Oxford, 29-31 January 2007: Kevin’s discussions with Gideon and Ros

Data compilation

Pa/Th:              Most data on Gideon’s website (U-Series Database). Mark Siddall’s may be more up-to-date. There are new records: Thomas et al., Scholten et al. Also old (Kumar 1995, Marcantonio 2001) records, considered it a proxy of productivity at those sites.

Cd/Ca:             Ros has most data required. Should be able to get a copy soon? And info about missing data.

13C:                 Compilation being done by Babette and Kevin (along with 18O, % carbonate etc), well underway. No formatting or processing (e.g. dating) yet.

14C dating is possible in a minority of cores. Wiggle-matching necessary (ahem).

Plan for final dataset (much applies to 13C only):

No gridding of data except visualisation.

                        Final dataset should have the following information (fields in italics are optional and may be inactive or uniform, at least at first):

Lat

Lon

Water depth

Sediment depth

Age

Age uncertainty^1

Raw^2 value

Uncer-tainty^3

Species flag^4

Meta-data flag^4a

¹ As age PDF is likely to be highly non-Gaussian, percentiles may be better than a single number.

² Vital effect may be “removed” from “raw” data in some cases. This may need to be flagged.

³ Exclude model uncertainty and uncertainty associated with vital effect. Do we include Mackensen effect here? Not specifically – no “correction” in publications. Should be kept separate from “vital effect”, for if we use sedgem.

⁴ Include planktonic/benthic flag, e.g. 00-09 = planktonic, 10-19 = epifaunal, 20-29 = infaunal (no clear distinction between epifaunal and infaunal)

^4a Separate table will associate publication and any other metadata (lab, standard, size fraction) to metadata code.

This should be provided in combination with species-specific information, converting the following inputs to the following outputs:

INPUTS			OUTPUTS
Species flag	Lat	Lon	Vital effect correction	Vital effect uncertainty	Season flag^5

⁵ For planktonic species, seasonal rather than annual-mean value may be recorded.

The data directly used in model-data comparison should be:

Lat

Lon

Water depth

Age

Age uncertainty^1

Correctedvalue

Data unceratinty

Season flag

¹ As age PDF is likely to be highly non-Gaussian, percentiles may be better than a single number.

The goodness-of-fit of the model to the data can then be estimated using a model snapshot (or annual/equilibrium field), taken to represent a given point (or period) based on the following model output:

Lat

Lon

Water depth

Age

Model location uncertainty (x,y,z)^6

Modelled value

“Model error”

Covar-iance infor-mation^7

Other weight-ing^8

⁶ There is uncertainty that a location in the model represents the same location in reality due to e.g. coarse topography, thick western boundary current.

⁷ The method described will strongly weight towards data rich areas, which is only partly desirable. Model convariance could be used to objectively mitigate this. However, Gideon pointed out that this covariance may not exist in reality (and therefore in the data). A method that is independent of model behaviour may be preferable?

⁸ We may wish to up-weight some locations if we think they are more pertinent to the particular question we’re addressing?

Proxies in GENIE (biogem)

Initial plan is to take water-column values. We probably won’t need sedgem until long after it’s uploaded.

Cd:                   Andy’s put this in GENIE as function of P (and optionally Fe) at present. The temperature effect is not there.

13C:                 This is in GENIE, but the equations all need checking. Although Andy’s got them from papers, the authors may have plucked them from thin air….

Pa/Th:              I actually have to put this in myself. Initial plan is to reproduce Siddall results. Different ways of representing falling particles in Bern3D-GENIE is problem in principle, but unlikely to affect results and GENIE scheme can be modified. The presence of an active carbon cycle should enable an immediate advance on Siddall study.

Early papers

Given that a Cd/Ca paper might put us in competition with Andy, there are the following possibilities:

1) The 13C compilation, as a pure data paper (i.e. no GENIE). In that case, we want to do as little interpretation as possible, but apparently G³ accept such papers? Very useful (if tedious) and fairly straightforward.

2) A 13C paper using GENIE that is analogous to the Siddall et al. Pa/Th paper. In this context we could probably have Cd/Ca and/or Pa/Th in the study as well. Probably, we will find some proper science in this (it won’t just be “look GENIE does 13C quite well”).

3) Given that we will be in a position to use Cd and Pa/Th data before 13C, is there a study we can do using these proxies together that doesn’t just duplicate other people’s work?