Extreme events: the floods that displace us from our homes, the high waves that wash out coastal roads, or the toppling of trees and power poles from a passing storm. For locations around the Pacific Basin, where remote island chains sit perilously close to sea level and where rainfall is the primary source of water, questions arise concerning the return frequency and duration of such events, and whether or not they are getting more extreme. Understanding the long-term variability and change in coastal climate extremes has grown in public awareness given the potentially severe impacts related to sea-level rise coupled with coastal storms. To reduce vulnerability to the economic, social, and environmental risks associated with these phenomena, decision makers in coastal communities need timely access to accurate and contextually relevant information that affords them an opportunity to plan and respond accordingly.
To address this need, the Pacific Storms Climatology Products (PSCP) project—or Pacific Storms—was established under the direction of the NOAA National Climatic Data Center (NCDC). Pacific Storms is focused on improving our understanding of patterns and trends in storm frequency and intensity—“storminess”—within the Pacific Basin. Pacific Storms is exploring how the climate-related processes that govern extreme storm events are expressed within and between three thematic areas: strong winds, heavy rains, and high seas.
Theme-specific data integration and product development teams were formed to conduct analyses and create a broad suite of derived data products, which are publicly available online (www.pacificstormsclimatology.org). These teams included representatives from NCDC, as well as NOAA's Center for Operational Products and Services (CO-OPS), Coastal Services Center (CSC), and National Weather Service (NWS), and the University of Hawaii, University of Alaska, University of Guam, and Oregon State University. Sources of information include NOAA's Integrated Surface Hourly (ISH) mean sea-level pressure and wind speed data, the Global Historical Climate Network-Daily (GHCN-D) precipitation dataset, the National Water Level Observing Network (NWLON) tide gauge records, the University of Hawaii Sea Level Center (UHSLC) Joint Archive for Sea Level research quality dataset and Global Sea Level Observing System (GLOSS)/Climate Variability and Predictability (CLIVAR) “fast delivery” sea level dataset, the National Data Buoy Center (NDBC) wave buoy records, the U.S. Army Corps of Engineers' Coastal Data Information (CDIP) buoy data, and other data sources.
The data analysis and product development frame-work and guidelines outlined below are innovative in a number of ways. First, they focus on extreme events, and integrate data and products across a range of storm-related phenomena. Furthermore, they also paint a comprehensive picture of changes and variation in extreme-event magnitude and frequency for a mix of theme-specific parameters on seasonal, annual, and interannual time frames. The resulting extremes climatology datasets are unique, as are some of the specific products. Finally, success of the project is fundamentally tied to the collaborative efforts of the data integration and product development teams.
The development of extremes climatology derived data products was limited to the analyses of historical records obtained from in situ stations (e.g., meteorological stations, rain gauges, sea-level stations, and wave buoys) located throughout the Pacific Basin, defined for this project to encompass all terrestrial and buoy stations within 100 km of any coast—continental or island—ranging from South America to Australia. Individual station datasets were placed into subsets for analysis and product development on the basis of their suitability with respect to record length and continuity. The “3/5 rule” for calculating monthly and annual 30-yr climate normals as set by the World Meteorological Organization (WMO) in 1989—which considers a period of record incomplete if more than 3 consecutive observations or 5 daily totals in a month are missing—was incorporated such that the station datasets could be broken into three distinct levels (Table 1):
Only years and months with at least 80% of the record and days with at least four regularly spaced hourly observations and a record length of at least 30 years;
Only years and months with at least 75% of the record and days with at least one observation in every 6-hour observing interval and a record length of at least 20 years; and
Only years and months with 66% of the record and days with at least four hourly observations and a record length of at least 15 years.
In all cases, a year must have at least 50% of the “winter season” data, as most occurrences of strong climate impact parameters occur during this season, and at least two of the years in the record must appear since 1998. This requirement is intended to ensure that trends or “regime shifts,” which may have occurred more recently in the record, are accounted for. From a statistical perspective, the stations fitting the Level 1 criterion are the most desirable, most complete, and ideal for use in computing valid statistics. Conversely, the Level 3 stations reflect the reverse; any less data is not worthy of consideration. Level 2 stations represent the compromise; they could be better but are not too bad. If only Level 1 stations were considered, there would be very few stations that qualified for analysis. The use of Level 2 and 3 stations provides a broader set of data to work with, albeit with limitations.
