Nothing Special   »   [go: up one dir, main page]
Next Article in Journal
Comparison of UAV RGB Imagery and Hyperspectral Remote-Sensing Data for Monitoring Winter Wheat Growth
Previous Article in Journal
Local Scale (3-m) Soil Moisture Mapping Using SMAP and Planet SuperDove
You seem to have javascript disabled. Please note that many of the page functionalities won't work as expected without javascript enabled.
 
 
Search for Articles:
Title / Keyword
Author / Affiliation / Email
Journal
Article Type
 
 
Advanced Search
 
Section
Special Issue
Volume
Issue
Number
Page
 
Journals
Remote Sensing
Volume 14
Issue 15
10.3390/rs14153805
remotesensing-logo
Submit to this Journal Review for this Journal Propose a Special Issue
► Article Menu

Article Menu

share Share announcement Help format_quote Cite question_answer Discuss in SciProfiles thumb_up
...
Endorse textsms
...
Comment
first_page
settings
Order Article Reprints
Font Type:
Arial Georgia Verdana
Font Size:
Aa Aa Aa
Line Spacing:
Column Width:
Background:
Article

Time-Series Monitoring of Dust-Proof Nets Covering Urban Construction Waste by Multispectral Images in Zhengzhou, China

by
Zilu Li
1,2,3,
Huadong Guo
1,2,3,
Lu Zhang
1,2,3,*,
Dong Liang
1,2,3,
Qi Zhu
1,2,3,
Xvting Liu
1,2,3 and
Heng Zhou
1,2,3
1
Key Laboratory of Digital Earth Science, Aerospace Information Research Institute, Chinese Academy of Sciences, Beijing 100094, China
2
International Research Center of Big Data for Sustainable Development Goals, Beijing 100094, China
3
University of Chinese Academy of Sciences, Beijing 100049, China
*
Author to whom correspondence should be addressed.
Remote Sens. 2022, 14(15), 3805; https://doi.org/10.3390/rs14153805
Submission received: 5 July 2022 / Revised: 27 July 2022 / Accepted: 3 August 2022 / Published: 8 August 2022
(This article belongs to the Section Urban Remote Sensing)
Browse Figures
Figure 1
<p>Study area. (<b>a</b>) Location of Henan Province in China; (<b>b</b>) Location of the study area in Henan Province; (<b>c</b>) Sentinel-2 remote sensing image of Zhengzhou (22 February 2018) and delineation of the eastern/western suburbs and developing/built-up zones.</p> "> Figure 2
<p>Field investigation pictures of (<b>a</b>) construction waste covered with dust-proof nets, (<b>b</b>) bare soil covered with dust-proof nets captured by drones and (<b>c</b>) the process of dumping bare soil with a shovel.</p> "> Figure 3
<p>The workflow of this study.</p> "> Figure 4
<p>Example data of features from Google Earth. (<b>a</b>) Bare soil; (<b>b</b>) Green and blue dust-proof nets; (<b>c</b>) Regularly arranged temporary buildings with roofs similar in color to blue dust-proof nets, commonly found on construction sites.</p> "> Figure 5
<p>Two subsets of construction waste samples from western Zhengzhou in Sentinel-2 images. (<b>a</b>) The yellow and green parts represent bare soil and dust-proof nets, respectively (22 February 2018); (<b>b</b>) The yellow, blue, green and purple parts represent bare soil, blue dust-proof nets, green dust-proof nets and temporary buildings, respectively (7 July 2019).</p> "> Figure 6
<p>The reflectance plot showing the spectral characteristic of four classes derived from the Sentinel-2 image on 7 July 2019.</p> "> Figure 7
<p>Construction waste and dust-proof nets map of Zhengzhou city from 2015 to 2020. (The green, blue and yellow parts represent green dust-proof nets, blue dust-proof nets and bare soil, respectively. The temporary buildings in the classification results are not displayed after post-processing).</p> "> Figure 8
<p>Comparison of ground truth and classification results. (<b>A</b>–<b>D</b>) Different regions of this study area. In each group, the left is the ground truth, and the right is the prediction result.</p> "> Figure 9
<p>The area of construction waste in western suburbs, eastern suburbs and the whole area from 2015 to 2020.</p> "> Figure 10
<p>The variation of construction waste in the developing zone and built-up zone, which are represented by the yellow and red line, respectively. The green line named “Reference density” is the density of construction waste in the developing zone relative to the built-up zone. A comparison of the red and green lines (with the difference in the area of partitions removed) highlights the true disparity between the two regions.</p> "> Figure 11
<p>The coverage of dust-proof nets in the built-up zone and the whole area from 2015 to 2020.</p> "> Figure 12
<p>The linear regression models of (<b>a</b>) air quality index, (<b>b</b>) concentration of PM2.5, (<b>c</b>) concentration of PM10 and the dust-proof net’s coverage. The dashed lines denote the best-fit lines from linear regression.</p> ">
Review Reports Versions Notes

Abstract

:
Rapid urbanization has produced a huge amount of construction waste. The operations and consequences of construction and demolition can lead to windblown dust problems, profoundly affecting the living environment of residents. Fortunately, dust-proof nets have been used in construction sites to reduce and prevent pollution by fine particles such as dust, so it is important to monitor and evaluate their efficacy. In this study, Earth observation techniques were used for the extraction and monitoring of solid waste and dust-proof nets. In order to fully perceive the validity and necessity of dust-proof nets for urban air health, we conducted a case study in Zhengzhou, China. We explored the potential of multispectral remote sensing available for monitoring urban construction waste and proposed a multi-layer classification method to identify construction waste and dust-proof nets based on Landsat-8 OLI and Sentinel-2 MSI data, with an average identification accuracy and Kappa coefficient of 96.27% and 0.93 for construction waste in the study area from 2015 to 2020, respectively. In addition, our study revealed the driving factors and impact of temporal variations in regional construction waste areas and dust-proof nets coverage. The results indicate the classification can track municipal solid waste management and changes in air pollutant concentrations and is useful for achieving SDG 11.6 (reduce the adverse per capita environmental impact of cities, including by paying special attention to air quality and municipal and other waste management). This study has the potential to monitor construction waste and dust-proof nets, paving the way for better urban environmental governance and surveillance actions in the future, especially involving big data.

