The wide-spread availability of ensembles of high-performance clocks has motivated interest in time-scale algorithms. There are many such algorithms in use today in applications ranging from scientific to commercial. Although these algorithms differ in key aspects and are sometimes tailored for specific applications and mixtures of clocks, they all share the goal of combining measured time differences between clocks to form a reference time scale that is more stable than any of the clocks in the ensemble. A new approach to forming time scales is presented here, the multiscale ensemble timescale (METS) algorithm. This approach is based on a multiresolution analysis afforded by the discrete wavelet transform. The algorithm does not assume a specific parametric model for the clocks involved and hence is well-suited for an ensemble of highly disparate clocks. The approach is based on an appealing optimality criterion which yields a reference time scale that is more stable than the constituent clocks over all averaging intervals (scales). The METS algorithm is presented here in detail and is shown in a simulation study to compare favorably with a time-scale algorithm based on Kalman filtering.

There has been considerable recent interest in using wavelets to analyze time series and images that can be regarded as realizations of certain 1-D and 2-D stochastic processes on a regular lattice. Wavelets give rise to the concept of the wavelet variance (or wavelet power spectrum), which decomposes the variance of a stochastic process on a scale-by-scale basis. The wavelet variance has been applied to a variety of time series, and a statistical theory for estimators of this variance has been developed. While there have been applications of the wavelet variance in the 2-D context (in particular, in works by Unser in 1995 on wavelet-based texture analysis for images and by Lark and Webster in 2004 on analysis of soil properties), a formal statistical theory for such analysis has been lacking. In this paper, we develop the statistical theory by generalizing and extending some of the approaches developed for time series, thus leading to a large-sample theory for estimators of 2-D wavelet variances. We apply our theory to simulated data from Gaussian random fields with exponential covariances and from fractional Brownian surfaces. We demonstrate that the wavelet variance is potentially useful for texture discrimination. We also use our methodology to analyze images of four types of clouds observed over the southeast Pacific Ocean.

The ability to accurately forecast potential hazards posed to coastal communities by tsunamis generated seismically in both the near and far field requires knowledge of so-called source coefficients, from which the strength of a tsunami can be deduced. Seismic information alone can be used to set the source coefficients, but the values so derived reflect the dynamics of movement at or below the seabed and hence might not accurately describe how this motion is manifested in the overlaying water column. We describe here a method for refining source coefficient estimates based on seismic information by making use of data from Deep-ocean Assessment and Reporting of Tsunamis (DART) buoys (tsunameters). The method involves using these data to adjust precomputed models via an inversion algorithm so that residuals between the adjusted models and the DART data are as small as possible in a least squares sense. The inversion algorithm is statistically based and hence has the ability to assess uncertainty in the estimated source coefficients. We describe this inversion algorithm in detail and apply it to the November 2006 Kuril Islands event as a case study.

Water temperature measurements from Wivenhoe Dam offer a unique opportunity for studying fluctuations of temperatures in a subtropical dam as a function of time and depth. Cursory examination of the data indicate a complicated structure across both time and depth. We propose simplifying the task of describing these data by breaking the time series at each depth into physically meaningful components that individually capture daily, subannual, and annual (DSA) variations. Precise definitions for each component are formulated in terms of a wavelet-based multiresolution analysis. The DSA components are approximately pairwise uncorrelated within a given depth and between different depths. They also satisfy an additive property in that their sum is exactly equal to the original time series. Each component is based upon a set of coefficients that decomposes the sample variance of each time series exactly across time and that can be used to study both time-varying variances of water temperature at each depth and time-varying correlations between temperatures at different depths. Each DSA component is amenable for studying a certain aspect of the relationship between the series at different depths. The daily component in general is weakly correlated between depths, including those that are adjacent to one another. The subannual component quantifies seasonal effects and in particular isolates phenomena associated with the thermocline, thus simplifying its study across time. The annual component can be used for a trend analysis. The descriptive analysis provided by the DSA decomposition is a useful precursor to a more formal statistical analysis.

The wavelet variance is a scale-based decomposition of the process variance for a time series and has been used to analyze, for example, time deviations in atomic clocks, variations in soil properties in agricultural plots, accumulation of snow fields in the polar regions and marine atmospheric boundary layer turbulence. We propose two new unbiased estimators of the wavelet variance when the observed time series is 'happy,' i.e., is sampled at regular intervals, but certain observations are missing. We deduce the large sample properties of these estimators and discuss methods for determining an approximate confidence interval for the wavelet variance. We apply our proposed methodology to series of gappy observations related to atmospheric pressure data and Nile River minima.

