NG23A-1104
National Geophysical Data Center Historical Natural Hazard Event Databases
After a major event such as the 2004 Indian Ocean tsunami or the 2008 Chengdu earthquake, there is
interest in knowing if similar events have occurred in the area in the past and how often they have occurred.
The National Geophysical Data Center (NGDC) historical natural hazard event databases can provide
answers to these types of questions. For example, a search of the tsunami database reveals that over 100
tsunamis have occurred in the Indian Ocean since 416 A.D. Further analysis shows that there has never
been such a deadly tsunami anywhere in the world. In fact, the 2004 event accounts for almost half of all the
deaths caused by tsunamis in the database. A search of the earthquake database shows that since 193
B.C., China has experienced over 500 significant earthquakes that have caused over 2 million deaths and
innumerable dollars in damages. The NGDC global historical tsunami, significant earthquake, and significant
volcanic eruption databases include events that range in date from 4350 B.C. to the present. The database
includes all tsunami events, regardless of magnitude or intensity; and all earthquakes and volcanic eruptions
that either caused deaths, moderate damage, or generated a tsunami. Earthquakes are also included that
were assigned either a magnitude >= 7.5 or Modified Mercalli Intensity >= X. The basic data in the
historical event databases include the date, time, location of the event, magnitude of the phenomenon
(tsunami or earthquake magnitude and/or intensity, or volcanic explosivity index), and socio-economic
information such as the total number of deaths, injuries, houses damaged, and dollar damage. The tsunami
database includes an additional table with information on the runups (locations where tsunami waves were
observed by eyewitnesses, tide gauges, or deep ocean sensors). The volcanic eruptions database includes
information on the volcano elevation and type. There are currently over 2000 tsunami source events, 12500
tsunami runup locations, 5700 earthquakes, and 460 volcanic eruptions in the databases. The natural
hazard event databases are stored in a relational database management system (RDBMS) which facilitates
the integration and access to these related databases. For example, users can search for destructive
earthquakes that preceded a volcanic eruption that then generated a damaging tsunami. The databases are
accessible over the Web as tables, reports, and interactive maps. The maps provide integrated web-based
GIS access to individual GIS layers including the natural hazard events and various spatial reference layers
such as topography, population density, and political boundaries.
http://www.ngdc.noaa.gov/hazard/hazards.shtml
NG23A-1105
Record-Breaking Earthquakes
A record-breaking earthquake has a larger magnitude than any previous earthquake in the study region. A starting date and minimum magnitude must be specified. The first earthquake to satisfy this condition is by definition a record- breaking earthquake. The next record-breaking earthquake has a larger magnitude than the first and so forth. In this paper we utilize the global CMT catalog for the years 1977 to 2006. We consider both the entire period and the set of two- year sub intervals. We determine the number of record-breaking earthquakes during the specified intervals, their magnitudes, and their times of occurrence. We compare the results with the predictions for a random (iid) process. One prediction is that the magnitude differences between successive record-breaking earthquakes is a constant. A second prediction is that the difference between the logarithms of the times of occurrence of successive record-breaking earthquakes is a constant. Good statistical agreement between the observations and the predictions is obtained. It is interesting to note that an aftershock cannot be a record breaking earthquake by definition.
NG23A-1106
Exploring the statistical convergence of earthquake inter-event times
Seismic activity is routinely quantified using mean event rates or mean inter-event times. Standard estimates of the error on such mean values implicitly assume that the events that are used to calculate the mean are independent. However, earthquakes can be triggered by other events and are thus not necessarily independent. As a result, the errors on mean earthquake inter-event times do not exhibit Gaussian convergence with increasing sample size according to the Central Limit Theorem [1]. In this presentation we investigate how the errors decay with sample size in earthquake catalogues. We demonstrate that the errors on mean inter-event times, as a function of sample size, are well estimated by defining an effective sample size using the autocorrelation function to estimate the number of pieces of independent data that exist in samples of different length. This allows us to accurately project estimates of error as a function of sample size, which are further verified using extended simulations of the ETAS model. This is a generic technique that can be used to assess errors on a wide variety of correlated datasets. [1] Naylor, M., Main, I.G. and Touati, S., (In press) Quantifying uncertainties on mean earthquake inter-event times, JGR.
