IEEE Geoscience and Remote Sensing Magazine - December 2021 - 87

Observing Satellite Advanced Visible and Near-Infrared Radiometer
2 (10 m), pan-sharpened WorldView-2 (0.5 m), and lidar]
were generated and compared. As described in Figure 16,
the Landsat image cannot accurately discriminate the mangrove
extent, due to the mixed-pixel problem [Figure 16(e)],
and more fine-scale mangrove features, i.e., tree-crown-level
species, can be captured only by pan-sharpened WorldView-2
imagery [Figure 16(l)-(p)]. By summarizing the current
literature, it can be found that most studies focus on
detecting the extent of wetland change but ignore species
change. For instance, Hu et al. [201] monitored land cover
changes in the Hangzhou Xixi wetland from 2000 to 2013
using IKONOS, QuickBird, and WorldView-2 images. It was
shown that the nonwetland area increased by approximately
100%, mostly in the form of herbaceous zones, followed
by forests, ponds, cropland, marshes, and rivers. Wu et al.
[202] integrated lidar data and multitemporal aerial imagery
(1 m) to map wetland inundation dynamics in the Prairie
Pothole region of North America, which is characterized
by millions of small depressional wetlands.
The difficulties of species change detection in wetlands lie
in the following aspects. On the one hand, tidal and phenological
changes make different plant species highly dynamic
on daily and seasonal frequencies, respectively. On the other
hand, many species have a similar spectral reflectance during
the peak biomass in complex wetland landscapes [203],
and the spectral signature of the same species can be influenced
by many complex factors, such as the off-nadir angle,
sun-viewing geometry, crown porosity, leaf clumping, and
ground surface scattering [204]. For instance, in [200], mangrove
species were categorized from WorldView-2 images
using the nearest-neighbor classifier to extract object-based
spectral and textural features within tree crowns, but a low
overall accuracy of around 54% was reported. As demonstrated
in Figure 16(p), misclassified open scrub Avicennia
marina can be clearly observed. To improve the discriminative
power among various species, the potential of VHR hyperspectral
images, dense time-series data, and vertical information
for characterizing detailed spectral, phenological,
and height attributes needs to be explored.
ECOSYSTEM SERVICES MONITORING
Ecosystem services link ecosystems to human welfare by
regarding nature as a stock providing a flow of services
(e.g., local climate regulation and water purification) [205].
Monitoring urban ecosystem services is of great value for
investigating ecological function changes and can help improve
the understanding of urbanization impacts on local
ecological benefits. VHR satellite data can monitor spatially
explicit ecosystem services at fine scales. Generally speaking,
there are two categories of methods to derive ecosystem
services: 1) statistical regression and radiative transfer models
and 2) land use/cover-based methods [206]. Since in situ
observations are not always available and the validity of
statistical regression and radiative transfer models is affected
by time inconsistencies between ground and remotely
DECEMBER 2021 IEEE GEOSCIENCE AND REMOTE SENSING MAGAZINE
sensed measurements, land use/cover-based methods are
often preferred. For example, in [207], land use/cover maps
of Shanghai's urban core from 2000 to 2009 were classified
using IKONOS and GeoEye-1 images, and the classes
were then transformed into ecosystem service supply and
demand budgets, including regulating, provisioning and
cultural services, and ecological integrity. An increase of
at least 20% in ecosystem service supply budgets was observed,
which was mainly attributed to the replacement of
continuous urban fabric and industrial areas by high-rise
commercial/residential areas despite a slight increase in urban
green sites.
Huang et al. [144] assessed ecosystem service change
in Shenzhen from 2005 to 2017 using Gaofen-2 (4-m) and
QuickBird (2.4-m) images. In the study, multitemporal land
use maps were generated by a transferred deep CNN (as
shown in Figure 12), based on which ecosystem service
supply and demand values were estimated. It was found
that supply capacity had decreased by 13.7% due to a reduction
in woodlands, water, farmland, and so on, but, on
the other hand, demand values had grown by 23.5% because
urban expansion and redevelopment had increased
the amount of residential, commercial, and infrastructure
land. The results clearly demonstrated the ecosystem degradation
of Shenzhen during the previous 10 years. Ren et al.
[208] evaluated the ecosystem services of Guyuan City in
2003, 2009, and 2014 via VHR satellite imagery (e.g.,
QuickBird and Gaofen-1) and showed that VHR images were
advantageous in the dynamic, quantitative, and visual examination
of ecological changes. With VHR remote sensing
images, fine-scale ecosystem services within urban areas
can be effectively quantified. However, most of the current
works focus on urban areas and ignore the ecosystem services
of natural scenes, such as forests and wetlands. Moreover,
these works present only case studies, and large-scale
examinations are still lacking.
IMPERVIOUS-SURFACE CHANGE DETECTION
The change detection of impervious surfaces is important
in monitoring and understanding urban development and
has been extensively studied in the remote sensing literature.
However, most of the existing studies monitor the
change of impervious surfaces based on coarse- and medium-spatial-resolution
satellite imagery, such as MODIS and
Landsat [209], [210], which, on the other hand, have difficulty
dealing with areas that have low impervious-surface
intensities and mixed pixels [211]. During recent decades,
images with high spatial resolution have provided new opportunities
for subtle impervious-surface monitoring at
very fine scales. However, impervious-surface monitoring
using VHR imagery is a challenging task. VHR multitemporal
images exhibit a large number of details (e.g., buildings,
roads, driveways, and sidewalks), greater spatial heterogeneity
(e.g., different viewing geometries), and occlusion by
urban trees, shadow, and vertical structure layover [212].
To address the problem caused by shadow, Li et al. [213]
87
IEEE Geoscience and Remote Sensing Magazine - December 2021

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