A detailed quality assessment of this data set has not been completed. However, it is possible to make some generalizations, drawing heavily on two studies based on EWG Atlas data (Dedrick et al., 2001, and Partington et al., 2003). The history of changes in data sources and charting practices is also relevant.

Accuracy and Precision

The accuracy and precision of the original charts is not known with certainty. Partington (2003) cites ±5 percent to ±10 percent as the accuracy of ice concentration estimates. Total ice concentration for an individual polygon is recorded in a range. That range is expressed in tenths. The mean value of that range is used in the EASE-Grid files. Thus, the precision can vary from grid cell to grid cell depending on the range with which the concentration was originally charted.

An analysis by H. Stern, University of Washington (personal communication, 2005) indicated that fast ice information is unreliable prior to 1976.

Geiger (2006) published an error analysis of NIC's Antarctic charts.

Meier et al. (2006) compared NIC charts with passive microwave concentrations in a poster titled Operational sea ice charts: An integrated data product suitable for observing long-term changes in Arctic sea ice? available on the NSIDC Posters and Presentations Web page.

Strengths and Weaknesses

The primary strength of the NIC chart series is that the charts were created by specialists using manual analyses of data from many sources. Where the ice concentration might be difficult to gauge in the summer in some places using SAR imagery because of surface melt, for example, the analyst has the option of checking visible band imagery. If the ice edge cannot be located in visible band imagery because of clouds, the analyst can use scatterometry. NIC software allows image manipulation to enhance imagery. To date, this mostly manual form of multisensor data fusion or data assimilation produces ice information more accurately than using any single data source or automated approach.

Manual analysis does have weaknesses. One is the subjective nature of image interpretation for ice concentration and for ice type. Ice concentration is generally estimated visually by looking at what are generally bright ice floes against a dark background. Ice concentration estimated in this way is estimated to be accurate to within about +- 10% (Partington, 2003). Ice type (stage of development) is more difficult to gauge. High resolution visible and IR imagery, and especially SAR imagery used to distinguish MY from younger forms, along with a time series of charts and imagery from the area being analyzed, makes it possible.

The primary weakness of the chart series is inconsistency in the quality of the analyses. Such inconsistency arises from a number of sources. However, it is important to note that the operational source data from NIC received additional quality control by NIC before being added to the archive for this product.

Sources of inconsistency include the following:

1. Increasing number of data sources for analysis, increasing resolution and volume of satellite data, evolving analysis systems.

In the 1970s, analysts often had to make educated guesses based on climatology or persistence. As data sources grew to include active as well as passive sensor data, analysts could chart ice in increasing detail, as suggested by Figures 8 and 9.

From 1972 through 1979, only total concentration was recorded. Analysts depended heavily on visible, infrared, and single channel passive microwave imagery. Passive microwave data from SMMR and SSM/I were added in 1980 and 1989 respectively. In 1990, NIC's first computer assisted production system (DIFAS1) was fully installed. In 1991, NIC began to use OLS visible and IR imagery, with roughly .5-km resolution, versus the 1-km resolution of already available AVHRR. In 1992, increasing use of digital data along with improved prediction models made it possible for analysts to break down the amount of ice at each stage of development; or in other words, report partial concentrations for each ice type (this advance happened later, in 1994, for Antarctic data, although it has been restricted to the Ross Sea region since 2000).

NIC began to use ERS-1 SAR (at 240 m resolution) in 1995, and RADARSAT-1 (at 200 m, 100 m, and 25 m resolutions) in 1996, while continuing to add to the capabilities of its computer system for image analysis and beginning to move toward a GIS production environment. Scatterometer data (QuikSCAT) was added in 2004.

More recently, beginning in 2005, Moderate Resolution Imaging Spectroradiometer (MODIS) and Envisat Advanced SAR (ASAR) Global Monitoring Mode (GMM) data have been available to analysts. In January 2006, the NIC installed the Satellite Image Processing and Analysis System (SIPAS), an almost completely ESRI-based analysis and production system, which allows the analyst to both analyze imagery and produce spatial data (ice chart) files in a common environment.

2. The operational nature of chart production.

Charts produced to support operations or in areas of high national interest are naturally created with more attention to detail and often with greater frequency. Usually, the ice edge; marginal ice zone; and seas such as the Bering, Chukchi, and Beaufort that have important U.S. economic activity are given a higher priority than the central Arctic, where operations take place less frequently.

A large proportion of the NIC analyses are produced by analysts who are military personnel on two to three year rotations. Analysts vary in skill and level of training, thus a consistent bias in a particular analyst's output is possible. In order to mitigate this effect, the analyst training curriculum was codified and standardized in the early 2000s.

3. Changes in analysis conventions or procedures

Prior to 1976 and after 1986, pack ice in the central Arctic was charted as 9 to 10 tenths concentration, if not known to be otherwise (Partington et al., 2003). Between those years, such ice was labeled 10 tenths. Therefore the central Arctic area concentration may be labeled in the SIGRID code as 92, meaning 100 percent, or 91 for between 90 and 100 percent. In the EASE-Grid files, this is either 95 percent or 100 percent concentration.

