MAXIMIZING THE LEGACY: MINING THE COLUMBIA RIVER BASALT GROUP DATA AND SAMPLE ARCHIVE

Thanks largely to the work of Peter Hooper and his many students over the years, including Vic Camp, Columbia River flood basalt province in the inland Pacific Northwest USA is the best-documented continental flood basalt province on Earth. High-quality maps exist for the flood lavas and their feeder dikes, plus thousands of whole-rock chemical analyses. Very few dikes are visibly connected to surface lava flows. Using the lava chemistry data base, we have developed a machine learning model (CRB ID model) to assign the probability of any unknown sample to a specific lava member, and to validate previously assigned flows. The current CRB ID model is trained on 10 normalized major elements and 19 trace elements analyzed with the current XRF method at Washington State University, and has a prediction accuracy of 99%. Here, we outline an ongoing project to process all known analyzed dike samples through the CRB-ID model with the goal of assigning them to known lava flow members, in order to maximize the value of the sample and data archive. However, much of the dike chemical data are old, and many dike analyses report only major elements, with a corresponding reduction in model prediction accuracy to 81%. Consequently, we are currently re-analyzing dike samples held as fused beads for XRF analysis in the WSU archive. The aims of the project are to improve understanding of the relationship of dikes to lavas, specifically to (1) identify dikes that have no correlative flow; (2) evaluate any dependence of eruption probability upon composition; (3) estimate dike volumes for each formation; (4) better characterize proposed northward migration of eruption sites during the lifetime of the province; (5) expand the lava and dike data base to include trace elements from LA-ICP-MS analysis of the archived beads, and including these data to train future CRB ID models to improve accuracy.