Characterization of cloud property retrieval uncertainty using a Markov chain Monte Carlo algorithm

Description

The multiple active and passive remote sensing measurements collected at the Department of Energy’s ARM Climate Research Facility sites enable simultaneous retrieval of cloud and precipitation properties and air motion. Recent research (Deng and Mace 2006a, 2006b, 2008) has demonstrated success retrieving cirrus cloud properties using Gaussian least-squares-based optimal estimation (OE) techniques. These methods not only return a best estimate of the cloudy state but also an estimate of the uncertainty in the retrieval. While OE algorithms are computationally efficient and return a robust solution in linear (and sometimes moderately nonlinear) cases, their general performance is unclear for strongly non-Gaussian measurement-retrieval relationships.

Markov chain Monte Carlo methods (MCMC, Posselt 2013) can be used to produce a robust estimate of the probability distribution of a retrieved quantity for nonlinear, non-Gaussian cases (Posselt et al. 2008). In this work, we highlight the utility of MCMC methods for exploring the error characteristics and information content of ARM-based cloud property retrievals. We compare the probability distributions produced by OE and MCMC algorithms, and comment on the ability of a multi-sensor retrieval to effectively constrain vertical profiles of the cloud particle size distribution.

Deng, M, and G Mace. 2006. “Cirrus microphysical properties and air motion statistics using cloud radar Doppler moments: Part I—Algorithm description.” Journal of Applied Meteorology and Climatology 45: 1690–1709.

Deng, M, and GG Mace. 2008a. “Cirrus cloud microphysical properties and air motions statistics using cloud radar Doppler moments Part II—Climatology.” Journal of Applied Meteorology and Climatology 47: 3221–3236.

Deng, M, and GG Mace. 2008b. “Cirrus cloud microphysical properties and air motion statistics using cloud radar Doppler moments: Water content, particle size and sedimentation relationships.” Geophysical Research Letters 35: L17808, doi:10.1029/2008GL035054.

Posselt, DJ, TS L’Ecuyer, and GL Stephens. 2008. ‘Exploring the Error Characteristics of Thin Ice Cloud Property Retrievals Using a Markov Chain Monte Carlo Algorithm.” Journal of Geophysical Research 113: D24206, doi:10.1029/2008JD010832.

Posselt, DJ. 2013. “Markov chain Monte Carlo Methods: Theory and Applications.” In Data Assimilation for Atmospheric, Oceanic and Hydrologic Applications, 2nd Edition, ed. SK Park and L Xu. Springer, in press.