1. Introduction

In the context of rapid global industrialization and urbanization, combating environmental pollution is a huge challenge for all countries. Air pollution has received a lot of attention from researchers as the type of pollution most closely related to human health, such as epidemic infection rates [1]. In particular, severe atmospheric haze is occurring in many industrialized cities. Fine particulate matter with a particle size of less than 2.5 μ m (PM2.5) is the main air pollutant that causes haze [2]. There are two sources of fine particles in the air: primary particles are mainly caused by combustion, roads or wind, while secondary particles are the nucleation and formation of various compounds in the air through secondary reactions [3]. Dust, which accounts for the majority of primary particles [3,4], is one of the main sources of PM2.5 pollution.
At the same time, with the acceleration of urbanization, construction waste has been the largest source of urban waste, accounting for more than 80% of the total urban waste in China [5]. Since the output of construction waste far exceeds the capacity of legal disposal sites, most of the waste is dumped on the outskirts of cities, resulting in the phenomenon of a “construction waste siege” [6]. On the one hand, these illegally disposed construction waste sites occupy numerous land resources while aggravating human-land conflicts. On the other hand, the dust raised by solid wastes in the processes of removal, deposition and incineration seriously effect the urban environment [7]. To give an example, in Zhengzhou, China, the fugitive dust caused by construction waste accounts for 60% of the total amount of dust in the city, which designates dust as the main source of dust pollution. Therefore, it is highly necessary to systematically manage construction waste in urban areas to minimize the effects of pollution. Understanding construction waste is central to UN Sustainable Development Goal target 11.6—aiming to reduce the negative environmental impacts of urbanization, particularly with regard to municipal waste and air quality. The extraction of illegal disposal site locations of construction waste is a key condition of this endeavor. However, the latest accumulation sites and generation times are difficult to predict, and the location of construction waste has no regularity [8], which means that it is inefficient and difficult to obtain the latest data about urban construction waste using traditional manual inspection.
Remote sensing technology has the characteristics of wide coverage, strong timeliness, and low acquisition cost. It can objectively and intuitively reflect the situation on the earth’s surface and provide stable data support [9], which has good application prospects in monitoring construction waste [10]. For example, Bagheri et al. used aerial imagery to identify hazardous waste sites in New Jersey in 1988 [11]. Thereafter, there are a number of studies on the extraction of landfill areas through traditional visual interpretation methods [12,13]. With the rapid development of image processing technology, the deep learning models for the classification of construction waste are endless. Lei Zhou et al. studied the best parameters of three typical machine learning methods for detecting construction waste [14]. Wu et al. proposed a CNN-based informal garbage dump detection framework, which reduced the average detection time of a single scene sample on top of detection accuracy of 85.92% [15]. Other indicators retrieved from remote sensing images can also be used to effectively segment construction waste [16]. In addition, the DEM obtained by stereo image data can also be used to analyze the accumulation of construction waste [17]. In summary, the benefit of remote sensing technology in extracting urban construction waste is scientifically feasible, and it is of great significance in terms of accuracy and efficiency compared to traditional monitoring methods.
In order to improve the atmospheric environment, it is recommended to cover all bare soil and construction waste with dust-proof nets to reduce dust pollution, which is a precursor of haze [18,19]. In addition to effectively reducing dust, monitoring the dust-proof nets’ coverage of construction waste can also help urban waste supervision and environmental protection. However, the studies on solid waste based on remote sensing mainly focus on exposed construction waste, while there is still a big gap in the research of dust-proof nets. Meanwhile, most of the studies have focused on identifying construction waste from a particular landfill site or a small area, and limited research has been published globally on the change of construction waste in the spatial distribution pattern at the urban scale.
Therefore, based on the status of research on construction waste, this study explored the feasibility of using multi-temporal and multispectral remote sensing images to monitor construction waste, solving the following specific problems.
(1)
A multi-layer identification method based on remote sensing was proposed to extract both dust-proof nets and uncover construction waste.
(2)
Combined with urban construction planning, the variations in the area and distribution of construction waste were analyzed in terms of time and space.
(3)
Comparing the coverage rate of dust-proof nets and AQI data of the same period, it is concluded that mulching dust-proof nets have a positive impact on urban air quality.
In addition, the effectiveness of the dust-proof nets was verified by the concentration of fine particulate matter in the atmosphere under different coverage rates. Considering the prominent problem of construction waste pollution in Zhengzhou City due to accelerated urbanization, we treat it as a relevant case study. This study holds promise as a means to help contemporary governments dedicated to urban air governance and construction waste supervision. Meanwhile, it provides ideas for remote sensing technology to contribute to the construction of a “smart city”, which is of great significance.

2. Materials

2.1. Study Area

The study area is Zhengzhou city, located in the central plain of China. Zhengzhou is located in the north-central part of China, with a temperate continental monsoon climate, and the average annual precipitation is above 600 mm [20,21]. In recent years, the city has been experiencing accelerating urbanization, with the problem of construction waste pollution among the most prominent. Between 2013 and 2017, nearly 100 million cubic meters of construction waste have been generated in Zhengzhou’s new urbanization construction [22]. The rectangular area covers the main municipal districts of Zhengzhou, with a total area of about 1010 km2 (Figure 1). For a more detailed analysis of the distribution of construction waste, we performed two kinds of partitions in the study area based on urban planning, namely the eastern/western suburbs (blue/green border) and the developing/built-up zones (orange/purple border). The built-up zone is bordered by the third ring road of Zhengzhou city, while the developing zone is between the third and fourth ring roads. The eastern and western suburbs are delineated by the boundaries of built-up zones and some major roads.

2.2. Data

2.2.1. Remote Sensing Data

We selected Sentinel-2 images from 2017 to 2020 and Landsat-8 images from 2015 to 2016 to extract the construction waste and dust-proof nets in this study. Sentinel-2 is a high-resolution multispectral imaging constellation launched by the European Space Agency, consisting of two satellites: Sentinel-2A (23 June 2015) and -2B (7 March 2017), with revisit periods of 5 days [23,24]. Sentinel-2 MultiSpectral Instrument (MSI) has a total of 13 spectral bands with different spatial resolutions (10, 20, 60 m); only bands at 10 or 20 m resolution were used [25]. The Level-1C product of Sentinel-2 has been orthorectified and geographically registered, which was selected for this study.
Landsat-8 Operational Land Imager (OLI) has 9 spectral bands, with a temporal resolution of 16 days. The Level-2SP products are geometrically corrected and atmospherically corrected, which can be directly used in this study. The bands and the spatial resolution selected in this article are shown in Table 1. Sentinel-2 MSI and Landsat-8 OLI have been obtained from the ESA and the US Geological Survey (USGS), respectively.

2.2.2. Fieldwork Data

The real state and primary features of construction waste were obtained using drones and other means in certain areas of Zhengzhou in 2020. According to the survey, the main and most dust-prone type of exposed construction waste is bare soil, which is the reason why it is used to represent the exposed construction waste in Section 3.2. During fieldwork, we found that there are two types of dust-proof nets: green and blue, with the blue one appearing after 2019. Figure 2 shows some survey photos.

2.2.3. Atmospheric Data

In order to verify the role of dust-proof nets in atmospheric governance, we introduced the monthly air quality index (AQI) calculated in terms of urban areas officially designated by the government for evaluation. This index comprehensively considers the pollution levels of six pollutants, including SO 2 , NO 2 , PM10, PM2.5, CO and O 3 , which are used to assess the air quality status in China [26]. It is calculated by summing the individual quality indices of each pollutant. The larger the value of AQI, the more serious the pollution level is [27]. Only the AQI for the month in which the satellite acquisition time is located was acquired for correlation analysis.
In addition, we used Ground-level PM2.5 L2 Daily 0.01 Deg Product (ChinaHighPM2.5) and Ground-level PM10 L2 Daily 0.01 Deg Product (ChinaHighPM10) released by Dr. Wei et al., which are the series of long-term, high-resolution (about 1 km) and high-quality datasets of ground-level air pollutants for China (97.3°E–131.39°E, 18.11°N–46.40°N) [28,29]. They are generated from MODIS/Terra+Aqua MAIAC AOD products together with other auxiliary data (e.g., ground-based measurements, satellite remote sensing products, atmospheric reanalysis and model simulations) using artificial intelligence by considering the spatio-temporal heterogeneity of air pollution. The daily PM10 has a cross-validation coefficient of determination (CV- R 2 ) of 0.86 and a root-mean-square error (RMSE) of 24.28 μ g/m 3 on a daily basis, while the daily PM2.5 estimates agree well with ground-based PM2.5 measurements (i.e., CV- R 2 = 0.86–0.90), with average RMSE values ranging from 10.0 to 18.4 μ g/m 3 . Both the PM2.5 and PM10 concentration data in the study area 30 days after the satellite image acquisition time are selected to verify the effectiveness of dust-proof nets.

