HR: 0800h
AN: B51A-0049 [Abstracts]
TI: Impacts of Pixel Deformation and Misregistration on Cross-calibration of Vegetation Index Data Records
AU: * Miura, T
EM: tomoakim@hawaii.edu
AF: University of Hawaii at Manoa, 1910 East-West Rd., Sherman 101, Honolulu, HI 96822,
United States
AU: Suzuki, T
EM: tomokosu@hawaii.edu
AF: University of Hawaii at Manoa, 1910 East-West Rd., Sherman 101, Honolulu, HI 96822,
United States
AU: Yoshioka, H
EM: yoshioka@ist.aichi-pu.ac.jp
AF: Aichi Prefectural University, 1522-3 Kumabari, Nagakute, 480-1198, Japan
AB:
The development of a long-term, seamless vegetation index (VI) data record requires data assemblage from
multiple sensors and their cross-calibration. The latter could be performed by directly comparing data from an
overlapping period of observations on a per-pixel or per-window (e.g., 5-by-5 window) basis. Due to differences
in orbital, scanning, geolocation accuracy characteristics of sensors, however, pairs of observations to be cross-
compared had different footprint sizes/scales and were acquired at slightly different locations, resulting in
different coverage of surface areas. In this study, we characterized the effects of these footprint deformation and
misregistration on cross-sensor VI comparisons. The objectives were to establish error bounds in cross-
calibration results due to these effects and to develop recommendation for reducing the impacts of pixel
deformation and misregistration on cross-calibration. Orbital, scanning, and geolocation error characteristics of
three satellite sensors, Terra Moderate Resolution Imaging Spectroradiometer (MODIS), NOAA-14 Advanced Very
High Resolution Radiometer (AVHRR), and SPOT-4 VEGETATION, were modeled and footprints of these three
sensors were predicted for a 16-day compositing period (June 1998 and June 2002) over agricultural fields in
Bondville, IL. An atmospherically-corrected Landsat Enhanced Thematic Mapper (ETM) image acquired within the
compositing period was acquired and spatially aggregated to simulate normalized difference vegetation index
(NDVI) and enhanced vegetation index (EVI) values for the modeled observation/footprint conditions. The results
showed that root mean square errors (RMSE) of the NDVI and EVI values varied from day to day with the mean
RMSE values of 0.02 for the NDVI and 0.015 for the EVI. After examining a series of spatial averaging with various
window sizes, we found that taking 5-by-5 to 7-by-7 averages of neighboring pixels would effectively reduce
RMSE to obtain reliable cross-calibration results. These error bounds and number of pixels to be averaged are
expected to vary depending on surface heterogeneity and the same procedures need to be applied to other areas
for obtaining general recommendation.
DE: 0480 Remote sensing
SC: Biogeosciences [B]
MN: 2007 Fall Meeting