NG23A-1107
Observations Favor BASS, the Self-Similar Limit of ETAS
What is the ETAS model? It is a stochastic model for the generation of aftershock sequences. It is based on the concept of parent and daughter earthquakes. The number of daughter earthquakes that a parent earthquake generates is determined randomly from a productivity relation. The magnitude and time of occurrence of each daughter earthquake is determined randomly from the Gutenberg-Richter and Omori laws. Each first generation daughter earthquake becomes a parent for second generation daughters, and so forth until the sequence dies out. What is the BASS model? The BASS model is the self-similar limit of the ETAS model. Instead of the productivity relation, the modified form of Bath's law is used. The two arbitrary parameters in the productivity relation are replaced by the b-value in the GR law and the magnitude difference Delta m* between the parent earthquake and the largest expected daughter earthquake. Why is the BASS model preferred to the ETAS model? Because the BASS model is in better agreement with observations than the ETAS model. Specifically: (1) The BASS model generates Bath's law statistics since they are an input; (2) The BASS model generates inverse GR statistics for foreshock generation. The distribution of magnitudes of foreshocks is independent of the mainshock magnitude. The ETAS model has an exponential dependence of foreshock magnitude on the mainshock magnitude which is not in agreement with observations. Why do ETAS model proponents reject the BASS model? Because the BASS model is inherently unstable generating infinite numbers of aftershocks. However, this instability is easily removed by making the physically reasonable hypothesis that the excess magnitudes of daughter earthquakes over the parent earthquake cannot exceed a specified difference.
NG23A-1108
Do Burst-over-threshold Distributions and Structure Functions allow us to Infer the Coexistence of SOC and Intermittent Turbulence in Natural Systems ?
Space plasma physics provides an important arena for the study of natural hazards, because of the threat posed by space weather to space-based and ground based communications and other infrastructure. Extreme fluctuations are thus of interest, and there is by now abundant evidence for scaling in many quantities in the coupled solar-terrestrial system (solar wind, magnetosphere and ionosphere). Direct physical explanations for scaling have been sought through descriptions such as low dimensional chaos, intermittent turbulence (IT) and self-organised criticality (SOC). We have however advocated consideration of a complementary approach (Watkins [NPG, 2002]; Watkins et al. [Space Science Reviews, 2005]). This is the use of deliberately oversimplified mathematical "testbeds" to separate the proprties of the diagnostics used to infer IT or SOC from those of the models themselves. To demonstrate the need for this we consider a recent claim by Uritsky et al ([PRL, 2007]; U07) of direct observational evidence for the coexistence of SOC and IT in the magnetized plasma of the solar corona. By analyzing two dimensional (2D) EUV snapshots (typically 3-4000) of the solar corona, U07 found coexisting power law avalanche statistics and multiscaling of the structure functions. Avalanches were defined by "bursts" for which the signal exceeded a given threshold. These properties were asserted to be robust signatures of SOC and IT respectively. U07 took their coexistence to imply new physics with elements of both SOC and IT. We first point out that U07 assumed that their chosen signatures were unique to SOC and IT. We show however i) that a standard 1D multifractal model of IT, the p-model, straightforwardly generates U07's IT and SOC signatures simultaneously, and ii)that a stochastic process, linear fractional stable motion or LFSM can give the IT signatures and nonlinearity in the structure functions. We infer that not only may it not be necessary to invoke SOC to explain U07's observations, but also that our result has wider implications, which will be discussed.
NG23A-1109
Investigating earthquake scaling relationships from a 15 year archive of InSAR-derived earthquake models
In the 15 years since the first InSAR study of the 1992 Landers earthquake, the first event to be studied using InSAR, over 50 events have been studied wholly or jointly using InSAR. This constitutes a rich archive of published studies that can be mined for information on earthquake phenomenology. Empirical earthquake scaling relationships, as can be inferred from estimates of fault dimensions, slip and moment for multiple earthquakes, are extensively used in seismic hazard forecasting, and also constitute a means of placing constraints on the bulk mechanical behaviour of the seismogenic upper crust. As a source of such data, studies that utilise information from InSAR have an advantage over seismic methods in that in many cases, a key parameter, the fault length, can be measured directly from the observations. In addition, in cases of good interferogram coherence, the high spatial density of surface deformation observations that InSAR affords can place tight constraints on fault width and other important parameters. We present here a preliminary survey of earthquake scaling relationships as supported by the existing archive of InSAR earthquake studies. We find that for events with Mw > 6, the data support moment scaling with the square of fault length, in keeping with the studies of Scholz and others, and imply proportionality between fault average slip and fault length. There are currently too few datapoints for great earthquakes (Mw > 8) to assess any proposed change in scaling for such events. Scatterplots of average slip versus fault length show two broad fields -- an area of high slip-to-length ratios (> 2 × 10-5) which are predominantly associated with faults with low long-term slip rates, predominantly from intraplate settings, and an area of lower slip-to-length ratios (< 2 × 10-5) which typically are larger events from faults with higher long-term slip rates (e.g. the North Anatolian and Kunlun faults, and the Peru-Chile subduction zone). In addition, we compare InSAR-derived 'geodetic' moment estimates with 'seismic' moment estimates from body wave studies and from the Global CMT catalogue; we find that any bias towards larger moments from InSAR studies is within the standard deviation of the difference between moment estimates from InSAR and seismology.