3. Method

This paper proposed a multi-layer classification method to detect different types of construction waste. Satellite data covering Zhengzhou were processed, and the sample collection was constructed based on the spectral and geometric features of construction waste by visual interpretation. The multi-layer classification was built using two machine learning algorithms. The features and classifier were optimized by adjusting the principles and parameters for the second layer of the method. The identification method was evaluated by the classification accuracy calculated by the confusion matrix. The technical flow chart for construction waste monitoring and dust-proof nets effectiveness verification developed in this study is shown in Figure 3, which mainly includes preprocessing and sampling, multi-layer classification, accuracy assessment and correlation analysis with air quality.

3.1. Preprocessing

The Level-1C product of Sentinel-2 we used is Top of Atmosphere (TOA), which needs atmospheric correction to obtain the surface reflectance data, while the Level-2SP product of Landsat-8 does not require atmospheric correction. Atmospheric correction of Sentinel-2 data is performed using the plug-in (Sen2cor) released by ESA.
Due to the different spatial resolutions of the two kinds of data, for the convenience of subsequent processing and statistics, we used the nearest neighbor algorithm to resample the spatial resolution of all bands of Landsat-8 and some bands of Sentinel-2 to 10 m to match the bands of the Sentinel-2 image with a spatial resolution of 10 m [30].

3.2. Sample Data Acquisition

There are various types of objects in the city, which can be roughly divided into three types: water, vegetation and impervious surface (the farmland around the city in the study area was masked). The impervious surface consists of roads, buildings, construction waste, etc., and is very complicated. This study focuses on the dust pollution caused by construction waste and aims to monitor the dust-proof net’s coverage of urban construction waste. Therefore, the targets mainly include exposed and mulching construction waste.
Bare soil is one type of construction waste as a product of the construction process. Given that it is the most dust-prone type of construction waste without dust-proof nets, we took bare soil to represent exposed construction waste for monitoring (Figure 4a).
Field investigations revealed that there are two types of dust-proof nets: green and blue. In the true color remote sensing image, the construction waste covered with green and blue dust-proof nets is generally light green and blue, respectively (Figure 4b), while the bare soil will be light yellow or orange depending on the water content.
In addition, temporary buildings with blue roofs built during the construction process are very similar to the blue dust-proof nets in color, as are vegetation and green dust-proof nets, which are prone to confusion (Figure 4c). Construction waste has an irregular contour, blurred boundaries and is often distributed in a concentrated manner [31]. In contrast, the shape of temporary buildings is mostly rectangular, so they can be distinguished by visual interpretation. The classification accuracy can be improved by comparing the temporary building with non-construction waste ground objects.
Thus, the sample data includes vegetation, water, impervious surfaces and four types of construction waste belonging to impervious surfaces: bare soil, green and blue dust-proof nets and temporary buildings. Since there were no blue dust-proof nets before 2019, construction waste covered with blue dust-proof nets could not be sampled from 2015 to 2018.
In order to avoid intra-sample bias, pure pixels with a uniform distribution were selected (Figure 5). The numbers of pixels in a construction waste category satisfied statistical significance (see Table 2). The separability of the samples was tested using the Jeffries-Matusita distance and Transformed Divergence parameters. Each class of samples was randomly divided into two parts, 60% for training and 40% for validation [30].

3.3. Multi-Layer Classification

Supervised classifiers require prior knowledge to learn features of different targets and identify these learned features in the unlabeled data, which is more robust than model-based methods [32]. In this study, we adopted the hierarchical classification method consisting of supervised classifiers to extract construction waste information.
The random forest (RF) classifier has been widely used in remote sensing image classification due to its strong robustness and high accuracy [33]. It is an algorithm that integrates a combination of decision trees, and each decision tree is going to judge the unlabeled pixels using the random sample collection from the original training data. Then, the unlabeled pixel categories are determined by the majority voting results of all decision trees [34].
RF has more advantages in large-scale land use land cover mapping, which is not suitable for construction waste segmentation because of its small size and scattered distribution [35]. Therefore, it is used as the first layer of multi-layer classification for primary classification, and the number of trees is set to 50 because previous studies have verified that this is the best parameter for construction waste segmentation [14]. Thus, the surface is divided into water, vegetation, buildings and construction waste.
The first layer of the multi-layer classification method performs well in segmenting vegetation and water, while the separability of construction waste and buildings is poor. Therefore, we reclassified the first-layer classification results with water and vegetation removed using detailed types of construction waste samples. Because the construction waste data in this article is unimodal and distinctive, as shown in Figure 6 (in Section 3.4.1) [36], the traditional supervised classification such as Maximum Likelihood Classification (MLC) was chosen for finer classification.
In order to determine a suitable classifier, we selected Sentinel-2 images and sample datasets on 7 July 2019, as materials to test the overall accuracy and Kappa coefficient of the following four classifiers: Parallelepiped Classification, Mahalanobis Distance Classification, MLC and Support Vector Machine (SVM).
  • Parallelepiped Classification is generally used when there is no overlap between point clouds of different categories in the feature space. The decision boundary of each class is limited by a parallelepiped consisting of its mean vector and standard deviation. If an unlabeled pixel vector falls inside the parallelepiped, it is assigned to the category [37].
  • The discriminant function of Mahalanobis Distance Classification is the spectral distance; that is, the unlabeled pixel is classified into the class with the smallest spectral distance. Unlike Euclidean distance, which is also spectral distance, Mahalanobis distance is weighted by the covariance matrix between different bands, and so it is a direction-sensitive distance classifier [38,39].
  • Maximum Likelihood Classification assumes that the statistics for each class in each band are normally distributed. The main idea is to predict the category label y that maximizes the likelihood of our observed data; that is, pixels are classified into the most probable type after calculating the attribution probability of pixels belonging to various types of features. If the highest probability is less than the threshold you specify, the pixel remains unclassified [39,40].
  • Support Vector Machine is a machine learning method based on statistical learning theory. It can automatically find those support vectors that have a greater ability to distinguish between classifications and establish classification rules based on them to maximize the interval between classes. When the sample data cannot be divided linearly, the method of increasing dimension mapping is adopted to map the vectors to a higher dimensional space for division. However, this process requires a lot of computing resources [41].
In the above comparison (Table 3), SVM has the highest accuracy, slightly larger than MLC, and the accuracy is greater than 97%. Parallelepiped Classification and Mahalanobis Distance Classification have lower accuracy. Due to the time consumption of SVM being much more than that of MLC, MLC is considered to be the most time-efficient method that can guarantee accuracy and thus serves as a second layer of the multi-layer classification method suitable for this study.

