradiance sets are considered, as explained in section 2. Comparisons of modeled TOA irradiances with those derived empirically from CERES broadband radiance measurements, and surface irradiances computed with and without CALIOP- and CPR-derived properties are presented in section 3. Uncertainties in the inputs used for the irradiance computations are estimated in section 4. Comparisons between modeled and observed surface longwave irradiance are presented in section 5. Last, how CALIOP- and CPR-derived cloud properties improve irradiance computations is discussed in section 6.

2. Data Set

  1. Top of page
  2. Abstract
  3. 1. Introduction
  4. 2. Data Set
  5. 3. Results
  6. 4. Uncertainty in Inputs Used for Surface Downward Longwave Irradiance Computations
  7. 5. Comparison With Surface Observations
  8. 6. Discussion
  9. 7. Conclusions
  10. Appendix A:: CALIOP, CPR, and MODIS Derived Cloud Properties and Flux Computations With Them
  11. Acknowledgments
  12. References
  13. Supporting Information

[7] In order to assess the irradiance estimate improvement, we use four sets of surface irradiance computations extracted from three different data products in this study. These four sets of modeled irradiances use nearly identical radiative transfer models. One data product is the CCCM (CALIPSO CloudSat CERES and MODIS, Edition B1 [Kato et al., 2010], hereafter CCCM_CC). CCCM contains cloud and aerosol vertical profiles derived from CALIOP (Version 3) and CPR (Revision 4), as well as cloud properties derived from MODIS radiances by a CERES cloud algorithm [Minnis et al., 2011; Trepte et al., 2010]. It also contains the vertical irradiance profile computed with cloud properties determined using a combination of CALIOP, CPR, and MODIS data. In addition, it contains the TOA and surface irradiances and irradiances at three atmospheric pressure levels (500, 200, and 70 hPa) computed with cloud properties derived from MODIS only (hereafter this irradiance set is referred as CCCM_MODISonly). The algorithm that uses only MODIS radiances is hereafter referred to as the B1 cloud algorithm. By comparing CCCM_CC and CCCM_MODISonly computed over the CALIOP and CPR ground track, we assess the improvement of instantaneous irradiances due to inclusion of the CALIOP- and CPR-derived cloud vertical profiles in the computations. The third set of modeled irradiances is from the CERES standard product Cloud and Radiative Swath (CRS, Terra edition 2G) [Charlock et al., 2006], which uses only MODIS-derived cloud [Minnis et al., 2011] properties (hereafter this irradiance set is referred as CRS). Because MODIS covers the entire globe daily, we can assess the error caused by the nadir view sampling by subsetting irradiances computed for nadir view CERES footprints and comparing them with irradiances computed for the full swath using CRS. The fourth set of modeled irradiances is from the CERES standard product AVG (Terra Edition 2C). AVG contains monthly mean gridded modeled TOA and surface irradiances. The monthly mean irradiance is computed from daily mean values that account for a diurnal cycle of cloud properties retrieved from geostationary satellites and temperature and humidity profiles from reanalysis [Young et al., 1998] (hereafter this irradiance set is referred as AVG). Therefore, we can assess the improvement of the global mean surface irradiance estimated with CALIOP- and CPR-derived cloud profiles with the use of these four sets.

[8] Modeled irradiances of CCCM used for this study are from January 2008 through December 2008, modeled irradiances of CRS are from January, April, July, and October 2008, and modeled irradiances of AVG are from January 2001 through December 2004. Note that the use of AVG data from a different year does not cause a significant problem, because we only use irradiances from AVG averaged annually and globally.

2.1. CALIPSO and CloudSat Merged Cloud Profiles and Irradiance Computations for CCCM_CC

[9] Because the method of integrating the CALIOP- and CPR-derived cloud masks is described in Kato et al. [2010], only a brief description is provided here. Cloud vertical profiles from CALIOP and CPR are initially merged into 1 km horizontal resolution vertical cloud profiles (Figure 1). Starting with a CALIOP-derived cloud profile, we add cloud boundaries from CPR in the following cases: (i) when CPR detects a cloud boundary more than 480 m above or below cloud boundaries detected by CALIOP, the CPR-derived boundary is inserted; (ii) when the CALIOP signal is completely attenuated by clouds (attenuation level), and the CPR-derived cloud base is lower than the attenuation level, the CPR-derived cloud base is used; (iii) otherwise, the attenuation level is used as the cloud base. As a result of these processes, CALIOP provides approximately 85% of cloud top heights and 77% of cloud base heights for the merged cloud profiles.

image

Figure 1. Schematic of a CERES footprint containing the CALIPSO and CloudSat ground track.

Download figure to PowerPoint

[10] The resulting merged cloud profiles are then collocated with CERES footprints, which are approximately 20 km in size (Figure 1). To maintain the horizontal resolution used in the original CALIPSO and CloudSat products, 1 km horizontal atmospheric columns that contain the same cloud vertical profiles are grouped together (cloud group). The cloud-grouping process is described in detail by Kato et al. [2010].

[11] Irradiance vertical profiles are computed for each cloud group by the use of a radiative transfer model (FLux model of CERES with k-distribution and correlated-k for Radiation (FLCKKR) [Fu and Liou, 1993; Fu et al., 1997; Kratz and Rose, 1999; Kato et al., 1999, 2005; Rose et al., 2006] with a two-stream approximation using the independent column approximation [Stephens et al., 1991]. The hierarchy of cloud optical property sources used in the irradiance computations and the details of the irradiance computations are explained in Appendix A. In brief, cloud properties derived from CALIOP, CPR, or MODIS are used in the computations. In retrieving cloud properties from MODIS radiances, the B1 cloud algorithm is forced to retrieve the uppermost cloud top effective height given by the collocated merged CALIOP and CPR cloud profile (hereafter, enhanced cloud algorithm, see Appendix A for detail) when a single-layer cloud is present in the pixel. As a consequence, the enhanced cloud algorithm uses a better cloud top effective temperature, which leads to a better estimate of the emission contribution in near-infrared (IR) channels. Note that when multilayer clouds are present (about 50% of cloudy cases [Kato et al., 2010]), the enhanced algorithm is the same as the B1 cloud algorithm. Therefore, the CALIOP- and CPR-derived cloud properties affect the clouds used in irradiance computations in two ways; first, by directly providing better cloud mask and profiles, and second, by improving cloud retrievals within the enhanced cloud algorithm.

[12] In the order of their use in the irradiance computations, the aerosol optical thickness sources are CALIOP, MYD04 [Remer et al., 2005], and the Model of Atmospheric Transport and Chemistry (MATCH [Collins et al., 2001]). The aerosol optical thickness is averaged over a CERES footprint. A CERES footprint could contain multiple vertical aerosol layers having different optical thicknesses, but there is no horizontal aerosol optical thickness variability within a given aerosol layer. CALIOP-derived aerosol optical thicknesses are averaged by excluding values with “the column optical depth aerosol” with an uncertainty of 99.99 (i.e., default value for cases when the CALIPSO extinction calculation failed). In addition, noise in the lidar signal can produce negative extinction values when aerosol loading is low and background noise is high. These negative values occasionally result in a small, negative column optical thickness. Although they are rare, large negative optical thickness (less than −0.1) can produce erroneous retrievals and are excluded in the averaging process. Aerosol optical properties, such as wavelength dependence of the aerosol optical thickness, asymmetry parameter, and single scattering albedo, are determined by assigning aerosol types based on Optical Properties of Aerosols and Clouds (OPAC) [Hess et al., 1998] and Tegen and Lacis [1996]. Aerosol types include small dust, large dust, sulfate, sea salt, soot, soluble particles, and insoluble particles. The aerosol type is chosen mostly based on MATCH, except for dust aerosols. When the CALIOP detects dust and polluted dust, large and small dust aerosol models, respectively, are used.