NG23A-1110
The Weibull - log Weibull transition of interoccurrence times for synthetic and natural earthquakes
We have studied interoccurrence time distributions by analyzing the synthetic and three natural catalogs of the Japan Meteorological Agency (JMA), the Southern California Earthquake Data Center (SCEDC), and Taiwan Central Weather Bureau (TCWB) and revealed the universal feature of the interoccurrence time statistics, Weibull - log Weibull transition. This transition reinforces the view that the interoccurrence time statistics possess Weibull statistics and log- Weibull statistics. Here in this paper, the crossover magnitude from the superposition regime to the Weibull regime mc2 is proportional to the plate velocity. In addition, we have found the region-independent relation, mc2/mmax = 0.54 ¥pm 0.004.
NG23A-1111
Time, Size and Space Patterns of Triggered Landslides in Different Tectonic Settings.
We compare the respective time, size and space patterns of the 1996-2004 New Zealand landslide database with landslides in other tectonic and weathering settings, including scale effects, i.e. Grenoble cliffs, French Alps, Val d'Arly cliff, French Alps, Yosemite massif, California and Australia. For the NZ catalogue, the size, time and space distributions of 1943 landslides, is compared to (i) random catalogues and to (ii) a NZ seismicity catalogue with the same number of events. First, we found that both NZ landslide and earthquake daily rates are power law distributed over 2 orders of magnitude, with a rough exponent ~2.5. Second, we confirm that the cumulated volume distribution of the NZ landslide catalogue follows a power law with a ~0.4 exponant, for V≥50m3. Third, the interevent time distribution of NZ landslides is more clustered than the earthquake and random distributions. Signals for t<10 days are mainly driven by the V<50m3 landslides. The interevent space distribution shows that both NZ landslides and earthquakes are clustered in space. Last, a significant seasonal trend was found in the rate of NZ landslide occurrences, with a peak from June to October. The NZ landslide catalogue is mainly driven by the rainfall triggered landslides. To reproduce the time and space correlation we observed for NZ landslides, we need either the rainfall events to be correlated in time and space or the landslide response to rainfall to be delayed and/or shifted. Using other landslide catalogues, we test the relative influence of tectonic, weathering and scale effects on the landslide clustering patterns.
NG23A-1112
Pattern Informatics in Mining Induced Seismicity and its Applications to Rockburst Hazard Assessment
The Pattern Informatics (PI) (Rundle et al., 2002; Tiampo et al., 2002) was applied to over 23000 events recorded from August 2004 to November 2007 at Falconbridge's Kidd Creek D mine in Timmins – ON to estimate regions prone to rockburst hazard. To determine the optimal lengths of training periods and other important parameters, the effective ergodicity of the data was accessed using the Thirumalai-Mountain (TM) fluctuation metric (Thirumalai and Mountain, 1989) in order to determine the appropriate meshes and time bins for which the system is considered stationary and in a state of metastable equilibrium. Retrospective forecasts were made to test the PI for the latter configurations and these results are compared to relative intensity (RI) maps (Holliday et al., 2006). Both binary analysis methods are quantified using a receiver operating characteristic (ROC) diagram and initial results suggest that the PI outperforms the RI for similar configurations, although both methods need further adjustment for the optimal assessment of rockburst hazard.
NG23A-1113
Using Cluster Analysis to Optimize Tsunami Evacuation Zones
There are many factors which affect how strongly a distant-source tsunami will impact on a particular stretch of coast, such as: the angle of tsunami approach, the wave-guiding influence of undersea ridges, and resonant interactions with bays and harbors. One approach to issuing tsunami warnings is to have the coast divided into a number of zones, and to issue a threat-level forecast for each zone when an event occurs. When choosing how to divide the coast into zones it is very useful to be able to identify which sections of coast will in general respond in similar ways. We have used cluster analysis to try and identify sections of the New Zealand coast which tend to have similar responses to incoming tsunami. Since historical data is too sparse for this task we have used a suite of modelled events, representing tsunami caused by Mw 9.0 earthquakes on the subduction zones around the Pacific Rim. In this talk we present the method which has been developed, the results obtained, and the insights gained. The advantages and disadvantages of the technique will be discussed, as well as the potential for the method to be applied to other hazards.