3.4. Model Optimization

Although MLC displayed satisfactory accuracy, it classified some objects that were not part of the study and should have been designated as “Unclassified” into the study category, making the classification results erroneous. For example, streets and parts of buildings are all classified as dust-proof nets, which makes the accuracy of dust-proof nets very high, improving the overall classification accuracy. Hence, the classifier needs to be optimized.

3.4.1. Feature Selection

The more features involved in classification, the more information can be obtained. However, the consequent increase in redundancy among features can reduce the extraction accuracy. Thus, it is necessary to conduct feature selection before MLC to improve the separability of the targets and other objects [42].
(1)
Spectral Features Method
The spectral characteristic curve can visualize the separability of ground objects, which is a common manner of selecting features. The Spectral Response Functions (SRF) of samples on Sentinel-2 images in 2019 are shown in Figure 6. The curves of temporary buildings are completely different from construction waste, which peaks at Band 12. Band 2 was selected as a classified feature because the two kinds of dust-proof nets have similar trends in curves, except for this band. The difference between the response values of the four types at Band 6 is the largest, which means that the separability is good. Similarly, analyzing the Landsat-8 images led us to determine that Bands 3 and 4 are useful classified features.
(2)
Band Statistic Features
The selection of features should contain as much information as possible; that is, a band with a larger standard deviation is usually selected.
The standard deviations of Landsat-8 and Sentinel-2 were calculated separately. Table 4 shows the first four bands with the largest standard deviation, and it was found that the SWIR-2 band is the largest on both sensors, so the SWIR-2 band should be preferred in feature selection.
Considering that there is vegetation and water in the city, spectral indices are introduced to enhance the discrimination between them and the target. Here, we adopted the two spectral indices: Normalized Difference Vegetation Index (NDVI) and Normalized Difference Water Index (NDWI). The formulas for each are listed below:
N D V I = B a n d N I R − B a n d R E D B a n d N I R + B a n d R E D
N D W I = B a n d G R E E N − B a n d N I R B a n d G R E E N + B a n d N I R
where B a n d N I R , B a n d R E D and B a n d G R E E N represent the values of the near-infrared band, red band and green band, respectively.
In summary, for the purpose of this study, the optimal feature subset on Landsat-8 images are Bands 3, 4, 7, NDVI and NDWI, and Bands 2, 6, 12, NDVI and NDWI on Sentinel-2.

3.4.2. Thresholds Selection

The minimum similarity threshold of MLC ensures that when a pixel does not belong to any class, it will not be classified into the class with the highest attribution probability, although there will be some loss of accuracy. At the same time, it can avoid black spots that do not belong to any defined category in continuous pixels. Setting a minimum similarity threshold directly affects the accuracy of the maximum likelihood method.
Table 5 shows the overall accuracy and Kappa coefficient at different minimum similarity thresholds. With the decrease in threshold, the overall accuracy and Kappa coefficient increase. By fitting this set of data, it is found that the rate of accuracy change first increases and then decreases. The rate of change is the largest when the threshold is 0.005, and it tends to be flat when the threshold is 0.001. Therefore, we set the minimum similarity threshold to 0.001.

3.5. Verification Method

In this study, a confusion matrix was calculated to evaluate the accuracy of the method. Based on the test samples, the results are evaluated using six indicators: overall accuracy (Equation (3)), Kappa coefficient (Equation (4)), producer accuracy (Equation (6)), user accuracy (Equation (7)), commission (Equation (8)) and omission (Equation (9)) [43,44].
O A = ∑ k = 1 n p k k p
K a = O A − p e 1 − p e
p e = ∑ k = 1 n a k b k ∑ k = 1 n a k ∑ k = 1 n b k
P A = p k k a k
U A = p k k b k
O m i s s i o n = 1 − P A
C o m m i s s i o n = 1 − U A
where p k k represents the value in the kth row and kth column of the confusion matrix, which is the number of samples correctly classified for each type; a k is the number of true samples of each type, and b k is the number of samples of that type in the classification result.

3.6. Atmospheric Data Integration and Correlation Analysis

Utilizing the method of identifying construction waste with remote sensing images described above, the distribution and area of different types of construction waste are obtained, which is the proportion of dust-proof nets that can be derived. Thus, we can compare the mulch coverage with atmospheric data to verify the role of dust-proof nets.
Weather conditions have a significant impact on diffusion and the transport of air pollutants [45]. Precipitation and higher wind speeds are favorable to the dilution of particulate matter in the air [46], in addition to temperature and relative humidity, which are also important parameters affecting the concentration of particles [47]. Therefore, in order to minimize the effect of occasional wind and rain on particulate matter concentrations, we used the monthly value of AQI and average concentrations of PM2.5 and PM10 for thirty days after the extraction of construction waste as a proxy for air quality after laying dust-proof nets to verify its effectiveness.
Considering the strong fluidity of atmospheric particulate matter [48], it is difficult to conduct a co-location analysis of dust-proof net areas and pollutant concentrations at a finer scale than the study area, although data on PM concentrations with a spatial resolution of 1 km have been obtained. Therefore, we performed per-pixel statistics for the thirty-day average PM data to obtain the average pollutant concentration value for the study area, eliminating the effects of wind. The monthly AQI and integrated pollutant concentration data are shown in Table 6.
The AQI is a reference index of urban air quality that aggregates the concentrations of six matters. Compared to other matters, dust is most likely to accumulate as fine particles in the air. Therefore, in order to further verify the effectiveness of dust nets, linear regression analysis was performed on AQI, PM10 concentration and PM2.5 concentration data separately for dust-proof net coverage, not only for AQI. The coefficients of determination ( R 2 ) obtained were able to illustrate the relevance between them and the effect of dust nets.

4. Results and Analysis

4.1. Time-Series Urban Construction Waste Map in Zhengzhou from 2015 to 2020

Figure 7 shows the distribution of construction waste in the study area from 2015 to 2020, where the yellow, green and blue parts represent bare soil, green dust-proof nets and blue dust-proof nets, respectively. The class of temporary buildings is not displayed in the results because it was only utilized to improve the accuracy of the blue dust-proof nets and is not relevant to this study. Obviously, the urban construction waste is roughly characterized by a circular outward divergence, most of which is around the city and showing more sparsely towards the city center. As the area and the proportion of various construction waste are constantly changing, the bare soil represented by yellow accounted for the vast majority in 2017.

4.2. Accuracy Assessment

The classification results were examined using ground-truth data derived from visually interpreted high-resolution remote sensing images on 22 February 2018 to verify the classification effect. As shown in Figure 8, we selected certain areas with a more concentrated accumulation of construction waste for more detailed analysis.
Edge recognition works well for objects with regular shapes, as in regions B and D. However, due to the mixed pixels, it is not sensitive to the identification of edges with complex shapes and many internal details of construction waste, which is also an impact of the post-processing performed to deal with spots, as in regions A and C.
As illustrated in the confusion matrix (Table 7), the Kappa coefficient of all the results is greater than 0.89, and the overall accuracy is above 91% (accuracy of temporary buildings is used to calculate the overall accuracy like other types). The accuracy of Sentinel-2 images for July 2019 and 2020 were lower than those for February 2017 and 2018, which may be due to the confounding of the two newly added classes, blue dust-proof nets and temporary buildings, rather than seasonality, as dust nets are not affected by seasons. In general, these results are acceptable compared to other land surface classifications based on remote sensing [24,49].