The wavelet variance provides a scale-based decomposition of the process variance for a time series or a random field and has been used to analyze various multiscale processes. Examples of such processes include atmospheric pressure, deviations in time as kept by atomic clocks, soil properties in agricultural plots, snow fields in the polar regions and brightness temperature maps of South Pacific clouds. In practice, data collected in the form of a time series or a random field often suffer from contamination that is unrelated to the process of interest. This paper introduces a scale-based contamination model and describes robust estimation of the wavelet variance that can guard against such contamination. A new M-estimation procedure that works for both time series and random fields is proposed, and its large sample theory is deduced. As an example, the robust procedure is applied to cloud data obtained from a satellite.

We consider the problem of detecting changes in a multivariate data stream. A change detector is defined by a detection algorithm and an alarm threshold. A detection algorithm maps the stream of input vectors into a univariate detection stream. The detector signals a change when the detection stream exceeds the chosen alarm threshold. We consider two aspects of the problem: (1) setting the alarm threshold and (2) measuring/comparing the performance of detection algorithms.

We assume we are given a segment of the stream where changes of interest are marked. We present evidence that, without such marked training data, it might not be possible to accurately estimate the false alarm rate for a given alarm threshold. Commonly used approaches assume the data stream consists of independent observations, an implausible assumption given the time series nature of the data. Lack of independence can lead to estimates that are badly biased. Marked training data can also be used for realistic comparison of detection algorithms. We define a version of the receiver operating characteristic curve adapted to the change detection problem and propose a block bootstrap for comparing such curves. We illustrate the proposed methodology using multivariate data derived from an image stream.

Naval submarines have collected operational data of sea-ice draft (93% of thickness) in the Arctic Ocean since 1958. Data from 34 U.S. cruises are publicly archived. They span the years 1975 to 2000, are equally distributed in spring and autumn, and cover roughly half the Arctic Ocean. The data set is strong: we use 2203 values of mean draft, each value averaged over a nominal length of 50 km. These values range from 0 to 6 m with a standard deviation of 0.99 m. Multiple regression is used to separate the interannual change, the annual cycle, and the spatial field. The solution gives a climatology for ice draft as a function of space and time. The residuals of the regression have a standard deviation of 0.46 m, slightly more than the observational error standard deviation of 0.38 m. The overall mean of the solution is 2.97 m. Annual mean ice draft declined from a peak of 3.42 m in 1980 to a minimum of 2.29 m in 2000, a decrease of 1.13 m (1.25 m in thickness). The steepest rate of decrease is –0.08 meters per year (m/a) in 1990. The rate slows to –0.007 m/a at the end of the record. The annual cycle has a maximum on 30 April and a peak-to-trough amplitude of 1.06 m (1.12 m in thickness). The spatial contour map of the temporal mean draft varies from a minimum draft of 2.2 m near Alaska to a maximum just over 4 m at the edge of the data release area 200 miles north of Ellesmere Island.

Second-order structure functions are widely used to characterize turbulence in the inertial range because they are simple to estimate, particularly in comparison to spectral density functions and wavelet variances. Structure function estimators, however, are highly autocorrelated and, as a result, no suitable theory has been established to provide confidence intervals for turbulence parameters when determined via regression fits in log/log space. Monte Carlo simulations were performed to compare the performance of structure function estimators of turbulence parameters with corresponding multitaper spectral and wavelet variance estimators. The simulations indicate that these latter estimators have smaller variances than estimators based upon the structure function. In contrast to structure function estimators, the statistical properties of the multitaper spectral and wavelet variance estimators allow for the construction of confidence intervals for turbulence parameters. The Monte Carlo simulations also confirm the validity of the statistical theory behind the multitaper spectral and wavelet variance estimators. The strengths and weaknesses of the various estimators are further illustrated by analyzing an atmospheric temperature time series.

A new Ricean parameter estimator is considered in the context of wireless communications. In comparison to existing methods, the proposed estimator is especially useful in low signal-to-noise environments. It is less reliant on knowledge of the transmitted signal frequency than existing methods, and it performs well in communications environments where the frequency is varying in time.