NG23A-1114
Quasi-periodic traveling waves of large-scale Earth's global seismicity
Using the global Centroid Moment Tensor catalog of earthquakes, the statistical properties of a coarse- grained spatio-temporal field, describing the Earth's global seismicity, have been investigated. A large-scale pattern, in the form of quasi-periodic traveling waves, has been recognized. This can be heuristically viewed as a westward continuous migration of seismicity on Earth, a coherent phenomenon on time scales of the order of few years. The above migration corresponds to a global phenomenon of stress transfer along the main faults, probably due to the global westward delay of the lithosphere with respect to the mantle. The average speed of the traveling wave is about Vs ~eq 100 ÷ 250 Km/yr, corresponding to a period of few hundred years in wich the stress ships around the whole Earth.
NG23A-1115
Seismic Hazard Statistics: Implications for Risk Assessment
The history of instrumental seismology is rather short, so that many well-known quantitative relations of earthquake size and recurrence are being suggested at the times of lacking data and require a serious revision. There is growing evidence that many of the "old good" paradigms are simply misleading to surprises like the Great 2008 Wenchuan Earthquake. The applicability of commonly used probabilistic estimates of seismic hazard is really questionable, if not unacceptable. Recent publications demonstrate that the PSHA is affected by severe shortcomings, deriving from wrong mathematical assumptions and naive physical considerations. A viable alternative capable of minimizing the drawbacks of PSHA is so-called neo-deterministic approach that allows integrating unified scaling law for earthquakes, pattern recognition techniques aimed at earthquake predictions (from termless identification of earthquake prone areas down to intermediate-term medium-range or better accuracy), and realistic earthquake scenarios. Scenario-based seismic hazard maps are purely based on the empirical geophysical and seismotectonic features of a region and take into account the recurrence of earthquakes only for their classification into exceptional (catastrophic), rare (disastrous), sporadic (very strong), occasional (strong) and frequent events. Therefore they may provide an upper bound for the ground motion levels to be expected for most regions of the world, more appropriate than exceedance probabilities. The newly available high quality data from earth observation (e.g., GPS and InSAR) may permit to compile real-time displacement/deformation maps within the alerted areas and to combine the analysis of real-time deformation patterns with routinely updated information from seismic monitoring.
NG23A-1116
On the Detection of Long-Term Memory in Short Records
Long-term memory is ubiquitous in nature and has important consequences for the occurrence of natural hazards, but its detection often is complicated by the short length of the considered records. Here we study synthetic long-term correlated records of length N that are characterized by a correlation exponent γ, 0<γ<1. We show that the autocorrelation function CN(s) has the general form CN(s)=(C∞(s)-E)/(1-E), where C∞(0)=1, C∞(s>0)=B· s-γ and E=E(B,γ,N)= 2B/((2-γ)(1-γ))· N-γ+O(N-1). Due to the finite-size correction E, a direct determination of γ is difficult to achieve and generally leads to an enhanced value of γ. The parameter E also occurs in related quantities, that are characterized by the Hurst-exponent, which describe how the fluctuations of the records in time windows of length s decay with s, for example the variance of the local mean in time windows of length s. We show how to estimate E from a given data set which then allows a more accurate determination of γ. This approach can be applied also to long-term correlated records in the presence of additional white noise, where the determination of γ is particularly difficult (which is the case, e.g. in the records of return intervals between consecutive events above a certain threshold).