The data were broken down by year and season. For the annual series, the “analysis year” is defined as the period starting on 1 May of one year and ending on 30 April of the following year. For the seasonal series, the winter season is defined as the 1 October through 30 April subset of the annual series, and the summer season is the 1 May through 30 September subset of the annual series. This seasonal definition has broad regional applicability; it serves to distinguish between the wet and dry seasons in the Hawaiian Islands and the cold and warm seasons in Alaska. Correspondingly, this definition has implications for the types of storms impacting each region during these periods. For example, in Hawaii, winter is the wet season, when cold fronts from the midlatitudes, along with Kona storms and upper-level low pressure systems, impact the islands. By contrast, summer is dominated by the trade winds and most rainfall is topographically enhanced.
A further breakdown of the data was performed within each theme. Specifically, rainfall analyses were done for 1-, 5-, and 30-day “events,” and wind speed and direction analyses were done in total and by quadrants. Water levels were done for observed water levels, nontidal residuals, and high seas. High seas–waves were characterized by significant wave height and power. The 5- and 30-day rainfall events are calculated using a running mean for the period in question. Specifically, the 5- and 30-day totals are derived from 1-day totals by summing the current day with the d-1, d-2, d-3, and d-4 (for the 5-day sum), and so on for the 30-day sum. As each day increments by 1, the sum window moves along with it.
Parameters extracted for product development from each of these theme-based datasets were the mean, maximum, and extreme “event.” Following the method for determining long-term variability of extreme events used by F. J. Mendez and colleagues, special attention was given to extracting extreme values at the level of the storm event, rather than the level of the individual observation, to avoid contaminating the dataset with nonindependent observations. Specifically, extreme “events” were defined by first selecting all values above the annual 95th percentile value and then filtering the hourly or daily time series to eliminate all but the largest reported values within a 72-h period, a time scale longer than the duration of typical Pacific Basin storms. This is a similar approach utilized by D. E. Atkinson to identify storm events in a wind speed record. For the sea level extremes, the approach parallels that used by M. A. Merrifield et al., where extreme event climatologies are formed for a global set of sea level stations by deconstructing the records into seasonal, tidal, and high-frequency water level components.
The incorporation of nonstationary Generalized Extreme Value (GEV) and Peaks-Over-Threshold (POT) analysis is a significant aspect of the Pacific Storms products. A nonstationary process has characteristics that change through time. Following S. G Coles's statistical modeling of extreme values, the GEV analysis provides a model for the smallest or largest extremes among a large set of independent identically distributed random values representing measurements or observations. Block maxima or minima data are then fit to the GEV distribution to explain the probability distribution of the extremes Conversely, the POT analysis, also called the Generalized Pareto Distribution (GPD), requires a sample o data exceeding a sufficiently high threshold, and the differences between the data and the threshold are used to determine the probability distribution of the extremes. Standard practice has been to apply extreme value statistical models to project the 25- through 100-yr return levels, respectively, having progressively lower probabilities of occurrence, where the 100-yr event has a 1% probability during any given year Those analyses have, for the most part, assumed that on average the data being analyzed are “stationary,” and long-term trends of progressively increasing or decreasing parameters have not been accounted for However, considering climate change and variability the corresponding high waves, strong winds, and heavy rains climates in many cases have been found to be nonstationary. This is specifically true for the wave buoys along the Oregon and Washington coasts, where several studies note that both wave heights and wave periods have been increasing since the mid-1970s.
Therefore, projections of the extreme values used in applications must take into account both trends and patterns in the measured elements, and to the extent possible acquire an understanding of their underlying climate controls. A variety of procedures have been developed to undertake extreme value assessments for time-varying data populations, directed in many cases toward environmental changes that are a result of climate change and variability, such as increased rainfalls and river discharges. These approaches range from simply extrapolating extreme value projections using trends derived from the historical record to fitting separate extreme value distributions to different periods within the data record, and, in the most sophisticated analyses, applying statistical models such as those used by X. Zhang and colleagues as well as F. J. Mendez et al. that directly incorporate nonstationary processes in assessments of extreme values.
The primary purpose of the product line is to delineate the patterns and trends of extremes within and between locations and regions, how they have been expressed historically, and how they may be (expected to be) expressed in a changing climate. The products also establish a common framework to support regional and site-specific intercomparisons. Furthermore, the products employ a common look and feel across the themes to make them easier to understand and use, and clear technical oversight via product peer-review during the development process was exercised to ensure suitability for large-scale web delivery. The datasets derived through the process identified in the previous section formed the basis for product development.
