The global surface and atmosphere radiation budget: an assessment of accuracy with 5 years of calculations and observations

We compute the surface and atmosphere radiation budget (SARB) and validate with simultaneous, collocated broadband observations at TOA and surface. The differences of TOA and surface fluxes determine the net radiative cooling of the atmosphere. If known accurately over the globe at all times, this net radiation would exactly balance the sum of the global mean latent and sensible heating - other benchmarks sought by the GCM and climate diagnostics community. Our radiative transfer calculations are a bit more sophisticated than those in GCMs. By comparing computations and observations at TOA and surface, we find specific examples of the strengths and weaknesses of the radiative transfer code, the inputs for the code, and the satellite and ground-based broadband fluxes. Cloud and aerosol inputs are based on MODIS, and temperature and humidity are from NWP. Over land, we use broadband TOA observations from clear FOVs to assign surface albedo; hence the calculations for SW at TOA over land are partly contaminated with observations. No broadband radiometric observations at the surface are used for radiative transfer inputs, adjustments, or tuning; the flux comparison at the surface is a “cold” test. The extensive comparison of broadband computed and observed fluxes at TOA over the ice-free ocean is a cold test. Over the mean of all ground sites, computed surface downward LW is too small, because the surface air temperature input from NWP is biased; the time mean cloud forcing to downward LW (based on MODIS clouds) is surprisingly good. The small biases for surface upward LW mark the quality of the skin temperatures from MODIS. Biases for OLR are small, but a daytime-only drift of ~1 Wm-2 in 5 years (and none in the window OLR) for the global mean suggests a refinement of the broadband instrument record is needed. Agreement for surface insolation in clear skies suggests that in the gross mean, inputs for aerosol (i.e., direct forcing) are better than expected. There are signficant, unresolved discrepancies (~3%) for reflected SW at TOA, especially over ocean; but the geographical patterns of interannual variability in calculations and observations match very well.

For each CERES cross track FOV on Terra, the Langley Fu-Liou radiative transfer code is run for profiles of SW and LW fluxes; and aerosol and cloud forcings at surface and TOA. Inputs include MODIS cloud retrievals (Minnis et al.) matched with CERES FOVs, soundings from GEOS-4, and aerosols from MODIS (Atmosphere Team) and the NCAR Model for Atmospheric Transport and Chemistry (MATCH). Except for the determination of surface albedo over land and cryosphere clear FOVs, CERES broadband observations are not used as inputs for the "untuned" version of these calculations. "Tuned" calculations are made by adjusting inputs to produce a closer match of broadband computations and observations at TOA. We focus on untuned results here. Results at the time of Terra overpass are in the public archive over the globe for March 2000 to June 2005.

This summary table for 2001 (NOT corrected for official CERES Rev1 instrument adjustments to SW at TOA) at a representative group of BSRN, SURFRAD, and ARM ground sites is from the URL www-cave.larc.nasa.gov/cave. “Obs” denotes mean observed value (Wm-2). Bias as untuned calculation minus observation (Wm-2) at ~1030 LST and ~2230 LST Terra overpasses. SFC denotes surface.

Parameter____Obs__{Bias}__(sample size)

All-sky

LW down SFC 286.1 {-6.1} (22420)

LW up__ SFC 353.5 {-3.6} (10938)

SW down SFC 444.3 {13.1} (11204)

SW up__ SFC 112.8 {-18.4} (5152)

LW up__ TOA 218.8 {1.4} (22885)

SW up__ TOA 261.0 {10.7} (10873)

Clear-sky

LW down SFC 291.5 {-8.7} (3500)

LW up__ SFC 400.0 {-0.7} (2263)

SW down SFC 726.1 {-0.4} (1801)

SW up__ SFC 154.1 {-22.7} (1048)

LW up__ TOA 274.8 {-0.3} (3597)

SW up__ TOA 196.5 {-0.2} (1844)

The computed LW down SFC has an all-sky (clear-sky) bias of -6.1 Wm-2 (-8.7 Wm-2); the main cause is the GEOS-4 input for surface air temperature (it's too cold). The skin temperature retrieved by CERES from MODIS is great: the clear-sky bias for LW up SFC is only -0.7 Wm-2. The clear-sky bias for insolation SW down SFC is only -0.4 Wm-2; a small bias is obtained at many individual sites, suggesting good inputs for aerosols in the time mean (this holds up for many sites in the annual mean, but it does not for others). The large biases for SW up SFC mean that the retrieved surface albedo needs some explanation. The bias for SW up TOA is 10.7 Wm-2, which is almost 4%. The application of SARB "tuning" reduces the bias in SW up TOA but increases the bias in SW down SFC for cloudy sky. By applying a new “Rev1” adjustment to the CERES SW record, as officially recommended by the Science Team, the discrepancy of almost 4% in 2001 would still remain above 3%.

In the global mean (rather than just at the ground sites above), computed SW reflection to TOA exceeds CERES observations by ~3%. We have not been able to reconcile this discrepancy. An error of this scale could be due to several factors, including the simplified scheme for broadband radiative transfer, the MODIS input data used to retrieve cloud optical properties, the assumptions in the cloud optical property retrieval, the inversion from measured radiance to observed flux, or procedures that yield CERES radiances. It is signficant, because if the earth were to suddenly reflect 3% more SW than it does at present, the planet would be out of radiative balance by ~7 Wm-2; and for the global annual mean of recent decades, we have surely have no such imbalance. The error in reflected SW for clear (cloud-free) ocean is typically 1% to 3%; here MODIS calibration has only a tiny effect on the computation, whose ocean surface albedo has been well validated with independent platform observations.