[13] Temperature and humidity profiles used in CCCM_CC, CCCM_MODISonly, and CRS irradiance computations are from the Goddard Earth Observing System (GEOS-5) Data Assimilation System reanalysis [Rienecker et al., 2008], while the profiles used in AVG irradiance computations are from GEOS-4 [Bloom et al., 2005]. The GEOS-4 and -5 temperature and relative humidity profiles have a temporal resolution of 6 h. Spatially, the profiles are regridded to 1° × 1° maps. Skin temperatures are from both GEOS-4 and GEOS-5 at a 3-hourly resolution, the native temporal resolution of GEOS-4 skin temperature, although the GEOS-5 product has a higher 1-hourly native resolution available. Gridded 6-hourly and 1° × 1° temperature and humidity profiles and the 3-hourly and 1° × 1° skin temperature are further linearly interpolated in space and time to the CERES footprint observation times and locations. Note that the effect on GEOS-5 and -4 temperature and humidity differences yields a global mean surface downward shortwave and longwave irradiances difference of −0.7 W m−2 and 1.2 W m−2, respectively. Because the longwave irradiance difference is smaller than the uncertainty discussed in section 4, we neglect the GEOS-5 and -4 differences in this study.

[14] CCCM_MODISonly irradiances are computed using cloud properties derived by the B1 cloud algorithm over the entire CERES footprint using MODIS radiances collocated with the CERES footprint. The B1 algorithm and irradiance computations are similar to those used for CRS, which are explained in section 2.2.

2.2. Irradiance Computations in CERES CRS and AVG Products

[15] The CRS product contains instantaneous modeled irradiances computed with MODIS-derived cloud properties. The CERES Ed2 cloud algorithm [Minnis et al., 2011], the precursor to the B1 algorithm, is used to derive cloud properties from MODIS 1 km resolution spectral radiances. A cloud within a 1 km MODIS pixel is assumed to be a horizontally uniform single-layer overcast cloud. Because the size of a CERES footprint is 20 km at nadir, there are more than 150 sets of retrieved cloud properties (retrieved from one out of two scan lines and one out of two pixels in a scan line, i.e., 25% of pixels within a footprint) within a CERES footprint. Cloud properties derived within a CERES footprint are averaged using the CERES instrument point spread function [Smith, 1994] as a weighting function. In averaging the cloud properties, two cloud top heights within a CERES footprint are retained and cloud properties of high and low clouds are averaged separately.

[16] AVG contains monthly gridded modeled irradiances computed with cloud properties derived from MODIS and 3-hourly geostationary satellites. Footprint-level cloud properties are gridded in 1° × 1° spatial grids and in 1-hourly temporal grids (hour boxes). Up to four cloud heights (cloud types) are retained for each hour box within a 1° × 1° grid box. Cloud properties for hour boxes other than those for the Aqua overpass time are derived from geostationary satellites. Both linear and logarithmic means of cloud optical thicknesses are computed for each cloud type. The distribution of cloud optical thickness expressed as a gamma distribution is estimated from the linear and logarithmic cloud optical thickness means [Barker 1996; Oreopoulos and Barker, 1999; Kato et al., 2005]. Once the distribution of cloud optical thickness is estimated for each cloud type, the gamma-weighted two-stream radiative transfer model is used to separately compute the shortwave irradiance vertical profile for four cloud types in AVG and two clouds types in CRS. A detailed description of the gamma-weighted two-stream radiative transfer model used for the irradiance computation is given by Kato et al. [2005]. The logarithmic mean optical thickness is used in the longwave irradiance computation with a modified two-stream approximation [Toon et al. 1989; Fu et al., 1997]. In addition, irradiance under a clear-sky condition is always computed for every grid box in AVG and for every footprint for CRS. The cloud base height, which largely influences the surface downward longwave irradiance in midlatitude and polar regions, is estimated by an empirical formula described by Minnis et al. [2011] for CCCM_MODISonly, CRS, and AVG.

[17] Other inputs to the two-stream models are ozone amount [Yang et al., 2000] and ocean spectral surface albedo from Jin et al. [2004], which are used for all four sets. Broadband land surface albedos are inferred from MODIS narrowband albedos [Moody et al., 2005] for CCCM_CC and CCCM_MODISonly and from the clear-sky TOA albedo derived from CERES measurements [Rutan et al., 2009] for CRS and AVG.

3. Results

  1. Top of page
  2. Abstract
  3. 1. Introduction
  4. 2. Data Set
  5. 3. Results
  6. 4. Uncertainty in Inputs Used for Surface Downward Longwave Irradiance Computations
  7. 5. Comparison With Surface Observations
  8. 6. Discussion
  9. 7. Conclusions
  10. Appendix A:: CALIOP, CPR, and MODIS Derived Cloud Properties and Flux Computations With Them
  11. Acknowledgments
  12. References
  13. Supporting Information

[18] Before investigating surface irradiances, a reasonable agreement of the modeled TOA irradiance with the CERES-derived irradiance by angular distribution models is a good consistency check of modeled irradiances. We therefore compare modeled TOA irradiances with CERES-derived irradiances and investigate how CALIOP- and CPR-derived cloud profiles improve the modeled TOA instantaneous irradiance.

3.1. TOA Irradiance

[19] Figure 2 shows the difference between modeled and CERES-derived shortwave and longwave monthly zonal mean irradiances for January and July 2008. Note that Ed3 CERES calibration constants and Ed2 angular distribution models [Loeb et al., 2005] are used to derive CERES irradiances. CCCM_CC TOA reflected shortwave irradiance is smaller than CCCM_MODISonly TOA reflected shortwave irradiance. As a result, CCCM_CC TOA reflected shortwave irradiances agree better with CERES-derived irradiances, particularly in the summer hemisphere. CALIOP- and CPR-derived cloud properties increase the monthly zonal mean TOA longwave for most latitudes (Figure 2). As a result, the CCCM_CC TOA longwave irradiances also agree better with the CERES-derived longwave irradiance. Table 1 summarizes the modeled TOA reflected shortwave and longwave irradiance differences from those derived from CERES-observed radiances. The global and annual mean improvement due to the use of CALIPSO- and CloudSat-derived cloud vertical profiles is 12.5 W m−2 (5.0%) for reflected shortwave and 2.5 W m−2 (1.1%) for longwave. The reason for a better agreement of the CCCM_CC TOA irradiance with CERES-derived irradiance is discussed in section 5.2 after the surface irradiance section because the reason for both TOA and surface irradiance improvements is primarily caused by the more accurate cloud vertical profile.

image

Figure 2. Monthly zonal TOA shortwave and longwave modeled irradiance differences from CERES-derived irradiances for (top) January 2008 and (bottom) July 2008. CC indicates the flux computed with CALIOP-, CPR-, and MODIS-derived cloud and aerosol properties. Shortwave irradiance is the daytime average of instantaneous irradiances, and the longwave irradiance is the day and nighttime average of instantaneous irradiances.

Download figure to PowerPoint

Table 1. Global Annual Mean (2008) of TOA Instantaneous Irradiance Difference
 Reflected Shortwavea (W m−2)Longwave (OLR) (W m−2)
With CALIOP and CPR cloudsMODIS OnlyWith CALIOP and CPR CloudsMODIS Only
  • a

    Daytime only.

