Sea Ice Modeling: A Mini-Workshop
4. Rapporteur Summaries
Session 5. Global Sea Ice Data Sets
(Rapporteur: Claire Parkinson)
This session focused on century-long sea ice data sets from a variety of sources and 18-20-year sea ice data sets from satellite data. Amongst the highlighted concerns were the huge scarcity of data for the pre-satellite period, especially in the southern hemisphere, and the difficulties of merging data sets from different sources (including different satellite instruments, even when of the same type). The session began with talks by John Walsh of the University of Illinois and Nick Rayner of the Hadley Centre for Climate Prediction and Research on century-long sea ice data sets. These were followed by a talk by Tom Smith and Richard Reynolds of the National Weather Service on sea ice and sea surface temperature (SST) from satellites, a talk by Claire Parkinson and Don Cavalieri of Goddard Space Flight Center on sea ice concentrations and extents from satellites, and two talks on sea ice motion from satellites, by Ron Kwok and Tim Liu of the Jet Propulsion Laboratory and by Tony Liu of Goddard Space Flight Center. The session concluded with a general discussion on the various data sets.
|Table 2. Significant sources of sea ice data for 20th century: Northern Hemisphere|
|1978 onward||SMMR/SSMI (passive microwave) Ð hemispheric|
|1972 onward||US National Ice Center — hemispheric|
|1890s-1950s||Danish ice charts — North Atlantic (summer)|
|1960 onward||UKMO charts — Arctic /Atlantic|
|1930 onward||Russian ice charts -- Eurasian Arctic|
|1900 onward||Norwegian ice charts -- eastern No. Atlantic|
|1950-1970s||US Navy -- Alaskan region|
|1950s onward||Canadian regional charts|
|(1950s)-1970s||Icelandic regional charts|
|1970s onward||Japanese charts -- Sea of Okhotsk|
|1800s onward||Finland -- Baltic Sea ice charts|
John Walsh listed significant sea ice data sources for the twentieth century (Table 2), mentioning in particular newly available Russian ice charts for the Eurasian Arctic for 1930 onward and some Norwegian ice charts for the northern North Atlantic going back to the 1500s. He mentioned on-going attempts to create a merged data set, including the Global sea Ice and Sea Surface Temperature (GISST) effort, and showed time series of ice-extent seasonal values for 1900-1997. The time series show an overall downward trend in the summer ice cover starting in the middle of the century. Periods of missing data were filled in on the time series with climatological values, resulting in far less variability revealed in the first half of the century than in the second half, when many more data points were available. The use of climatological values also resulted in a strong anomaly for the World War II years, as the climatological values inserted for those years are well below the ice amounts plotted for the surrounding years. Walsh identified the following four key problem areas for these long-term data sets: (1) The heterogeneity of the different data sets. (2) The existence of regions and time periods without data, especially in the Southern Hemisphere ice region prior to the 1970s. Southern Ocean ice extents inferred from whale-catch records (de la Mare, 1997) are questionable but might be the best estimates we have for the early part of the century. (3) The lack of ice area data for the first half of the century, exceeding even the lack of ice extent data. (4) The need to quantify the uncertainties in the derived sea ice trends.
Nick Rayner discussed the sea ice data sets in GISST, used at the Hadley Centre for model forcing, model validation, and variability studies. The GISST data set incorporates a variety of sea ice data sources. In the Arctic, these center on compilations by John Walsh for 1901-1990 and on satellite data for the 1990s. In the Antarctic, they center on a mixture of German and Russian values for the pre-satellite era, then on satellite sources. The difficulties for the Antarctic in the pre-satellite era are typified by the use of only two Russian maps, together giving one climatological ice-extent contour per month, for the entire period 1947-1962. Recognizing the inhomogeneities in the data sources, a working group has been formed to produce a more homogenized data set, with more consistent ice concentration indications. For instance, for the satellite era, the wintertime ice charts of the National Ice Center are being adjusted to match more closely the passive-microwave ice concentrations, and the summertime apparent "meltpond bias" in the passive-microwave data is being adjusted for through comparisons with the National Ice Center charts. The preliminary homogenized blend in the Southern Hemisphere now yields no trend in sea ice coverage, in contrast to the earlier downward trend suggested before the homogenization. Rayner noted that the choices made in the homogenization effort are not necessarily improving the absolute values of the various data sets, but instead are centered on making them more consistent amongst themselves. The importance of this point was clear in the discussion, as the specific passive-microwave ice concentrations used in the Antarctic have lower values than other ice concentrations derived from the same data using revised algorithms.
Tom Smith explained that the National Weather Service has traditionally emphasized latitudes between 60 N and 60 S, constituting by far the bulk of the global oceans, in their sea surface temperature (SST) analyses based on Advanced Very High Resolution Radiometer (AVHRR) data. Recently, however, they have directed increased attention to the higher latitudes and consequently have become more concerned about the impact of sea ice. Earlier, they uniformly set the SSTs to -1.8 C wherever ice concentrations exceeded 0.5. Now they are attempting greater sophistication, based on a quadratic fit of SST versus ice concentration, resulting in many temperatures being raised above -1.8 C that had formerly been set exactly at -1.8 C. Smith explained that further improvements are expected in the future and that there is a major need for validating the satellite data with ground observations. He also explained differences between the ice concentrations of the National Ice Center and those of Nomura/Grumbine and the difficulty, in view of the lack of validation data, of knowing which values are closer to being correct. Problems appear to be particularly large in summer, highlighting the need to get an improved handle on meltponding and on understanding how the meltponds impact the satellite sea ice signatures.
