Manuscript Number:	bg-2012-487 
Title:			A novel method for diagnosing seasonal to inter-annual surface ocean carbon dynamics from bottle data using neural networks
Authors:		Tristan P. Sasse, Ben I. McNeil and Gab Abramowitz 
Affiliations:		Climate Change Research Centre, Faculty of Science, University of New South Wales, Sydney, Australia
Corresponding author: 	T. P. Sasse, Climate Change Research Centre, Faculty of Science, University of New South Wales, Sydney, NSW, 2052 Australia. (t.sasse@unsw.edu.au)

File name:	Supplementary_material.pdf
contains:	Details relating to the following issues (note: the order presented here matches that of the pdf file)
		A)	Identifying mixed-layer measurements - Details the method used to identify mixed-layer measurements.
		B)	Identifying coastal data	- Summarizes the method employed to discriminate coastal measurements from open-ocean samples.
		C)	Anthropogenic correction for Ct measurements - Outlines the method used to normalize bottle dissolved inorganic carbon measurements to year 2000.
		D)	Significance of anthropogenic Ct correction - Describes and presents results to determine the influence of our Ct normalization approach.
		E)	Supervised SOM			- Detailed information relating to the supervised SOM adaption.
		F)	Principle Component Regression	- Outlines the PCR method.
		G)	Evaluating the effectiveness of a bathymetric approach in identifying coastal data - Tests the effectiveness of using a bathymetric approach to identify coastal data.
		H)	Are the neurons capturing the system? - Investigation of the potential biases relating to our primary independent test approach to constrain the optimal parameter combination.
		I)	SOMLO model without Arctic measurements - Testing the influence of Arctic Measurements on the global model.
		J)	Stochastic nature of the SOM	- Evaluating the stochastic nature of the SOM initialization.
		Table S1) Skill comparison between coastal and open-ocean predicted measurements.
		Table S2) RSE values (mol kg-1) for models under optimal configurations and two increases in neuron map size.
		Table S3) Independent test RSE values for data below 70N.
		Table S4) RSE results for stochastic initialization test.
		Figure S1) Global and Mauna Loa site CO2 difference between in-situ year and the year 2000. 
		Figure S2) Correction factor applied to Ct measurements.
		Figure S3) Annual delta(RSE) between CT models trained with and without anthropogenic corrections.
		Figure S4) Principle Component Regression schematic.
		Figure S5) Distribution of measurements assigned to a neuron containing at least one Arctic measurement. 
		
File name:	Supplementary_tables.docx
Contains:	Table T1) Ad-hoc and universal CT regression equations with interaction terms (Int.).
		Table T2) Ad-hoc and universal AT regression equations with interaction terms (Int.).

File name:	Supplementary_figures.pdf
Contains:	Figure F1)	Global map after Longitude shift.
		Figure F2)	Geographical separation of independently predicted dataset into 14 regions.



	
		   