Some new and exciting capabilities are in the next IRIS proposal, and include the management of one copy of most of the data IRIS manages on on-line disk based storage systems; enhanced techniques for users to access data at the IRIS DMC; the acquisition of a significant computational resource in the form of a cluster attached directly to the on-line data holdings, and the development of workflow systems that will enable users to apply algorithms directly to the data holdings of IRIS. We believe that the opportunity for significant activities in data mining and complex algorithmic execution should assist users in more fully exploiting the rich observational data archive at the IRIS DMC.


John Rundle (UC Davis) and Andrea Donnellan (Jet Propulsion Laboratory)
Process, Pattern, Prediction: Using Space Data to Understand and Predict Complexity in Driven Dynamical Earth Systems

Thursday 4:30-5:00, Fountain II

Abstract:

With the advent of new space data types such as Synthetic Aperture Radar Interferometry for the measurement of earth crustal motions, data will be available in rapidly increasing quantities for the study and interpretation of the internal dynamics of the earth. Transforming data into information, and furthermore into forecasts and predictions of future states of the system, implies the development and use of new types of nonlinear models and simulations of the fundamental processes. However, Edward N. Lorenz discovered that chaos and unpredictability are hallmarks of even simple driven systems. Predicting the future evolution of a variety of driven nonlinear systems is further complicated by the fact that their dynamical processes are 1) often not amenable to direct observation; and 2) are strongly multi-scale, so that length and time scales range from very much smaller and shorter than human perception, to very much larger and longer. An example of such systems is the atmosphere, in which, from a practical standpoint, it is impossible to measure the temperatures, pressures, and humidity at all locations at all times. Other important systems include neural networks and earthquake fault systems, both of which are examples of driven threshold systems. In systems such as these, we can only observe the space-time patterns of extreme events. Using these space-time patterns, and whatever is known about the dynamics of these high-dimensional nonlinear earth systems, it often possible to construct numerical simulations that can be used to make predictions about the future space-time evolution of the system and the possible occurrence of extreme events. The accuracy of these predictions and forecasts is limited by the proximity and similarity of the model trajectory through state space, to that of the actual system. The existence of flexible new Grid computing techniques made possible by the World Wide Web has opened new avenues for the realization of sophisticated, state-of-the-art numerical simulations. Thus our ability to forecast the extreme events of the future is limited by a range of issues originating from the dynamical process of interest, the space-time patterns we can observe, and the accuracy of the predictions that are desired.


Carola Tiede (Fraunhofer IPSI)
Some Ideas about the Use of Global Sensitivity Analyses in Geoscience Applications

Thursday 5:00-5:30, Fountain II

Abstract:

The talk gives an overview about some sensitivity analysis applications, their need and some aspects on how to improve modelling attempts with this tool. Generally, in the geosciences, sensitivity analyses are computed before the optimization step, where the unknown parameters of the anticipated model are to be determined. In this case, the sensitivity analysis provides information about the ability to quantify the unknown parameters from the measured data. This is equivalent to determining whether or not the measured data are sensitive to changes in the anticipated model parameters. In particular, two specific applications of global sensitivity analyses are discussed here:

1) For the detection of changes in highly active regions (e.g. earthquake or volcanic), geodetic networks are installed in order to monitor changes originating from the active source. The installation and maintenance of such monitoring networks is very time consuming. Consequently, their configuration is of essential importance in order to reduce the time factor and increase measurement efficiency as much as possible. Often, some prior information about the location as well as about the physical parameters of such an active source (e.g. a fault zone for seismic active regions or a magmatic source for volcanic monitoring) are known. Normally, the network points are installed according to this prior anticipation, with the aim of determining the unknown parameters of the expected underlying mathematical model. Global sensitivity analyses can assist in the estimation of the most effective configuration of these monitoring networks, due to their ability to rank the sensitivity of each observation point relative to the changes in the unknown parameters of the anticipated model.

2) Often the mathematical model for an active region is not determined by a single kind of measured data. Rather, various and different data types have to be fused into one model which has the ability to explain all data sources in common. For such a data fusion, appropriate weight factors of the different data sources become indispensable. Global sensitivity analyses can provide information about the most appropriate weight factors of these data. This method allows the determination of a common model by the use of all data sources with respect to the sensitivity of the different data concerning changes in the unknown parameters of the model.