CERES-derived irradiance237.8237.8240.1240.1
Model - CERES−0.4912.02−2.61−5.13

3.2. Surface Irradiance

[20] Figure 3 shows the surface downward shortwave (Figure 3, left) and longwave (Figure 3, right) irradiance differences computed with and without CALIOP- and CPR-derived cloud properties. The difference is defined as CCCM_MODISonly minus CCCM_CC. Both the surface downward shortwave and longwave irradiances computed with CALIOP and CPR cloud properties are larger than those computed without them. A larger downward shortwave irradiance increase occurs in the tropics, while a larger downward longwave irradiance increase occurs in polar regions. The seasonal variation of the difference is small, except in polar regions. The downward longwave irradiance difference in the Arctic and Antarctic increases in their respective winter seasons (Figure 4) because the active sensors can detect clouds better during polar night (section 6.1). While these results suggest that the CALIOP and CPR cloud properties increase both downward shortwave and longwave irradiances at the surface (for the reason explained in section 6.2), the differences are computed only using instantaneous irradiances for the nadir CERES footprint. In the following sections, we address sampling issues of the CALIOP and CPR ground track.

image

Figure 3. Monthly zonal mean shortwave and longwave downward surface irradiance difference. The difference is defined as the irradiance computed with MODIS-derived cloud properties minus the irradiance computed with CALIOP-, CPR-, and MODIS-derived cloud properties. Green, red, magenta, and blue lines are for January, April, July, and October 2008, respectively. Shortwave irradiance is the daytime average of instantaneous irradiances, and the longwave irradiance is the day and nighttime average of instantaneous irradiances.

Download figure to PowerPoint

image

Figure 4. Monthly mean surface longwave downward irradiance difference gridded in 1° latitude by 30° longitude grids. The difference is defined as the irradiance computed with MODIS only minus the irradiance computed with CALIOP and CPR.

Download figure to PowerPoint

3.3. Ground Track (1 km Width) Versus Full Footprint Coverage (20 km)

[21] To check how the cloud properties over the ground track (∼1 km width) represent the entire CERES footprint (20 km), we compare the cloud fraction defined over the ground track within a CERES footprint and that defined over the entire CERES footprint. Even though the CALIOP and CPR miss clouds present outside their ground track within a CERES footprint, we do not expect any systematic bias, because clouds along the ground track should be a random, unbiased sample of clouds across the footprint. The issue is, therefore, how many footprints are necessary for the ground track sampling error to become negligible to represent the entire footprint. Figure 5 shows the zonal mean cloud fraction root mean square (RMS) difference computed with three different averaging processes. It indicates that the cloud fractions over the CALIPSO-CloudSat ground track and over the CERES footprint are almost identical when computed with one day of footprints averaged over a 1° latitudinal zone; the RMS difference is less than 0.01 for most latitudes (green line in Figure 5).

image

Figure 5. Cloud fraction root mean square (RMS) difference computed with MODIS-retrieved clouds by the B1 algorithm over the CALIPSO and CloudSat ground track within a CERES footprint and over the whole CERES footprint. Top (red) line is RMS difference of the instantaneous cloud fraction over a CERES footprint computed for a month; middle (blue) line is the mean RMS difference of 1° latitude by 30° longitude daily mean cloud fractions (i.e., RMS of 30 daily means, which are an average of approximately 12 footprints); and bottom (green) line is the mean RMS difference of daily 1° zonal mean cloud fractions (i.e., RMS of 30 zonal means, which are an average of approximately 120 footprints). One month of data from July 2008 was used for the plot.

Download figure to PowerPoint

3.4. Nadir Footprint (20 km) Versus Full Swath (∼2000 km)

[22] CERES footprints containing the ground track of CALIOP and CPR cover a small fraction of the globe every day. With a 16-day repeat cycle [e.g., Stephens et al., 2002], the longitudinal width of a grid box needs to be at least 30 degrees to have daily samples. Hence, we assess in this section whether or not the nadir-view-only sampling is adequate to provide an unbiased zonal or global monthly mean irradiance compared with the value derived from full-swath sampling. Modeled instantaneous irradiances included in the CRS product use only the Ed2 MODIS-derived clouds and aerosol properties, but cover almost the entire globe daily. We can, therefore, estimate the zonal TOA and surface downward irradiance biases by subsetting to obtain near-nadir view irradiances and then comparing them with full-swath irradiances.

[23] Figure 6 shows the difference of the zonal TOA reflected shortwave (Figure 6, top) and longwave (Figure 6, middle) irradiances computed with nadir-view-only CERES footprints and with all footprints. The nadir view is defined as the viewing zenith angle of the CERES instrument less than or equal to 5 degrees. Note that the CERES instrument viewing zenith angle containing the CALIOP and CPR ground track is not exactly nadir because CALIPSO and CloudSat fly about 200 km to the east of the Aqua ground track at the equator to avoid sun glint [Kittaka et al., 2011]. We, however, ignore the effect of this shift on this sampling study.

image

Figure 6. Annual zonal mean TOA reflected (top) shortwave, (middle) longwave, and (bottom) cloud fraction difference. The difference is defined as the nadir-only mean minus the full-swath mean (solid line). One year of data from 2002 is used. The shading indicates the maximum and minimum monthly mean difference within a 1° latitudinal zone. Shortwave irradiances are converted to the daily mean value using the method discussed by Kato et al. [2008].

Download figure to PowerPoint

[24] TOA irradiances are derived from CERES radiances by angular distribution models [Loeb et al., 2005]. Note that CERES angular distribution models minimize viewing angle dependent errors [e.g., Loeb et al., 2007, Figure 12] by accounting for the radiance anisotropy of many scene types. A larger reflected shortwave irradiance difference toward the poles is a result of full-swath footprints reaching regions poleward of the limit reached from nadir. A small difference of zonal monthly mean TOA irradiances for all latitudes indicates that nadir-view sampling is adequate in computing monthly zonal mean irradiances when CERES angular distribution models are used. Figure 6 (bottom) shows the zonal cloud fraction difference between nadir-view-only and full-swath footprints. The difference occurs because cloud sides observed from oblique views increase the cloud area projected along the line-of-sight of the instrument [e.g., Minnis, 1989], optical thickness along the line-of-sight increases with viewing angle (clouds are more likely to be detected [Maddux et al., 2010]), and the pixel size increases with viewing angle (more likely to have clouds in a pixel [Ackerman et al., 2008]).

[25] Similar to the TOA irradiance, the bias in the modeled surface downward irradiance (Figure 7, bottom) caused by the nadir-view-only sampling is small with a global mean bias of 1.0 W m−2 (out of 202.8 W m−2) for shortwave and −0.5 W m−2 (out of 342.1 W m−2) for longwave irradiances. The difference can be explained by the bias of the cloud fraction derived by the Ed2 cloud algorithm. Therefore, we conclude that, for the modeled surface downward shortwave and longwave irradiances, the bias error caused by nadir-view sampling is negligible when cloud properties derived from CALIOP and CPR are used.

image

Figure 7. Annual mean surface downward (top) shortwave and (bottom) longwave irradiance differences. The difference is defined as the mean irradiance computed for nadir-view-only instantaneous CERES footprints minus the mean irradiance computed for full-swath CERES footprints. Shortwave irradiances are converted to the daily mean value using the method discussed by Kato et al. [2008]. Four seasonal months, January, April, July, and October 2008, are used to compute the annual mean value.

Download figure to PowerPoint

3.5. Global Mean Irradiance Estimate

[26] The two sets of surface irradiances, CCCM_CC and CCCM_MODISonly, provide the effect of CALIOP and CPR cloud properties on instantaneous irradiance. In the CERES AVG data product, the diurnal cycle of irradiances is taken into account by combining MODIS and 3-hourly geostationary satellite-derived cloud properties together. To convert the mean of instantaneous surface downward longwave irradiances computed with CALIOP- and CPR-derived cloud properties to an annual global mean irradiance, we compute the ratio of global mean instantaneous irradiance computed with CALIOP- and CPR-derived cloud properties equation imageCC to the irradiance computed with B1 MODIS-derived cloud properties only equation imageM and scale the global annual mean radiance from AVG equation image by

  • equation image

[27] The angle brackets indicate the mean irradiance over the complete diurnal cycles. This scaling leads to 345.4 W m−2 for equation image compared with 342.0 W m−2 for equation image, which is the annual mean surface downward longwave irradiance from the AVG product from January 2001 through December 2004. Note that using the annual mean from January 2001 through December 2004 for equation image and from 2008 for equation imageCC and equation imageM has no significant impact on the estimate (Table 3), because the interannual variability of global mean TOA irradiance is small [Kato, 2009], and the maximum and minimum values of 4 years of the annual mean surface downward longwave irradiance from AVG is 342.6 and 341.1 W m−2, respectively. Therefore, the global annual mean surface downward longwave irradiance is increased by 3.6 W m−2 because of a better cloud mask and vertical profile of clouds by the active sensors.

