There is also inconsistency in ice area (Figure 1, right), but this occurs during where the chart areas are mostly lower than PM; before and after this period, there is closer agreement with the chart areas generally slightly higher than PM. The inconsistencies are most pronounced in the September extent timeseries (Figure 2), where the NIC trend is less extreme than in the passive microwave data and there is greater difference between the trends after Area trends for September show similar relationships. Operational Sea Ice Charts: An Integrated Data Product Suitable for Observing Long-term Changes in Arctic Sea Ice? W. N. Meier and F. Fetterer (Nat’l Snow and Ice Data Center), C. Fowler (Univ. of Colorado), P. Clemente-Colón and T. Street (U.S. Nat’l Ice Introduction The location and character of sea ice has been charted in various locations of the arctic at least since the early 20 th century. Standardization of charting procedures and the regular production of charts began in the early 1950s with the advent of national ice centers and involvement of the World Meteorological Organization. With the development of satellite sensors in the early 1970s it became possible to produce complete and accurate analyses of the entire Arctic. Satellite technology has improved, culminating with Radarsat-1 since 1995, which provides high resolution information on not only ice concentration and extent, but also ice type. The ice charts provide an estimate of the arctic ice cover that is complementary to passive microwave (PM) sources. The value in the charts is that they extend further back in time and they use a a variety of sources (satellite [including passive microwave] and other) to produce the best possible estimate, thus avoiding some of the biases and errors inherent in the passive microwave estimates. A limitation of the charts is that they are not based on consistent sources and the quality of the charts is dependent of the quality of information used to produce them. Here, we compare ice charts produced during the satellite era by the U.S. National Ice Center (NIC) in Suitland, MD with two passive microwave derived estimates, one from the NASA Team algorithm and on from the Bootstrap algorithm. We investigate (1) the consistency of the NIC charts by comparing with the consistent passive microwave timeseries, and (2) biases and errors in the passive microwave data by comparing with the more accurate NIC charts. Methodology The NIC chart data employed here is a newly released data set, “National Ice Center Arctic Sea Ice Charts and Climatologies in Gridded Format”, available online at NSIDC ( The data set was produced from the NIC charts by converting the original GIS vector data into a gridded binary format on the 25 km EASE projection. The charts are weekly spanning (bi-weekly after 11 June 2001). For only total ice concentration (including a landfast ice flag) are available. Since 1995, partial ice concentrations of multiyear, first-year, and thin ice are available. Along with the weekly chart data, climatologies of maximum, median, minimum, and quartiles were produced. The comparison passive microwave estimates from NASA Team (NT) and Bootstrap (BT) algorithms are also archived at NSIDC and available online ( The fields were produced at NASA Goddard and contain substantial quality control to remove erroneous ice as well as to assure consistency throughout the timeseries. Daily and monthly data spanning are currently available in a 25 km polar stereographic grid. For comparison with the NIC charts, the daily data corresponding to the day of each ice chart were obtained and regridded to the same EASE grid for direct comparison with the charts. Due to small differences in pixels denoted as land in the two sources, a combined land mask was used for both data sets. Also, for area comparisons, the region near the pole not observed by the satellite sensors was masked out in both the passive microwave and the chart estimates. For extent comparisons, the unobserved region was assumed be ice- covered. Weekly Standardized Arctic Sea Ice Anomalies, (52-week running mean filter) Year # St. Dev. from Mean Sea Ice ExtentSea Ice Area Year Figure 1. Standardized anomalies are computed by subtracting the mean ice extent or area by each weekly value and dividing by the standard deviation. Means and standard deviations were computed for thirds of a month and were computed for each individual data source based on a reference period Total Ice Comparison The first objective of this project is to investigate the consistency of the NIC charts. We have confidence that the PM timeseries are consistent because they rely on the same type of sensor, at nearly the same frequencies over the entire period. Careful intersensor calibration has been done to minimize any bias introduced due to different frequencies and different satellite orbits. Thus marked changes in the NIC ice chart estimates relative to the passive microwave estimates indicate a discontinuity in the chart data. An example can be seen in the total extent in Figure 1 (left) when the NIC extent became much higher than the passive microwave extents beginning in Before 1995, the estimates agree reasonably well. But two things occurred beginning around First, Radarsat-1 SAR data began being used around this time, which allowed better detection of ice near the ice edge, in new ice regions, and in melt regions. Second, NIC