Multi-Variable Parameterizations of Whitecap Fraction

Oceanic whitecaps, formed by wave breaking with air entrainment, are the major source of sea spray, the dominant natural aerosols in remote marine air. Sea spray aerosols (SSA) contribute to radiative and chemical processes affecting climate. SSAs affect global climate directly by contributing to the planetary albedo; indirectly by acting as cloud condensation nuclei (CCN) thus altering cloud albedo; and chemically by removing methane and surface ozone. Under certain conditions, SSAs alter the cooling effect of the sulfate aerosols by removing natural and anthropogenic sulfates from the atmosphere and by preventing the activation of sulfate aerosols as CCN. SSAs also contribute to both heat fluxes (sensible and latent), thus affecting storm intensification, and to air-sea gas exchange, including carbon dioxide (CO₂). Parameterizations of all these air-sea processes rely on the quantification of the SSA production (flux). Improved representation SSA flux can thus improve the accuracy of weather prediction, chemical, and climate models.

Parameterizations of SSA flux contain two major dependencies: (i) On the size of the sea spray droplets quantified with a size distribution (or shape factor); and (ii) On forcing variable(s) determining the flux magnitude (or scale factor). Whitecap fraction is a suitable forcing factor because whitecaps are the most direct expression of breaking waves producing sea spray. Whitecap fraction W is typically modeled as a function of wind speed U. However, field measurements do not constrain well the SSA flux predicted using the relationship W(U) alone. Other variables, beside wind, affect the formation of whitecaps and SSA flux. Satellite remote sensing observations of W can help to introduce additional variables in parameterizations of SSA flux. This, in turn, will allow global, long-term monitoring of SSA flux and processes associated with it.

We use satellite W retrievals to develop parameterizations in the form W(U, other factors). As such, these parameterizations capture not only the trend with the wind speed but also the spread of whitecap values due to natural variability and with this the variations in magnitude of the SSA flux.

We used two yearlong data sets for 2014 and 2013 as training and verification data. We consider data in different regions of the ocean, including the Tropics, the Southern Ocean, and the Gulfstream. These choices capture wide range of conditions from, respectively, moderate Trade winds to high winds over long fetches to strong atmospheric stability variations. Parameterizations were developed with two different fitting procedures: conventional multi-variable linear regression (LR) and neural network (NN). We did a sensitivity study for deep (4) vs. shallow (2) number of layers of the NN and the number of neurons in a layer (8, 16, 128). We have chosen and worked with a shallow NN with 128 neurons with an input layer having different forcing variables and an output layer giving predicted W. For each fitting case (LR or NN, training or verification), we summarized the results in a table of root-mean-square errors (RMSEs). We used the RMSEs to assess the utility and effectiveness of different W(U, other factors) parameterizations. We will describe the methodology and discuss the results of developed W(U, other factors) parameterization(s).