IEEE Geoscience and Remote Sensing Magazine - September 2021 - 118

can be used to reconstruct remote sensing images affected by
phenological and significant land cover changes [54]. Therefore,
reconstruction and spatiotemporal fusion are closely
interconnected, which is the main reason for reviewing these
methods here.
RECONSTRUCTION FOR SPATIAL CONTINUITY
RECONSTRUCTION OF MISSING LST PIXELS
Although partially cloudy pixels are often treated as missing
pixels, there are obvious differences when estimating LST for
cloudy pixels and missing pixels from neighboring cloudfree
pixels. In this review, missing pixels are regarded only as
the case in which measurements are actually missing owing
mainly to defective sensors (e.g., Landsat ETM+ SLC-off data)
[13] and scanning gaps between orbits (e.g., Aqua/AMSR-E,
Global Change Observation Mission/AMSR2 data), which
can occur in TIR and PMW imagery.
Reconstruction techniques can be used to complete the
missing data and enhance the usability of incomplete LST data.
In the literature, several approaches have been proposed that
can be roughly classified into three types [55]: 1) spatial methods,
in which no additional reference information is required;
2) temporal techniques, in which the reference information of
Coarse-Spatial-Resolution
LST Data
one area was obtained at different observational times; and 3)
spatiotemporal approaches, in which the reference information
is from additional spatial and temporal information.
An example of the spatiotemporal reconstruction of
missing AMSR-E LST pixels is given in Figure 5. Proposed
by Zeng et al. [56], the adopted spatiotemporal reconstruction
method consists of three steps. First, a classification map
should be obtained to differentiate the land surface types; the
missing AMSR-E LST values of the target image (observed on
1 January 2011, in this case) can then be reconstructed using
robust regression with the spatiotemporal information
from other auxiliary LST images in each classification (such
as those observed on 2 January 2011, in this case). Finally, a
postprocessing procedure with spatial information is applied
to eliminate outliers.
Fine-Spatial-Resolution
LST Data
Spatially Continuous?
No
Reconstruction of LST
Under Missing Pixels
Yes
Reconstruction of LST
Under Cloudy Pixels
Yes
SPATIAL RECONSTRUCTION METHODS
Reconstruction techniques based exclusively on spatial information
are the most traditional of the three approaches.
These methods reconstruct missing information using the
existing or remaining valid LST values and assume that they
share the same statistical and geometrical structures with the
missing values [57]. The most basic reconstruction methods
are spatial interpolation approaches, e.g., spline function
[58] and geostatistical methods. Motivated
by the high spatiotemporal heterogeneity
of LST, some researchers
have considered more environmental
variables and used multivariate interpolation
methods, e.g., cokriging [59]
and regression tree analysis [60].
Spatiotemporal Fusion
SCHR-LST
Applications
FIGURE 4. A schematic diagram showing the generation of SCHR-LST and the relationship
between LST reconstruction and spatiotemporal fusion.
118
TEMPORAL RECONSTRUCTION
METHODS
Reconstruction methods based on
multitemporal or time-series information
use temporal images of the same
region at different times to reconstruct
missing pixels. The algorithms primarily
employed include the linear temporal
approach [61]-[63], harmonic
analysis method [64], temporal Fourier
analysis approach [65], wavelet
transformation method [66], asymmetric
Gaussian function fitting method
[67], singular spectrum analysis
algorithm [68], discrete cosine transform
[69], diurnal temperature cycle
(DTC)-based method [8], [70], [71],
and multitemporal dictionary learning
[72]. Multitemporal methods generally
ignore data from geographically
neighboring pixels and are appropriate
when the differences are linked
mainly to regular changes, such as
those in observation conditions and
IEEE GEOSCIENCE AND REMOTE SENSING MAGAZINE SEPTEMBER 2021
IEEE Geoscience and Remote Sensing Magazine - September 2021

Table of Contents for the Digital Edition of IEEE Geoscience and Remote Sensing Magazine - September 2021