started using digital methods to create the analyses, instead of using paper charts, yielding better quality control and greater consistency. These two changes made the NIC charts more accurate, but at the expense of consistency with the earlier part of the record. A third factor is the conversion of the NIC charts to EASE-Grid took place in two distinct ways, leading to two distinct data records for , and (see the documentation for the NIC series for more information). The resulting discontinuity limits the accuracy with which long term trends can be tracked using NIC data alone. Year Extent (10 6 km 2 ) September Sea Ice Extent Anomalies and Trends NIC NT BT Trend YrsNICNTBT Trend in % decade -1 Figure 2. September extent anomaly for NIC charts and passive microwave. Trend lines are included for (NIC only) and The inset table indicates the trend values for various time periods in % per decade. Note how the NIC trend agrees more closely with the passive microwave data before 1995 (trend lines not shown for and ) Total and Partial Ice Comparison Here we use the NIC charts to evaluate biases and errors in the passive microwave data. We employ the post-1994 period because (1) it avoids the discontinuity in the early part of the NIC record, (2) it relies mainly on Radarsat-1 data and thus is reasonably independent of the PM record, and (3) partial concentrations are available that can help illuminate the differences between the chart estimates and the passive microwave data. An error in the charts was discovered in that the sum of the partial concentrations was found to be less than the total concentration. This “missing” ice was found to be primarily first-year ice and thus was added to the FY ice category so that the sum of the partial concentrations equals the total concentration in each chart. This change is reflected in the images and timeseries below (Figures 3-5). NIC NT BT NIC Sea Ice Multiyear (MY), First-year (FY) and Thin Ice Areas and PM-NIC Total Area Difference Area (10 6 km 2 ) Year Figure 4. Sea ice partial concentrations and the passive microwave minus chart total concentrations for The dates of the ice charts shown above are denote by thin black lines. The seasonal cycle in FY, MY, and Thin ice is well represented in the chart data, both in the basin charts (Figure 3) and in the timeseries of the seasonal cycle (Figure 4). The passive microwave estimates are almost always biased low. As freeze-up begins (4 Oct) the BT estimates are low near the ice edge, while the NT estimates are low throughout. After freeze-up has begun (11 Nov) and through winter maximum, there tends to be overestimation within the pack (because the charts denote 95% ice throughout most of the arctic, except for 100% in fast ice regions) and underestimation near the ice edge. With summer melt (24 Jul), the NT has a large low bias throughout; BT also has a significant low bias, but it is less than NT and occurs mostly near the ice edge. NT-TOTALBT-TOTALTOTALMULTIYEARFIRST-YEARTHIN 4 OCT NOV FEB JUL % % 4 Oct 8 Nov 28 Feb 24 Jul Correlation Coefficient Year Correlation of PM vs. NIC Total and Partial Conc. Figure 3. Sea ice partial concentrations and the passive microwave minus chart total concentrations for four days over the season. 4 Oct is near the minimum extent and just after FYI, by convention, is set to zero on 1 Oct. 11 Nov is at the point of max. thin ice extent. 28 Feb is the maximum total ice extent. 24 Jul is during the peak melt season when the PM underestimation is largest. Though there is substantial interannual variability, the season shown above is representative of the other years (Figure 4), with the maximum PM underestimation occurring between late June and early August and the minimum underestimation at the onset of freeze-up. Meltponds and surface melt likely cause the underestimation during summer. When the surface of the existing ice refreezes but before substantial new ice has formed, the PM algorithms have the smallest bias. Then as new ice forms and thickens into first-year ice, the bias increases. The spatial correlation between PM and the charts is also illuminating, with a clear seasonal signal (Figure 5). Figure 5. Spatial correlation of PM concentrations with NIC chart partial concentrations. Correlation with FYI and Total (top) is highest in winter and lowest in summer. Correlation with MY and Thin (bottom) is highest during summer and lowest in winter. BT correlations are consistently higher than NT. NT vs. Total BT vs. Total NT vs. FYI BT vs. FYI NT-Total BT-Total MY Ice FY Ice Thin Ice Acknowledgment: This work was supported through a contract with the U.S. National Ice Center (N P-0169), through funding for the NASA Polar Distributed Active Archive Center, and through support from the NOAA National Geophysical Data Center. Thanks to Jeff Smith (NSIDC) for conversion of PM data to EASE-Grid. 4 Oct 8 Nov 28 Feb 24 Jul NT vs. MYI BT vs. MYI NT vs. Thin BT vs. Thin NT consistently underestimates ice area more than BT. Both PM biases appear to be consistent from year to year, making both suitable for tracking long term trends; the NIC charts contain discontinuities, limiting their usefulness for trend detection. However, the NIC ice charts provide an assessment of the PM errors and are useful for obtaining more accurate measurements at a given point and time, and by providing estimates of partial concentration. Conclusion