NG23A-1117
Statistical analysis of earthquake event correlations in Virtual California
The combination of advanced computer simulation tools and statistical analysis methods has yielded promising improvements in our understanding of the earthquake process. The Virtual California simulation tool can be used to study fault and stress interaction scenarios for realistic California earthquakes. The simulation generates a synthetic earthquake catalog of events with a minimum size of M 5.0 that can be evaluated using statistical analysis methods. Virtual California is a Monte Carlo based simulation code that utilizes realistic fault geometries and a rate and state friction model in order to drive the earthquake process by means of stress interactions between and slip deficits on faults within the model. Like any complex system, Virtual California may generate emergent phenomena unexpected even by its designers. In order to investigate this, we have developed a statistical method that analyzes the interaction between Virtual California fault elements and thereby determine whether events on any given fault elements show correlated behavior. Our method examines events on one fault element and then determines whether there is an associated event within a specified time window on a second fault element. Note that an event in our analysis is defined as any time an element slips, rather than any particular "earthquake" along the entire fault length. Results are then tabulated and then differenced with an expected correlation (calculated by assuming either 1) a uniform distribution of events in time or 2) a random distribution of events in time). We generate a correlation score matrix, which indicates how weakly or strongly correlated each fault element is to every other in the course of the VC simulation. We calculate correlation scores by summing the difference between the actual and expected correlations over all time window lengths and normalizing by the time window size. The correlation score matrix can focus attention on the most interesting areas for more in-depth analysis of event correlation vs. time. We have performed this analysis on 59 faults (639 elements) in the model, which includes all the faults save the creeping section of the San Andreas. The analysis spans 40,000 yrs of Virtual California-generated earthquake data. Preliminary statistical analysis of the data indicates promising insights into emergent behavior, such as interactions between fault elements that include long-range interaction between faults in different geographical regions (i.e. fault elements in northern California interacting with those in southern California). In addition, analysis indicates possible triggering and quiescence relationships between events (i.e. between the southern San Andreas and the Eastern California Shear Zone). We will carry out further investigations to compare model results to geologic observations.
NG23A-1118
Landslide Susceptibility Assessment of the Tahang River Catchment in northern Taiwan
This study analyzed shallow landslide induced by the 2004 typhoon Aere at Tahang river catchment in northern Taiwan. Landslide inventories were interpreted from SPOT5 imageries and checked in the field. Landslide susceptibility factors including slope, slope aspect, slope roughness, terrain curvature, slope high, terrain wetness index were derived from a 10m resolution DEM. Lithology data were obtained from a 1:50000 geological map and NDVI produced from SPOT5 imageries. Rain gauge data were used to calculate rainfall intensity. Landslide data were divided into two parts randomly by object, one part was used to establish a susceptibility model and the other was used for validation. Discriminant analysis (DA), logistic regression (LR) and artificial neural network (ANN), were used in the susceptibility analysis. This study completed the susceptibility analysis and validation. The prediction rate curves indicate that the area under the curve (AUC) for three methods are 0.83, 0.86 and 0.88 for DA, LR and ANN, respectively. All of the three methods obtained stable and good results in interpreting landslides.
NG23A-1119
A Feasibility Study of Medium-term Earthquake Forecasting Using Numerical Earthquake Simulators: Comparison to the WGCEP
We have carried out simulations for earthquakes on models of California's fault system (Virtual California -- "VC") for simulation runs over time intervals from tens of thousands of years to millions of years. Using these simulations, we have now developed techniques to assimilate observed earthquake variability into the simulations. Our technique is based on mining the simulation data to identify time intervals that look most like the recent past history of earthquakes on the California fault system. We then use these optimal time intervals to "look into the future" and forecast the likely locations of future major earthquakes. We note that the parameters that enter into the model are set using the long term average properties of the fault system -- earthquake and plate rate variability are not used at this stage of the simulation. Here we describe this method and carry out a feasibility study of its application. We develop fault-based relative spatial probabilities that can be compared with recent results from the Working Group on California Earthquake Probabilities (WGCEP 2008). Both VC and WGCEP forecast elevated relative probabilities for the Southern San Andreas fault (40.4% VC; 35.5% WGCEP). However, the relative probabilities are significantly different for the Northern San Andreas fault (22.6% VC; 12.7% WGCEP); the Calaveras fault (13.5% VC; 4.2% WGCEP); the Hayward-Rodgers Creek faults (5.0% VC; 18.7% WGCEP); and the San Jacinto fault (10.5% VC; 18.7% WGCEP). An important qualification is that since our model has not been systematically validated, these first probabilistic results should be treated with caution.
NG23A-1120
A Statistical Method Linking Geological and Historical Eruption Time-series: Applications to the Hazard Assessment of Active Volcanoes
The volcanic-eruption time series contain information of complex volcanic processes, and they represent one of the main tools for the assessment of the volcanic hazard. Unfortunately, most of the available series, capable to describe the whole range of eruption magnitudes, are usually confined to historical data, since the smaller eruptions do not leave long-lasting deposits. On the other hand, major-eruptions are usually confined to geological records, which may be incomplete and with uncertain datings and magnitudes. Additionally, these sequences may include rare or extreme events, involving at most a few events. On the other hand, the time series may be non-stationary, with time-dependant eruption rates, making the assessment of the volcanic hazard awkward. A general methodology to analyze such sequences is proposed in this work using an integral statistical methodology that include a Weibull analysis of the repose-times distributions, and a No-Homogeneous Poisson Process with a Generalized Pareto Distribution as its intensity function (NHPPP), with the purpose of calculating more precise values of the volcanic eruption hazard linking the historical and geological data series. We use a power law criterion relating the occurrence rates of a given class of eruptive magnitudes inversely to the log of the energy released by the eruptions in that magnitude range, to assemble the complete eruptive history of a given volcano, complementing the historical eruptive time series with geological eruption data and thus expanding the data population. The hazard or eruption probabilities of some selected active volcanoes are then calculated with this methodology and compared with the results obtained with other methods.