Measured sea-ice draft exhibits variations on all scales. We regard draft profiles up to several hundred kilometers in length as being drawn from a stationary stochastic process. We focus on the estimation of the mean draft of the process. This elementary statistic is typically computed from a profile segment of length L and has some uncertainty, or sampling error, that is quantified by its variance. How efficiently can the variance of mean draft be reduced by the use of more data, that is, by increasing L? Three properties of the data indicate the need for a non-standard statistical model: the variance of mean draft falls off more slowly than L^-1; the autocorrelation sequence does not fall rapidly to zero; and the spectrum does not flatten off with decreasing wave number. These indicate that ice draft exhibits, as a fundamental geometric property, "long-range dependence." One good model for this dependence is a fractionally differenced process, whose variance is proportional to L ^-1+2δ. From submarine ice draft data in the Arctic Ocean, we find δ = 0.27. Mean draft estimated from a 50-km sample has a sample standard deviation of 0.29 m; for 200 km, it is 0.21 m. Tabulated values provide the sample standard deviation for various values of L for samples both in a straight line and in a rosette or spoke pattern, allowing for the efficient design of observational programs to measure draft to a desired accuracy.

Discrete wavelet transforms (DWTs) are mathematical tools that are useful for analyzing geophysical time series. The basic idea is to transform a time series into coefficients describing how the series varies over particular scales. One version of the DWT is the maximal overlap DWT (MODWT). The MODWT leads to two basic decompositions. The first is a scale-based analysis of variance known as the wavelet variance, and the second is a multiresolution analysis that reexpresses a time series as the sum of several new series, each of which is associated with a particular scale. Both decompositions are illustrated through examples involving Arctic sea ice and an Antarctic ice core. A second version of the DWT is the orthonormal DWT (ODWT), which can be extracted from the MODWT by subsampling. The relative strengths and weaknesses of the MODWT, the ODWT and the continuous wavelet transform are discussed.

This study describes winter climate during the last 500 yr for the greater Baltic Sea region through an examination of well-documented time series of ice cover, sea level pressure, and winter surface air temperatures. These time series have been the focus of previous studies, but here their covariation over different time scales is analyzed based on two modern descriptive statistical techniques, matching pursuit and wavelet analysis. Independently, 15 time periods were found during the last 500 yr with different climatic signatures with respect to winter severity, circulation patterns, and interannual variability. The onsets of these periods are presumably caused largely by perturbations within the system, although correspondences with solar and volcanic activity can be identified for certain of the periods. The Baltic region climate has changes on both centennial and decadal time scales, often with rapid transitions. Major warmer periods were the first half of the eighteenth century and the twentieth century. A common feature for warm (cold) periods is low (high) variability on shorter time scales. Century-scale variability and the modulation of interannual and decadal signals are quite diverse in the temporal records and do not suggest strong periodicities. An "event" type conceptual model therefore appears adequate for characterizing Baltic climate variability.

OBJECTIVES:
The purpose of this study was to examine the pattern of and factors that influence hot flash severity across the menopausal transition (MT) and early postmenopause (PM).

METHODS:
Women from the Seattle Midlife Women's Health Study (N=302) provided data for these analyses: at least one annual health questionnaire and a menstrual calendar. A subset of women provided a first morning voided urine specimen from 1997 through 2005. Urine samples were assayed for estrone glucuronide and FSH. Linear mixed effects modeling was used to identify change in hot flash severity scores over time, including the relationship to age, MT-related, psychosocial and lifestyle factors.

RESULTS:
Increases in hot flash severity were associated with late transition stage, early postmenopause, use of HRT, duration of early transition stage, age of entry into early PM and level of FSH. Age of entry into early transition and estrone levels were associated with decreased hot flash severity. Not associated with hot flash severity were being in early transition stage, age of entry into or duration of late transition stage and all of the psychosocial (anxiety, stress, depressed mood) and lifestyle variables (BMI, activity level, sleep, alcohol use).

CONCLUSIONS:
Variables associated with reproductive aging independently predicted changes in hot flash severity; psychosocial and lifestyle variables did not. The effect of age dropped out when factors associated with reproductive aging were considered. Use of HRT ameliorated but did not eliminate severe hot flashes suggesting that there is room for alternative approaches less likely to cause harm.

OBJECTIVE:
To determine whether hot flashes, depressed mood, sleep, cognitive and sexual symptoms correlate with urinary follicle-stimulating hormone (FSH), estrone (E(1)G), and testosterone (T) and with each other during the menopausal transition and early postmenopause (PM).
METHODS:
Forty-one women who transitioned from middle or late transition stage to PM rated symptoms and provided monthly urine specimens as part of a longitudinal study of the menopausal transition.
RESULTS:
Correlations between endocrine levels and symptom severity ratings over time revealed that hot flash severity was significantly and positively related to FSH and negatively to E1 G. Vaginal dryness was positively correlated with FSH and negatively correlated with T. Decreased sexual desire was correlated negatively with E(1)G levels. Forgetfulness was positively correlated with FSH; difficulty concentrating was negatively correlated with T. Severity of sleep symptoms and depressed mood were not correlated with E(1)G, FSH, or T. Correlations among the symptoms revealed that severity of hot flashes was associated with sleep disruption and forgetfulness. Depressed mood was correlated with sleep disruption, difficulty concentrating, and decreased sexual desire but not with hot flashes or vaginal dryness. Awakening during the night was correlated with decreased sexual desire and vaginal dryness, as well as hot flashes. Forgetfulness was associated with hot flashes and difficulty concentrating, whereas difficulty concentrating was associated with depressed mood and early awakening.
CONCLUSIONS:
Symptoms many women experience during the menopausal transition and early PM are related to different endocrine levels (FSH, E(1)G, and T).

