A New Normalized Difference Cloud Retrieval Technique Applied to Landsat Radiances over the Oklahoma ARM Site

Abstract

The authors propose a new cloud property retrieval technique that accounts for cloud side illumination and shadowing effects present at high solar zenith angles. The technique uses the normalized difference of nadir reflectivities (NDNR) at a conservative and an absorbing (with respect to liquid water) wavelength. It can be further combined with the inverse nonlocal independent pixel approximation (NIPA) of Marshak et al that corrects for radiative smoothing, thus providing a retrieval framework where all 3D cloud effects can potentially be accounted for. The effectiveness of the new technique is demonstrated using Monte Carlo simulations. Real-world application is shown to be feasible using Thematic Mapper (TM) radiance observations from Landsat-5 over the Southern Great Plains (SGP) site of the Atmospheric Radiation Measurement (ARM) Program. For the moderately oblique (45°) solar zenith angle of the available Landsat scene, NDNR gives similar regional statistics and histograms when compared with standard independent pixel approximation (IPA), but significant differences at the pixel level. Inverse NIPA is also applied for the first time on observed high-resolution radiances of overcast Landsat subscenes. The dependence of the NIPA-retrieved cloud fields on the parameters of the method is illustrated and practical issues related to the optimal choice of these parameters are discussed. It is natural to compare novel cloud retrieval techniques with standard IPA retrievals. IPA is useful in revealing the inadequacy of plane parallel theory in certain situations and in demonstrating sensitivities to parameter choices, parameterizations, and assumptions. For example, it is found that IPA has problems in matching modeled and observed band-7 (2.2 ?m) reflectance values for ?6% of the pixels, most of which are at cloud edges. For simultaneous cloud optical depth?droplet effective radius retrievals (where a conservative and an absorptive TM band are needed), it is found that the band-4 (0.83 ?m)?band-7 pair was the most well behaved, having less saturation, smaller changes in nominal calibration, and better overall consistency with modeled values than other bands. Mean values of optical depth, effective radius, and liquid water path (LWP) for typical IPA retrievals using this pair are τ = 22, re = 11 ?m, and LWP = 157 g m?2, respectively. Inclusion of aerosol scattering above clouds results in ?8% decrease in mean cloud optical depth for an aerosol optical depth of 0.2. Degradation of instrument resolution up to ?2 km has small effects on the optical property means and histograms, suggesting small actual cloud variability at these scales and/or radiative smoothing. Comparisons with surface instruments (microwave radiometer, pyranometer, and pyrgeometer) verify the statisitical adequacy of the IPA retrievals. Last, cloud fractions derived with a simple threshold method are compared with those from an automated procedure called Automatic Cloud Cover Assessment now in operational use for Landsat-7. For the northernmost 2000 scanlines of the scene, the cloud fraction Ac is 0.585 from thresholding, as compared with Ac = 0.563 for the automated procedure, and the full scene values are Ac = 0.870 and Ac = 0.865, respectively. This suggests that the Landsat-7 automated procedure will likely give reliable scene-averaged cloud fractions for moderately thick clouds over continental U.S. scenes similar to SGP.