There have been no changes in the convention of charting the ice edge as the boundary between no ice and ice of any concentration including concentrations of less than 10 percent.

Fast ice has been charted continuously, but the information appears to be unreliable prior to 1976 (H. Stern, personal communication, 2005). When fast ice breaks up, it is classified as either FY or MY ice, but this designation may not be consistent with the ice type it was assigned before becoming fast ice, if any. If the ice was MY when the fast ice formed, then the MY stage would be carried through in the resulting polygon for the fast ice area. If the fast ice formed from young or new ice, normally it would hold together until reaching a first year stage of development, thus it would naturally not be the same stage during breakup as during freezeup (P. Seymour, personal communication, October 2006.)

After 1994, all polygons carried partial concentrations (See Figure 6). Only total concentration was consistently reported over the entire operational chart period.

There have been no changes through 2004 in the convention of recoding all ice in stages of development younger than second year to old ice on 1 October of each year. (However, beginning in October 2005, NIC adopted the CIS practice of recoding younger types to second year, and second year to MY on 1 October in much of the Western Arctic)

For 1972 through 1994, NIC produced separate Eastern and Western Arctic charts. The Western Arctic chart was produced on Tuesday and the East Arctic on Wednesday. (On the EWG Atlas these are combined into one Arctic-wide file). By 1996, NIC began producing thirty-nine non-overlapping regional charts like those in Figure 16. These were combined into an Arctic-wide file for the online archive, with boundaries between regions matched for consistency. Beginning in January 2006, however, NIC adopted a new architecture whereby the hemispheric analysis is created first with the regional charts cut out after the entire hemisphere is completed.

NIC changed from producing charts on a weekly basis to biweekly chart production in mid June 2001.

In June 2003, the North American Ice Service (NAIS) was formed. Under the NAIS umbrella, CIS, NIC, and the International Ice Patrol (IIP) are working together toward harmonizing Canadian and U.S. charting practices and operations in support of the joined U.S. and Canadian ice requirements. More changes in charting practices over the coming years will result.

4. Introduction of error from data transformations and regridding.

The transformation of chart data from paper or vector format to gridded format results in some unavoidable loss of information and introduction of error. These have not been quantified. The documentation related to the precursor data sets offers information relevant to understanding this source of error. Also see the Note on Resolution in the Data Processing section.

The climatology products on the EWG Atlas precursor data set used 25-km resolution weekly data, obtained by hand digitizing paper ice charts. Later, post 1994 weekly data in vector interchange (.e00) format allowed ice concentration to be represented at a much finer resolution. We encountered difficulties, partially resolved at this time, in merging these two data sources related to differences in the representation of land. Land is masked (actually embedded in the source data) at differing resolutions (and may have been determined differently in the EWG products); and when transferred to a common 25-km grid for this data set, differences can remain. Figure 10 shows an example for the Canadian Queen Elizabeth Islands. At the time of publication of this data set (October 2006), our solution is to merge the land masks by creating a new land mask where either of the old masks had land. Preliminary testing has shown this to be only a partially acceptable solution, especially in the study of relatively small areas of fast ice near the coast.

Discontinuities

The factors listed above result in inconsistency, but not necessarily in discontinuities. Based on preliminary analysis, a discontinuity probably resulted from changes taking place over the period 1994-1997. We believe that total concentrations before 1997 are biased low relative to those after, due primarily to the lack of high resolution active microwave data for tactical use before that year (T. Street, personal communication, September 2005). Other changes from 1994 to 1997:

• NIC transitioned to digital imagery analysis tools for OLS and AVHRR imagery in 1996 and 1997.
• NIC began creating regional analyses rather than hemispheric or semi-hemispheric analyses.
• Source files for this data set changed from the EWG Atlas EASE-Grid files to ArcInfo interchange (.e00) files from NIC.

Other discontinuities are noted in the Changes in Analysis Conventions or Procedures section of this document.

Errors Discovered After the October 2006 Publication of the Data Set

Users working with the chart data have found the following errors, as of 20 January 2007.

• There are errors in partial concentration fields. It is important to note that the corresponding total concentration fields do not have errors. Thus, while the total concentration fields are correct, the thin, FY, and MY fields do not sum to the total concentration values when errors in one or more of the partial concentration fields are present.
• There may also be some error in the MY fields, particularly during 1997 through 1998 where the values are higher than other years. This needs to be investigated further. Examples include MY ice appearing and disappearing suddenly along the coasts of Labrador and Newfoundland in 1995 (starting 21 March), in 1997 (starting 17 March), in 1999 (starting 31 May) and in 2002 (starting 18 March).
• Throughout 1997, thin ice appears and disappears in an unrealistic fashion. In Baffin Bay and Hudson Bay in 2001, thin ice disappears on 12 March, 19 March, and 2 April.
• On the chart of 1978_01_22, there is a large section in Hudson Bay and Baffin Bay with pixels coded with a value of 157. This value represents undigitized in SIGRID code. On the chart of 1979_05_29, in about the same area, there is a large area of 0 percent or 5 percent concentration, when the concentration should be much higher over most of the area. These errors, combined, resulted in a 1 to 3 mill. sq. km. underestimate of total extent (W. Meier, personal communication, 17 Nov 2006).

