V13E-2154 INVITED
Drowning in Geochemical Data: The Good, the bad, and the Ugly
Geochemical databases are placing unprecedented amounts of geochemical data at the fingertips of professionals and students. How these data are being used is taking an increasingly important role in shaping our thinking about the Earth. Databases have helped to expose (and eventually kill?) some long- cherished myths, such as the idea of the well-homogenized upper-mantle reservoir, and and they have made geochemical data accessible to geophysicists and enabled them to look at geochemistry with fresh eyes, leading to genuinely new insights. Yet, their very accessibility also makes them "dangerous tools" in the hands of the inexperienced. Statistical treatment of masses of geochemical data without, or with excessive, filtering can yield all sorts of "answers" we would probably be better off without. Data that are severely flawed (due to alteration or poor analytical quality, errors in published data, or errors during data entry) might not be easily identified by, say, a geodynamicist. Other dangers stem from overrepresentation of over-sampled locations and the general, but faulty, assumption of random sampling of the Earth. We will show examples where raw downloads of data from databases without extensive screening can yield data collections where the garbage swamps the useful information. We will also show impressive but meaningless correlations, e.g. upper-mantle temperature versus atmospheric temperature. The lesson is that screening is necessary. On the other hand, sound database compilations now demonstrate that average incompatible-element concentrations in global MORB are two to five times higher than published estimates. This fundamentally changes 30-year-old geochemical mass balance estimates of the mantle. OIBs are fundamentally similar to MORBs but are isotopically shifted, on average, to more "enriched" values. Mantle geochemistry is now fully consistent with dynamic models of "whole-mantle" circulation, with the likely exception of a relatively small, reservoir probably hidden near the base of the mantle. Among the most important changes triggered by the new databases are the following: (1) Driven by the database community, the standards required for reporting geochemical data in publications are changing in fundamental ways. Whereas it was common practice, especially in high-profile journals, to treat the data supporting the claims made in the published articles as relatively unimportant technical details, journal editors now require much more careful geochemical data reporting that includes complete data in numerical rather than graphical representation, information on sample locations, analytical methods, and uncertainty estimates. (2) Given the ever-increasing production rate of geochemical data, comprehensive use of geochemical data without relational databases is, or will soon be, impossible.
V13E-2155
MetPetDB: New Directions for Metamorphic Studies
It is estimated that less than 1% of the data collected on metamorphic rocks is published, and MetPetDB (database for metamorphic geochemistry) is being developed and populated to preserve these data and to foster new and innovative directions for scientific research and education. The data model is based on a sample of metamorphic rock and includes information about location, rock type, mineral assemblage, fabric, plus images of all types and mineral composition data. Mineral analyses are linked to locations on appropriate images so the spatial integrity of the data is preserved. Tools will be available for mineral recalculation, plotting, and thermobarometric applications. Derivative data such as peak P-T conditions, metamorphic P-T path, and cooling rate will also be stored. The database will be searchable based on any number of data fields, permitting rapid location of samples that can be used to test hypotheses and discover new relationships. For example: A student is designing a thesis project and MetPetDB will be a first resource to determine the types of rocks present in a region, the work that has been done on them, and links to the published findings. The Fe/Mg zoning in migmatitic garnets has been used to infer cooling rates. What is the range of cooling rates recorded by migmatitic garnets, and is there a correlation between peak metamorphic temperature and cooling rate? Is it possible that melting triggers rapid thrusting that causes the rapid cooling? A search on: rock type = migmatite plus Fe and Mg X-ray maps of garnet would reveal all samples that could be used in this study. A new geobarometer based on a specific mineral assemblage is proposed that permits pressures to be estimated to within 50 MPa. A search of the database for all samples with this assemblage plus analyses of the necessary minerals would provide a set of samples to which this new barometer can be applied. Recalculating pressures and temperatures for an entire region using new, internally consistent geothermobarometers will be possible with heretofore unprecedented ease. It has been proposed that the "stiffening" of continental crust is a function of the metamorphic temperature distribution, but it is just as likely a function of the "drying out" of the crust by metamorphic processes. Creating a regional map of the crustal "dryness" could be achieved by searching for samples with specific mineral assemblages, or for samples that reach particular peak metamorphic temperatures. Regional trends could be readily identified by plotting assemblages on Google Earth and color coding the plotting icons.