NG23A-1121
Climatic Variability of Hurricane-Size Statistics
The influence of climate variability and global warming on the occurrence of tropical cyclones (roughly speaking, hurricanes, or simply, storms) is a complex issue,which faces the extra complication of the relative lack of reliability of the existing databases. But a more fundamental hindrance in order to address this problem is the absence of a basic understanding of the intrinsic nature of tropical-cyclone genesis and evolution. It is known that tropical cyclones are not just a passive response to changing external forcing, but it is not clear which kind of dynamical process tropical cyclones define. In this presentation, we discuss different measures of hurricane size and study their fluctuations in successive occurrences. The changes of these statistical properties in different climatic conditions can shed light on hurricane variability.
NG23A-1122
Non-Linear Time Series Analysis of Dissolved Oxygen in Five Diverse Aquatic Environments
Temporal variations in the concentration of Dissolved oxygen (DO) can create catastrophic conditions for organisms that rely on aerobic metabolic processes for survival. Dissolved oxygen (DO) is an aquatic parameter whose concentration is controlled by physical, biological, and chemical processes. The concentration of DO in an aquatic system is important to organisms that rely on aerobic metabolic processes for survival. A power-spectral-density analysis of time series of DO concentration is used to quantify persistence (the degree of internal correlation) over durations of 3 months to 19 years. The interval between data points was either 15 minutes or one hour. The data are from ten different water bodies throughout the United States. Four of these sites are large, slow moving bodies of water including three estuaries: Chesapeake Bay (Virginia), Winyah Bay (North Carolina) and Elkhorn Slough (California); and one reservoir: the Cheney Reservoir in Kansas. The other six sites are small, fast moving water bodies. They included four rivers: Christina River (Delaware), St. Croix River (Maine), Ramapo River (New Jersey), and Passaic River, New Jersey; one stream: Green Pond Brook (New Jersey); and one man-made channel: Reynolds Channel (New York). The analysis quantifies persistence as the power scaling exponent (β), which for all ten water bodies β ranges between 1.2 and 1.6 meaning that the signal is persistent and non-stationary. Rivers and streams, exhibit higher β-values of 1.5 < β<1.6 (greater persistence) than estuaries and lakes, which have β-values of 1.2< β <1.4t.
NG23A-1123
A Study of Earthquake Catalog Heterogeneity
Seismicity rate changes can result from both natural and man-made causes. It is important to differentiate them as much as possible in order to identify the meaningful and real rate changes in seismicity studies [Habermann, et al. (1987), BSSA 77, 141-159 & JGR 92, 9446-9450; Matthews and Reasenberg (1987), JGR 92, 9443-9445]. Man-made changes could be caused by transitions in the operation of a seismic network, which include changes in: station coverage, reported magnitude type, velocity-structure model, instrumentation and automation (e.g., analog to digital processing of waveform data), etc., and occasionally become apparent as systematic shifts in the frequency distribution of reported magnitudes in a catalog. For example, man-made seismicity rate changes can be generated when a catalog is cut to select events in a certain magnitude range and when such shifts are present in the distribution of reported magnitudes [e.g., Habermann (1982), BSSA 72, 93-111]. One way to detect systematic seismicity rate changes would be to directly compare time-avelaged rate changes of the entire catalog before and after a reference time, for various magnitude bands [e.g., Habermann (1987)]. Alternatively, one can detect systematic changes of a data set by means of 'energy fluctuation metric' or Thirumalai-Mountain metric (a measure of effective ergodicity) [Thirumalai & Mountain (1993), Phys. Rev. E 47, 479-489; Ferguson et al. (1999), Phys. Rev. E 60, 1359-1373; Klein, et al. (2000), Geophys. Monograph 120, AGU, pp. 43-71; Tiampo, et al. (2003), doi: 10.1103/ PhysRevLet t.91. 238501, Tiampo, et al. (2007), doi: 10.1103/PhysRevE.75.066107]. The metric, which measures if the time-averages would equal the ensemble averages, is more sensitive than the former to (either apparent or hidden) rate changes of seismicity in a catalog, as the stationarity of seismicity is a prerequisite for the process to be ergodic. Results of case studies using both natural and synthetic data will be presented.
