Decision Tree Analysis to Identify Factors that Impact Monomethylmercury Fraction in Wet Deposition
Dennis Jackson1 and Stephen P. Harris2
Numerous processes are postulated for the source of monomethylmercury (MMHg) in wet deposition. Competing theories on the source(s) of MMHg in wet deposition include volatilization of gaseous MMHg, evasion and demethylation of dimethylmercury, and direct methylation of Hg0 in the atmosphere. Hammerschmidt and others (2007) suggest that MMHg may be formed in the atmosphere through a reaction between labile Hg(II) complexes and an unknown methylating agent(s), potentially acetate or similar molecules. In this investigation we apply next generation data mining techniques to explore potential relationships between observations obtained from the NADP’s National Trends Network and the Mercury Deposition Network to identify potential factors that may be associated with elevated fractions of MMHg.
In this analysis a partition (decision tree) model was developed using paired weekly data for NADP sites that participate in both the National Trends Network and the Mercury Deposition Network, including the quantification of methyl mercury in precipitation. The partition modeling process recursively partitions data according to a relationship between the independent (NTN results) variables and the dependent (fraction of MMHg) variable, creating a tree of partitions. The analysis process exhaustively searches all possible cuts or groupings of the independent variables to identify cuts or groupings that best predict the dependent variable. These splits (or partitions) of the data are done recursively forming a tree of decision rules until the desired fit is reached.
1Savannah River National Laboratory, email@example.com 2Savannah River National Laboratory, firstname.lastname@example.org