4.3. Spatial Patterns of Urban Construction Waste

The distribution of construction waste directly reflects the state of urban arrangement. Analyzing its area in different regions can contribute to understanding the urbanization structure and provide data support for future planning.

4.3.1. Eastern and Western Distribution

As shown in Figure 9, in spite of some occasional fluctuations in 2016, the total area of construction waste in Zhengzhou City increased and then decreased from 2015 to 2020. The area remained stable during 2015–2016 and 2017–2019. Nevertheless, there was a significant surge between 2016 and 2017, and the area slumped to 45.3 km 2 in 2020.
In order to learn more details about the suburbs, the construction waste area in the eastern and western suburbs was separately counted. It is shown that the change trend of the area in the two regions is consistent with the overall state; that is, there is first an increase and then a decrease. Notably, the change in the west is relatively lagging behind it in the east. The difference may be attributed to the fact that Zhengzhou’s new urbanization is more focused on the eastern region in 2017–2019 rather than the west.

4.3.2. Developing Area and Built-Up Area Distribution

In addition, we analyzed the construction waste area in the built-up zone and developing zone. As can be seen in Figure 10, the area in the developing zone far exceeds that in the built-up zone, which is related to the difference in the area of partitions. The area of the developing zone is about three times the area of the built-up zone. To eliminate the influence of the regional area on construction waste statistics, the relative waste density of the developing zone with the area of the built-up zone as a reference (as the green line in Figure 10) is calculated. Overall, the variations in construction waste in both built-up and developing zones are consistent with the tendency and magnitude of variation revealed above. Before 2017, the state of the built-up zone and the developing zone stayed comparatively stable, and the built-up zone was slightly higher than that of the developing zone. Since 2017, the area of construction in the developing zone has drastically increased, and the relative density is 2–3 times that of the built-up zone. As of 2019, it has peaked at about 54 km 2 .
Although the area in the developing zone plunged to 20 km 2 in 2020, the relative density is still greater than that in the built-up zone. Thus, it can be seen that most of the construction waste is distributed around the city, which confirms the phenomenon of “construction waste siege”. By comparing the area in the developing zone with that of the whole city, it is found that the area in the developing zone has always accounted for 40–46% of the total in the 6 years. Therefore, it can be inferred that the focus of the new urbanization in Zhengzhou City is on promoting the construction of the developing zone.

4.4. Temporal Changes of Dust-Proof Nets Coverage

The percentage of dust-proof nets coverage reflects the surveillance and remediation of construction waste. After obtaining the statistics of bare soil and dust-proof nets area, the coverage of dust nets in Zhengzhou from 2015 to 2020 was obtained, showing 66.38%, 59.57%, 12.33%, 44.70%, 80.23% and 86.12%, respectively. The proportion was around 60% before 2017 and is on the decrease, which is the result of a lack of government regulation. Especially in 2017, the proportion reached a low, derived from two causes: a surge in the overall construction waste area and a plunge in the dust net area to its lowest value of 13.99 km 2 . Since then, the whole situation has gradually improved. Even though the total area of construction waste maintained a high level during the period from 2017 to 2019, the rate of dust-proof nets was still increasing. As of 2019, the coverage of dust-proof nets has reached 80% and continues to grow to 86.12% in 2020.
Likewise, we found the proportion in the built-up zone and made a comparison, as shown in Figure 11. The coverage in the built-up zone is higher than that overall; in other words, the coverage in the developing zone is lower than that of the whole. It is consistent with our daily cognition that in remote areas, due to reasons such as weak supervision, the implementation of policies is not in place. The result also reflects the advantages of using space observation technology to implement supervision.
The proportion of dust-proof nets is related to the measurement of SDG indicator 11.6.1, the proportion of municipal solid waste collected and managed in controlled facilities out of total municipal waste generated by cities. Although we do not actually count the proportion of managed waste, it can be considered that the waste covered with dust nets is monitored by the relevant authorities, which means that the proportion can be used to some extent as part of a proxy for this indicator.

4.5. Correlation of Dust-Proof Net Coverage and Atmospheric Quality

Mulching dust-proof nets has a positive effect on reducing dust. To verify the denotation of dust-proof nets, we applied a linear regression model to analyze the correlation between dust net coverage and different types of atmospheric data for the same period. As shown in Figure 12, through regression, the proportion of dust-proof nets was found to exhibit R 2 values of 0.86574, 0.90253 and 0.90546 with AQI and the concentrations of PM2.5 and PM10, respectively.
It can be noted that the temporal change of AQI and coverage are synchronous; that is, the higher the dust-proof net coverage, the smaller the value of AQI, and thus, the better the air quality. Hence, it can be presumed that there is a positive correlation between the proportion of construction waste and urban air quality. Similarly, both the R 2 of the PM2.5 and PM10 concentrations were higher than 0.9, which can be evidence that the dust-proof nets can effectively reduce the concentration of air pollutants.

5. Discussion

Based on the analysis of the experimental results, the conclusion can be drawn that the monitoring method we proposed can effectively identify construction waste. The method can quickly process time series city-wide images and obtain information such as the location and area of construction waste with high accuracy. However, the driving factors for how the area varies were unclear. Apart from that, it remains to be discussed whether the dust-proof nets are valid for reducing dust pollution. In this section, we will attempt to analyze the temporal variation of construction waste in combination with urban development planning and discuss the effectiveness of dust-proof nets as well as the limitations of this study.

5.1. Analysis of Construction Waste Changes in the Context of Urbanization

It is known that Zhengzhou City has been carrying out new urbanization planning since 2013, and, in 2017, it rapidly increased both in speed and scale. By 2020, the reform has achieved a milestone, and the speed of construction has slowed down. Our results, the variations of construction waste, are in line with the pace of urban construction mentioned above. In 2015–2016, urbanization was slow, and the area of construction waste was less compared to the later period. In 2017, the city entered a period of rapid change, with regions vigorously carrying out construction and demolition operations, leading to the emergence of many construction sites and a surge in construction waste. This state continued until 2019 when the rapid urbanization phase came to an end. Meanwhile, the coverage rate of dust-proof nets has increased in Zhengzhou due to the enhanced measures in construction waste management since 2017, which has promoted the improvement of the city’s air quality. By 2020, most operations were completed, and the area of construction waste has thus significantly reduced, falling back to a low level.
Typically, researchers use indicators such as land-cover classification, built-up density and urban heat island intensity to monitor urbanization [50,51,52]. However, we were surprised to find that the results of construction waste extracted in this study were consistent with the urbanization process. In other words, construction waste is as valid as known indicators in urbanization monitoring, which makes it a new medium for understanding the dynamics of urbanization.