This chapter is a collection of short contributions on environmental time series analysis.The focus is on nonparametric trend estimation, role of nonstationarity vs. long-memory or slowly decaying correlations, wavelets and extreme quantiles. Data examples illustrate the methods.

The acquisition and interpretation of increasingly high-resolution climate data from polar ice and firn cores motivates the question: What is the finest depth or timescale on which measurements on cores arrayed over a given area correlate? We analyze dated depth series of electrical and oxygen isotope measurements from a spatial array of firn cores with 3.5–7 km spacing in Dronning Maud Land, Antarctica, each with a temporal span of approximately 200 years. We use wavelet analysis to decompose the series into components associated with changes of averages on different scales, and thus deduce which scales are dominated by environmental noise, and which may contain a common signal. We find that common signals in electrical records have timescales of approximately 1–3 years. We identify only one electrical signal which rises significantly above the background in our 200-year records, evidently corresponding to the Tambora eruption. Several smaller signals correlate in a few of pairs of cores, one of which may correspond to a known volcanic event, but the others appear to be spurious. We present a simulation-based method for testing the significance of apparent electrical signal correlations, and highlight the importance of accurate relative dating between cores. In the case of oxygen-isotope records, we find, surprisingly, no significant correlation on any scale in the records, for any of the pairs of cores. There is, however, a weak trend toward positive correlation at longer timescales (up to 16 years). Statistical theory for the relevant confidence intervals and the observed statistics of the records permit estimation of the length of a data series necessary to reliably detect a hypothetical correlation equal to that observed. For the highest correlation observed on 16-year scales, core records of about 380 years (approximately 30 m at the Dronning Maud Land site) would be necessary to establish significance.

The circulant embedding technique allows for the fast and exact simulation of stationary and intrinsically stationary Gaussian random fields. The method uses periodic embeddings and relies on the fast Fourier transform. However, exact simulations require that the periodic embedding is nonnegative definite, which is frequently not the case for two-dimensional simulations. This work considers a suggestion by Michael Stein, who proposed nonnegative definite periodic embeddings based on suitably modified, compactly supported covariance functions. Theoretical support is given to this proposal, and software for its implementation is provided. The method yields exact simulations of planar Gaussian lattice systems with 10⁶ and more lattice points for wide classes of processes, including those with powered exponential, Matérn, and Cauchy covariances.

The circulant embedding method for generating statistically exact simulations of time series from certain Gaussian distributed stationary processes is attractive because of its advantage in computational speed over a competitive method based upon the modified Cholesky decomposition. We demonstrate that the circulant embedding method can be used to generate simulations from stationary processes whose spectral density functions are dictated by a number of popular nonparametric estimators, including all direct spectral estimators (a special case being the periodogram), certain lag window spectral estimators, all forms of Welch's overlapped segment averaging spectral estimator and all basic multitaper spectral estimators. One application for this technique is to generate time series for bootstrapping various statistics. When used with bootstrapping, our proposed technique avoids some — but not all — of the pitfalls of previously proposed frequency domain methods for simulating time series.

The statistical characterization of clock noise is important for understanding how well a clock can perform in applications where timekeeping is important. The usual frequency domain characterization of clock noise is the power spectrum. We present a primer on how to estimate the power spectrum of clock noise given a finite sequence of measurements of time (or phase) differences between two clocks. The simplest estimator of the spectrum is the periodogram. Unfortunately this estimator is often problematic when applied to clock noise. Three estimators that overcome the deficiencies of the periodogram are the sinusoidal multitaper spectral estimator, Welch's overlapped segment averaging estimator and Burg's autoregressive estimator. We give complete details on how to calculate these three estimators. We apply them to two examples of clock noise and find that they all improve upon the periodogram and give comparable results. We also discuss some of the uses for the spectrum and its estimates in the statistical characterization of clock noise.

