Latest Data Disclosure

Data Disclosure for the Actuaries Climate IndexTM Winter 2016-17 Release
October 5, 2017

In performing the work to produce the current values through winter 2017 for the Actuaries Climate IndexTM (ACI) and its six components, the American Academy of Actuaries (Academy), Casualty Actuarial Society (CAS), Canadian Institute of Actuaries (CIA), and Society of Actuaries (SOA) relied upon data and information provided by Solterra Solutions and a number of publicly available data sources: the National Oceanic and Atmospheric Administration (NOAA), CLIMDEX1, and Permanent Service for Mean Sea Level. We reviewed the data and information provided for reasonableness but did not perform detailed audits. We have, therefore, relied upon each of these sources to provide accurate and complete data and information

The underlying data for the ACI and its six components is based on measurements from an extensive network of meteorological stations and coastal tide stations within the United States (excluding Hawaii and unincorporated United States territories) and Canada. Some of the ACI data source providers have analyzed meteorological station observations to produce gridded data sets. For all ACI components except sea level, 2.5 degree latitude by 2.5 degree longitude grid data is used to develop ACI values.

The grid data and sea level station data are used to develop monthly and seasonal regional time series beginning in 1961 for the ACI and its components. Contrary to the USA region definition found on the ACI website, grid data and sea level station data for Alaska are included in developing USA ACI values.

In reviewing the underlying data, we noted that not all grids and sea level stations had observational data for every month or season in the time series under study—potentially impacting the accuracy of the analysis and results for the ACI and its components. Missing data is more common in the most recent months or seasons because the data used for the ACI and its components may not be updated to reflect the most recent station observations.

For sea level, because those stations included in the analysis are sparser than the meteorological stations used to produce the grid data for the other components, values for missing data are estimated by interpolation and extrapolation. For the other components, missing data are not imputed.

An in-depth analysis to measure the impact of the missing values on the results for every ACI component and region has not been performed. Given the large number of measurements from meteorological stations for each region, it is reasonable to expect only minor impacts for these components and the ACI. However, this may not be true for some regions:

  1. For some regions, the amount of grid data has declined across time. This is especially true for the Central Arctic (CAR) region. As a result, the lack of data in recent years may potentially distort ACI results. An example is the December 2016 monthly value for the precipitation component.
  2. The same is true for the Northwest Pacific (NWP) region for precipitation. NWP ACI values for precipitation have shown a significant upward movement in recent seasons as compared to past seasons. Completed analysis indicates data holes probably contributed to the increasing values.

Historical results for the ACI and its components may change with each website update because the underlying data may change. Data values may be updated by third parties, not only for recent months and years, but also for prior time periods.

In comparing the current data set to that used in the previous release, the following observations were made:

  1. Historical sea level data changed, as underlying data is now available through 2016, producing a noticeable difference in ACI sea level results for Canadian regions.
  2. Current ACI values for some regions and components are much higher than the prior recent periods. In general, these results occur for cases in which the standard deviation of the regional mean is relatively small. Examples include the following:
  • The warm temperature component, T90, November 2016 monthly values for Canada (CAN), Northern Plains (NPL), and combined U.S. and Canada (USC) regions;
  • Sea level September 2016 monthly value for Central East Atlantic (CEA) region; and
  • The precipitation component, Rx5-day, October 2016 monthly value for Canada (CAN) region.

Note that the scale of the ACI is in standard deviations for each component. Therefore, a component index value of 1.0 indicates that the index is one standard deviation above the mean value of that index during the reference period, based on the reference period standard deviation.

As an example, consider the temperature component T90, which describes the upper tail of the distribution of daily temperatures. Assuming temperatures (and also the exceedances represented by T10 and T90) are normally distributed, about one-third of the time one expects that T90std will be outside the interval ±1, and one-sixth of the time it will be greater than +1. But if it exceeds +2, this is indicative of a rare event, because it is expected only 2.5 percent of the time. Values exceeding +3 are very rare, and expected only approximately 0.125 percent of the time. Hence, the value of T90std is a direct reflection of the rarity of the events it tracks.

The composite index and its five-year rolling average are also shown on the same scale in the website graphs and documentation, for ease of comparison with the component indices, but the standard deviation for the composite is approximately 0.45 (depending on region). The lower standard deviation of the composite index results from its construction as the mean of the six components, which has the effect of lowering the variability of the composite relative to the variability of the components. The graph below illustrates this effect by showing two y-axis scales: standard deviations for the composite ACI should be read from the scale on the right and component standard deviations should be read from the left in order to properly assess the rarity or likelihood of index values being at a given level. Stated differently, because the composite index is calculated as the mean of the components, its values should be read from the left scale—but its standard deviations, for purposes of determining likelihood, should be read from the right scale.

The Academy, CAS, CIA, and SOA continue to explore ways to enhance the ACI and its components, which could impact historical results. Therefore, historical ACI results may change with each website update due to the implementation of a new data source or application of a new methodology/technique for the analysis of the ACI and its components. Areas for enhancement being considered include but are not limited to the following:

A. Data

  1. Comparing and contrasting various methodologies for addressing missing data in the analysis for the ACI and its components, such as the NWP region for precipitation and the CAR region for all components.
  2. Investigating new data sources for sea level and consecutive dry days (CDD).
  3. Identifying unexpected regional results per component and performing detailed data analysis. For the most current data, one regional result is under review:
    i. Central East Atlantic for wind power (WP), which has been a significant downward outlier in recent years.

B. Analysis

  1. Performing an audit of the WP calculation. In reviewing the ACI analysis, the average number of events over the base period is not 10% as expected, but 13%. The wind power component of the ACI is calculated by taking the standardized anomaly, i.e., the current observation less the mean in the reference period, and dividing by the standard deviation in the reference period. It was determined that the use of a higher threshold of 13% has little impact on the validity of the standardized anomaly calculation, and therefore the results have been released and are included on the website.
  2. Investigating alternative CDD methodology or other drought data sources/indicators. The input data for this component are on an annual basis and not as robust as what would be provided by a more frequently updated data source. Monthly CDD results are approximated by linear interpolation of the annual values.
    After the initial release in November 2016, a change was made to the methodology for producing the standardized anomalies for this component to more accurately reflect the variation found in the base period data. In the initial release, a separate reference period mean and standard deviation was calculated for each month, based on the 30 values for that month in the reference period. In the current methodology, the reference period mean and standard deviation is based on the 360 monthly CDD values.
  3. Investigating alternative methodologies for determining ACI regional values including regions USA, CAN, and USC.
  4. The sea level component excludes the impact of land motion. The impact of land motion on sea levels will be investigated.
  5. There is currently no sea level index for the Midwest (which has no ocean coastline) and Canadian Arctic (which has no historically reliable tide stations) regions. Proxies may be considered for these regions in the future but the overall ACI and country-level indices would not be affected.

For more information on the data and development of the ACI, see the Development and Design document and Terms of Use. Prior data disclosures are available upon request.


1CLIMDEX (Datasets for Indices of Extreme Weather) is developed and maintained by researchers at the Climate Change Research Centre (CCRC), The University of New South Wales (UNSW) and funded by the Australian Research Council and the Australian Department of Climate Change and Energy Efficiency through Linkage project LP100200690 and in collaboration with the University of Melbourne, Climate Research Division (Environment Canada) and NOAA’s National Climatic Data Center (USA). See also: Expert Team on Climate Change Detection and Indices at the World Meteorological Organization.

 

Last Updated: October 5, 2017

Save

Save

Save