3.3 Methods
3.3.2 Datasets
MODIS Composite Images
The main dataset used in this chapter, hereafter referred to as “the MODIS composites” or simply “composites”, was a series of 159 contiguous cloud-free MODIS composite images, generated from cloud-free portions of∼125,000 individual MODIS granules, covering the time interval 2000,061 to 2008,365 (in “year,day of year” for- mat). Each MODIS composite represents a 20-day cloud-free mean image of the surface (infrared during winter, visible during summer), rather than a daily snap- shot. This length interval was chosen for several reasons. Firstly, 20 days has been used as a fast-ice criterion in the Arctic (Mahoney et al., 2005), being long enough to exclude transient fast-ice growth events associated with the passage of synop- tic scale weather systems and short enough to resolve genuine fast-ice growth and
breakout events. Additionally, MODIS composite images generated from 20 days’ imagery are generally of a high quality, with few gaps due to persistent cloud. To fa- cilitate interannual comparisons, the compositing period was set at exactly 20 days for the first 17 composites in each year, with the final composite image covering the remaining days in the year (25 for non-leap years, 26 for leap years).
The techniques presented have been developed in order to classify fast ice by identifying ice which is contiguous with the coast, rather than ice which is not moving. The use of mean-value compositing was preferable to maximum/minimum value compositing in this case due to the difficulties of the latter in polar regions, arising from the similarities in brightness temperature and albedo between snow and polar cloud. Mean-value compositing is tolerant, to some degree, of imperfect cloud masking. Furthermore, due to the motion of the pack ice over the 20 day compositing period, the pack ice becomes blurred with mean value compositing, while fast-ice features typically remain more clearly defined (unless growth or breakout occurs), whereas this would not be the case for minimum/maximum value compositing. This enhances the utility of this technique for fast-ice detection. The details of the compositing procedure are given in Chapter 2, Section 2.3 (and also presented in
Fraser et al. (2009)).
Only 1 km resolution MODIS images were used in this chapter, despite the availability of higher-resolution 250 m imagery in some visible channels. The reason for this was twofold: i) in winter, few visible images are acquired at high latitudes, and no 250 m resolution infrared channels exist on MODIS; and ii) the CPU time required to produce these composite images at 250 m resolution is prohibitive (a factor of 16 higher in both CPU time and disk space requirements). Because the MODIS 1 km channel resolution is only 1 km at nadir (with pixels at the edges of the swath growing to ≃10 km2
a longer path to the swath edges, see Section 2.3.3), the output resolution of the composite images was set at 2 km.
Both the Terra and Aqua satellites provide one MODIS granule every 5 min- utes (288 per day). This equates to∼16,000 granules per platform per year over the region of interest i.e., over 32,000/year from both platforms, or a total of ∼250,000 granules for the entire time series (2000-2008). The average file size for a MODIS L1B 1 km-resolution granule is ≃100 MB (higher in summer when visible images are captured, and less during winter). Thus, the total size of the MODIS archive required to produce the 159 composites is on the order of 25 TB. This presents a challenge for both storage and processing of the data. One practical solution (adopted here) is to pre-select only the least cloudy MODIS granules for inclusion into the composites. This solution is possible because the MOD35 cloud mask prod- uct is a much smaller download than the equivalent L1B granule - averaging about 3MB. Thus, for this project, all ∼250,000 MOD35 granules were downloaded, and the cloud-content of each granule evaluated. These data were acquired from the Level 1 and Atmosphere Archive and Distribution Center (LAADS), available at http://ladsweb.nascom.nasa.gov.
To illustrate the cloud content of MODIS scenes along the East Antarctic coast, Figure 3.1 shows a histogram of cloud-free percentage, including all MODIS swaths covering the study region for the year 2007. The strong peak at a modal value of ≃7% cloud-free pixels confirms this region as one of the cloudiest on Earth (Spinhirne et al., 2004, 2005). A cutoff at 30% cloud-free pixels roughly bisects the histogram. After some experimentation, it was determined that using only those granules with 30% or more pixels classified as cloud-free produced composite images similar in quality to composite images produced using all granules, but requiring half the storage space and CPU processing time.
Figure 3.1: A histogram of “cloud-free percentage” (the percentage of pixels within a granule classified by the MOD35 product as “confident clear” or “probably clear”) for 2007. This histogram is representative of other years within the study period. In order to reduce storage and CPU requirements, only granules with≥30% cloud-free pixels were used in the compositing process. Resulting composites were similar in quality to those which used all available granules.
AMSR-E Sea-Ice Concentration Composite Images
Coincident sea-ice concentration data from the AMSR-E (Advanced Mi- crowave Scanning Radiometer - EOS) instrument onboard the NASA Aqua platform were occasionally used to supplement the MODIS composites during times when the MODIS composite image (and the previous/next MODIS composite images) were of poor quality. In fact, 30 out of 159 fast-ice maps (or ∼19%) required augmentation using AMSR-E imagery in this way. Daily AMSR-E ARTIST Sea Ice (ASI) 6.25 km resolution daily sea-ice concentration data were obtained from http://www.iup.uni- bremen.de:8084/amsr/ for this purpose (Spreen et al., 2008). AMSR-E composite images were then generated for the same time intervals as the MODIS composite images, and using the same projection.
Intercomparisons between MODIS and AMSR-E composite images show generally good agreement in the location of the fast-ice edge (see Figure 3.2), al- though the use of the lower resolution AMSR-E data leads to ambiguities in fast-ice detection at times. Additionally, the AMSR-E composite images erroneously show lower sea-ice concentration at the locations of large tabular icebergs and ice tongues (a phenomenon which arises due to the intrinsically different radiometric properties of sea ice and icebergs, noted by Kern et al. (2007)). This proved useful in the exclusion of large icebergs from the fast-ice maps in highly dynamic locations (e.g., east of the Mertz Glacier Tongue), though identification of the location of sub-pixel scale icebergs remains a problem.
Fusion of AMSR-E and MODIS data has been reported several times in the literature, e.g., using AMSR-E data to reduce the effects of cloud obscuration in the MODIS snow cover product at mid-latitudes (Gao et al., 2010), and using a combination of MODIS and AMSR-E products to retrieve Ice Water Path in clouds
over ocean (Huang et al., 2006). This is, however, the first effort to combine the two different satellite sensor datasets for fast-ice detection.
Despite the reasonable agreement between AMSR-E and MODIS composites on the location of the fast-ice edge noted above, the ASI algorithm appears to some- times underestimate the concentration of fast ice. Examples of this “behaviour” are shown in Figure 3.2 - particularly the area surrounding the large grounded iceberg D15. In this region, the ASI-reported fast-ice concentration is as low as∼55%. The 89-GHz channel used by the ASI algorithm to obtain high-resolution sea-ice concen- tration also has the potential to be adversely affected by cloud and water vapour (Ulaby et al., 1981). The ASI algorithm (Spreen et al., 2008) uses lower-frequency AMSR-E channels as weather filters to remove spurious sea-ice concentration in open water areas, where the contaminating effect of weather systems is largest. The algorithm is expected to be at its most accurate for retrieving high sea-ice concen- trations, e.g., in the fast-ice zone (Spreen et al., 2008).