Circulant embedding is a technique that has been used to generate realizations from certain real-valued Gaussian stationary processes. This technique has two potential advantages over competing methods for simulating time series. First, the statistical properties of the generating procedure are exactly the same as those of the target stationary process. Second, the technique is based upon the discrete Fourier transform and hence is computationally attractive when this transform is computed via a fast Fourier transform (FFT) algorithm. In this paper we show how, when used with a standard "powers of two" FFT algorithm, circulant embedding can be readily adapted to handle complex-valued Gaussian stationary processes.

Regimes and regime shifts are important concepts for understanding decadal variability in the physical system of the North Pacific because of the potential for an ecosystem to reorganize itself in response to such shifts. There are two prevalent senses in which these concepts are taken in the literature. The first is a formal definition and posits multiple stable states and rapid transitions between these states. The second is more data-oriented and identifies local regimes based on differing average climatic levels over a multi-annual duration, i.e. simply interdecadal fluctuations. This second definition is consistent with realizations from stochastic red noise processes to a degree that depends upon the particular model. Even in 100 year long records for the North Pacific a definition of regimes based solely on distinct multiple stable states is difficult to prove or disprove, while on interdecadal scales there are apparent local step-like features and multi-year intervals where the state remains consistently above or below the long-term mean. The terminologies climatic regime shift, statistical regime shift or climatic event are useful for distinguishing this second definition from the first.

To illustrate the difficulty of advocating one definition over the other based upon a relatively short time series, we compare three simple models for the Pacific Decadal Oscillation (PDO). The 104-year PDO record is insufficient to statistically distinguish a single preference between a square wave oscillator consistent with the formal definition for regime shifts, and two red noise models that are compatible with climatic regime shifts. Because of the inability to distinguish between underlying processes based upon data, it is necessary to entertain multiple models and to consider how each model would impact resource management. In particular the persistence in the fitted models implies that certain probabilistic statements can be made regarding climatic regime shifts, but we caution against extrapolation to future states based on curve fitting techniques.

Surface air temperatures (SAT) from three Alaskan weather stations in a north–south section (Barrow, Nome, and St. Paul) show that on a decadal scale, the correlation among the stations changed during the past century. Before the 1960s, Barrow and Nome were dominated by Arctic air masses and St. Paul was dominated by North Pacific maritime air masses. After the 1960s, the SAT correlation in winter between Barrow and St. Paul increased from 0.2 to 0.7 and between Nome and St. Paul from 0.4 to 0.8, implying greater north–south penetration of both air masses. The correlation change in the winter of the Barrow–St. Paul pair is significant at a 95% confidence level. The Nome–St. Paul pair in spring also shows some of this characteristic change in correlation. Relatively stable, high correlations are found among the stations in the fall; correlations are low in the summer. Our study shows a change in the climatological structure of the Bering Sea in the late twentieth century, at present of unknown origin and occurring earlier than the well-known 1976/1977 shift. These climatological results further support the concept that the southeast Bering Sea ecosystem may have been dominated by Arctic species for most of the century, with a gradual replacement by sub-Arctic species in the last 30 years.

Statistical tools based on the maximal overlap discrete wavelet transform (MODWT) are reviewed, and then applied to a dataset of aircraft observations of the atmospheric boundary layer from the tropical eastern Pacific, which includes quasi-stationary and non-stationary segments. The wavelet methods provide decompositions of variances and covariances, e.g., fluxes, between time scales that effectively describe a broadband process like atmospheric turbulence. Easily understood statistical confidence bounds are discussed and applied to these scale decompositions, and results are compared to Fourier methods for quasi-stationary turbulence. The least asymmetric LA(8) wavelet filter yields coefficients that exhibit better uncorrelatedness across scales than the Haar filter and is better suited for decomposition of broadband turbulent signals. An application to a non-stationary segment of our dataset, namely vertical profiles of the turbulent dissipation rate, highlights the flexibility of wavelet methods.

We consider the problem of estimating the parameters for a stochastic process using a time series containing a trend component. Trend, i.e., large scale variations in the series that are best modeled outside of a stochastic framework, is often confounded with low-frequency stochastic fluctuations. This problem is particularly evident in models such as fractionally differenced (FD) processes, which exhibit slowly decaying autocorrelations and can be extended to encompass nonstationary processes with substantial low frequency components. We use the discrete wavelet transform (DWT) to estimate parameters for stationary and nonstationary FD processes in a model of polynomial trend plus FD noise. Using Daubechies wavelet filters allows for automatic elimination of polynomial trends due to embedded differencing operations. Parameter estimation is based on an approximate maximum likelihood approach made possible by the fact that the DWT decorrelates FD processes approximately. We consider this decorrelation in detail, examining the between- and within-scale wavelet correlations separately. Better between-scale decorrelation can be achieved by increasing the length of the wavelet filter, whereas the within-scale correlations can be handled via explicit modeling by a low-order autoregressive process. We demonstrate our methodology by applying it to a popular climate dataset.

