IEEE Geoscience and Remote Sensing Magazine - September 2015 - 48

Abstract-Crop condition and natural vegetation monitoring require high resolution remote sensing imagery in
both time and space - a requirement that cannot currently
be satisfied by any single Earth observing sensor in isolation. The suite of available remote sensing instruments
varies widely in terms of sensor characteristics, spatial
resolution and acquisition frequency. For example, the
Moderate-resolution Imaging Spectroradiometer (MODIS)
provides daily global observations at 250m to 1km spatial
resolution. While imagery from coarse resolution sensors
such as MODIS are typically superior to finer resolution
data in terms of their revisit frequency, they lack spatial
detail to capture surface features for many applications.
The Landsat satellite series provides medium spatial resolution (30m) imagery which is well suited to capturing
surface details, but a long revisit cycle (16-day) has limited
its use in describing daily surface changes. Data fusion approaches provide an alternative way to utilize observations
from multiple sensors so that the fused results can provide
higher value than can an individual sensor alone. In this
paper, we review the Spatial and Temporal Adaptive Reflectance Fusion Model (STARFM) and two extended data
fusion models (STAARCH and ESTARFM) that have been
used to fuse MODIS and Landsat data. The fused MODISLandsat results inherit the spatial details of Landsat (30 m)
and the temporal revisit frequency of MODIS (daily). The
theoretical basis of the fusion approach is described and
recent applications are presented. While these approaches
can produce imagery with high spatiotemporal resolution,
they still rely on the availability of actual satellite images
and the quality of ingested remote sensing products. As a
result, data fusion is useful for bridging gaps between medium resolution image acquisitions, but cannot replace actual satellite missions.

Landsat

MODIS

e
Tim
Landsat
- 30m Spatial Resolution
- 16-Day Revisit Cycle

e
Tim

MODIS
- One or Two Revisit
Per Day
- 250m & 500m
Spatial Resolution

FIgURe 1. Landsat and MODIS data have complementary char-

acteristics in terms of spatial and temporal resolution. Although
they have been widely used for vegetation mapping and ecological
monitoring, they may not be sufficient for applications that require
high resolution imagery in both space and time.
48

I. INTRODUCTION
limate change and human activities are having a remarkable impact on the Earth's land cover and land
cover dynamics. Quantification of the type, magnitude, and
the spatial distribution of these changes is important for
climate mitigation and decision making processes [1], [2].
Satellite-based remote sensing provides valuable geospatial
data for characterizing land cover and land cover dynamics
at the globe scale. However, tradeoffs between spatial and
temporal resolution result in satellite sensor designs that
may be suboptimal for certain applications. For instance,
moderate resolution sensors (a few hundred meters resolution or coarser) such as the Moderate-resolution Imaging
Spectroradiometer (MODIS) sensors aboard the NASA EOS
Terra and Aqua satellites can provide daily global observations that are valuable for capturing rapid surface changes at
coarse spatial resolution. But the MODIS spatial resolution
of 250-1000m is often inadequate for highly heterogeneous
areas like agricultural landscapes. In contrast, Landsat data
provide sufficient spatial detail (30m, here referred to as a
medium spatial resolution) for monitoring land surface
condition and change, particularly at scales of human activity [3]-[5]. But the 16-day revisit cycle and the presence of
clouds have limited its use for studying global biophysical
processes, which can evolve rapidly during the growing season. Fig. 1 illustrates the temporal and spatial characteristics
of the Landsat and MODIS instruments. Remote sensing
data with daily frequency and medium spatial resolution
from a single sensor are not available today, largely as a result of prohibitive costs of such a system.
One way to construct daily observations at high spatial
resolution is to combine remote sensing data from multiple
sources. In order to fully use the complementary characteristics of different sensors, data fusion approaches have been
developed to combine the spatial resolution of Landsat with
the temporal frequency of coarse resolution MODIS. Traditional data fusion approaches were more focused on color
blending or sharpening for visual enhancement. These approaches are also known as downscaling or sharpening techniques. Many image sharpening approaches use data from
the same satellite platform or images from the same day. For
example, the pan-sharpening technique uses a fine resolution panchromatic band to sharpen multi-spectral bands
acquired at a coarser resolution by same instrument. Other
traditional image fusion approaches include intensity-huesaturation (IHS) modulation, as well as principal component
analysis (PCA) and wavelet transformation, which can be
treated as variants of the IHS technique [6]. In IHS, a multispectral image is converted from red, green and blue (RGB)
space into the IHS space using a defined conversion matrix.
The intensity image is then replaced by the finer resolution
panchromatic band image and converted back to the RGB
space using an inverse matrix. The resulting RGB image contains the spectral information (color) and the spatial details
from panchromatic band. This approach is more focused
on the image sharpness and better visualization; however,

ieee Geoscience and remote sensing magazine

september 2015

Table of Contents for the Digital Edition of IEEE Geoscience and Remote Sensing Magazine - September 2015