NG23A-1124
Scaling Analysis of Tide Gauge Data from the Atlantic, Gulf of Mexico, and Pacific Coasts of the United States
Most coastal regions are subject to inundation due to many periodic and non-periodic inputs including for example: diurnal and semi diurnal tides, storms, tsunamis, and global sea level change. Tide guage data provide a frequently sampled long term record of fluctuations in water level. A power-spectral-density analysis of tidal gauge data is used to quantify persistence (degree of internal correlation over various time intervals) in terms of the scaling exponent β and to identify temporal changes in persistence. The stations are located at different proximity to the open ocean, including bays, harbors, and channels. The datasets are from the NOAA CO-OPS Verified Hourly Station Datum. The length of the data sets ranges from 3 years to 101 years. The hourly data sets are decimated to one record every four hours. All data sets analyzed show three distinct regions of persistence with two inflection points at approximately one day and five days. For times less than one day, the scaling exponent ranges between 0 < β < 2.6. For the time interval 1 to 5 days, the scaling exponent ranges between 1.1 < β < 2.1. For times greater than 5 days, the scaling exponent ranges between 0.4 < β < 0.9. Persistence generally decreases as period increases but is stable between the inflection points. At Duck, NC, long term persistence in the tide gauge signal is 0.6 as compared to 0.9 for the biweekly shoreline position signal over twenty years, suggesting a strong correlation between the two and the possibility of using tide gauge data to quantify nearby shoreline mobility over similar time intervals.
NG23A-1125
Mitigating Large Fires in Drossel-Schwabl Forest Fire Models
We employ variations of the traditional Drossel-Schwabl cellular automata Forest Fire Models (FFM) to study wildfire dynamics. The traditional FFM produces a very robust power law distribution of events, as a function of size, with frequency-size slope very close to -1. Observed data from Australia, the US and northern Mexico suggest that real wild fires closely follow power laws in frequency size with slopes ranging from close to -2 to -1.3 (B.D. Malamud et al. 2005). We suggest two models that, by fracturing and trimming large clusters, reduce the number of large fires while maintaining scale invariance. These fracturing and trimming processes can be justified in terms of real physical processes. For each model, we achieve slopes in the frequency-size relation ranging from approximately -1.77 to -1.06.
NG23A-1126
Analysis of Water Level Dynamics in the Great Lakes of North America
Anthropogenic as well as natural fluctuations such as precipitation, runoff, snowmelt, retention time,
evaporation, and outflow all contribute to water levels observed in the Great Lakes. Verified hourly water
level data for five stations in Lake Michigan and four stations in Lake Superior were obtained from NOAA and
examined. For each station, an hourly time series ranging from 20 to 30 years in duration was decimated to
produce a time series with four hour intervals. A Fourier transform was performed on each time series and
the resulting frequency information displayed in a Power Spectral Density (PSD) graph. Water level records
in the Great Lakes are found to exhibit power law scaling between power spectral density and period. Four
distinct regions of scaling are observed with inflection points at approximately 1 day, 5 days, and 30 - 60
days. For time scales of less than one day, the power-scaling exponent (β) ranges from 0.1 to 0.5,
indicating a white noise. From 1 day to 5 - 7 days, β ranges from 1.5 to 2.6, indicating moderate to
strong persistence which we propose is due to frontal movements of weather systems. On timescales
between 5 days and 30 - 60 days, β ranges from 0.1 to 0.4, again indicating a white noise which we
propose is due to monthly and seasonal weather variations within the Great Lakes System. Beyond 30 - 60
days, all stations exhibit strong persistence, with β between 1.6 and 2.7. Analysis of physical
processes using nonlinear methods such as the Fourier transform allows one to determine the natural state
of the environment and if the natural state fluctuates randomly (white noise) or has some underlying order
(persistence). A parallel analysis approach, drawing from concepts in control theory and feedback systems,
uses Bode plots in the frequency domain and can be applied to explain variations in the Power Spectral
Density plots of water levels. By analyzing the pattern of change in amplitude and phase across frequencies
in a Bode plot, the dynamic properties of a system can be discerned. Bode plots draw a window into the
underlying internal dynamics of the system answering questions such as the presence of time delays, the
stability of the physical system, and the extent to which the system is acting as an integrator. For example,
the Power Spectral Density plot of water levels in which the Beta value is near β = 2 for time intervals
from 1 day to 5 - 7 days suggests that the system is acting as an integrator over this time scale.
