NG22A-01 INVITED 10:20h
Predicting Tomorrow's Weather Next Week: The Roles of Uncertainty, Probability, and Model Inadequacy in Weather and in Climate
The aim of constructing a forecast from the best model(s) simulations should be distinguished from the aim of improving the model(s) whenever possible. The common confusion of these distinct aims in earth system science sometimes results both in the misinterpretation of results and in a less than ideal experimental design. The motivation, resource distribution, and scientific goals of these two aims almost always differ in the earth sciences. The goal of this talk is to illustrate these differences in the contexts of operational weather forecasting and that of climate modelling. We adopt the mathematical framework of indistinguishable states (Judd and Smith, Physica D, 2001 & 2004), which allows us to clarify fundamental limitations on any attempt to extract accountable (physically relevant) probability forecasts from imperfect models of any physical system, even relatively simple ones. Operational weather forecasts from ECMWF and NCEP are considered in the light of THORPEX societal goals. Monte Carlo experiments in general, and ensemble systems in particular, generate distributions of simulations, but the interpretation of the output depends on the design of the ensemble, and this in turn is rather different if the aim is to better understand the model rather than to better predict electricity demand. Also, we show that there are alternatives to interpreting the ensemble as a probability forecast, alternatives that are sometime more relevant to industrial applications. Extracting seasonal forecasts from multi-model, multi-initial condition ensembles of simulations is also discussed. Finally, two different approaches to interpreting ensembles of climate model simulations are discussed. Our main conclusions reflect the need to distinguish the ways and means of using geophysical ensembles for model improvement from their applications to socio-economic risk management and policy, and to verify the physical relevance of requested deliverables like probability forecasts.
NG22A-02 INVITED 10:35h
Short-Term Earthquake Prediction Based on the Reverse Tracing of Lithosphere Dynamics
We describe the methodology for detecting short -term earthquake precursors with characteristic lead time months. Physical mechanism underlying this methodology comprises interaction of two processes in a fault network: accumulation of energy that the earthquake will release; and triggering that release by decrease of the network's integral strength. These processes are captured by an ensemble of seismicity patterns. Methodology detects such patterns by tracing dynamics of seismicity backwards in time; accordingly, it is named "Reverse Tracing of Precursors" (RTP). The principles of PTP are not specific to earthquakes and might be applicable to the wider class of hierarchical non-linear systems. Methodology is put to test by experiment in advance prediction of earthquakes in several regions.
NG22A-03 10:50h
Predicting Earthquake Occurrence at Subduction-Zone Plate Boundaries Through Advanced Computer Simulation
In general, predicting the occurrence of earthquakes is very difficult, because of the complexity of actual faults and nonlinear interaction between them. From the standpoint of earthquake prediction, however, our target is limited to the large events that completely break down a seismogenic zone. To such large events we may apply the concept of the earthquake cycle. The entire process of earthquake generation cycles generally consists of tectonic loading due to relative plate motion, quasi-static rupture nucleation, dynamic rupture propagation and stop, and restoration of fault strength. This process can be completely described by a coupled nonlinear system, which consists of an elastic/viscoelastic slip-response function that relates fault slip to shear stress change and a fault constitutive law that prescribes change in shear strength with fault slip and contact time. The shear stress and the shear strength are related with each other through boundary conditions on the fault. The driving force of this system is observed relative plate motion. The system to describe the earthquake generation cycle is conceptually quite simple. The complexity in practical modeling mainly comes from complexity in structure of the real earth. Recently, we have developed a physics-based, predictive simulation system for earthquake generation at plate boundaries in and around Japan, where the four plates of Pacific, North American, Philippine Sea and Eurasian are interacting with each other. The simulation system consists of a crust-mantle structure model, a quasi-static tectonic loading model, and a dynamic rupture propagation model. First, we constructed a realistic 3D model of plate interfaces in and around Japan by applying an inversion technique to ISC hypocenter data, and computed viscoelastic slip-response functions for this structure model. Second, we introduced the slip- and time-dependent fault constitutive law with an inherent strength-restoration mechanism as a basic equation governing the entire process of earthquake generation. Third, combining all these elements, we developed a simulation model for quasi-static stress accumulation driven by relative plate motion. Fourth, we also developed a simulation model for dynamic rupture propagation on a 3D curved plate interface by applying BIEM. Finally, to simulate the complete earthquake generation cycle, we couple these quasi-static and the dynamic models on the Earth Simulator, which is a high-performance, massively parallel processing computer system with 10 TB Memory and 40 TF peak speed. With this system, given the past slip history and the present stress state, we can predict the next step fault slip and stress changes through computer simulation. As an example of predictive simulation, we show the quasi-static process of stress accumulation at the source region of the 1968 Tokachi-oki earthquake, northeast Japan, and the subsequent dynamic process of rupture initiation, propagation and stop. In this simulation we forced dynamic rupture to start by giving an artificial stress drop, which corresponds to some external disturbance. The dynamic rupture is accelerated, if the stress state is in critical. Otherwise the started rupture is not accelerated. This indicates that the stepwise predictive simulation with the real-time data of stress states at plate interfaces is crucial for the prediction of large interplate earthquakes.