In recent years there has been much interest in the analysis of time series using a discrete wavelet transform (DWT) based upon a Daubechies wavelet filter. Part of this interest has been sparked by the fact that the DWT approximately decorrelates certain stochastic processes, including stationary fractionally differenced (FD) processes with long memory characteristics and certain nonstationary processes such as fractional Brownian motion. It is shown that, as the width of the wavelet filter used to form the DWT increases, the covariance between wavelet coefficients associated with different scales decreases to zero for a wide class of stochastic processes. These processes are Gaussian with a spectral density function (SDF) that is the product of the SDF for a (not necessarily stationary) FD process multiplied by any bounded function that can serve as an SDF on its own. We demonstrate that this asymptotic theory provides a reasonable approximation to the between-scale covariance properties of wavelet coefficients based upon filter widths in common use. Our main result is one important piece of an overall strategy for establishing asymptotic results for certain wavelet-based statistics.

In examining a plot of a time series of a scalar climate variable for indications of climate change, an investigator might pick out what appears to be a linear trend commencing near the end of the record. Visual determination of the starting time of the trend can lead to an incorrect conclusion that the trend is significant when the assessment is based on standard linear regression analysis; in fact, a presumed level of significance of 5% can be smaller than the actual level by up to an order of magnitude. An alternative procedure is suggested that is more appropriate for assessing the significance of a trend in which the starting point is identified visually.

Instrumental surface air temperature (SAT) records beginning in the late 1800s from 59 Arctic stations north of 64°N show monthly mean anomalies of several degrees and large spatial teleconnectivity, yet there are systematic seasonal and regional differences. Analyses are based on time–longitude plots of SAT anomalies and principal component analysis (PCA). Using monthly station data rather than gridded fields for this analysis highlights the importance of considering record length in calculating reliable Arctic change estimates; for example, the contrast of PCA performed on 11 stations beginning in 1886, 20 stations beginning in 1912, and 45 stations beginning in 1936 is illustrated.

While often there is a well-known interdecadal negative covariability in winter between northern Europe and Baffin Bay, long-term changes in the remainder of the Arctic are most evident in spring, with cool temperature anomalies before 1920 and Arctic-wide warm temperatures in the 1990s. Summer anomalies are generally weaker than spring or winter but tend to mirror spring conditions before 1920 and in recent decades. Temperature advection in the trough–ridge structure in the positive phase of the Arctic Oscillation (AO) in the North Atlantic establishes wintertime temperature anomalies in adjacent regions, while the zonal/annular nature of the AO in the remainder of the Arctic must break down in spring to promote meridional temperature advection.

There were regional/decadal warm events during winter and spring in the 1930s to 1950s, but meteorological analysis suggests that these SAT anomalies are the result of intrinsic variability in regional flow patterns. These midcentury events contrast with the recent Arctic-wide AO influence in the 1990s. The preponderance of evidence supports the conclusion that warm SAT anomalies in spring for the recent decade are unique in the instrumental record, both in having the greatest longitudinal extent and in their associated patterns of warm air advection.

Wavelets are relatively new mathematical tools that have proven to be quite useful for analyzing time series and spatial data. We provide a basic introduction to wavelet analysis which concentrates on their interpretation in the context of analyzing time series. We illustrate the use of wavelet analysis on time series related to vegetation coverage in the Arctic region.

In this article we consider trend to be smooth deterministic changes over long scales, and tackle the problem of trend estimation in the presence of long memory errors (slowly decaying autocorrelations). Using the fractionally differenced (FD) process as a motivating example of such a long memory process, we demonstrate how the discrete wavelet transform (DWT) is a natural choice at extracting a polynomial trend from such an error process. We investigate the statistical properties of the trend estimate obtained from the DWT, and provide pointwise and simultaneous confidence intervals for the estimate. Based on evaluating the power in the trend estimate relative to the estimated errors, we provide a test of nonzero trend. We finish by applying the methods to a climatological example.