There are eight primary product types available through Pacific Storms that depict patterns and trends of extremes in various forms—foundational, interannual, and annual:
Annual Time Series (foundational)
Cumulative Distribution Function (foundational)
Frequency Counts (interannual)
Exceedance Probabilities (interannual)
Extreme Value-Based Trends (interannual)
Daily Time Series (annual)
Monthly Polar Plots (annual)
Monthly Frequency Counts (annual)
In PSCP, the foundational product represents the most basic analyses: a time series and the frequency distribution. They are the foundation upon which all the other analyses are built. The foundational products result in a derived dataset from which sub-sequent analyses and thresholds can be determined. An example of a foundational product is shown in Fig. 1. Panel A represents the rainfall distribution for Snowtown, Australia, for the 30-day event (as described in the “Data Analysis” section), and panel B shows the wind speed analysis for Portland, Oregon. This product depicts the time history of these elements for the entire period of record for each station and is plotted for both the annual and seasonal series. For the rainfall, the 30-day maximum is determined by identifying the highest 30-day sum during the course of the year or period of record. For wind speed and rainfall, the five all-time events are the top five highest 30-day maximums in the period of record.
An example of an interannual product is shown in Fig. 2 for the National Data Buoy Center (NDBC) wave buoy southwest of Santa Monica, California. In contrast to typical extreme value analysis, this product is unique due to the application and incorporation of nonstationarity. This product displays the long-term trends calculated using modified nonstationary versions of the GEV (top graph) and GPD (lower graph) extreme value theories for each theme and indicator—the annual maxima and extreme event, respectively—for the annual and season series for the entire length of each station record. It is evident from the GPD plot that only a few events in a given year will exceed the calculated threshold.
Finally, an example of an annual product is shown in Fig. 3, which depicts the monthly water levels for Honolulu, Hawaii. Note that Fig. 3 also demonstrates the water levels along with the nontidal residuals (NTR)—contributions to water level other than oscillatory tides, such as storm surges, wind-driven water, etc. The NTR component of the observed water level was determined by subtracting the value of the hourly predicted tide from the matching value of the observed water level, where the predicted tide was computed using the harmonic tidal analysis program T_TIDE. Annual and semiannual harmonics are included in the tide fit; hence, seasonal water level changes are included in the predicted tide. Given the definition of the predicted tide, nonlinear interactions between the tide and storm-driven surge in shallow water are also included in the residual component.
The full suite of Pacific Storms products can be found at www.pacificstormsclimatology.org, an experimental product development site of NOAA. Overall, the site is geared toward a technical audience—in particular, to those that have some familiarity with statistical summaries and portrayals—though many of the product descriptions and regional climatologies will be of interest to a broader, more general audience. The main feature of the site is the ability to query and view products by station within each region. To view an interactive map showing all locations where a particular type of product is available, simply select a “Product Type” from among the list on the site. Once selecting a product, the user is then guided to a query results page where they can select an individual station. Doing so will reveal basic information about the station (latitude, longitude, elevation, etc.) and data and products available on the site.
Other subsections of the site include an overview of the project itself and a novel description of the regional synoptic climatology of each of the major geographic locations, both of which are geared toward a more general audience and that were developed for Pacific Storms. The synoptic climatology for Hawaii, for example, includes identification and discussion of the types of storm events that are associated with extremes in Hawaii, such as the frequency of Kona lows, cold fronts, and tropical cyclones. The site also includes a tutorial on how to identify key features from the plots and how to obtain the most information from the site.
In the future, the Pacific Storms project will continue to focus on developing an integrated picture of long-term trends and patterns of extreme events. A particular emphasis will be placed on the development of new products that explore the linkages of extremes climatologies to climate indices and tele-connections, such as correlations between extreme nontidal residuals and El Niño–Southern Oscillation (ENSO) occurrences. Area-wide (as opposed to individual station-based) products will also be added, like the ensemble of ENSO station climatologies for the various subregions. Attention is also being given to automation of the data analysis and product development process as a means to facilitate annual updates. While developed with a focus on the Pacific Basin, the comprehensive and integrated data analysis in concert with the product development framework and guidelines outlined here are readily applicable to other regions across the United State and the globe.