The Turkey Flat Blind Prediction Experiment for the September 28, 2004 Parkfield Earthquake: General Overview and Models Tested

A blind test has been conducted for site response of the Turkey Flat Site Effects Test Area, near Parkfield, California to the September 28, 2004 M6.0 Parkfield Earthquake. Acceleration time histories recorded on bedrock near one valley edge were provided along with detailed geotechnical properties at all recording sites to interested participants. Participants were asked to make predictions of the ground motions at the five other recording locations (3 surface and 2 downhole), while records were withheld by CSMIP until all predictions were received and officially logged. A second series of predictions were made based on ground motions recorded in bedrock beneath the valley floor. In all there were 15 participating teams from 4 countries that submitted a total of 92 sets of blind predictions, 45 for Part 1 and 47 for Part 2 of the test. Of the 92 sets of predictions, 18 are from industry, 54 from academia, and 20 from government sectors. Site response models tested fall into three categories, with 55 prediction sets based on equivalent linear methods, 33 based on non-linear, and 4 based on other approaches. Predictions generally overestimate valley response, with little improvement using independently interpreted verses standard geotechnical models. Results indicate a marked improvement for site-response predictions based on input of actual rock motions recorded immediately beneath the valley sediments. The strong-motion test demonstrates that surface motions at a valley edge may not always be reliable input motion for estimating site response of valley sediments. This paper provides an overview of the Turkey Flat strong-motion blind test, and summarizes the various soil models and computational models submitted for the test. A companion paper in these proceedings discusses quantitative results of the blind test and comparison with other data.