[28] The global mean surface downward longwave irradiance estimated by Zhang et al. [2004] from the International Satellite Cloud Climatology Project (ISCCP [Rossow and Schiffer, 1991]) is 345 W m−2. This estimate accounts for overlapping clouds with the use of climatological cloud vertical profiles [Wang et al., 2000]. The exact value of the surface downward longwave irradiance increase depends on the cloud base height estimated from passive sensors. According to Zhang et al. [2004], the increase in surface downward longwave irradiance due to overlapping clouds is 1.83 W m−2.

[29] Other surface irradiance components are also computed by scaling the irradiance from AVG. Table 2 summarizes the global annual mean surface irradiance estimates.

Table 2. Global Annual Mean Surface Irradiance Estimate
 With CALIOP and CPR CloudsaMODIS OnlyaAVGbScaled
  • a

    Average of instantaneous irradiance.

  • b

    Average of irradiances including diurnal cycle.

  • c

    Solar constant = 1365 W m−2.

Longwave down (W m−2)348.1344.7342.0345.4
Longwave up (W m−2)401.2401.1398.0398.1
Shortwave downc (W m−2)274.5270.2188.9191.9
Shortwave upc (W m−2)27.027.323.122.8

4. Uncertainty in Inputs Used for Surface Downward Longwave Irradiance Computations

  1. Top of page
  2. Abstract
  3. 1. Introduction
  4. 2. Data Set
  5. 3. Results
  6. 4. Uncertainty in Inputs Used for Surface Downward Longwave Irradiance Computations
  7. 5. Comparison With Surface Observations
  8. 6. Discussion
  9. 7. Conclusions
  10. Appendix A:: CALIOP, CPR, and MODIS Derived Cloud Properties and Flux Computations With Them
  11. Acknowledgments
  12. References
  13. Supporting Information

[30] All four modeled irradiance sets share some inputs such as temperature and humidity profiles. Errors in these profiles affect modeled surface irradiances but do not affect the irradiance differences discussed in section 3. In addition, even though CALIOP and CPR provide better cloud properties, cloud profiles derived from them contain uncertainties. For example, screening precipitation to identify the cloud base height is difficult. Understanding the error in inputs used for the irradiance computations is needed in order to determine the uncertainties in surface radiation budget.

4.1. Surface Skin Temperature and Near-Surface Air Temperature

[31] The surface skin temperature does not affect the surface downward longwave irradiance. However, a study by Zhang et al. [2007] indicates that, except for polar regions, the surface air temperature is within 3 K of the skin temperature despite much larger changes during the diurnal cycle [e.g., Minnis and Harrison, 1984]. Investigating the error in the surface skin temperature, therefore, provides an upper bound on the uncertainty in air temperature near the surface.

[32] The surface skin temperature retrieved by all cloud algorithms used in this study is derived from the 11 μm MODIS channel. Its uncertainty is caused by cloud contamination and uncertainties in surface emissivity, atmospheric humidity, and temperature profiles. When CALIOP and CPR are used to identify clear pixels, therefore, cloud contamination is nearly eliminated. A comparison by Zhang et al. [2006] indicates that the surface emissivities from two databases (ISCCP-FD and Wilber et al. [1999]) agree well over oceans and agree to within ±3% over land. Therefore, the uncertainty in the retrieved skin temperature over oceans is predominately caused by uncertainties in temperature and humidity profiles.

[33] Temperature and humidity profiles used in irradiance computations for this study are either from GEOS-5 (CCCM_CC, CCCM_MODISonly, and CRS) or GEOS-4 (AVG). To estimate the bias error in the GEOS-derived near-surface air temperature, we compare the skin temperature derived from MODIS and GEOS-5. This comparison is only valid for clear skies when a radiance-based skin temperature retrieval is possible. We therefore subset the GEOS-5 skin temperature using the clear-sky scene identified by CALIOP and CPR. Figure 8 shows the monthly zonal mean difference of skin temperature over land and ocean for four seasonal months. The global and annual mean difference is 0.63 K (GEOS-5 – MODIS-retrieved), and the differences over ocean and land are 0.96 K and −0.79 K, respectively.

image

Figure 8. Zonal monthly mean surface skin temperature differences over ocean and land. The difference is defined as the surface skin temperature used in irradiance computations, based on GEOS5, minus the skin temperature retrieved by the CERES cloud algorithm from MODIS radiances for pixels identified as clear by CALIOP and CPR. GEOS5 skin temperature is collocated with CALIOP and CPR ground track and clear-sky values are sampled. Zonal differences are averaged with a 5° moving window.

Download figure to PowerPoint

[34] The skin temperature retrieved from MODIS is the temperature at the overpass time (1:30 at the equator) and may not represent the mean skin temperature of the hour and grid boxes, especially over land where the spatial and temporal variability is larger than that over oceans. In the irradiance computations, skin temperature was interpolated from the 3-hourly from GEOS-5 reanalyses, which originally had a 1 h temporal resolution. Because of this coarse temporal interpolation, the daily maximum and minimum temperatures were missed in some locations especially over land. Therefore, spatial and temporal sampling by MODIS, temporal resolution used to interpolate GEOS-5 skin temperature, MODIS retrieval error, and the GEOS-5 skin temperature error cause the difference shown in Figure 8. Among these, the error due to MODIS sampling and retrieval errors does not affect the irradiance computations. We consider, therefore, the difference shown in Figure 8 as the upper limit of the uncertainty in the skin temperature.

[35] We estimate the uncertainty in the surface downward longwave irradiance caused by the uncertainty in near-surface air temperature and coarse 3-hourly interpolation separately. We use 0.96 K as the uncertainty in the skin temperature and assume that the uncertainty of near-surface air temperature is the same as the skin temperature uncertainty. When the temperature of tropical, midlatitude summer, and subarctic winter standard atmospheres below 800 hPa is reduced 0.96 K, the downward longwave irradiance decreases 4.3, 3.8, and 2.4 W m−2 for clear-sky and 5.4, 5.1, and 4.5 W m−2, respectively, for overcast conditions where a liquid-water cloud with an optical thickness of 10 is placed between 550 hPa and 850 hPa. Therefore, the uncertainty in the surface downward longwave irradiance caused by near-surface air temperature is approximately 4.5 W m−2 (an area-weighted average of the mean of clear and overcast values). Zhang et al. [1995] perturbed the lowest level of air temperature of 15 July 1985 atmosphere by 2 K and found that global surface downward longwave irradiance increases by 23 W m−2. In a revised estimate, Zhang et al. [2007] reported that an uncertainty in surface air temperature of 2.6 K resulted in an uncertainty in the surface downward longwave irradiance of 15 W m−2. Our uncertainty in downward longwave irradiance due to uncertainty in near-surface air temperature is smaller than estimates by Zhang et al. [1995, 2007]. This is because the near-surface air temperature uncertainty in the Zhang et al. results is derived from comparison of several reanalysis data sets and is larger than the uncertainty we find from direct comparison of GEOS-5 with retrieved values. Thus, we use the near-surface air temperature uncertainty of 0.96 K and corresponding downward longwave irradiance uncertainty of 4.5 W m−2.