V13E-2156 INVITED
Toward a Comprehensive Model of Volcanism in Central America
In 1987, MJ Carr and WI Rose published CENTAM, a database primarily of major element analyses, but now much expanded and called CAGeochem. Our motivation included a desire to bypass editors who wanted to publish only representative analyses, and not the extensive data sets starting to be produced by semi- automated instruments. We felt a commitment to publish data that had been obtained with public funds (primarily NSF). We also provided the data in digital format, which is now trivial but was not at the time. We did not predict that the database itself would help produce the explosion of new research into the geochemistry, geophysics and tectonics of Central America. Because CENTAM provided a comprehensive geochemical outline of an entire convergent margin, Central America was recognized for several large regional variations in geochemistry that helped make it a choice of the NSF Margins Subduction Factory program. Margins and the cooperating German research programs centered at GEOMAR have revolutionized our understanding of Central America. Data and samples from CAGeochem have assisted research efforts at several universities including, Columbia, Washington Univ., Rice, New Mexico, Caltec, Boston College, GEOMAR. The primary result of the regional database was the discovery of large geographic variations in elemental and isotopic ratios (e.g. Ba/La, 10B/9Be, U/Th, 87Sr/86Sr) that trace the cycling of elements from the subducted plate. Tracers of subducted material reach maxima in Nicaragua in the center of the margin and decrease outward toward Costa Rica and Guatemala as well as decreasing across the margin. Estimates of the flux of elements, e.g. Ba and U, indicate a constant volcanic output along the margin. The regional variation occurs in La, Th etc, the denominators of the ratios, all of which change with degree of melting. There is a margin-wide correlation between element and isotope ratios that trace subduction and degree of melting. The tectonic factor that most closely mimics this pattern is dip of the subducted slab. The most reasonable explanation then is for areas of steeper dip to focus the constant slab flux into smaller volumes of mantle wedge leading to higher degrees of melting and higher ratios for Ba/La, U/Th etc. The volcanic front is segmented into discrete right-stepping lines. A major geochemical characteristic of arcs, the extent of Nb depletion, varies sympathetically with this segmented pattern. Assuming a smooth subducted slab, the Nb depletion varies with depth to the slab. Volcanoes at the NW end of a volcanic segment have shallower depths and volcanoes at the SE end have greater depths. The volcanoes make a diagonal line across a mantle wedge partial melt zone and tap different parts of the zone. Large offsets in Nb depletion occur at the steps in the volcanic front. This requires that the volcanic lines are upper plate structures. In fact, the most pronounced right steps occur just trenchward of terrain boundaries and major strike-slip faults cutting the Caribbean plate. The size of volcanoes follows the segmented pattern because the largest volcanoes are within a segment, centered over the partial melt zone and collecting a maximum amount of melt. The segmentation is likely generated after last jump in the volcanic front between 6 and 4 Ma ago. A small amount of oblique convergence, coupled with preexisting upper plate structures and a weak zone caused by magma intrusion, generated the offset lines as an initial structure (en echelon, a right-lateral, strike-slip fault zone) that has subsequently persisted.
V13E-2157 INVITED
Mining the Western North American Volcanic and Intrusive Rock Database (NAVDAT) for Insights Into the Origin of Continental Intraplate Magmatism
Space-time-composition patterns in Late Cretaceous and Cenozoic igneous activity in southwestern North America are critical to models of the evolution of the deep continental lithosphere in this region. In the 1970's, for example, consideration of igneous rocks found some 1,000km inland from the continental margin in the southern Rocky Mountains led directly to suggestions that significant variations in the dip angle of subducting oceanic lithosphere must have occurred beneath this region during the Cenozoic. But no effort has been made to refine these models for at least the past 20 years, despite the large amount of high quality age and compositional data from Cenozoic igneous rocks that has been generated over that time. These data are now readily accessible through the on-line igneous rock database, NAVDAT. For example, NAVDAT includes published age and compositional data from over 3,000 samples of mid-Tertiary volcanic rocks from the Rocky Mountains and vicinity. Interrogation of these data in toto clearly illustrates that significant spatial variations exist in the major and trace element compositions of volcanic rocks produced at the major Rocky Mountain mid-Tertiary (20-50 Ma) volcanic fields. Volcanic rocks throughout the region span the entire compositional spectrum from basalt to high silica rhyolite, but volcanic rocks of the Trans-Pecos volcanic field in west Texas are more sodic than their "calc-alkaline" counterparts further north in the Mogollon-Datil, San Juan, Absaroka and Challis volcanic fields and show little evidence for their derivation from slab metasomatized mantle source regions. In contrast, the –calc-alkaline" centers typically have significantly lower Nb and Ti contents, and higher LIL/HFSE ratios that are consistent with their derivation from mantle rocks affected by fluids derived from dehydrating oceanic lithosphere. These observations generally support the concept that fluids derived from shallow subducting oceanic lithosphere were generated as far inboard as the present-day Colorado Rocky Mountains. In fact, volcanic rock LIL and Pb contents (at any given wt% SiO2) are highest in the San Juan volcanic field in Colorado, suggesting that lateral variations may have existed in the magnitude or duration of slab metasomatism beneath the interior portion of North America during Tertiary low-angle subduction. Maximum exposure to such fluids apparently occurred in Colorado and diminished both to the north and south, with no metasomatism occurring as far to the east as the Trans Pecos volcanic field. These variations are likely related to the disposition of the subducting Farallon plate through time beneath the interior portions of southwestern North America.