NG22A-04 11:05h
Matching physics-based models and data using wavelet-based fractal characterizations
Characterizing complex physical processes using the wavelet transform provides valuable insight into both the processes themselves, and the efficacy of simulation methods. Analyzing physical systems ranging from experimental systems exhibiting instabilities driven by shock dynamics to global ocean models, we use scaling information derived from 2-D wavelet transforms of images to determine where observations and simulations match, and how patterns evolve in space and time. The variety of fractal and multifractal techniques that are available to us through this wavelet-based approach has been crucial, as we have obtained useful metrics ranging from changes in monofractal (power-law) behavior to changes in higher-order moments of multifractal behavior, and taken full advantage of the spatial resolution to obtain estimates of local Holder characteristics. These metrics are critical to the process of determining where observations and simulations of diverse highly non-linear systems match. Having honed the techniques in repeatable, experimentally driven systems, we are now applying the same approach to comparing geophysical satellite data and simulations from the Global Ocean models run on massively parallel systems at Los Alamos National Laboratory (LANL). In particular, we are interested in how the temporal and spatial grid scales of the model and resolution of the satellite affects matches in dynamic regions of the ocean known to have substantial influence on global carbon fluxes. These methods of deriving physics-based metrics that allow data characterizations to inform simulations are a step toward ultimately improving predictive capability, scale by scale.
NG22A-05 INVITED 11:20h
Space Weather Forecasting: Integrated Model Based on Nonlinear Dynamics and Statistical Physics
The magnetosphere is an open system driven by the turbulent solar wind and exhibits complex behavior with global and multiscale characteristics. The multiscale behavior is characterized by power law distributions and arise due to, in part, the turbulence in the solar wind. On the other hand the overarching global dynamical behavior originate mainly from the internal dynamics and is evident in processes such as plasmoid formation and release. The global nature of the magnetosphere is characterized by low-dimensionality and is evident in the numerical simulations using global MHD models. The recognition of the capability of nonlinear dynamical models to capture the inherent features in the data forms the basis for the data-derived models of the coupled solar wind - magnetosphere system. The models of the global behavior use the dynamical trajectories in the reconstructed phase space and a mean field approach based on averages over nearest neighbors has been used for space weather forecasting. The multiscale aspects may not be predicted based on dynamical considerations. However the deviations from the mean field forecasts are closely related to the driver, i.e., the solar wind, and a Bayesian approach is used to compute the conditional probabilities from the solar wind and magnetospheric data. The probability density functions are computed using the leading eigenvalues from a principal component analysis. The predictions of the global features, based on nonlinear dynamical modeling, and the likelihood of the deviations from them, based on statistical physics considerations, provide an integrated space weather forecasting technique. The data from different phases of the solar cycle, and the corresponding levels of geospace disturbances are used to yield accurate predictions. Phase transitions, which exhibit global behavior (first order) and scale invariance (second order), provide a framework for the global and multiscale phenomena underlying the space weather forecasts.