[36] To estimate the bias errors caused by a 3-hourly skin temperature resolution, we compute a global mean downward longwave irradiance using a higher temporal resolution. When a 1-hourly temporal resolution is used for the computation of global mean irradiance for a 1 July 2008 atmosphere, the surface downward longwave irradiance increases by 2.6 W m−2 (from 357.7 W m−2). Although of opposite sign, a study by Zhang et al. [2004] indicates a similar sensitivity. In their study, including the cycle of surface temperature reduced the daily global mean surface downward longwave irradiance by 1.85 W m−2.

4.2. Cloud Base Height and Other Uncertainties

[37] A large uncertainty associated with the cloud base height derived from radar results from the difficulty in screening precipitation [e.g., Clothiaux et al., 2000]. To estimate the uncertainty in cloud base height derived from CALIOP and CPR, we compute the monthly zonal mean cloud base heights of nonprecipitating clouds (Figure 9, right). We use the precipitation flag from the CloudSat CLDCLASS product [Sassen and Wang, 2008] and average cloud profiles with the flag equal to 0 (nonprecipitating clouds). The global mean cloud base height difference between all clouds and nonprecipitating clouds is 0.5 km, where nonprecipitating clouds have a higher cloud base. The global mean difference of the cloud base estimated from CALIOP and CPR and from MODIS is 1.6 km. Therefore, the cloud base of nonprecipitating clouds is lower by 1.1 km than the cloud base estimated from MODIS cloud top heights with the empirical relationship [Minnis et al., 2011]. We estimate that the uncertainty in the cloud base height difference computed with and without CALIOP and CPR cloud properties is 30% (∼0.5/1.6). If we assume that the downward longwave irradiance changes linearly with cloud base height, a 30% uncertainty in the cloud base height corresponds to 1.1 W m−2 (0.3 times 3.6 W m−2 from section 3.5).

image

Figure 9. Zonal monthly mean cloud base height difference. The difference is defined as the (left) cloud base derived from MODIS minus the cloud base derived from CALIOP and CPR. (right) The cloud base height of nonprecipitating clouds minus the cloud base height of all clouds; both are derived from CALIOP and CPR. Zonal differences are averaged with a 5° moving window.

Download figure to PowerPoint

[38] The uncertainty in the column water vapor of 15% is according to Zhang et al. [2007]. We use the midlatitude summer water vapor profile and perturb the column water vapor amount by ±15% with clear and overcast conditions. The overcast cloud extends from 850 to 550 hPa and the optical thickness is 10. When clear and cloudy conditions are averaged, a ±15% column water vapor amount perturbation changes the surface downward longwave irradiance by 5.2 W m−2. L'Ecuyer and Stephens [2003], who perturbed water valor profiles in the tropics and subtropics, show similar surface downward longwave irradiance sensitivity to the water vapor amount.

[39] Table 3 summarizes the uncertainties in skin temperature, cloud base height, and other inputs, as well as the uncertainty in the modeled surface downward longwave irradiance. When we sum the bias errors and uncertainties listed in Table 3 assuming all uncertainties are independent, the bias error in the modeled global annual mean surface downward longwave irradiance of 345.4 W m−2 is −1.5 W m−2 with the uncertainty (1σ) of 6.9 W m−2.

Table 3. Surface Downward Longwave Uncertainty
VariablesGlobal Mean UncertaintyLW Irradiance Uncertainty With Known Sign (W m−2)LW Irradiance Uncertainty With Unknown Sign (W m−2)Reference
Near-surface temperature derived from skin temperature288.6 ± 0.96 K ±4.5 
Temperature and humidity temporal interpolation3 hourly to 1 hourly−2.6  
Cloud base height2.9–0.5 km1.1  
Precipitable water15% ±5.2Zhang et al. [2006]
Interannual variabilityfrom Jan. 2001 through Dec. 2004 ±0.8 
SamplingNadir versus full swath negligible 
Overall uncertainty −1.5±6.9 

5. Comparison With Surface Observations

  1. Top of page
  2. Abstract
  3. 1. Introduction
  4. 2. Data Set
  5. 3. Results
  6. 4. Uncertainty in Inputs Used for Surface Downward Longwave Irradiance Computations
  7. 5. Comparison With Surface Observations
  8. 6. Discussion
  9. 7. Conclusions
  10. Appendix A:: CALIOP, CPR, and MODIS Derived Cloud Properties and Flux Computations With Them
  11. Acknowledgments
  12. References
  13. Supporting Information

[40] To check if an increase of surface downward longwave irradiance computed with CALIOP- and CPR-derived cloud profiles is indeed an improvement, we compare instantaneous modeled irradiances with surface observations. Figure 10 shows the histogram of the modeled and observed surface downward longwave irradiance difference over three Arctic sites. To reduce sampling noise, modeled irradiances for all CERES footprints that fall within 150 km from the site in a day are averaged. The averaged irradiance is differenced with the mean surface observation taken within ±15 min of the overpass time. Even with this averaging process, sampling noise dominates, apparent from the wide range of differences. While the 3-year average of daytime difference with and without CALIOP- and CPR-derived cloud properties are −1.3 W m−2 and −2.3 W m−2, respectively, large negative differences are absent in the nighttime histogram when CALIOP- and CPR-derived cloud properties are used. As a result, the annual mean difference improves from −13.6 W m−2 without CALIOP and CPR cloud properties to −2.9 W m−2 with CALIOP- and CPR-derived cloud properties for nighttime.

image

Figure 10. Difference (modeled – observed) between modeled and observed surface downward longwave irradiance in the Arctic. Observed irradiances were taken at three sites (Ny Alesund Norway 78.93°N 11.95°E, Barrow Alaska 71.32°N 156.61°W, and Alert Canada 82.51°N 62.35°W). Modeled irradiances for CERES footprints that fall within 150 km from the surface sites in a day were averaged and compared with observations averaged over 15 min at satellite overpass time. Three years of data from July 2006 through June 2009 were used. Both modeled irradiances with CALIOP- and CPR-derived cloud properties (labeled CCCM) and without them (labeled MODIS only) are nadir view only.

Download figure to PowerPoint

6. Discussion

  1. Top of page
  2. Abstract
  3. 1. Introduction
  4. 2. Data Set
  5. 3. Results
  6. 4. Uncertainty in Inputs Used for Surface Downward Longwave Irradiance Computations
  7. 5. Comparison With Surface Observations
  8. 6. Discussion
  9. 7. Conclusions
  10. Appendix A:: CALIOP, CPR, and MODIS Derived Cloud Properties and Flux Computations With Them
  11. Acknowledgments
  12. References
  13. Supporting Information

[41] In section 5, we showed that the TOA reflected shortwave irradiance decreases and longwave irradiance increases when CALIOP- and CPR-derived aerosol and cloud properties are used in the irradiance computations. As a result, irradiances computed with CALIOP- and CPR-derived properties agree better with CERES observations at TOA. At the surface, both the downward shortwave and longwave irradiances increase when CALIOP- and CPR-derived properties are used. The change is caused by the differences in cloud properties used in the computations. In this section, we analyze how the surface downward longwave irradiance increases when using CALIOP- and CPR-derived cloud properties compared with that computed with MODIS B1 cloud properties.

