Surrogate gas prediction model as a proxy for Δ 14 C-based measurements of fossil fuel-CO 2

The measured C: C isotopic ratio of atmospheric CO (and its associated derived Δ C value) is an ideal tracer for determination of the fossil fuel derived CO enhancement contributing to any atmospheric CO measurement ( ). Given enough such measurements, independent top-down estimation of US fossil fuel-CO emissions should be possible. However, the number of Δ C measurements is presently constrained by cost, available sample volume, and availability of mass spectrometer measurement facilities. Δ C is therefore measured in just a small fraction of samples obtained by ask air sampling networks around the world. Here, we develop a Projection Pursuit Regression (PPR) model to predict as a function of multiple surrogate gases acquired within the NOAA/ESRL Global Greenhouse Gas Reference Network (GGGRN). The surrogates consist of measured enhancements of various anthropogenic trace gases, including CO, SF , and halo- and hydrocarbons acquired in vertical airborne sampling profiles near Cape May, NJ and Portsmouth, NH from 2005 through 2010. Model performance for these sites is quantified based on predicted values corresponding to test data excluded from the model building process. Chi-square hypothesis test analysis indicates that these predictions and corresponding observations are consistent given our uncertainty budget which accounts for random effects and one particular systematic effect. However, quantification of the combined uncertainty of the prediction due to all relevant systematic effects is difficult because of the limited range of the observations and their relatively high fractional uncertainties at the sampling sites considered here. To account for the possibility of additional systematic effects, we incorporate another component of uncertainty into our budget. Expanding the number of Δ C measurements in the NOAA GGGRN and building new PPR models at additional sites would improve our understanding of uncertainties and potentially increase the number of estimates by approximately a factor of three. Provided that these estimates are of comparable quality to Δ C-based estimates, we expect an improved determination of fossil fuel-CO emissions.