P4.1 RETRIEVAL OF WATER VAPOR OVER LAND SURFACES FROM MICROWAVE MEASUREMENTS Alan E. Lipton*, Jean-Luc Moncet, John F. Galantowicz, and Jennifer D. Hegarty Atmospheric and Environmental Research, Inc., Lexington, Massachusetts 1. INTRODUCTION Current satellite-derived water vapor analysis products have substantial shortcomings with respect to spatial coverage. Infrared instruments have very limited capabilities in cloudy areas, where the radiative signatures of water vapor content may be largely shielded from view by the opacity of the intervening clouds. Microwave measurements are skillful in many cloudy conditions, but have had marginal skill over land surfaces. Operational applications of retrievals from microwave data from sensors such as SSM/I and AMSU have been largely confined to ocean areas, where microwave signatures of water vapor conditions are strongest. The difference in retrieval skill between ocean and land has been generally attributed to differences in surface emissivity and its stability over time, but the relative importance of various factors has not been thoroughly documented. The sensitivity of microwave radiometric measurements to water vapor can be represented by the derivative with respect to water vapor amount in a layer, q l , which can be approximated as ( ) l l s s s l l l l B l B q T T q T q T ∂ ∂ ℑ - = ∂ ∂ ∂ ∂ = ∂ ∂ τ ε τ τ , where T B is brightness temperature, T is temperature, τ is the optical depth of a layer, ℑ is the transmittance to space, s denotes the surface, and ε s is the surface emissivity. The water vapor signal approaches zero whenever any of the terms on the right approach zero. The first term on the right is the contrast between the temperature of the layer and the effective temperature of the surface background. This term may be positive or negative, and may be small when T l ≈ T s and ε s ≈ 1, as is often the case over land surfaces. Provided that a significant water vapor signal is present, the ability to accurately retrieve the water vapor content depends on how distinct is that signal in relation to signals that may arise from other environmental variables. If a certain change in water vapor content would cause the same change in brightness temperatures as a plausible change in surface type (emissivity), it is impossible to determine which change occurred, unless some prior information is available. This ambiguity can be relieved somewhat by making measurements at several frequencies, but only when the emissivities vary smoothly over frequency so there is a strong correlation between emissivity in a water vapor channel and other channels. It is particularly * Corresponding author address: Alan E. Lipton, Atmospheric and Environmental Research, Inc., 131 Hartwell Ave, Lexington, MA 02421; e-mail: alipton@aer.com difficult to resolve the ambiguity between vapor and emissivity effect in cases where the vapor signal is weak. 2. EXPERIMENTAL APPROACH 2.1 Retrieval Algorithm Retrieval experiments were performed to elucidate the importance of surface emissivity in the retrieval of water vapor and cloud parameters for clear and cloudy conditions. For convenience, we examined precipitable water (PW) and cloud liquid water (CLW), as they provide a good metric for the impact on water vapor and cloud retrievals, respectively. The algorithm used for the retrieval experiments is the physical inversion method that functions as the Core Module in the algorithm set for the Conical-scanning Microwave Imager/Sounder (CMIS), which is under development for the National Polar-orbiting Operational Environ- mental Satellite System (NPOESS). Given a set of radiometric measurements of the atmosphere, the statistically most likely temperature profile is the one that minimizes the cost function ( ) ( ) ( ) ( ) 0 1 0 1 ) ( ) ( ) ( x x S x x x F y S x F y x - - + - - = - - x T y T J (Rodgers, 1976), where x is the atmospheric state vector that includes the temperature at discrete levels and may include other variables, y is a vector composed of the radiometric measurements, the operator F is a radiative transfer model that can be used to compute radiometric data from the state vector, and x 0 is an a priori estimate of x. The matrices S y and S x are the error covariances of the radiometric data and the a priori data, respectively. The matrix S y represents data noise and errors in the radiative transfer model, and is generally taken to be diagonal. The retrieved state vector x is composed of the profiles of temperature and water vapor, the difference between skin surface temperature and surface air (shelter) temperature, the cloud liquid water, cloud top pressure and thickness, and the surface emissivity at the frequency of each of the channels. In the CMIS algorithm, regularization is achieved by eigenvector transformations of the temperature profile, the water vapor profile, and the multichannel emissivities. Effectively, the algorithm retrieves the leading principal components of these variables, which are related to the original variables by a projection onto the leading eigenvectors of their respective covariance matrices. Those matrices are blocks along the diagonal of S x .