Longwave band-by-band cloud radiative forcing over the tropical oceans from 2003 to 2007: observations vs. simulations

Recent studies by the first author and his collaborators make it possible for the first time to directly compare the GCM-simulated CRF over each individual longwave spectral band with counterparts derived from collocated AIRS and CERES radiance measurements. Such evaluation of band-by-band CRF provides a unique way to understand the longwave cloud feedback from the spectral domain. Here we derived the band-by-band CRF over the tropical oceans from 2003 to 2007 directly from coincidental AIRS and CERES measurements, characterized both the mean states and variability, and then used them to evaluate two GCMs, GFDL AM2 and NASA GEOS-5. For averages over the entire tropical oceans, observational analysis show that fractional contributions from water vapor bands, the CO2band (560-800cm-1), the first window region (800-990cm-1), and the second window region (1070-1200cm-1) to the LW broadband CRF are 19.7%, 15.2%, 34% and 15.7% , respectively. Such partitioning among bands shows remarkably little change from year to year, with largest change no more than 0.38%. Over all timescales, the variation of CRF in each band is highly correlated with each other as well as with the variation of LW broadband CRF, which is the well-known Stefan-Boltzmann feedback. However, the fractional contribution to LW broadband CRF behaves differently. The CRF in the H2O band is significantly anti-correlated with those in window regions but only weakly correlate with that in the CO2 band, which can be interpreted primarily by the Wien’s displacement law. Though the tropical average changes little, the spatial distribution does change. EOF analysis shows that, for each band, the first two modes are related to ENSO and explain 59% of the total variance. The preliminary results from GFDL AM2 and NASA GEOS-5 show that, although their tropical annual-mean longwave broadband CRFs are similar to each other (28.13 vs 28.3Wm-2) and the band-by-band breakdown of the CRF is generally consistent with observations, relative difference in some bands could be as large as 25%-50%. The spatial maps of band-by-band CRF also show considerable difference between two GCMs as well as between GCMs and observations. These results indicate that GCMs produce the seemingly good agreement on broadband CRF by different tuning approaches, which can be determined by such band-by-band comparison. The interannual variations of simulated band-by-band CRF as revealed by EOF analysis are also discussed.