Bootstrap Sea Ice Concentrations Version 1 Processing Steps
The choice of tie-points and the justification for using them have been discussed in various publications (Comiso et al. 1997; Comiso 1995; Comiso and Sullivan 1986; Comiso 1984). The tie-points for ice and open water are the set of brightness temperature values that correspond to ice concentrations of 100% and 0%, respectively. With the Bootstrap Algorithm, the tie-point for ice is a point along a line AD, while the tie-point for open ocean clusters around a single data point (refer to figure 3 below). In previous retrievals, the line AD was fixed during fall and winter and allowed to fluctuate in spring and summer to account for large changes in emissivity during this period. In this data set, adjustments were made on a day to day basis by applying the results of a linear regression on the consolidated ice data points, that is all points above the line AD - 5K. The estimates are generally compatible with the previous ones and differences are typically small, but the technique allows for needed change when emissivities and temperatures fluctuate substantially due to adverse weather conditions. This was not done previously because of added requirements in processing time, but the advent of powerful computers has made the procedure practical.
Because the TB values vary from one sensor to another, tie-points were checked for compatibility in sea ice concentration during periods of sensor overlap. When applying the original tie-points (Comiso 1995), the frequency distribution of SMMR (in black) and SSM/I (in red) sea ice concentrations (Figure 1a) for the same day during overlap period are shown to be significantly different. In highly consolidated ice areas, the amplitude and standard deviation of the SSM/I peak is considerably different from the SMMR peak. When combining data from the two sensors, this discrepancy can cause problems in assessing long term changes in sea ice cover. However, the peaks can be made almost identical by slightly adjusting the sea ice tie-points. The use of the linear regression technique for both SMMR and SSM/I turned out to be appropriate for this purpose since when applied, the results show much better agreement (Figure 1b). This technique also improves the consistency in the estimates of sea ice extent and sea ice area from SMMR (solid line) and SSM/I (dashed line) during the period of overlap in 1987 (Figures 2a and 2b). This makes it unnecessary to apply a normalization on one set of data to make the two data sets consistent during periods of overlap. The regression technique provides a new slope and offset daily for the line AD, but the slope and offsets were found to change little during winter except during adverse weather conditions. These daily calculations improve data consistency not just between the SMMR and SSM/I sensors but also between the different SSM/I sensors (for example, DMSP-F8, -F11, and -F13). The use of this technique thus provides a better representation of sea ice cover and was adapted during the sea ice concentration reprocessing.
Figure 1. Histogram of SMMR and SSM/I concentrations in the Antarctic on 25 July 1987 (a, top) before tie-point adjustment, and (b, bottom) after tie-point adjustment.
Figure 2. Total Antarctic sea ice area during period of SMMR-SSM/I overlap (a, top) before tie-point adjustment, and (b, bottom) after tie-point adjustment.
Figure 3. Scatterplots of vertically polarized 18 GHz and 37 GHz SMMR brightness temperatures for (a, left) entire Antarctic, (b, middle) Weddell, Indian, and Pacific sectors, and (c, right) Ross and Bellingshausen/Amundsen sectors. Line AD represents the ice tie-point while O represents open water. Ice between 0 and 100% lies on a line between O and AD.
The signature of the Antarctic sea ice cover has been considered to be similar to that of a seasonal ice cover even in the Weddell Sea where a large fraction of the sea ice cover survives the summer, the signature is similar to other regions (Zwally et al. 1983). However, although the sea ice that survives the summer in the Bellingshausen/Amundsen Seas region appears to have a different signature in the subsequent winter, the sea ice algorithms, including the original Bootstrap Algorithm, did not take this into account. Accurate characterization of the sea ice cover in the region is desirable, especially since the region has been identified recently as a climatologically anomalous area. In Figure 3a, a scatter plot of 19V versus 37V data points from the entire Antarctic region on 17 June 1979 is shown. The tie-point for ice is labeled AD. The same set of points are shown in Figures 3b and 3c; but the data points in the Weddell, Indian, and West Pacific sectors were plotted separately from those in the Bellingshausen, Amundsen, and Ross Sea sectors, as indicated. It is apparent that the slope of the data points for consolidated ice (that is along AD) in one region is different from that of the other region. The use of one set of tie-points, as in Figure 3a, for the entire Antarctic region yielded the color coded ice concentration map shown in Figure 4a; while the use of a separate set of tie-points for the two regions, as indicated in Figures 3b and 3c, yielded the map in Figure 4b.
Figure 4. Antarctic sea ice concentration field for 17 June 1979 (a, top) before, and (b, bottom) after regional tie-point adjustments.
The ice concentration maps are basically the same in the two regions except towards the west of the Antarctic Peninsula where a large area with 70% ice concentration (brown) or less is apparent in the original data set (Figure 4a) but is shown to be much more consolidated and more consistent with observations from other sensors (for example, OLS and AVHRR) in the new one (Figure 4b). The historical SMMR data indeed shows that in the summer of 1979, highly concentrated ice survived in the same area identified as a relatively low concentration area (brown) in Figure 4a. This indicates that there was a dominance of multiyear ice in the region.
Data were interpolated in areas with missing pixels according to the following steps. First, data were spatially interpolated only for isolated empty pixels. An empty pixel was replaced by the average of four good surrounding pixels, or if four good pixels were not available, then a smaller number of pixels was selected. Second, a time interpolation was applied to the spatially interpolated map. Time interpolation was based on a weighting scheme, that is, the closer the good data were in time, the higher the weighted value. For each empty pixel, the algorithm searched forward in time for a good pixel, and backward in time for a second good pixel. The algorithm determined how many days ahead and behind the two good pixels occurred, and calculated the weight of each pixel. A weighted average of the two good pixels was then calculated, and the result was used for the empty pixel. During the SSM/I period, most temporal interpolations were conducted using only adjacent days. During the SMMR period, particularly in 1986 when larger gaps were present, temporal interpolations had separations of more days.
Comiso, J. C., D. Cavalieri, C. Parkinson, and P. Gloersen. 1997. Passive microwave algorithms for sea ice concentrations: A comparison of two techniques. Remote Sensing of the Environment 60(3):357-84.
Comiso, J. C. 1995. SSM/I concentrations using the Bootstrap Algorithm. NASA Reference Publication 1380. 40 pages.
Comiso, J. C. and C. W. Sullivan. 1986. Satellite microwave and in-Situ observations of the Weddell sea ice cover and its marginal ice zone. Journal of Geophysical Research 91(C8):9,663-81.
Comiso, J. C. 1984. Characteristics of winter sea ice from satellite multispectral microwave observations. Journal of Geophysical Research 91(C1):975-94.
Zwally, H. J., J. C. Comiso, C. L. Parkinson, W. J. Campbell, F. D. Carsey, and P. Gloersen. 1983. Antarctic sea ice 1973-1976 from satellite passive microwave observations. NASA Special Publication 459.