5.2. Effectiveness of Dust-Proof Nets on Ambient Air Quality

With the increasing awareness of environmental protection, air pollution problems caused by urbanization have attracted more and more attention [53]. As a natural consequence of construction projects, construction waste is one of the main sources of increasing resuspended soil particles in the air, affecting the pollution level of urban PM. The cover provided by dust-proof nets can effectively reduce the dust generated by construction waste in the process of stacking and transportation. The coefficient of determination between dust net coverage and AQI was 0.86574 in Section 4.5, indicating that there may be a strong correlation between coverage and AQI. What is more, the R 2 of PM2.5 and PM10, the two sub-indicators involved in AQI, are both greater than the R 2 of AQI (higher than 0.9), which shows a strong contribution of dust-proof nets with pollutant reduction. In other words, it may imply that other sub-indicators included in the AQI, such as SO 2 and NO 2 , are less affected by the dust nets; of course, it still needs to be further verified in the future.
Temperature, wind speed, water vapor pressure and precipitation frequency are all important factors that affect the accumulation of PM. Therefore, the method we proposed is applicable to dusty cities with a dry climate and low precipitation, such as Zhengzhou, where climate has little effect on urban deposition and is more likely to sustain the dust pollution caused by construction waste. For cities with more rain, dust pollution is much weaker, so it is less convincing to use dust-proof nets to supervise urbanization and municipal management.
In fact, PM shows a strong seasonal pattern due to factors such as winter heating [54,55]. It has also been suggested that PM pollution is related to the level of urbanization [56]. Therefore, the relationship between dust nets and atmospheric data corrected for parameters such as climate, season and urbanization level should be studied in more detail in the future.

5.3. Limitation of the Study

Spatial resolution is critical for the method proposed in this study. Finer imagery provides more surface detail but may not be suitable for all scenarios. When using GaoFen-2 images with better spatial resolution than Sentinel-2 to extract construction waste and dust-proof nets, there were a lot of misjudgments in the results, such as confusing construction waste with old house roofs, which would not appear in the 10 m resolution images. Therefore, the present method has limitations in selecting data. Exploring the reasons why the method is not suitable for finer images and improving the applicability are needed in the future.
We selected images for research based on the temporal resolution of one scene per year. However, this is not comprehensive enough. When we made a rough count of images from more time periods, we found that within a one-year time interval, there were sometimes quite different results from the present study. In fact, the rapid development of engineering may be the main reason for these variations. The current construction technology in China is mature, and the construction period of some small projects is short, so there are some fluctuations when a finer scale is considered for time. Nonetheless, the situation of construction waste in most periods is consistent with the results of this study for the same period. In future work, it is still required to conduct quarterly or monthly statistics of urban construction waste to find the period that is most appropriate to the engineering construction scenario for the study and analyze the change patterns on smaller time scales.

6. Conclusions

In this study, we proposed a multi-layer classification method for monitoring urban construction waste based on remote sensing. In particular, we identified the construction waste covered with dust-proof nets, which has a strong relationship with dust pollution in the city. Based on this method, we extracted the construction waste in Zhengzhou City from 2015 to 2020 with an accuracy of more than 91% and obtained the distribution and proportion of various construction waste. We concluded that the changes in areas and patterns in the past six years are in line with the urban construction plan, and, based on this, we presumed that the focus of urbanization in Zhengzhou is in the construction of the developing zone between the third and fourth ring roads, and more emphasis is placed on the eastern region in the suburbs. In addition, the dust-proof net’s coverage has been increasing since enhanced management was carried out in 2017 and reached more than 80% in 2019. Notably, as the urban area has better coverage than the suburban area, the mulching proportion in the urban area is actually higher than that of the total. What is more, the dust-proof net coverage is positively correlated with the urban air quality; that is, the air quality becomes better as the mulch coverage increases, which confirms the effectiveness of the dust net.
Compared with the existing methods, our study bridges the gap in large-scale monitoring studies of dust nets and serves as a proxy for SDG indicator 11.6.1 in a sense. It intuitively revealed and analyzed the relationship between the coverage of dust-proof nets and air quality. It provides a basis for environmental governance, industrial layout, urban planning and rational utilization of resources, which has important practical significance.

Author Contributions

Conceptualization, Z.L. and D.L.; methodology, L.Z. and Z.L.; software, L.Z., Z.L. and X.L.; validation, L.Z., Z.L. and D.L.; formal analysis, L.Z. and X.L.; investigation, Z.L. and L.Z.; data curation, L.Z. and H.Z.; writing—original draft preparation, L.Z.; writing—review and editing, Z.L. and Q.Z.; visualization, L.Z.; supervision, H.G.; project administration, H.G. All authors have read and agreed to the published version of the manuscript.

Funding

This research was funded by the Informatization Plan of Chinese Academy of Sciences, grant number CAS-WX2021PY-0107-02, and the National Natural Science Foundation of China, grant number 41876226.

Data Availability Statement

The data that support the findings of this study are available from the author upon reasonable request.

Acknowledgments

The authors thank the European Space Agency (ESA) and the US Geological Survey (USGS) for providing Sentinel-2 MSI and Landsat-8 OLI products, and are grateful to the anonymous reviewers for their helpful comments and suggestions to improve the quality of this paper.

Conflicts of Interest

The authors declare no conflict of interest.

