| Summary: | This study addresses the problems associated with each step of the process of retrieving radiation budgets at the top of atmosphere (TOA) and at the surface from satellite measurements in the Arctic. The data used are from the Earth Radiation Budget Experiment (ERBE) radiometer and Advanced Very High Resolution Radiometer (AVHRR), together with radiative transfer calculations. The limitations of the ERBE scene identification algorithm when applied to observations in the Arctic are investigated by comparing the scenes identified by ERBE with simultaneous and colocated AVHRR-based scenes. Considerable discrepancies are found, especially the misidentification by the ERBE scene identification method of clear fractional ice as being partly cloudy skies over open water. More accurate TOA fluxes and cloud forcing in the Arctic are derived by taking advantage of the broadband radiance measurements made by the ERBE radiometer and the more reliable scene identification deduced from narrowband multispectral measurements made by the AVHRR. Comparisons of the cloud forcing determined from the two scene identification methods show differences as large as 50 W m$ sp{-2}$ in some regions of the Arctic. The validity of the ERBE angular dependence models (ADMs) is also evaluated in the Arctic. It is found that the ADM for clear ice/snow is not reliable when applied in the Arctic during summer. The ADM for overcast skies contains systematic errors when applied to overcast conditions over ice/snow surfaces. The systematic error is removed by modifying the ADM. To estimate the surface absorbed flux from reflected flux at the TOA, a simple parameterized model is developed based on comprehensive radiative simulations. The model is independent of cloud optical thickness, surface type, and has only a moderate dependence on the presence or absence of cloud, cloud type and water vapour, but a strong dependence on solar zenith angle (SZA). Surface absorbed fluxes determined by radiative transfer calculations may generally be estimated to within 10 W m$ sp{-2}$ by the simple model from TOA reflected fluxes with knowledge of the SZA and precipitable water. Finally, narrowband-to-broadband reflectance conversion is carried out with careful attention paid to a particular statistical problem, namely, regression analysis with spatially autocorrelated satellite measurements. Both scene-dependent and scene-independent regressional models are developed to estimate broadband reflectance from the narrowband reflectances in channels 1 and 2 of AVHRR. The RMS errors in the percentage reflectances from the scene-dependent models are 1.0, 1.8, 2.0 and 3.1 for the ocean, land, ice/snow and cloud, respectively. Except for land, the scene-independent model does virtually as good a job as the scene-dependent models.
|
|---|