6.1. Surface Downward Longwave Irradiance

[42] The merged CALIOP and CPR cloud profiles improve the surface downward longwave irradiance by providing better cloud detection and cloud base heights. To understand how the cloud fraction and base height improvements influence the surface downward longwave irradiance, Figure 11 (right) shows the zonal difference of the cloud fraction derived from MODIS and from CALIOP and CPR. The global mean cloud fraction given by CALIOP and CPR is 0.761, which is 0.114 larger than the cloud fraction derived by the B1 cloud algorithm. Although the cloud fraction given by the B1 cloud algorithm for this comparison is derived from the nadir view only, the difference is larger than cloud fraction difference derived from the nadir view and full swath shown in Figure 6. The CERES Ed2 cloud mask typically misses clouds with optical thickness less than 0.3 [Minnis et al., 2008a, 2008b] and these optically thin clouds contribute a significant part of the cloud fraction derived from CALIOP and CPR (Figure 11, left). When the extinction coefficient derived by CALIOP is integrated and clouds having optical thickness less than 0.3 are neglected, the cloud fraction derived by the B1 cloud algorithm agrees with the cloud fraction derived from CALIOP and CPR to within 0.038. When the difference of cloud fraction retrieved from nadir and full swath is considered, the cloud fraction difference is further reduced to −0.004.

image

Figure 11. (left) Annual (2008) zonal mean cloud fraction derived from merged CALIOP and CPR cloud profiles. The dashed and solid lines are all clouds and clouds with the optical thickness greater than 0.3, respectively. (right) Annual zonal mean cloud fraction difference. The difference is defined as the cloud fraction derived from MODIS minus the cloud fraction derived from CALIOP and CPR. The dashed and solid blue lines are for the difference including all CALIOP- and CPR-derived clouds and clouds with an optical thickness greater than 0.3. The solid red line indicates the difference computed with the MODIS-derived clouds using full swath minus CALIOP- and CPR-derived clouds, excluding clouds with cloud optical thickness less than 0.3.

Download figure to PowerPoint

[43] The cloud base height difference depends on the empirical relationship used to estimate the cloud base height from the MODIS-derived cloud top height. Because the B1 cloud algorithm treats each cloud retrieval as if it were for a single-layer cloud, the cloud thickness algorithm used to retrieve the cloud base is applied to the uppermost cloud. The lower-level cloud and its base are missed. Thus, the higher cloud base altitude derived from the MODIS data is predominately caused by missing the lower-level clouds in multilayer cloud systems.

[44] While the surface downward longwave irradiance is sensitive to both cloud fraction and cloud base, Figure 12 indicates that the lower cloud base contributes slightly more to the improvement in the Arctic in July. In Figure 12, the surface downward longwave irradiance difference computed with and without CALIOP and CPR cloud properties are sorted by the differences in cloud base height and fraction derived from CALIOP and CPR and by the B1 cloud algorithm. The sensitivity of the surface downward longwave irradiance to cloud base height in July is larger than the sensitivity in January, which is indicated by more vertical contour lines in the right plot. In addition, CALIOP and CPR detect clouds that are missed by the B1 cloud algorithm during polar night in January, which is apparent in the lower left plot of Figure 12 showing a 2D histogram of the number of samples. A large improvement of cloud detection by CALIOP and CPR contributes mostly to increasing the surface downward longwave irradiance in fall and winter in the Arctic, apparent in Figure 13, which shows regional and monthly mean differences. In other regions, both a larger cloud fraction and lower cloud base height derived from CALIOP and CPR compared with the cloud fraction and base height derived by the B1 cloud algorithm contribute to a greater surface downward longwave irradiance (Figure 13).

image

Figure 12. (top) Surface downward longwave irradiance difference defined as the irradiance computed with MODIS-derived cloud properties only minus that computed with CALIOP- and CPR-derived cloud properties for the months of (left) January and (right) July. The irradiance difference is sorted by the cloud base height and cloud fraction differences between those derived from MODIS and from CALIOP and CPR. The difference is defined as MODIS minus CALIOP and CPR-derived cloud properties. (bottom) Same as Figure 12 (top), but the logarithm (base 10) of the number of samples is plotted.

Download figure to PowerPoint

image

Figure 13. Monthly mean cloud fraction, cloud base height, and surface downward longwave irradiance difference as a function of month for four regions. The difference is computed as MODIS-derived minus CALIOP CPR-derived values. Four regions are, from left to right, 60°S to 30°S, 30°S to 30°N, 30°N to 60°N, and 60°N to 82°N. Blue and red bars in the cloud fraction difference plot are computed with all clouds and clouds having optical thickness greater than 0.3, respectively. Blue and red bars in the cloud base height difference plot are computed with all clouds and non-precipitating clouds, respectively. MODIS-derived cloud fraction and cloud base height are unaltered in computing the difference shown by blue and red bars. Note that the y-axis of the middle plot changes from positive to negative.

Download figure to PowerPoint

6.2. Surface Downward Shortwave and TOA Irradiances

[45] A larger cloud fraction derived from CALIOP and CPR compared with the cloud fraction derived from MODIS does not explain the larger downward shortwave irradiance at the surface, smaller TOA reflected shortwave and larger TOA longwave irradiances shown in Figures 2 and 3. To understand the reason for the change qualitatively, Figure 14 shows the differences in the four irradiance components as a function of the cloud top pressure of a single-layer cloud. The four components are TOA reflected shortwave (solid red line with closed circle), TOA longwave (solid blue line with closed circle), surface downward shortwave irradiance (dashed red line with closed square), and surface downward longwave irradiance (dashed blue line with closed square). The difference is defined as the modeled irradiance with a single-layer overcast cloud minus the modeled irradiance with two vertically overlapping cloud layers. Both layers of the two-layer cloud are overcast. The upper- and lower-layer cloud tops are 200 and 850 hPa on the two left plots and 550 and 850 hPa on the two right plots, respectively. The upper layer is an ice cloud with the optical thickness of 0.1 and the effective diameter of 60 μm. The lower layer is a liquid-water cloud with the optical thickness of 5 and the effective radius of 10 μm. The single-layer cloud in the top two plots is a water cloud with the effective radius of 15 μm, and in the bottom two plots it is an ice cloud with the effective diameter of 100 μm. The optical thickness of the single-layer water cloud is 5.3 and of the single-layer ice cloud is 3.7. The optical thickness of the single-layer cloud is determined by matching the 0.55 μm albedo of the two-layer cloud at TOA, which crudely simulates the optical thickness retrieval from visible reflectance and maintains a constant scaled optical thickness between two cloud systems.

image

Figure 14. TOA reflected shortwave (solid red line), longwave (solid blue line), surface downward shortwave (dashed red line), and surface longwave downward (dashed blue line) irradiance differences as a function of single-layer cloud top pressure. The difference is defined as the irradiance computed with a single-layer cloud minus the irradiance computed with two-layer cloud. Cloud top pressures of the two-layer cloud are (right) 200 and 850 hPa and (left) 550 and 850 hPa, indicated by the vertical dotted lines with a depth of 50 hPa. The upper layer is an ice cloud with optical thickness of 0.1 and effective diameter of 60 μm. The lower layer is a liquid-water cloud with optical thickness of 5.0 and effective radius of 10 μm. The depth of both upper and lower layers is 50 hPa. The single-layer cloud in the top two plots is a liquid-water cloud with effective radius of 15 μm and, in the bottom two plots, the single-layer cloud is an ice cloud with effective diameter of 100 μm. The single-layer cloud optical thickness is (top) 5.3 for the water cloud and (bottom) 3.7 for the ice cloud both with a depth of 100 hPa. The cosine of the solar zenith angle is 0.8. The tropical standard atmosphere and an ocean surface are used for the computations.