Changes in water levels and tides have been used as an index for physical parameters such as temperature,
density, and circulation (Keeling and Whorf, 1997; Denny and Paine, 1998). Long term 1/f noise
(environmental noise) in the physical environment has been shown to affect populations of species
embedded in these environments. Variations observed in the changing β of water levels
(environmental noise) may have biological impacts on population dynamics of organisms, including rates of
survival or extinction (Batchhelder and Powell 2002). The application of Bode analysis and control theory
concepts to population dynamics may provide additional insight into the underlying dynamics of the response
of a population to noise found within the physical environment. Knowledge of the biological-physical
coupling and the impact of environmental noise (as observed in water level data) in this aquatic environment
are needed to understand the complex ecosystem dynamics.
NG23A-1127
Spatio-Temporal Fluctuations of the Earthquake Magnitude Distribution: Robust Estimation and Predictive Power
We establish positive correlation between the local spatio-temporal fluctuations of the earthquake magnitude distribution and the occurrence of regional earthquakes. In order to accomplish this goal, we develop a sequential Bayesian statistical estimation framework for the b-value (slope of the Gutenberg-Richter's exponential approximation to the observed magnitude distribution) and for the ratio a(t) between the earthquake intensities in two non-overlapping magnitude intervals. The time-dependent dynamics of these parameters is analyzed using Markov Chain Models (MCM). The main advantage of this approach over the traditional window-based estimation is its "soft" parameterization, which allows one to obtain stable results with realistically small samples. We furthermore discuss a statistical methodology for establishing lagged correlations between continuous and point processes. The developed methods are applied to the observed seismicity of California, Nevada, and Japan on different temporal and spatial scales. We report an oscillatory dynamics of the estimated parameters, and find that the detected oscillations are positively correlated with the occurrence of large regional earthquakes, as well as with small events with magnitudes as low as 2.5. The reported results have important implications for further development of earthquake prediction and seismic hazard assessment methods.
NG23A-1128
Statistics and Correlations of Seismic and Tectonic Moment Rate in California and the Great Basin
The largest regional earthquakes provide the most significant contribution to the earthquake-related damage, they also responsible for the most of the long-term regional seismic moment release. Thus, in order to develop physically based earthquake hazard assessments, one should be able to connect the moment rate from the observed seismicity to its long-term predictions based on geodetic and geological information. We study here a most detailed spatial distribution of geodesy- and geology-based expected moment rate in California and the Great Basin (see Blewitt et al., "GPS Velocity and Strain Rate Fields in the Great Basin and California", Session G2) and compare it with the observed seismic moment rate on different temporal and spatial scales. The comparison clearly demonstrates a moment deficiency at intermediate and small spatial scales that can be explained, at intermediate scales, by the heavy-tailed Pareto distribution of seismic moment and the typically short time-span of catalogs. On short spatial scales the observed moment deficiency is significantly larger than that predicted by a simple Pareto model. We present a statistical framework for studying these and other related phenomena connected with the seismic moment release.
NG23A-1129
"Geo-statistics methods and neural networks in geophysical applications: A case study"
The study is focus in the Ebano-Panuco basin of northeastern Mexico, which is being explored for hydrocarbon reservoirs. These reservoirs are in limestones and there is interest in determining porosity and permeability in the carbonate sequences. The porosity maps presented in this study are estimated from application of multiattribute and neural networks techniques, which combine geophysics logs and 3-D seismic data by means of statistical relationships. The multiattribute analysis is a process to predict a volume of any underground petrophysical measurement from well-log and seismic data. The data consist of a series of target logs from wells which tie a 3-D seismic volume. The target logs are neutron porosity logs. From the 3-D seismic volume a series of sample attributes is calculated. The objective of this study is to derive a set of attributes and the target log values. The selected set is determined by a process of forward stepwise regression. The analysis can be linear or nonlinear. In the linear mode the method consists of a series of weights derived by least-square minimization. In the nonlinear mode, a neural network is trained using the select attributes as inputs. In this case we used a probabilistic neural network PNN. The method is applied to a real data set from PEMEX. For better reservoir characterization the porosity distribution was estimated using both techniques. The case shown a continues improvement in the prediction of the porosity from the multiattribute to the neural network analysis. The improvement is in the training and the validation, which are important indicators of the reliability of the results. The neural network showed an improvement in resolution over the multiattribute analysis. The final maps provide more realistic results of the porosity distribution.