(1) Field work
One of the main achievements of the two main current geochemical data bases, GEOROC and PETDB lies in the fact that they have at least an approximate latitude and longitude for every sample. How they achieved this I confess I know not, many published samples are given sweeping locations such as "East Iceland" or some such. You who are are contempating of career in determining the composition of the Earth and it's components, please remember that any sample collected may one day be of key importance. Some later worker should be able to relocate it. When we were able to obtain samples of Tibetan rhyolites from the Swedish Museum collected by the Swedish explorer Sven Hedin a hundred years before, Kevin Burke was able to relocate them all by a combination of sateliite pix and Hedin's diary. Hedin was a meticulous kind of person. In years long ago I used to rely on things like, "Foot of prominent bluff, 2 miles east of Cinco Ribeiras"! Unfortunately "Cinco Ribeiras" is about as common a name as " Trois Riviere".
These days I use a GPS and advise that all do the same. Pray remember also that not all the world may be familar with a location such as "Halmundershraun". One may guess Iceland but little more.
If anyone has done a perfect field job on one visit I would like to meet him. In general one does a recconnaissance, collects a range of samples, goes back to the lab, analyses them, plots them up, reads a few papers about similar series, thinks about them for a few months, and goes back and does a proper job. I may be stupid but I goofed up a number of times by trying to perform marvels on one field site visit. I could name twenty sites world wide I would give an arm to go back to. Anyone offer me a lift back to to the Crozet Islands? What would I not give for a day or two back on Steens Mountain? Do plenty of sketches of rock relationships, ones memory fades. And always take large kilo+ multiple samples. Much of the scatter we see in diagrams shown here is due to poor sampling and coarse grainsize.

(2) Sample Numbers
Some people use samples numbers 20 characters long ( I kid you not) and no two alike. This is designed to drive lab workers nuts, who often have to allocate a lab number as well as a field number. We used to assign a 2 letter 'Project code', ie all samples from Mont Pelee would be numbered "MP0001, MP0002 etc while those from Morne Jacob would be MJ0001, MJ9999 etc. At least we never mixed samples up and I do not believe we ever analysed more than 9999 samples from one volcano! Replicates would be MPA0123 and MPB0123 etc. Some ODP sample numbers are 25 char long. Record the date, who collected it, the core number, the distance along the core, coordinates and altitude or depth somewhere else!

(3) Sample Quality
The finer grained and more glassy the sample the better. When working on submarine rocks, use chilled glasses or nothing! Never use less than a 6lb hammer when collecting. I once chipped a sample off a nephrite boulder! (It's tough stuff!). Alkali basalts, trachytes and syenites or granites are especially variable, always multiple sample each flow or outcrop. We cannot solve the problem of the range of parental magmas for OIBs, or Icelandic series or even arc andesites finally, until we have more analysed glasses at least as many as we have for the ORBs. In general it is a waste of time collecting basalts showing green with chlorite, or streaks of calcite except for elements heavier than La. Collecting granites with 5 cm crystals is in general a waste of time unless you intend to mash up a truckload.

(4) Sample Names
It does help to know what a rock is, and one should have no doubts that it the right one. When one finds that of about 4000 analysed samples named "Andesite" that the silica content ranges from 35% to 78% in fairly even array, it begins to appear as though rocks are sometimes named rather loosely. A thousand or two "rhyolites" have the same range. Rocks should be named either for their petrography or from their analysis. All dark rocks are not basalts. I once wrote a computer program in Fortran tacked on to our NORM program which looked at feldspar composition, Fe/Mg range, silica content, qtz-ne content etc etc and applied a name. It was seldom wrong. I wish I still had it at times, I do not recall it ever naming a rock of 47% silica and 11% MgO a "black dacite" as appeared in one file recently! Rocks, eg from ORBs with less than 10% normative qz are not rhyolites, I guess they are commendites. Just because an ORB has normative andesine, does not make it an "andesite" which has calc-alkaline connotations. Call it an "Icelandite". An EMORB with normative ne is not an AOB or Basanite or Tephrite. EMORBs lie on a quite different lineage, we really need some new names.