Download figure to PowerPoint

[46] Figure 14 indicates that when two-layer overlapping clouds are present, the TOA shortwave irradiance is larger and the other three components are lower, if the irradiance is computed with a single-layer cloud that gives the same TOA albedo at 0.55 μm as the two-layer overlapping clouds and the cloud top located between two overlapping layers. Note that the sign of the surface downward shortwave irradiance difference is sensitive to the retrieved particle size of the single-layer cloud. For example, when the ice effective diameter of the single-layer cloud is reduced to 60 μm or smaller from 100 μm shown in the bottom two plots in Figure 14, the surface downward shortwave irradiance computed with the single-layer cloud is larger than the irradiance computed with the two-layer overlapping clouds. This result indicates, therefore, that a possible reason for the lower TOA reflected shortwave irradiance and higher TOA longwave, surface downward shortwave and longwave irradiances when using the CALIOP- and CPR-derived cloud properties is, in the presence of multilayer clouds with optically thin ice clouds vertically overlapping with low-level water clouds, the B1 algorithm retrieves single-layer clouds of which the cloud top height is lower than the true upper-layer cloud top height.

6.3. Cloud Property Retrieval Improvement

[47] In this study, two sets of cloud properties were retrieved from MODIS radiances over the CALIPSO and CloudSat ground track. The first set was produced by the B1 cloud algorithm without the use of CALIOP- and CPR-derived properties. The second set was produced by the enhanced cloud algorithm, which used the B1 algorithm constrained by the uppermost cloud effective height (and mask) derived from CALIOP and CPR when a single-layer cloud is present (i.e., two algorithms are the same when multilayer clouds are present). Forcing the algorithm to retrieve the cloud at the height detected by CALIOP and CPR eliminates the error in cloud detection and cloud top height and improves cloud property retrievals [e.g., Cooper et al., 2003]. While the difference between these two sets of cloud properties is not indicative of the error caused by assumptions in the algorithm, cloud properties derived from the enhanced algorithm are consistent with accurate cloud mask and top height derived from the active sensors. The differences in cloud properties, therefore, can be considered as uncertainties in the cloud properties derived from the B1 cloud algorithm due to the limitation of detecting clouds and in determining cloud top height. The differences in cloud properties derived by the enhanced and B1 algorithms are summarized in Table 4.

Table 4. Global Annual Mean Cloud Properties Derived From CALIOP, CPR, and MODIS and Their Differences From Using MODIS Only
 Mean Derived From CALIOP, CPR, and Enhanced Cloud AlgorithmMODIS Only Minus (CALIPSO, CPR, and Enhanced Algorithm)
AllOceanLandAllOceanLand
  • a

    Includes all clouds.

  • b

    1 = water clouds, 2 = ice clouds.

  • c

    Derived from 3.7 μm channel.

Fractiona (all)0.7610.7830.655−0.114−0.088−0.134
Fraction (τ > 0.3)0.6850.7180.559−0.038−0.022−0.037
Optical thickness (linear mean)a7.867.718.110.660.391.62
Optical thickness (logarithmic mean)a0.9430.9640.8350.280.230.48
Top heighta (km)8.07.69.3−2.2−2.2−2.0
Base heighta (km)2.92.93.01.61.32.6
Phasea,b1.491.451.58−0.04−0.04−0.06
Water cloud effective radiusa,c (μm)14.014.512.0−0.89−0.99−0.55
Ice cloud effective diametera,c (μm)53.555.747.15.564.987.20

6.4. Implication for Hydrological Cycles

[48] As mentioned at the beginning of this paper, the global surface radiation budget needs to balance against other surface energy fluxes. Therefore, we briefly discuss the effect of the net surface irradiance increase suggested by the result of this study and how this relates to the uncertainty of surface fluxes estimated in earlier studies.

[49] The standard deviation of annual global mean precipitation derived from Global Precipitation Climatology Project (GPCP) and Climate Prediction Center Merged Analysis of Precipitation (CMAP) data from 1988 through 1999 is about 1% [Schlosser and Houser, 2007]. The difference of annual global mean precipitation based on GPCP and CMAP is approximately 10% [Schlosser and Houser, 2007]. Trenberth et al. [2009] suggest an underestimate of precipitation over land by 17.9%, which is equivalent to an approximately 5% global underestimate, due primarily to gauge undercatch and interpolation in areas with steep and complex topography. Inconsistency among the surface energy fluxes, precipitation, and divergence is 15 W m−2 over ocean [Edwards, 2007] and among the surface energy fluxes and precipitation is 9 to 20 W m−2 over land and ocean [Lin et al., 2008], where the positive sign indicates more energy deposited to the surface. Therefore, earlier studies indicate that a smaller net irradiance at the surface is needed to close the inconsistency.

[50] We show that including CALIOP- and CPR-derived cloud profiles increases the global annual mean surface downward longwave irradiance to 346.9 W m−2 (345.4 + 1.5 W m−2). Although the uncertainty due to the diurnal cycle of the shortwave irradiance was not considered, a larger downward surface shortwave irradiance computed with CALIOP- and CPR-derived properties adds to the positive net flux at the surface. Once other surface irradiance components from Table 2 are combined, the surface net irradiance is 116.4 W m−2 (including the surface downward longwave irradiance bias error of 1.5 W m−2). When we use the sensible heat flux of 17 W m−2 [Trenberth et al., 2009], and the surface latent heat flux of 75.5 W m−2, the value computed from the annual mean global value of 2.61 mm day−1 [Adler et al., 2003], the sum of the latent and sensible heat fluxes is 92.5 W m−2. Stephens and Kummerow [2007] listed three sources of uncertainty in retrieving precipitation from microwave measurements from space; distinguishing precipitating cloudy scenes from nonprecipitating scenes, atmospheric model used in forward calculations, and microphysical approximations used in forward calculations. Because the effects of these on the precipitation error are complex, they point out why error estimates in precipitation estimated from passive sensors have been so elusive in the past. If we assume the uncertainty of the precipitation estimate is 10%, and both the surface downward longwave irradiance and precipitation uncertainties are 1σ, this study indicates that 2σ surface energy flux uncertainties estimated from the surface downward longwave irradiance and latent heat flux uncertainties overlap (Figure 15).

image

Figure 15. Global annual mean surface net irradiance on the left (labeled 1) and latent heat flux estimated from the Global Precipitation Climatology Project (GPCP) [Adler et al., 2003] plus surface sensible heat flux (17 W m−2) [Trenberth et al., 2009] on the right (labeled 2). Open circles indicate mean values, while boxes indicate 1σ uncertainty and vertical lines extend 2σ uncertainty. Only the (left) surface downward longwave irradiance uncertainty and (right) latent heat uncertainty are included in the uncertainty estimate.

Download figure to PowerPoint

[51] In this study, however, we did not estimate the uncertainty in upward longwave and upward and downward shortwave irradiances. In addition, we relied on geostationary satellite-derived cloud properties to account for a diurnal cycle of irradiances (AVG). Although the process to account for the diurnal cycle minimizes the geostationary satellite calibration effects to the irradiance computation [Young et al., 1998], and diurnal cycle correction alters the global annual mean TOA irradiance very little [Loeb et al., 2005], the uncertainty due to geostationary satellite-derived cloud properties needs to be estimated, especially for the surface downward shortwave irradiance. A more rigorous closure study among the surface net irradiance, surface sensible, and latent heat fluxes is currently taking place within the NASA Energy Water Cycle Study (NEWS) project.