We present a case study in modeling the North Pacific (NP) index, which is a time series related to atmospheric pressure variations at sea level. We consider three statistical models, namely, a Gaussian stationary autoregressive process, a Gaussian stationary fractionally differenced (FD) process, and a "signal plus noise" process consisting of a square wave oscillation with a pentadecadal period embedded in Gaussian white noise. Each model depends upon three parameters, so all three models are equally simple. Statistically each model fits the NP index equally well. The fact that this index consists of just a hundred observations makes it unrealistic to expect to be able to clearly prefer one model over the other. Although the models fit equally well, their implications for the long term behavior of the NP index can be quite different in terms of, e.g., generating regimes of characteristic lengths (i.e., stretches of years over which the NP index is predominantly either above or below its long term average value). Because we cannot determine a preferred model statistically, we are faced with either entertaining multiple models when considering what the long term behavior of the NP index is likely to be or using physical arguments to select one model. The latter approach would arguably favor the FD process because it has an interpretation as the synthesis of first order differential equations involving many different damping constants.

In this primer we first give an overview of stochastic models that can be used to interpret clock noise. Because of their statistical tractability, we concentrate on fractionally differenced (FD) processes, which we relate to fractional Brownian motion, discrete fractional Brownian motion, fractional Gaussian noise and discrete pure power law processes. We discuss several useful extensions to FD processes, namely, composite FD processes, autoregressive fractionally integrated moving average processes and time-varying FD processes. We then consider the statistical analysis of clock noise in terms of how these models are manifested via the spectral density function (SDF) and the wavelet variance (WV), the latter being a generalization of the well-known Allan variance. Both the SDF and the WV decompose the process variance with respect to an independent variable (Fourier frequency for the SDF, and scale for the WV); similarly, judiciously chosen estimators of the SDF and WV decompose the sample variance of clock noise. We give an overview of the statistical properties of estimators of the SDF and the WV and show how these estimators can be used to deduce the parameters of stochastic models for clock noise.

Discrete fractional Gaussian noise (dFGN) has been proposed as a model for interpreting a wide variety of physiological data. The form of actual spectra of dFGN for frequencies near zero varies as f^{1–2H, where 00 rather than f1–2H. For finite N and small H, the expected value of the periodogram can in fact exhibit a local power-law behavior with a spectral exponent of 1–2H at only two distinct frequencies.}

As part of the effort to characterize the acoustic environment during the high frequency sediment acoustics experiment (SAX99), fine-scale variability of sediment density was measured by an in situ technique and by core analysis. The in situ measurement was accomplished by a newly developed instrument that measures sediment conductivity. The conductivity measurements were conducted on a three-dimensional (3-D) grid, hence providing a set of data suited for assessing sediment spatial variability. A 3-D sediment porosity matrix is obtained from the conductivity data through an empirical relationship (Archie's Law). From the porosity matrix, sediment bulk density is estimated from known average grain density. A number of cores were taken at the SAX99 site, and density variations were measured using laboratory techniques. The power spectra were estimated from both techniques and were found to be appropriately fit by a power-law. The exponents of the horizontal one-dimensional (1-D) power-law spectra have a depth-dependence and range from 1.72 to 2.41. The vertical 1-D spectra have the same form, but with an exponent of 2.2. It was found that most of the density variability is within the top 5 mm of the sediment, which suggests that sediment volume variability will not have major impact on acoustic scattering when the sound frequency is below 100 kHz. At higher frequencies, however, sediment volume variability is likely to play an important role in sound scattering.

We consider the problem of testing for homogeneity of variance in a time series with long memory structure. We demonstrate that a test whose null hypothesis is designed to be white noise can, in fact, be applied, on a scale by scale basis, to the discrete wavelet transform of long memory processes. In particular, we show that evaluating a normalized cumulative sum of squares test statistic using critical levels for the null hypothesis of white noise yields approximately the same null hypothesis rejection rates when applied to the discrete wavelet transform of samples from a fractionally differenced process. The point at which the test statistic, using a nondecimated version of the discrete wavelet transform, achieves its maximum value can be used to estimate the time of the unknown variance change. We apply our proposed test statistic on five time series derived from the historical record of Nile River yearly minimum water levels covering 622%u20131922 A.D., each series exhibiting various degrees of serial correlation including long memory. In the longest subseries, spanning 622%u20131284 A.D., the test confirms an inhomogeneity of variance at short time scales and identifies the change point around 720 A.D., which coincides closely with the construction of a new device around 715 A.D. for measuring the Nile River. The test also detects a change in variance for a record of only 36 years.

A major difficulty in investigating the nature of interdecadal variability of climatic time series is their shortness. An approach to this problem is through comparison of models. In this paper a first-order autoregressive [AR(1)] model is contrasted with a fractionally differenced (FD) model as applied to the winter-averaged sea level pressure time series for the Aleutian low [the North Pacific (NP) index] and the Sitka winter air temperature record. Both models fit the same number of parameters. The AR(1) model is a "short-memory" model in that it has a rapidly decaying autocovariance sequence, whereas an FD model exhibits "long memory" because its autocovariance sequence decays more slowly.

