Spectrally resolved fluxes and cloud radiative forcing from collocated AIRS and CERES measurements: derivation and application in climate studies

Spectrally resolved outgoing thermal-IR flux and cloud radiative forcing (CRF) have unique values in climate studies and evaluating climate model simulations. Here we describe an algorithm for deriving such spectral flux and spectral CRF through the entire thermal-IR spectrum from the collocated Atmospheric Infrared Sounder (AIRS) and the Clouds & the Earth's Radiant Energy System (CERES) measurements over the tropical oceans. Based on the predefined scene types in the CERES Single Satellite Footprint (SSF) dataset, spectrally-dependent angular distribution models are developed and used to estimate the spectral flux at each AIRS channel. A multivariate linear prediction scheme is then used to estimate spectral fluxes at frequencies not covered by the AIRS instrument. The whole algorithm is validated using synthetic spectra as well as the CERES OLR measurements. Next, we show their application in climate studies by examining variations of spectral fluxes and CRFs at different timescales and comparing with counterparts from the GFDL model simulation. By comparing the observed spectral fluxes and CRFs and simulated ones over the eight longwave bands used in the GFDL model, compensating errors in the simulated OLR (or the simulated CRF) from different absorption bands are revealed. Discrepancies between the simulated and observed spatial distributions and seasonal evolutions of the spectral fluxes are further discussed. The simulated fluxes in the water vapor rotational band and ν2 vibrational- rotational band are higher than the observed through the whole tropics, while the flux difference in the window regions is correlated with large-scale circulation features. The modeled seasonal cycles of spectral fluxes show distinctly difference from the observed ones, especially in the water vapor band and CO2 band. The observed and simulated interannual variation from 2003 to 2005 is examined and the implication for model improvement is further discussed.