7. Conclusions

  1. Top of page
  2. Abstract
  3. 1. Introduction
  4. 2. Data Set
  5. 3. Results
  6. 4. Uncertainty in Inputs Used for Surface Downward Longwave Irradiance Computations
  7. 5. Comparison With Surface Observations
  8. 6. Discussion
  9. 7. Conclusions
  10. Appendix A:: CALIOP, CPR, and MODIS Derived Cloud Properties and Flux Computations With Them
  11. Acknowledgments
  12. References
  13. Supporting Information

[52] Improvements in modeled TOA and surface irradiances using CALIOP- and CPR-derived cloud and aerosol properties from those computed with MODIS only are investigated. When one year of TOA instantaneous irradiances computed with CALIOP- and CPR-derived properties are averaged globally, the shortwave irradiance decreases by 12.5 W m−2 (5.0%) and the longwave irradiance increases by 2.5 W m−2 (1.1%). As a consequence, both the shortwave and longwave irradiances computed with CALIOP- and CPR-derived properties agree better with CERES-derived irradiances to within 0.5 W m−2 (out of 237.8 W m−2 CERES value) for shortwave and 2.6 W m−2 (out of 240.1 W m−2 CERES value) for longwave irradiances. The difference of monthly zonal mean irradiance is larger in some latitudes, however, and a good agreement with CERES-derived irradiance is, in part, achieved by compensating errors. The global annual mean of instantaneous surface downward longwave irradiances increases by 3.4 W m−2 (1.0%) because of a larger cloud fraction and lower cloud base height derived from CALIOP and CPR compared with those derived from the B1 cloud algorithm that uses only MODIS-derived properties. In addition, the global mean of instantaneous surface downward shortwave irradiances increases by 8.6 W m−2 (1.6%). Therefore, the global annual mean net surface irradiance increases when the CALIOP- and CPR-derived properties are used in the irradiance computations. The largest improvement of the surface downward longwave irradiance is in the Arctic during fall and winter because of better cloud detection provided by CALIOP and CPR. The estimated global annual mean downward longwave irradiance is 345.4 + 1.5 ± 6.9 W m−2. The estimated uncertainty in the surface downward longwave irradiance caused by the cloud base height because of precipitation is 1.1 W m−2. As pointed out by Zhang et al. [2007], the uncertainty is primarily caused by the near-surface air temperature and column water vapor amount uncertainties. A further study needed to estimate the surface radiation budget more accurately is a rigorous treatment of the diurnal cycle in the modeled shortwave irradiance computed with the CALIOP- and CPR-derived cloud properties.

Appendix A:: CALIOP, CPR, and MODIS Derived Cloud Properties and Flux Computations With Them

  1. Top of page
  2. Abstract
  3. 1. Introduction
  4. 2. Data Set
  5. 3. Results
  6. 4. Uncertainty in Inputs Used for Surface Downward Longwave Irradiance Computations
  7. 5. Comparison With Surface Observations
  8. 6. Discussion
  9. 7. Conclusions
  10. Appendix A:: CALIOP, CPR, and MODIS Derived Cloud Properties and Flux Computations With Them
  11. Acknowledgments
  12. References
  13. Supporting Information

[53] The hierarchy of cloud optical thickness and extinction coefficient vertical profile sources for irradiance computations follow. When the CALIOP-derived extinction coefficient (532 nm) for the cloud layer is available within the CERES footprint with the extinction QC flag of 0, 1, 2, 16, or 18 and with the CAD score ranging from −100 to 100, the CALIOP-derived extinction profile is used. When the CALIOP-derived extinction profile is not available within the CERES footprint, CPR-derived (2B CWC-RO, Revision 4) ice-water content (IWC) and liquid-water content (LWC) are converted to the extinction coefficient profile [Fu and Liou, 1993; Fu et al., 1999]. To minimize the conversion error when mixed-phase clouds are present in a column and when the extinction vertical profile is determined from CALIOP- or CPR-derived properties, the scaled optical thickness integrated from vertical extinction profile is normalized by the MODIS-derived scaled optical thickness by

  • equation image

where τM and gM are the optical thickness and asymmetry parameter, respectively, derived from MODIS, βi is the extinction coefficient in the ith layer, and gi is the cloud particle asymmetry parameter in the ith layer. In (1), α is a free parameter to scale the CALIOP- or CPR-derived scaled cloud optical thickness by MODIS-derived scaled cloud optical thickness by the enhanced cloud algorithm. The enhanced cloud algorithm is the same as the B1 cloud algorithm with one important difference. The enhanced cloud algorithm uses CALIOP- and CPR-derived cloud height as a constraint in the following way.

[54] When a single-layer cloud is present in the pixel, then the effective cloud top is placed at a height based on the following criterion. When the uppermost cloud layer optical thickness is greater than 0.3 and less than 2, cloud top is placed at a height equal to one half of the layer cloud optical thickness. When the layer optical thickness is greater than 2, the cloud top is placed at a height equal to the optical thickness of 1 from the cloud top. This adjustment is made before emission contribution is subtracted from the 3.7 μm channel.

[55] When CALIOP did not retrieve an extinction coefficient profile and CPR did not retrieve IWC and LWC, then the MODIS-derived cloud optical thickness by the enhanced cloud algorithm is distributed among cloud layers proportional to their geometrical thickness, i.e., the extinction coefficient is constant within all cloud layers. When both the enhanced and B1 cloud algorithms did not provide cloud properties (optical thickness, particle size, and phase) of the cloud group for various reasons, such as a large solar zenith angle or the observed reflectance exceeded the range used for look-up tables, we neglect the cloud group. We then increase the cloud fraction of other cloud groups within the CERES footprint that have retrieved cloud properties so that the total cloud fraction over the CERES footprint is not altered. If no MODIS-retrieved cloud properties are available within the CERES footprint, we use either CALIOP- or CPR-derived cloud optical thickness without the normalization of the scaled optical thickness assuming an effective radius of 10 μm. The frequency of occurrence of cases where all the enhanced and B1 cloud algorithms and CALIOP- and CPR-detected clouds have no retrieval is less than 1%.

Acknowledgments

  1. Top of page
  2. Abstract
  3. 1. Introduction
  4. 2. Data Set
  5. 3. Results
  6. 4. Uncertainty in Inputs Used for Surface Downward Longwave Irradiance Computations
  7. 5. Comparison With Surface Observations
  8. 6. Discussion
  9. 7. Conclusions
  10. Appendix A:: CALIOP, CPR, and MODIS Derived Cloud Properties and Flux Computations With Them
  11. Acknowledgments
  12. References
  13. Supporting Information

[56] We thank Bing Lin, Tristan L'Ecuyer, and Stefan Kinne for useful discussions and Amber Richards for proof reading the manuscript. The work was supported by the NASA Energy Water Cycle Study (NEWS) project. Also two of the authors (S.K. and P.M.) received support from the NASA cryosphere IPY program for this study.

References

  1. Top of page
  2. Abstract
  3. 1. Introduction
  4. 2. Data Set
  5. 3. Results
  6. 4. Uncertainty in Inputs Used for Surface Downward Longwave Irradiance Computations
  7. 5. Comparison With Surface Observations
  8. 6. Discussion
  9. 7. Conclusions
  10. Appendix A:: CALIOP, CPR, and MODIS Derived Cloud Properties and Flux Computations With Them
  11. Acknowledgments
  12. References
  13. Supporting Information

Supporting Information

  1. Top of page
  2. Abstract
  3. 1. Introduction
  4. 2. Data Set
  5. 3. Results
  6. 4. Uncertainty in Inputs Used for Surface Downward Longwave Irradiance Computations
  7. 5. Comparison With Surface Observations
  8. 6. Discussion
  9. 7. Conclusions
  10. Appendix A:: CALIOP, CPR, and MODIS Derived Cloud Properties and Flux Computations With Them
  11. Acknowledgments
  12. References
  13. Supporting Information
FilenameFormatSizeDescription
jgrd17298-sup-0001-t01.txtplain text document0KTab-delimited Table 1.
jgrd17298-sup-0002-t02.txtplain text document0KTab-delimited Table 2.
jgrd17298-sup-0003-t03.txtplain text document1KTab-delimited Table 3.
jgrd17298-sup-0004-t04.txtplain text document1KTab-delimited Table 4.

Please note: Wiley Blackwell is not responsible for the content or functionality of any supporting information supplied by the authors. Any queries (other than missing content) should be directed to the corresponding author for the article.