Quantitative Evaluation of Retrievals of Cloud Properties from Spectral Transmittance and Reflectance

With passive measurements of shortwave radiation at discrete channels the unique spectral features from clouds and other drivers of climate, such as aerosols, water vapor, and different surface types, cannot be completely discriminated from each other. This lack of discrimination is one source of uncertainty in the retrieved products. Approaches to effectively characterize retrieval uncertainty to measurement error and uncertainty in forward model inputs such as atmospheric state and surface type are essential to improve our understanding of clouds and their relation to radiation. To this end, we apply the GEneralized Nonlinear Retrieval Analysis (GENRA) approach, which utilizes the Shannon information content and maximum likelihood estimate as diagnostics to quantify the information inherent in a measurement and the retrieval accuracy. Similar approaches can also guide the use of large amounts of satellite data anticipated with the movement of Earth-viewing satellite missions towards hyperspectral measurements. In the near term, we focus on spectral measurements and simulated observations of reflected and transmitted cloud radiation by the Solar Spectral Flux Radiometer.

In this work, we briefly outline the methodology and present results that guide the efficient and effective use of spectrally resolved cloud measurements at approximately 400 independent channels spanning 300 nm to 2500 nm. We examine the dependencies of the information content of cloud albedo to variability in gas absorption amounts, surface features, and cloud top height. We also quantitatively analyze and assess the cumulative information of a cloud retrieval statistic.

We present examples of cloud retrieval bias and reductions in retrieval information content that occur when the spectral features of different climate drivers cannot be uniquely extracted in the measured signal. These conditions encompass the challenges of current discrete-band spectral imagers and retrieval statistics based on a handful of measurement channels. In particular, we investigate reflected cloud radiation that interacts with an absorbing aerosol layer above, but physically distinct from, a cloud and the impacts of variable surface type and atmospheric state on ground-based measurements of radiation transmitted through a cloud.

We conclude by investigating a broad range in measurement accuracy on the uncertainty in retrieved cloud properties using simulated measurements of reflected cloud radiance above a bright (snow) surface at the discrete shortwave channels of the Moderate Resolution Imaging Spectroradiometer (MODIS). The range in measurement accuracy spans 3% to 0.3%; the typical performance of current imagers and the expected performance for future passive imagers like the HyperSpectral Imager for Climate Science (HySICS).