References

  1. Liang, D.; Shi, L.; Zhao, J.; Liu, P.; Sarnat, J.A.; Gao, S.; Schwartz, J.; Liu, Y.; Ebelt, S.T.; Scovronick, N.; et al. Urban Air Pollution May Enhance COVID-19 Case-Fatality and Mortality Rates in the United States. Innovation 2020, 1, 100047. [Google Scholar] [CrossRef] [PubMed]
  2. Dong, F.; Yu, B.; Pan, Y. Examining the synergistic effect of CO2 emissions on PM2.5 emissions reduction: Evidence from China. J. Clean. Prod. 2019, 223, 759–771. [Google Scholar] [CrossRef]
  3. Guo, S.; Hu, M.; Levy Zamora, M.; Peng, J.; Shang, D.; Zheng, J.; Du, Z.; Wu, Z.; Shao, M.; Zeng, L.; et al. Elucidating severe urban haze formation in China. Proc. Natl. Acad. Sci. USA 2014, 111, 17373–17378. [Google Scholar] [CrossRef] [PubMed] [Green Version]
  4. Zhang, R.; Khalizov, A.; Wang, L.; Hu, M.; Xu, W. Nucleation and growth of nanoparticles in the atmosphere. Chem. Rev. 2012, 112, 1957–2011. [Google Scholar] [CrossRef] [PubMed]
  5. Hong, S.; Chen, Q.; Jia, Z.; Li, Y. Research on Remote Sensing Feature Selection and Information Extraction of Urban Construction Waste Based on Jilin-1 Image. IOP Conf. Ser.-Earth Environ. Sci. 2020, 615, 012039. [Google Scholar] [CrossRef]
  6. Liu, Y.; Huang, X.; Li, H.; Liu, Z.; Chen, C.; Wang, X. Extraction of Irregular Solid Waste in Rural based on Convolutional Neural Network and Conditional Random Field Method. J. Geo-Inf. Sci. 2019, 21, 259. [Google Scholar] [CrossRef]
  7. Li, C.; Yang, L.; Liu, X.; Yang, Y.; Qin, L.; Li, D.; Liu, G. Bridging the Energy Benefit and POPs Emission Risk from Waste Incineration. Innovation 2021, 2, 100075. [Google Scholar] [CrossRef]
  8. Chen, Q.; Cheng, Q.; Wang, J.; Du, M.; Zhou, L.; Liu, Y. Identification and Evaluation of Urban Construction Waste with VHR Remote Sensing Using Multi-Feature Analysis and a Hierarchical Segmentation Method. Remote Sens. 2021, 13, 158. [Google Scholar] [CrossRef]
  9. Liu, Y.; Ren, Y.; Wei, C.; Wang, A.; Zhou, H.; Chi, Y. Study on monitoring of informal open-air solid waste dumps based on Beijing-1 images. J. Remote Sens. 2009, 13, 320–326. [Google Scholar]
  10. Glanville, K.; Chang, H. Remote sensing analysis techniques and sensor requirements to support the mapping of illegal domestic waste disposal sites in Queensland, Australia. Remote Sens. 2015, 7, 13053–13069. [Google Scholar] [CrossRef] [Green Version]
  11. Bagheri, S.; Hordon, R.M. Hazardous waste site identification using aerial photography: A pilot study in Burlington County, New Jersey, USA. Environ. Manag. 1988, 12, 119–125. [Google Scholar] [CrossRef]
  12. Dai, J.; Yang, Q.; Lu, D.; Zhang, F. Application of remote sensing to garbage dump recognizing in Nanjing City. Geomat. Spat. Inf. Technol. 2012, 35, 127–135. [Google Scholar]
  13. Li, S.; Luo, X.; Long, Y.; Yao, T.; Liu, J.; Wang, W. Remote Sensing-based Monitoring for Spatial Expansion of Municipal Waste Landfills –A Case Study of Gaoyan Site in Guiyang City, China. Earth Environ. 2017, 45. [Google Scholar] [CrossRef]
  14. Zhou, L.; Luo, T.; Du, M.; Chen, Q.; Liu, Y.; Zhu, Y.; He, C.; Wang, S.; Yang, K. Machine Learning Comparison and Parameter Setting Methods for the Detection of Dump Sites for Construction and Demolition Waste Using the Google Earth Engine. Remote Sens. 2021, 13, 787. [Google Scholar] [CrossRef]
  15. Wu, T.; Peng, L.; Hu, Y. Informal garbage dumps detection in high resolution remote sensing images based on SU-RetinaNet. Remote Sens. Land Resour. 2020, 32, 90–97. [Google Scholar]
  16. Yan, W.Y.; Mahendrarajah, P.; Shaker, A.; Faisal, K.; Luong, R.; Al-Ahmad, M. Analysis of multi-temporal landsat satellite images for monitoring land surface temperature of municipal solid waste disposal sites. Environ. Monit. Assess. 2014, 186, 8161–8173. [Google Scholar] [CrossRef]
  17. Jia, Z.; Chen, Q.; Hong, S.; Cheng, Q. Remote Sensing Recognition of Construction Waste Accumulation Based on Resources satellite three Multi-view Stereoscopic Image. IOP Conf. Ser.-Earth Environ. Sci. 2020, 615, 012027. [Google Scholar] [CrossRef]
  18. Ministry of Ecology and Environment of the People’s Republic of China. Law of the People’s Republic of China on the Prevention and Control of Atmospheric Pollution. 2018. Available online: http://www.mee.gov.cn/ywgz/fgbz/fl/201811/t20181113_673567.shtml (accessed on 1 February 2021).
  19. Chen, Y.; Wu, Y.; Ma, J.; An, Y.; Liu, Q.; Yang, S.; Qu, Y.; Chen, H.; Zhao, W.; Tian, Y. Microplastics pollution in the soil mulched by dust-proof nets: A case study in Beijing, China. Environ. Pollut. 2021, 275, 116600. [Google Scholar] [CrossRef] [PubMed]
  20. Geng, N.; Wang, J.; Xu, Y.; Zhang, W.; Chen, C.; Zhang, R. PM2.5 in an industrial district of Zhengzhou, China: Chemical composition and source apportionment. Particuology 2013, 11, 99–109. [Google Scholar] [CrossRef]
  21. Feng, H.; Duan, G.; Zhang, R.; Zhang, W. Time series analysis of hand-foot-mouth disease hospitalization in Zhengzhou: Establishment of forecasting models using climate variables as predictors. PLoS ONE 2014, 9, e87916. [Google Scholar] [CrossRef]
  22. General Office of Zhengzhou Municipal People’s Government. Policy Interpretation of the Notice of the General Office of Zhengzhou Municipal People’s Government on the Issuance of the Implementation Plan for Strengthening Construction Waste Management and Promoting Resourceful Utilization in Zhengzhou. 2017. Available online: http://public.zhengzhou.gov.cn/interpretdepart/245483.jhtml (accessed on 15 February 2021).
  23. Phiri, D.; Simwanda, M.; Salekin, S.; Nyirenda, V.; Murayama, Y.; Ranagalage, M. Sentinel-2 Data for Land Cover/Use Mapping: A Review. Remote Sens. 2020, 12, 2291. [Google Scholar] [CrossRef]
  24. Ranagalage, M.; Morimoto, T.; Simwanda, M.; Murayama, Y. Spatial Analysis of Urbanization Patterns in Four Rapidly Growing South Asian Cities Using Sentinel-2 Data. Remote Sens. 2021, 13, 1531. [Google Scholar] [CrossRef]
  25. Hu, L.; Xu, N.; Liang, J.; Li, Z.; Chen, L.; Zhao, F. Advancing the Mapping of Mangrove Forests at National-Scale Using Sentinel-1 and Sentinel-2 Time-Series Data with Google Earth Engine: A Case Study in China. Remote Sens. 2020, 12, 3120. [Google Scholar] [CrossRef]
  26. Ministry of Ecology and Environment of the People’s Republic of China. Technical Regulation on Ambient Air Quality Index (on Trial). 2016. Available online: http://www.mee.gov.cn/ywgz/fgbz/bz/bzwb/jcffbz/201203/t20120302_224166.shtml (accessed on 12 February 2021).
  27. Department of Ecology and Environment of Henan Province. Ambient Air Quality Ranking-Composite Index Calculation Methodology. 2015. Available online: http://sthjt.henan.gov.cn/2015/03-20/996685.html (accessed on 3 February 2021).
  28. Wei, J.; Li, Z.; Xue, W.; Sun, L.; Fan, T.; Liu, L.; Su, T.; Cribb, M. The ChinaHighPM10 dataset: Generation, validation, and spatiotemporal variations from 2015 to 2019 across China. Environ. Int. 2021, 146, 106290. [Google Scholar] [CrossRef] [PubMed]