Claire Parkinson discussed efforts to obtain consistent data sets of sea ice concentrations and extents over the period since 1978, using satellite passive-microwave data. She illustrated how well the passive-microwave data can reveal the large-scale location of the ice edge but discussed also the greater uncertainties in deriving ice concentrations and the difficulties arising from such phenomena as "land contamination," produced because data for coastal ocean areas are contaminated by radiation from the adjacent land, resulting in a miscalculation of ice along the coasts. The problems with land contamination become especially apparent in time series spanning different satellite instruments and hence must be adjusted for in order to create consistent, merged data sets. Weather filters, tie points, and instrument drifts also must be considered on an instrument-by-instrument basis when trying to create long-term merged data sets. Such a merged data set from four satellite passive-microwave instruments spanning the period from October 1978 through the end of 1996 has recently been generated and is available through the National Snow and Ice Data Center. This data set reveals an overall negative trend in ice extents in the Northern Hemisphere and a lesser positive trend in the Southern Hemisphere. The Northern Hemisphere's negative trend proved robust to changes in the definition of ice extent, from including all pixels with ice concentrations of at least 15% to including all with ice concentrations of at least 30%. Passive-microwave data also exist for much of the period 1973-1976, but the lack of data overlap between the 1976 data and the October 1978 data has hindered the generation of a merged data set extending back prior to October 1978.
Ron Kwok discussed sea ice motions from various satellite data sets. Passive-microwave data can provide ice-motion fields in wintertime with displacement errors of about 10.5 km and sampling periods of 1-2 days (1 day for 85 GHz data; 2 days for 37 GHz data). These data are less useable in summer because of problems produced by wet surfaces and meltponding. AVHRR data can be used to derive ice motions in all seasons, although only under non-cloudy conditions. The AVHRR data provide 1-day sampling and 1-4 km displacement errors. Synthetic Aperture Radar data (from ERS 1 and 2 and Radarsat) can be used to derive ice motions with displacement errors of only 0.3 km, comparable to the displacement errors from the buoys of the International Arctic Buoy Program, and 3-day sampling (versus sampling on the order of hours for buoys). Recognizing the differing limitations of the various data sets, work is underway to generate blended ice-motion data sets for both hemispheres, from all the various sources for the period from 1978 to the present. These data sets are likely to be available in about September 1999. Kwok also mentioned the determination from satellite data of dates of freeze-up and melt, plus innovative attempts to combine radar data and modeling to infer ice age and ice thickness from areal distribution changes.
Tony Liu discussed sea ice motion determinations from wavelet analysis of satellite scatterometer and passive-microwave data, taking advantage of the ability of wavelets to track features in the ice cover. The first attempt to use wavelet analysis to obtain large-scale ice motions from satellite data used measurements of the passive-microwave Special Sensor Microwave Imager (SSMI). This attempt successfully revealed both the Beaufort Sea gyre and the Transpolar Drift Stream. However, the likelihood of atmospheric effects in the 85 GHz passive-microwave data led to the use of an alternative data set, specifically the NASA Scatterometer (NSCAT) measurements. Comparisons between SSMI-derived ice speeds and buoy-derived ice speeds for six months of daily data, October 1992 through March 1993, yielded an RMS of 2.6; similar comparisons for two months of NSCAT and buoy data, November-December 1996, yielded an RMS of 2.8. The NSCAT, SSMI, and buoy data are now being combined in a merged data product felt to be improved over each of the individual data sets. From the ice velocities, streamlines are being constructed, as are shear and divergence maps. Results are being compared with model simulations of Sirpa Hakkinen.
The talks were followed by a lively discussion, with many important points introduced or reemphasized. Jim Hansen began the discussion period by noting, from the perspective of climate modeling, the huge problems caused by data discontinuities such as those shown by Walsh where the World War II sea ice data gap was filled in with climatological data that are clearly out of line with the ice data before and after the war, and by unrealistic long-term trends such as appear to exist in the GISST3 record. Kim Partington mentioned that in comparing the National Ice Center ice charts with ice concentrations derived from satellite passive-microwave data, he has found that in some regions the passive-microwave data appear to underestimate ice concentrations by as much as 50%, perhaps because of interpreting meltponds as open water. Don Cavalieri agreed that the passive-microwave data have errors introduced by meltponds but expressed greater concern with the National Ice Center's policy of sometimes inserting assumed ice values in preference to calculated values with high error bars. Peter Stone questioned Parkinson about the satellite-derived ice extent trends, and she reiterated the importance of having done the trend calculations with different definitions of the ice edge, showing that the trend values at least are robust with respect to whether the ice extent is defined based on a 15%, 20%, or 30% ice-concentration cutoff. David Rind expressed the importance of ice thickness, currently not obtained from satellite measurements, and Prasad Gogineni, agreeing, indicated that eventually a space-based technique for ice-thickness measurements should be developed. Regarding validation of ice concentration and extent measurements, Parkinson noted that sometime within the next year, the Earth Observing System (EOS) program aims to put out a NASA Research Announcement (NRA) aimed toward validation studies for many EOS measurements. This NRA will give scientists in the sea ice community an opportunity to undertake studies to validate, for instance, sea ice coverages obtained from the upcoming Moderate Resolution Imaging Spectroradiometer (MODIS) and Advanced Microwave Scanning Radiometer (AMSR) and sea ice motions obtained from AMSR and other EOS instruments.
To conclude: (1) Sea ice data in the pre-satellite era are very sparse in the Northern Hemisphere and even more sparse in the Southern Hemisphere, but important attempts are being made to accumulate whatever data exist both in the pre-satellite and satellite eras and to create from them a homogenized sea ice data set for the twentieth century. (2) Satellite data are allowing the creation of detailed sea ice concentration, extent, and motion data sets since 1978, although the data sets are in need of much more ground-truth validation, followed by appropriate adjustments in the satellite algorithms. (3) A critical variable not yet obtained by satellite measurements is sea ice thickness.