NG22A-06 11:35h
Elasto-Dynamic and CA Simulation Model Studies of the Critical Point Hypothesis
Elasto-dynamic numerical simulation models and CA provide a means to study the dynamic and complex system behaviour of fault systems. Various observational, theoretical and computational evidence suggests that crustal fault systems exhibit critical point like behaviour in which correlations in the stress field grow as smaller earthquakes modify the stress field and prepare the system for the occurrence of a large event. Here, we present and compare results of three different simulation models, 2D and 3D finite element simulations of interacting fault systems, 2D and 3D lattice solid particle simulations of granular zones, and CA models of fault systems. The results suggest that stress within fault systems does evolve and with some characteristics of critical point systems, and that there may be only two regimes of phase space, one in which earthquakes are forecastable and a second in which they are not. Comparisons between the CA and elasto-dynamic models suggest that typical elasto-dynamic systems may lie near the transition between these regimes determined by tuning parameters such as viscosity, fault density and geometry, and deformation rates.
NG22A-07 INVITED 11:50h
Pattern informatics and its application for forecasting large earthquakes in Japan
The 17 January 1995 Kobe, Japan, earthquake was only a magnitude 7.2 event and yet produced an estimated \$200 billion loss. The magnitude of potential loss of life and property is so great that reliable earthquake forecasting should be at the forefront of research goals, especially in Japan. An approach to earthquake forecasting is Pattern Informatics (PI). The PI technique can be used to detect precursory seismic activation or quiescence and make earthquake forecasts. Application to earthquake data from southern California shows that this method is a powerful technique for forecasting large events. Here, we attempt to forecast Japan earthquakes using the PI method. To insure the completeness of earthquake catalog maintained by Japan Meteorological Agency, events in 1955-1994 around the epicenter of the Kobe event are used for our analyses. This is done for forecasting the occurrence of large future events that are the earthquakes of magnitude greater than 5 for the time period 1995-present, including the Kobe event. Optimizing parameters of the PI method needs to be performed. We also change the extent of our study area to determine the optimal application of the method. Our results show that the method has skill for forecasting the spatial and temporal distribution of the large future earthquakes. Specifically, we find that the occurrence of the Kobe event can correspond to a seismically anomalous region. We further use two statistical tests to evaluate the accuracy for forecasting the large future events. The results of these tests also support that the method has some forecast skill.
NG22A-08 INVITED 12:05h
Forecasting Shoreline Position
Analysis of historical shoreline positions on sandy coasts, in the geologic record, and study of sea-level rise curves reveals that the dynamics of the underlying processes produce temporal/spatial signals that exhibit power scaling and are therefore self-affine fractals. Self-affine time series signals can be quantified over many orders of magnitude in time and space in terms of persistence, a measure of the degree of internal correlation in the stochastic portion of a time series. Fractal statistics developed for self-affine time series are used to forecast a probability envelope bounding future shoreline positions. The envelope is the (+-) standard deviation as a function of three variables: persistence, a constant equal to the value of the power spectral density when 1/period equals 1, and the number of time increments. The persistence of a twenty-year time series of the mean-high-water (MHW) shoreline positions was measured for four profiles surveyed at Duck, NC at the Field Research Facility (FRF) by the U.S. Army Corps of Engineers. The four MHW shoreline time series signals are self-affine with persistence ranging between 0.8 and 0.9, which indicates that the shoreline position time series is weakly internally correlated (where zero is uncorrelated), slightly non-stationary (mean and standard deviation are not constant), and has highly varying trends for all time intervals sampled. Forecasts of a probability envelope for future MHW positions are made for the 20 years of record and beyond to 50 years from the start of the data records. The forecasts describe the twenty-year data sets well and indicate that within a 96% confidence envelope, future decadal MHW shoreline excursions should be within ± 14.6 m of the position in 1981, i.e. this is a stable-oscillatory shoreline. The forecasting method developed here includes the stochastic portion of the time series while the traditional method reduces the time series to a linear trend line fit to historic shoreline positions and extrapolated linearly to forecast future positions with a linearly increasing mean that breaks the confidence envelope eight years into the future and continues to increase. The traditional method is a poor representation of the observed shoreline position time series and is a poor basis for extrapolating future shoreline positions.