Integrating Landsat with MODIS Products for Vegetation Monitoring - Satellite-based Applications on Climate Change - page 247

16.3.2.1 STAARCH for Mapping Reflectance Change

Hilker et al. ( 2009b ) developed a new data fusion model STAARCH for detecting
reflectance changes associated with land cover change and disturbance. STAARCH
produces both a spatial change mask from two Landsat images as well as an image
sequence from the MODIS time series which describes the temporal evolution of
disturbance events. The algorithm includes functionality for the prediction of
surface reflectance based on an extended version of STARFM. The STAARCH
approach constrains the optimal image pair to be used in the STARFM prediction.
Hilker et al. ( 2009b ) applied the STAARCH approach over a 185

185 km area
in southern Alberta, Canada. Results show that STAARCH predictions agreed well
with field-based observations (93% for spatial accuracy of the disturbed area).
Temporal changes in the landscape were correctly predicted for 87-89% of
instances for the total disturbed area (Hilker et al. 2009b ). The change sequence
derived from STAARCH was used to fuse Landsat images for each available date of
MODIS imagery. The fused results were improved when compared to existing
Landsat observations.
STAARCH requires a minimum of two image pairs to develop the change mask
required as input to the algorithm. It focuses on detecting changes at Landsat scale
from the MODIS time series. ESTARFM also requires two image pairs but focuses
more on improving predictions for the mixed pixels.

16.3.3 Products Normalization

Remote sensing data products from different instruments may not consistent due
to many factors such as the differences in spectral bandwidth, spectral response
function, spatial resolution, and processing approach (Roy et al. 2008 ). Even
using same algorithm, Landsat and MODIS data products may be different. The
normalization approach produces a consistent data product using MODIS data
products as references. Gao et al. ( 2010 ) developed a generalized reference-based
approach to build a MODIS-consistent data set from multiple Landsat-like
sensors. The generalized reference-based empirical approach was tested for
converting medium resolution data product from digital number (DN) to a stan-
dard surface reflectance product. The globally available, consistent MODIS
surface reflectance products were used as the reference. As opposed to a physi-
cally based atmosphere correction approach (e.g., LEDAPS), this empirical
approach is a relative correction, and therefore, the corrected surface reflectance
is a kind of “MODIS-like” surface reflectance. It provides a way to standardize
satellite data from different medium resolution sensors to one standard and thus
allow continuous time-series analysis and land cover change detection. This
approach builds on a long history of regression-based image normalization
procedures from the remote sensing literature (Schott et al. 1988 ;Duetal.
2001 ;Olthofetal. 2005 ) and provides a practical “operational” framework for
merging information from multiple sensors.