(5) Lab Procedures
About 95% of all data published is more gap than data. Suppose 40 samples are analysed for 40 elements. It is the exception rather than the rule for any one sample to be analysed for all 40, the data sets more often look like fish nets. Don't ask me to explain why! Why leave out 5 determinations for Sr when Sr can be determined to within a ppm in a time of 2 minutes by XRF? Why leave out 4 TiO2s when these take 40sec if you want precision to 3 dec. places, or 20 sec if 2 will suffice. Why leave out half the Zr's? Why do Ta but not Nb? Why don't reviewers send these papers back?
Thousands of Electron probe analyses of submarine glasses have left out the P, the Mn and sometimes even the K. At least 400 have left out either Fe, CaO, or one or other or both of Na2O and K2O. Now the three elements that vary most and are most useful in the MORB major element array are P2O5, TiO2, K, and Fe. Leave out 2 or 3 of them and you have simply wasted a great deal of time. Leave out a few hundred CaO or FeO and a nice FMA plot may look untidy!
We have about 4000 EMP analyses for ORB olivines and about 2000 for plagioclases. We do not need any more, BUT, we badly need much more on the TE. There is almost NO data for Sr, Zr, Ba on ORB plagioclases for example.
Many studies are done on one or two elements only. Suppose they are U and Th. Now we don't know whether the sample is a basalt, a phonolite, a tholeiite or a basanite. So all that has been established is "Some rocks found in a marine environment may have this range of U, Th". Not a really profound investigation. Some people do Ce, leave out La, do Nd, Sm and leave out the rest of the REE. This upsets me because I use La/Ce a lot, because it (a) is a good test of how good the data is, ie if there is scatter, the analyst is no good or the samples are awful or the rocks are not remotely related; (b) the slope give an idea of the degree of alkalinity and (c) it pleases me to see a dead straight line of dots,(when this happens). The relationship between Ce and Nd is very instructive for ORBs. If the Nd exceeds Ce you have a REALLY depleted rock. Find a key data set is missing the Nd and one has to very firmly recall our Christian heritage. Worst of all are the meteorites where the REE seem to be done only for identification puposes, usually for La,Ce, Sm, Eu and Yb,Lu only. With a great deal of interpolation sometimes use can be made of them.
Don't bother to analyse a group of unknown rocks for a single element. Guess the maximum and use a random number generator! Nevertheless it has been done more than once! If there is any point in spending a great deal of time and some institution's money doing Pb 206/204 etc for unknown, unnamed and unanalysed rocks I have yet to discover it. Especially if it is named "Mafic Inclusion!" or better still "Mafic Erratic". Some one recently analysed about 50 samples from Dominica for B only, with no indication as to whether the samples were rhyolites or basalts. Yet a referee passed it!
Others seem to get a perverse pleasure in leaving out some critical element like Nb plus one or two others like Cs Rb and Ba. So we cannot then tell if the rocks is a MORB or an IAB (in some cases). I have just been corresponding with one worker who claims a very late entry on the solidus for clinopyroxene. I don't see any wiggle in the CaO trends but if he had included Sc (which has a marked preference for CPX) the answer would be obvious. Within the last year an important paper appeared with the XRF done on one set of samples, some ICPMS done another set and some isotopes done on a third set. So not even a normalised fingerprint diagram could be had.
Moral: Do all the elements you can and do every one for each sample (except maybe for the isotopes). If you produce a fishnet data array, hope it does not come to me to be refereed or I may politely send it back with a kindly suggestion that the work would look better completed. All referees should reject partial data and if they don't, the editors should, but seemingly do not. Picking a set of samples from the major elements which give good trends and doing these only for trace elements and REE is downright unethical, but one sees it done daily.

(6) Wild Points
Many wild points turn out to be mere typos, it is especially suspicious if an analysis shows 9.18% P2O5 and the whole adds to 107.17%. On occasion whole element columns have been transposed but especially if this happens with REE it is immediately obvious. Any apparently real wild point should be double checked, if necessary on a fresh reground sample. Due to slackness of referees and editors this is seldom done.
Thousand of wild points in OIB, ORB and andesites files are due to the inclusion of xenoliths. These are often "leuco-gabbro" or "pyroxenite" or even "sedimentary hornfels" etc, they are depleted of minor elements and usually plot completely at random. Mineralogically they may be interesting, chemically, not a bit! Now that we know the compositional limits of ORBs any wild point that lies outside the envelope should be analysed at least twice. Many high-Mg "glasses" with more than 10.25% MgO have proved to be in fact fused samples done by DCP and contain ground up cumulative olivine.

