Cloud three-dimensional effects evidenced in Landsat spatial power spectraand autocorrelation functions

An analysis of nadir reflectivity spatial Fourier power spectra and autocor relation functions for solar wavelengths and cloudy conditions is presented . The data come from Landsat thematic mapper. (TM) observations, while Mont e Carlo (MC) simulations are used to aid the interpretation of the observat ions and to examine sensitivity to various factors. We show that shortwave radiative processes produce consistent signatures in power spectra and auto correlation functions. Power spectra take a variety of shapes not shown or explained in previous observational studies. We demonstrate that TM spectra call potentially be affected by radiative "roughening" at intermediate, sc ales (similar to 1-5 km) and radiative "smoothing" at small scales (<1 km). These processes are wavelength-dependent, with systematic differences betw een conservative (for cloud droplets) TM band 4 (similar to 0.8 mu m) and a bsorbing band 7 (similar to 2.2 mu m). Band 7 exhibits moro roughening and less smoothing than band 4 and faster decrease in autocorrelation. Rougheni ng is more prevalent at large solar zenith angles due to optical and/or geo metrical side illumination and shadowing. MC spectra illustrate that scale invariant optical depth fields can produce complex power spectra that take a variety of shapes under different conditions. Radiative roughening increa ses with decreasing single scattering albedo and increasing solar zenith an gle (as in the observations). For low solar zenith angles, there is a clear shift in the radiative smoothing scale to smaller values as droplet absorp tion increases. Power spectra also show stronger decorrelations between opt ical depth and reflectivity when cloud top variations are more pronounced. Finally, it is shown that power spectral analysis is a useful tool for eval uating the skill of novel optical depth retrieval techniques in removing th ree-dimensional radiative effects. New techniques using inverse nonlocal in dependent pixel approximation and normalized difference of nadir reflectivi ty yield optical depth fields which better match the scale-by-scale variabi lity of the true optical depth field.