Statistical tests cannot distinguish the superiority of one model over the other when fit with 100 NP or 146 Sitka data points. The FD model does equally well for short-term prediction and has potentially important implications for long-term behavior. In particular, the zero crossings of the FD model tend to be farther apart, so they have more of a "regimelike" character; a quarter century interval between zero crossings is 4 times more likely with the FD than the AR(1) model. The long-memory parameter δ for the FD model can be used as a characterization of regimelike behavior. The estimated δs for the NP index (spanning 100 yr) and the Sitka time series (168 yr) are virtually identical, and their size implies moderate long-memory behavior. Although the NP index and the Sitka series have broadband low-frequency variability and modest long-memory behavior, temporal irregularities in their zero crossings are still prevalent. Comparison of the FD and AR(1) models indicates that regimelike behavior cannot be ruled out for North Pacific processes.

A fractionally differenced (FD) process is used to model aerothermal turbulence data, and the model parameters are estimated via wavelet techniques. Theory and results are presented for three estimators of the FD parameter: an "instantaneous" block-independent least squares estimator and block-dependent weighted least squares and maximum likelihood estimators. Confidence intervals are developed for the block-dependent estimators. We show that for a majority of the aerothermal turbulence data studied herein, there is a strong departure from the theoretical Kolmogorov turbulence over finite ranges of scale. A time-scale-dependent inertial range statistic is developed to quantify this departure.

Multiscale analysis of univariate time series has appeared in the literature at an ever increasing rate. Here we introduce the multiscale analysis of covariance between two time series using the discrete wavelet transform. The wavelet covariance and wavelet correlation are defined and applied to this problem as an alternative to traditional cross-spectrum analysis. The wavelet covariance is shown to decompose the covariance between two stationary processes on a scale by scale basis. Asymptotic normality is established for estimators of the wavelet covariance and correlation. Both quantities are generalized into the wavelet cross covariance and cross correlation in order to investigate possible lead/lag relationships. A thorough analysis of interannual variability for the Madden-Julian oscillation is performed using a 35+ year record of daily station pressure series. The time localization of the discrete wavelet transform allows the subseries, which are associated with specific physical time scales, to be partitioned into both seasonal periods (such as summer and winter) and also according to El Nino-Southern Oscillation (ENSO) activity. Differences in variance and correlation between these periods may then be firmly established through statistical hypothesis testing. The daily station pressure series used here show clear evidence of increased variance and correlation in winter across Fourier periods of 16–128 days. During warm episodes of ENSO activity, a reduced variance is observed across Fourier periods of 8–512 days for the station pressure series from Truk Island and little or no correlation between station pressure series for the same periods.

Many physical processes are an amalgam of components operating on different scales, and scientific questions about observed data are often inherently linked to understanding the behavior at different scales. We explore time-scale properties of time series through the variance at different scales derived using wavelet methods. The great advantage of wavelet methods over ad hoc modifications of existing techniques is that wavelets provide exact scale-based decomposition results. We consider processes that are stationary, nonstationary but with stationary dth order differences, and nonstationary but with local stationarity. We study an estimator of the wavelet variance based on the maximal-overlap (undecimated) discrete wavelet transform. The asymptotic distribution of this wavelet variance estimator is derived for a wide class of stochastic processes, not necessarily Gaussian or linear. The variance of this distribution is estimated using spectral methods. Simulations confirm the theoretical results. The utility of the methodology is demonstrated on two scientifically important series, the surface albedo of pack ice (a strongly non-Gaussian series) and ocean shear data (a nonstationary series).

Procedures for detecting change points in sequences of correlated observations (e.g., time series) can help elucidate their complicated structure. Current literature on the detection of multiple change points emphasizes the analysis of sequences of independent random variables. We address the problem of an unknown number of variance changes in the presence of long-range dependence (e.g., long memory processes). Our results are also applicable to time series whose spectrum slowly varies across octave bands. An iterated cumulative sum of squares procedure is introduced in order to look at the multiscale stationarity of a time series; that is, the variance structure of the wavelet coefficients on a scale by scale basis. The discrete wavelet transform enables us to analyze a given time series on a series of physical scales. The result is a partitioning of the wavelet coefficients into locally stationary regions. Simulations are performed to validate the ability of this procedure to detect and locate multiple variance changes. A 'time' series of vertical ocean shear measurements is also analyzed, where a variety of nonstationary features are identified.