KUP DATES: 
i] Satellite Visible 
: 81 selected lakes 
105°W to 40°N, 
of climate change 
anced greenhouse 
| context, in which 
cloud cover, poor 
are ice from open 
erence of 3.2 days 
iation and reduced 
«e ice breakup can 
toring. 
  
  
  
  
ces. 
| in an explicitly 
sults are included 
the interpretation 
process. This paper is organized chronologically, 
beginning with a description of the equipment, image 
processing and storage, followed by sections on 
image analysis, classification schemes and data 
storage. We conclude with remarks on the potential 
for operational monitoring of lake ice, including the 
required time and resources, suitability of the GOES 
images, and the spatiotemporal coherence of ice-off 
dates as an aid to optimizing lake selection. 
2. METHODS 
2.1 Hardware, Software and Imagery 
We used a dedicated Silicon Graphics Indigo? 
Extreme® workstation at ERSC for all image 
preparation and analysis. The system is based on the 
MIPS R4000 microprocessor that runs at 100 MHZ 
internal clock speed and 50 MHZ external. It is 
configured with 64 MB RAM, a 1GB internal system 
disk, a 2GB external drive, an internal 4 mm digital 
audio SCSI tape drive, and an internal double speed 
CD-ROM SCSI drive. Image processing/GIS 
software utilized included ERDAS Imagine (8.10), 
ARC/INFO (6.1), and McIDAS-X (2.0), all running 
under IRIX (5.2). 
The ERSC hardware environment also 
includes numerous specialized peripherals such as 
tape drives, optical disks, digitizing tables, and film 
recorders. Through a link to the University of 
Wisconsin-Madison Campus Area Network, access is 
provided to Internet and Bitnet. Similarly, ERSC has 
a local network connection to the Space Science and 
Engineering Center (SSEC), which is in the same 
building as ERSC. SSEC maintains facilities for the 
reception of real-time AVHRR and GOES data and 
ingestion into the Man-Computer Interactive Data 
Access System (McIDAS). The SSEC holds the 
national GOES archive. ERSC's local network 
connection to SSEC enabled ready downloading of 
the archival GOES images used in this study. 
We interpreted 122 scenes from the 0.9-km 
resolution visible band (0.54-0.70 um) of the GOES- 
VISSR for each of the 15 years in the SSEC archives 
(1980-1994). Each year of data consisted of daily 
images acquired from March 1 through June 30 at 
approximately 19:00 Coordinated Universal Time 
(UTC), translating into 12-2 p.m. local time across 
the image. Spatial coverage included the U.S. upper 
Midwest as well as portions of Saskatchewan, 
Manitoba, and Ontario, Canada south of Hudson Bay 
(Figure 1, previous page). 
2.2 Preparation and Storage of Images 
Most images were received on 4-mm DAT 
tape in McIDAS file format. The McIDAS images 
were converted to the ERDAS 7.5 image format 
using a program written for this purpose by Randolph 
H. Wynne. Those images were then filtered using a 
fast Fourier transform (FFT) notch filtration program 
to eliminate horizontal banding, written by Frank L. 
Scarpace (for additional details on this program 
please refer to Wynne et al. 1995). 
The FFT (filtration program required 
geometrically-constrained image dimensions; we used 
1024 rows by 2048 columns. Since most of the pre- 
filtered images had sizes of approximately 1300 rows 
by 2400 columns, part of each image was lost in the 
filtering process. The filtered area of each image was 
therefore selected to capture the spatial extent of ice 
change — more southerly closer to March 1, more 
northerly toward June 30. Also, image column 
boundaries were selected to include the same set of 
81 lakes for each year. 
Some of the filtered images in ERDAS 7.5 
format were converted to the ERDAS Imagine format 
using the Import/Export option in Imagine. This was 
done primarily for the ease of image manipulation, 
such as contrast stretches, in Imagine. 
All of the GOES images were archived on 
duplicate sets of five 4-mm DAT tapes in both their 
unfiltered McIDAS and filtered ERDAS 7.5 or 
ERDAS Imagine file formats. 
The major limitation on image preparation 
and storage was available disk space. While most of 
the image processing programs could be run 
overnight in batch files, a continual struggle for 
available disk space put limitations on the number of 
images that could be stored simultaneously on disk 
and archived overnight with a batch file operation. 
2.3 Lake Selection 
Lake selection was governed by size, 
identifiability, availability of ground-derived ice-off 
dates (to check our interpretation), and proximity to 
other lakes. 
Lakes had to be large enough to identify on 
the 0.9-km resolution GOES-VISSR images, yet 
small enough to exhibit a discrete ice-off date. They 
also had to be large enough to be named in an atlas, 
and unique enough in shape and spatial context to be 
easily identifiable on the image. Ground-derived ice- 
off dates were available for very few of the lakes, so 
almost all of the lakes with these data were selected.