V13E-2158
Optimisation methods of internally consistent thermodynamic databases for minerals: Towards realistic uncertainties
Computation of phase diagrams in mineral systems and quantitative geothermometry thrive on the availability and accuracy of internally consistent thermodynamic datasets. The prevailing two methods to derive them, mathematical programming (MAP) and least square regression (REG) have their very specific advantages and deficiencies which are to some extent complementary. Bayes Estimation (BE), the novel approach proposed recently for obtaining internally consistent thermodynamic database can combine the advantages of both MAP and REG, but avoid their drawbacks. It optimally uses the information of thermochemical, thermophysical and volumetric properties of phases and experimental reaction reversals to refine the thermodynamic data and returns their uncertainties and correlations. Therefore, BE emerges as the method of choice. However, although BE is conceptually simple, it can be computationally demanding, restricting the use of this approach in multi component systems with solid-solutions parameters, calorimetric and/or volumetric properties to be refined. Moreover, in order to simplify resulting data, previous BE studies approximated the optimized properties by normal probability distributions that allowed correlations to be estimated, but which permitted some properties to be partly outside of the feasible domain. In this study, we propose to circumvent the computationally demanding problem by rotating the feasible domain in the main component analyses directions, and by sampling it using a uniform distribution. The weight of each sampling is attributed afterward. We also propose to use directly the entire resulting samples to propagate uncertainties in phase diagrams and geothermobarometry calculations. We will compare both methods in terms of accuracy and precision. This new approach has the advantages to considerably reduce the time needed to get the optimized properties by two to three orders of magnitude and to precisely fit the probability density of the feasible domain. The potential of BE using the new mathematical concepts detailed in this study, and its future perspective for application to multi-component systems including solid-solution and aqueous species is very promising.
V13E-2159
Using Self Organizing Map with geochemical compositional data.
The Self Organizing Map algorithm (SOM) is a very well known artificial neural networks (ANN) with properties
of vector quantization and vector projection. The procedure converts complex, nonlinear statistical
relationships between high--dimensional data into simple geometric relationships on a low--dimensional
display. The basic SOM algorithm works as a clustering method by unsupervised competitive learning
that adapts the network parameters according to the presentation of new input data to the network. One of
the main results of the clustering is the reduction of the amount of data and the formation of a topologically
ordered map, by means of a nonparametric regression, of the data. This map gives the SOM the capability of
a projection method by showing the input space into a low dimensional (commonly one or two
dimensions) regular array of model vectors (neurons). In addition, the process in which such mapping is
formed allows a non--linear regression and projection of the data. The amount of applications carried out
using SOM are vast. In particular, it can be used to describe, quantize and classify in detail complex
multivariate geochemical (compositional) data.
Compositional data are vectors whose sum of their components is constrained to a constant value, for
instance, for geochemical data the 100 (wt %) closure value. This condition means that compositional data
yield only relative information about the magnitude of its components and, therefore, standard multivariate
statistical methodologies may not be directly suitable for analyzing it. In order to circumvent this problem, the
log--ratio transformation was introduced during the 80's producing non--constrained data, but inducing in the
space of the data a non--Euclidean geometric structure. Since the original SOM has been developed within
the framework of the Euclidean geometry, analysis of geochemical compositional data using SOM may not
yield coherent results.
In this contribution the original SOM algorithm and a modified version for compositional data (in which the
usual vectorial sum and scalar multiplication has been replaced by log--ratio analogue operators), are tested
with geochemical data. We found that both versions of the algorithm can be applied on compositional data
providing that the initial model vectors (neurons) are initialized also compositional. Although both
methodologies can be successfully applied, the results differ, reflecting the particular characteristics of the
geometry assumed as a framework.