The above errors in partial concentrations are in the E00 source files and are not a result of processing. Two errors that apparently resulted from processing have been noted and further investigation is underway:

• The thin ice and total concentration files for 1999_12_20 are erroneous: thin ice is falsely present over the entire Baltic Sea and the Gulf of Bothnia, and total concentration is erroneous.
• The fast ice and total concentration files for 2000_04_24 are erroneous: Fast ice is falsely present over the entire Baltic Sea and the Gulf of Bothnia, and total concentration is erroneous.

See the poster titled Operational sea ice charts: An integrated data product suitable for observing long-term changes in Arctic sea ice? by Meier et al., available on the NSIDC Publications, Posters and Presentations Web Page, for more information on data set inconsistencies.

Sea ice charts are maps with areas containing ice of similar concentration and type outlined and labeled with a code. This code (known as the egg code, for its shape) has information on total sea ice concentration, usually given as a range (for example 7-9, meaning 7 to 9 tenths or 70 percent to 90 percent concentration). The egg code for a particular area may also have partial concentrations (that is, concentration by ice type), information on ice stage of development (often called ice type, usually expressed as ice age, and a proxy for ice thickness), and information on ice form as well. Fast ice, pancake ice, and brash ice are examples of ice form. As a rule, NIC only uses the fast ice form in its charts. World Meteorological Organization (WMO) designations are used for sea ice stage of development and ice form. See WMO (1970) and the NSIDC All About Sea Ice pages and Glossary for more information on ice types. The Canadian Ice Service (CIS) Manual of Standard Procedures for Observing and Reporting Ice Conditions, or MANICE (Environment Canada, 2005), has a good explanation of how ice information is encoded using the egg code. Figure 15 summarizes the information that can be encoded.

Ice services around the world implement the egg code in differing ways depending on what information the charts routinely capture. NIC charts, for example, have never contained C_d, S_e, F_d, or F_e information, but NIC often uses S_o and S_d. See MANICE (Environment Canada, 2005) for more information on these codes.

In the early 1980s, the international ice charting community discussed ways of encoding and digitally preserving information in hand-drawn ice charts. A format called Sea Ice Grid (SIGRID) (Thompson, 1981) was developed and adopted as a WMO format. Essentially, a sea ice chart encoded in SIGRID is a text file where the information in each egg code is a string of text. There is information for each grid point. Grid resolution is nominally 15 minutes in latitude and a variable amount in longitude.

One limitation of SIGRID is that ice charts are more like vector (point, line, polygon) data, with no inherent resolution, than gridded data. In converting chart information to a grid with fixed points, information is lost. Several ice services have adopted a vector archive format, SIGRID-3, that overcomes this limitation (Intergovernmental Oceanographic Commission, 2004). See GDSIDB Data Formats for an overview of SIGRID development and related references.

As NIC moved to producing ice charts in a GIS environment, NIC continued to label areas of ice having similar characteristics (these areas are called polygons in the GIS environment) using the egg code (Figures 16 and 17). However, the attributes for each polygon in the attribute table for each ice chart file use the SIGRID string that encodes that polygon's egg (Figure 17).

Originally, ice charts were digitized to archive them and to make them easier to use for research. Now, with charts produced in a digital environment, there is less of a distinction between operational or vector format and archive or research format. However, many NSIDC users continue to request charts in gridded rather than the operational vector format. A further distinction in the case of this data set is that the source charts received additional quality control by NIC before being added to the archive.

In 2005, NIC contracted with NSIDC to update the EWG Atlas climatology (which was based on an ice chart series that ended with 1994). The first goal of the project was to create 5-year, 10-year, and 33-year total ice concentration climatology products for the Arctic region using the weekly NIC ice chart data from 1972 to 2004 and provide climatology products to NIC that matched the appearance and functionality of the climatology products on the EWG Atlas. (One difference, however, is that this new climatology, unlike that on the EWG Atlas CD-ROM, does not incorporate any information from classified sources.) The second goal of the project was to make these climatology products, along with the entire series of weekly or biweekly ice charts, available in EASE-Grid. The equal area projection and gridded format of EASE-Grid data (Brodzik and Knowles, 2002), along with the existence of many other polar data sets in EASE-Grid, makes EASE-Grid data more useful for many researchers than are ice chart data in the GIS format distributed by NIC.

We plan to continue extending these climatologies through periodic updates in cooperation with NIC.