IEEE Geoscience and Remote Sensing Magazine - December 2021 - 85

angles, completeness, and time differences) [164]. Therefore,
successful 3D building change detection relies on more advanced
models that can produce accurate multitemporal 3D
data in an economical and effective way. Very recently, deep
learning has been explored for 3D reconstruction from multiview
images. For example, a CNN-based method was proposed
for dense image matching in [165]. This novel technique
may provide a new research orientation for 3D urban
change detection when vertical and height information can
be accurately derived from multiview satellite images.
VEGETATION CHANGE DETECTION
Analysis of vegetation change is important to understanding
ecological transitions [166]. Using VHR imagery, vegetation
change can be investigated at a much finer scale, e.g.,
from forest stands to individual trees. In general, there are
three types of vegetation changes: 1) seasonal, caused by
plant phenology; 2) gradual, caused by interannual climate
variability, land management, and land degradation; and 3)
abrupt, caused by disturbances, e.g., urbanization, deforestation,
and fires [167]. In [168], to assess seasonal changes,
both spectral and textural information extracted from multiseasonal
Pléiades imagery (2 m) was used for multiseasonal
leaf area index (LAI) mapping. The results showed that the
highest LAI occurred in midsummer, followed by late spring,
autumn, and winter, and the observed seasonal change trend
was similar to that based on the in situ measured LAI. Seasonal
changes in the crown scale in an Amazon tropical evergreen
forest were assessed by Wang et al. [169] using Planet
constellation imagery with a spatial resolution of 3 m. The
crown scale fraction of nonphotosynthetic vegetation showed
large seasonal trend variability from June to November.
As for gradual changes, Gärtner et al. [170] used QuickBird
and WorldView-2 imagery to quantify tree crown diameter
changes in a degraded riparian tugai forest in northwestern
China, and their results indicated that the diameter increased
by 1.14 m, on average, during 2005-2011. Tian et al. [171] explored
DSMs from satellite stereo sensors to monitor vertical
tree growth and found that periodic annual increments at the
study sites were in the range of 0.3-0.5 m. In the case of abrupt
change, Dalagnol et al. [172] quantified tree canopy loss and
gap recovery in tropical forests where there was low-intensity
logging by using WorldView-2 and GeoEye-1 images. Their study
showed that VHR satellite imagery has potential for tracking
small-scale human disturbances. Ardila et al. [173] identified
bi-temporal tree crown elliptical objects through the iterative
surface fitting of a Gaussian model to crown membership in
two urban residential areas in The Netherlands using QuickBird
and aerial images. A detection rate of 77% was reported
for both removed and planted trees.
In addition to coverage, tree crown diameters, and
canopy heights, species types are an essential parameter of
vegetation community structures. In particular, VHR imagery
is able to identify small and highly mixed species. Since
different vegetation types exhibit similar spectral characteristics,
textures are often used to identify various species.
DECEMBER 2021 IEEE GEOSCIENCE AND REMOTE SENSING MAGAZINE
For instance, Lu and He [174] investigated seasonal species
variations in a tall grassland in Ontario, Canada, during
the growing season (from April to December) in 2015 using
UAV images. The reflectance value, vegetation indices, and
GLCM textures were used in the classification, and temporal
change analysis revealed the growing process and succession
of different species. Notably, some advanced methods,
e.g., deep features [175], photogrammetric-derived
DSMs from stereo images [176], phenological characteristics
[177], and data fusion (e.g., lidar and airborne hyperspectral
images) [178], have been considered for the change
analysis of vegetation species. Moreover, some researchers
attempted to discriminate vegetation function types, e.g.,
park, roadside, and residential-industrial trees in urban areas
[179]. Likewise, vegetation function-type change monitoring
is of great significance but has not been addressed in
the current research.
MONITORING CROP CHANGES
Information about agricultural land changes, crop type
conversions, and crop growth, critical for precision agriculture,
can be effectively captured using VHR images. In
[180], land cover data for Guanlin, Yixing City, China, in
2006, 2009, 2012, and 2015 were generated using QuickBird
images, and they showed a decrease followed by an increase
in the agricultural land area that was observed. Malinverni
et al. [181] quantified the temporal variation of main crop
rotations on the Capitanata plain of Southern Italy using
WorldView-2 images, and the textural features (e.g., the
GLCM and the Gabor wavelet) were employed to improve
the classification accuracy. The study suggests that multitemporal
classification is preferred in crop mapping, due to
its rich phonological characteristics. Furthermore, frequent
crop growth monitoring is extremely important for timely
decision making in precision agriculture. Therefore, timeseries
data are recommended, although dense time series of
VHR images are relatively difficult to acquire.
Recently, new generation micro-/nanosatellites (e.g.,
Planet) and UAV systems have become available and are
able to obtain time-series VHR images, which has potential
for agricultural applications. For example, Sadeh et al.
[182] detected sowing dates using dense time-series Planet
CubeSat data with an interval of two days. As shown in
Figure 15, a partly sown field was successfully detected,
implying that detailed processes on a near daily basis can
be monitored by dense time series of VHR data. Likewise,
Bendig et al. [183] monitored plant growth based on crop
surface models using stereo UAV images. Notably, height
differences between cultivars and their increased trend during
the growing season can be observed.
Crop change caused by disease and insect damage can
also be located. VHR images are able to identify small-extent
disease and insect damage, which is beneficial for controlling
problems at early stages. Generally, diseases and insects
can result in various kinds of harm to crop canopies,
such as the removal of leaves, skeletonizing of leaf tissue,
85
IEEE Geoscience and Remote Sensing Magazine - December 2021

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