(9) Data Accuracy.
We can only advance in knowledge if the data is the best we can possibly get. Ones see duplicate columns of data done by XRF and some amateur effort at ICPMS that differ by 50 - 150%. No check standards, and no "reccomended values". This as much as anything shows that referees are no longer doing their job. We used to get agreement with ID for 1-2ppm Rb within 0.1 ppm Rb about 35 years ago! I have gone into XRF labs and asked "When did you last check your Nb peak, and what pulse height descrimination do you use?" and got blank looks, "It is set in the factory!"
Factory Techs are not geochemists! In Quebec the factory Phillips senior salesman in XRF was Maurice Bonet, a good man. He was used to selling XRF to mining companies and thought that a limit of 0.1% Ni or 1% Fe was good. When I showed us measuring 2 ppm Ni and 0.25ppm Rb, he used to stand there shaking his head, "I don't see how you can do it!"! He used to show us off to customers as though we were circus freaks!
Currently there are about 6 or maybe 8 really good ICPMS analysts in the world, it is obvious who they are when you look at their results.

(8) Data plotting
One should plot out the covariance of every element with every other element plus a few differentiation indices such as Mg#. Plus a few fingerprint diagrams against Mantle or MORB, plus a few REE plots against Chondrite, Mantle, MORB etc plus a few ternary diagrams. If you have a decent set of plotting routines and using automatic plotting, this should not take more than an hour. Very few people do this which results in some awful bloopers. You data also wants to be plotted alongside several other data sets of similar rocks from elsewhere. So you may spend all day plotting, this is time well spent, believe me. Put the key ones out on a colour printer, sit down again and write it all up! This will only take 3 months! It will then be returned for rewriting by the editor and the average time before an MS is sent back the second time is six months! Sometimes more than a year! (see any journal). Remember, you can NEVER do enough plotting.
We play around with GEOKEM for various reasons. One is to explore what the textbook of the future will be like. An other is to provide much needed summaries of what is known and equally important, what is NOT known about this lump of rock and dirt we subsist on. A third reason is try to introduce a little more sophistication into data handling, plotting, and presentation. So far in 11 years we seem to have made little impact, one must confess.

(9) Other Data.
No paper should be published if it ignores all other data in the same field, yet this is done daily. Because a referee knows an author must "publish or perish" it has become standard that ALL papers are published, almost regardless of original content. We now see papers based an a dozen or so samples which claim to have solved all the problems of petrology, and which completely ignore tens of thousands of lines of existing data which would show the new data to be the same as thousands of others. The result is that we advance much more slowly than we should. I think we will attach the Macquarie Parental Magma file somewhere, so it can be used for comparison. Probably in "Oceanic Basalts, Summary". Papers based on, say 30 ORB samples often make sweeping claims for them, based on about 0.01% of the data available. This is more akin to witchcraft than science.
We should be refining the compositional envelopes of the different rock types with every project we do. Yet ask the average petrologist what rocks have the most Zr or most Nb in them and they cannot answer. Not good teaching!

(10) Illustrations.

Thanks to the digital camera, field and sample descriptions can be illustrated with a detail unheard of five years ago. Yet for reasons we cannot fathom, illustrations on net sites rate somewhere below "abysmal". Of the may thousands of sites on "Antarctica" for example the images shown are usually tiny, (small post-card size at best) fuzzy, flat-lit, and almost NEVER located. One or two interesting shots of ranges, glaciers, rocks, have appeared on the "Net". The photgrapher when contacted for a location usually replies "Between McMurdo and the Pole" which is about 800 miles. One said of a whole mountain range, "Between the Wisconsin Range and McMurdo" which is over a thousand miles. The USGS spend some time on the problem but were unable to locate it. It would be a good idea to record time and image number for every shot. If you have already recorded takeoff time and get the airspeed from the pilot, one can at least have some idea. Unfortunately GPS's do not work inside the average aircraft. About 90% of all pix taken in the Antarctic by amateurs is either unlocated or mislocated. eg one sees Mt Erebus described as "Trans-Antarctic Mountains". see www.RossSea.info An NOAA image site describes Mt Cook in New Zealand as "Part of the Trans-Antarctic Mountains!"