Algorithm for validation of sea ice concentration

Validation vs. ice chart data

Full document (pdf file)


We choose not to construct intermediate level products on a refined resolution; i.e. we integrate and average, rather than interpolate and extrapolate.

Starting point

A sea ice concentration (SIC) analysis is available at a very high spatial resolution as an ice chart, with this discrete set of values:

Class Values Mapping
Open water <0.1 0.05
Very open drift ice 0.1, 0.2, 0.3 0.25
Open drift ice 0.4, 0.5, 0.6 0.55
Close drift ice 0.7, 0.8 0.8
Very close drift ice 0.9, 1.0 0.95
Fast ice 1.0 1.0

Model results are available at an intermediate spatial resolution, using a continuous set of values in the range 0.0-1.0; Fast ice is not modeled. The model assimilates SIC, but presently, the data being assimilated are independent from the data used in the ice chart analysis.


SIC ice chart data are available with a resolution that is order of 10x finer than the model product, so SIC from the ice chart analysis is mapped onto the model product grid, and averaged over each model grid cell.

Both of these coarse resolution product are categorized into the discrete set of classes used in the ice chart analysis, but the Fast ice category is discarded, and Fast ice values in the ice chart analysis are re-set to the class 0.9-1.0/Very close drift ice.

Validation products

  • Time series for SIC root-mean-square (RMS)
  • Time series for ice edge length, position bias and position RMS
  • Time series for areas of classes Open Water, Very open drift ice, Open drift ice, and Close drift ice
  • Contingency tables for observation vs. model ice class match-ups
  • Maps depicting geographical extent of sea ice classes

Validation vs. OSI SAF sea ice concentration data

Similarities and contrasts to ice chart data

    The main differences between the OSI SAF data and the ice chart data are
  1. Product resolution OSI SAF: 10km; ice charts: 1km
  2. Domain OSI SAF: entire Arctic; ice charts: European Arctic
  3. Data source OSI SAF: Microwave data (SSMIS); ice charts: multiple, primary source is radar data (SAR)
  4. Dependency OSI SAF data are assimillated in hte TOPAZ analysis; ice chart data are not assimillated

In addition to the fact that the two observational products represent the same physical variable, a similarity is also the update frequency which is daily for both products (albeit presently only on working days for ice charts).


Condensed statistics in the form of confusion matrices for sea ice categories has proven to be very useful for the case of validation vs. ice chart data. Here, we adopt the same approach by introducing an analogous approach. The categorization is defined in the table below.

Class Values Mapping
Ice free <0.01 0
Open water 0.01-0.1 0.05
Very open drift ice 0.1-0.4 0.25
Open drift ice 0.4-0.7 0.55
Close drift ice 0.7-0.9 0.8
Very close drift ice >0.9 0.95

Validation products

The same products as listed for validation vs. ice charts above, with the exception of time series for selected ice classes, and maps. NOTE! Results for sea ice edge position bias and position RMS are only computed provided that the lengths of the two representation of the sea ice edges that are investigated, are both at least 10 grid cells.

See Met Privacy