  29. Wei, J.; Li, Z.; Lyapustin, A.; Sun, L.; Peng, Y.; Xue, W.; Su, T.; Cribb, M. Reconstructing 1-km-resolution high-quality PM2.5 data records from 2000 to 2018 in China: Spatiotemporal variations and policy implications. Remote Sens. Environ. 2021, 252, 112136. [Google Scholar] [CrossRef]
  30. Xu, F.; Li, Z.; Zhang, S.; Huang, N.; Quan, Z.; Zhang, W.; Liu, X.; Jiang, X.; Pan, J.; Prishchepov, A.V. Mapping Winter Wheat with Combinations of Temporally Aggregated Sentinel-2 and Landsat-8 Data in Shandong Province, China. Remote Sens. 2020, 12, 2065. [Google Scholar] [CrossRef]
  31. Zhang, F.L.; Du, S.H.; Guo, Z. Extracting municipal solid waste dumps based on high resolution images. Guang Pu Xue Yu Guang Pu Fen Xi = Guang Pu 2013, 33, 2024–2030. [Google Scholar]
  32. Niemeyer, J.; Rottensteiner, F.; Soergel, U. Contextual classification of lidar data and building object detection in urban areas. ISPRS J. Photogramm. Remote Sens. 2014, 87, 152–165. [Google Scholar] [CrossRef]
  33. Pal, M. Random forest classifier for remote sensing classification. Int. J. Remote Sens. 2005, 26, 217–222. [Google Scholar] [CrossRef]
  34. Rodriguez-Galiano, V.F.; Ghimire, B.; Rogan, J.; Chica-Olmo, M.; Rigol-Sanchez, J.P. An assessment of the effectiveness of a random forest classifier for land-cover classification. ISPRS J. Photogramm. Remote Sens. 2012, 67, 93–104. [Google Scholar] [CrossRef]
  35. Nguyen, H.T.T.; Doan, T.M.; Radeloff, V. Applying random forest classification to map land use/land cover using Landsat 8 OLI. Int. Arch. Photogramm. Remote Sens. Spat. Inf. Sci. 2018, 42, 363–367. [Google Scholar] [CrossRef] [Green Version]
  36. Belgiu, M.; Drăguţ, L. Random forest in remote sensing: A review of applications and future directions. ISPRS J. Photogramm. Remote Sens. 2016, 114, 24–31. [Google Scholar] [CrossRef]
  37. Schowengerdt, R.A. CHAPTER 9—Thematic Classification. In Remote Sensing, 3rd ed.; Schowengerdt, R.A., Ed.; Academic Press: Burlington, UK, 2007; pp. 387–456, XXVII–XXXIII. [Google Scholar] [CrossRef]
  38. Marghany, M. Chapter 8—Mahalanobis classifier and neural network algorithms for mineral exploration. In Advanced Algorithms for Mineral and Hydrocarbon Exploration Using Synthetic Aperture Radar; Marghany, M., Ed.; Elsevier: Amsterdam, The Netherlands, 2022; pp. 225–249. [Google Scholar] [CrossRef]
  39. Richards, J.; Jia, X. Remote Sensing Digital Image Analysis; Springer: Berlin/Heidelberg, Germany, 1999. [Google Scholar] [CrossRef]
  40. Ahmad, A.; Quegan, S. Analysis of maximum likelihood classification on multispectral data. Appl. Math. Sci. 2012, 6, 6425–6436. [Google Scholar]
  41. Awad, M.; Khanna, R. Support Vector Machines for Classification. In Efficient Learning Machines: Theories, Concepts, and Applications for Engineers and System Designers; Apress: Berkeley, CA, USA, 2015; pp. 39–66. [Google Scholar] [CrossRef] [Green Version]
  42. Che, J.; Yang, Y.; Li, L.; Bai, X.; Zhang, S.; Deng, C. Maximum relevance minimum common redundancy feature selection for nonlinear data. Inf. Sci. 2017, 409–410, 68–86. [Google Scholar] [CrossRef]
  43. Pal, S.; Ziaul, S. Detection of land use and land cover change and land surface temperature in English Bazar urban centre. Egypt. J. Remote Sens. Space Sci. 2017, 20, 125–145. [Google Scholar] [CrossRef] [Green Version]
  44. García-Balboa, J.; Alba-Fernández, V.; Ariza-Lopez, F.; Rodríguez-Avi, J. Analysis of Thematic Similarity Using Confusion Matrices. ISPRS Int. J. Geo-Inf. 2018, 7, 233. [Google Scholar] [CrossRef] [Green Version]
  45. Feng, X.; Wang, S. Influence of different weather events on concentrations of particulate matter with different sizes in Lanzhou, China. J. Environ. Sci. 2012, 24, 665–674. [Google Scholar] [CrossRef]
  46. Dao, X.; Lin, Y.C.; Cao, F.; Di, S.Y.; Hong, Y.; Xing, G.; Li, J.; Fu, P.; Zhang, Y.L. Introduction to the National Aerosol Chemical Composition Monitoring Network of China: Objectives, Current Status, and Outlook. Bull. Am. Meteorol. Soc. 2019, 100, ES337–ES351. [Google Scholar] [CrossRef]
  47. Marcazzan, G.M.; Vaccaro, S.; Valli, G.; Vecchi, R. Characterisation of PM10 and PM2.5 particulate matter in the ambient air of Milan (Italy). Atmos. Environ. 2001, 35, 4639–4650. [Google Scholar] [CrossRef]
  48. Xu, W.; Jin, X.; Liu, M.; Ma, Z.; Zhou, Y. Analysis of spatiotemporal variation of PM2. 5 and its relationship to land use in China. Atmos. Pollut. Res. 2021, 12, 101151. [Google Scholar] [CrossRef]
  49. Anderson, J.R. A Land Use and Land Cover Classification System for Use with Remote Sensor Data; US Government Printing Office: Washington, DC, USA, 1976; Volume 964. [Google Scholar]
  50. Taubenböck, H.; Wegmann, M.; Roth, A.; Mehl, H.; Dech, S. Urbanization in India—Spatiotemporal analysis using remote sensing data. Comput. Environ. Urban Syst. 2009, 33, 179–188. [Google Scholar] [CrossRef]
  51. Xiong, Y.; Huang, S.; Chen, F.; Ye, H.; Wang, C.; Zhu, C. The Impacts of Rapid Urbanization on the Thermal Environment: A Remote Sensing Study of Guangzhou, South China. Remote Sens. 2012, 4, 2033–2056. [Google Scholar] [CrossRef] [Green Version]
  52. Melchiorri, M.; Florczyk, A.J.; Freire, S.; Schiavina, M.; Pesaresi, M.; Kemper, T. Unveiling 25 Years of Planetary Urbanization with Remote Sensing: Perspectives from the Global Human Settlement Layer. Remote Sens. 2018, 10, 768. [Google Scholar] [CrossRef] [Green Version]
  53. Ameen, R.F.M.; Mourshed, M. Urban environmental challenges in developing countries—A stakeholder perspective. Habitat Int. 2017, 64, 1–10. [Google Scholar] [CrossRef]
  54. Bell, M.L.; Dominici, F.; Ebisu, K.; Zeger, S.L.; Samet, J.M. Spatial and temporal variation in PM2. 5 chemical composition in the United States for health effects studies. Environ. Health Perspect. 2007, 115, 989–995. [Google Scholar] [CrossRef] [PubMed] [Green Version]
  55. Querol, X.; Alastuey, A.; Ruiz, C.; Artiñano, B.; Hansson, H.; Harrison, R.; Buringh, E.; Ten Brink, H.; Lutz, M.; Bruckmann, P.; et al. Speciation and origin of PM10 and PM2. 5 in selected European cities. Atmos. Environ. 2004, 38, 6547–6555. [Google Scholar] [CrossRef]
  56. Ji, X.; Yao, Y.; Long, X. What causes PM2. 5 pollution? Cross-economy empirical analysis from socioeconomic perspective. Energy Policy 2018, 119, 458–472. [Google Scholar] [CrossRef]
Remotesensing 14 03805 g001 550
Figure 1. Study area. (a) Location of Henan Province in China; (b) Location of the study area in Henan Province; (c) Sentinel-2 remote sensing image of Zhengzhou (22 February 2018) and delineation of the eastern/western suburbs and developing/built-up zones.
Figure 1. Study area. (a) Location of Henan Province in China; (b) Location of the study area in Henan Province; (c) Sentinel-2 remote sensing image of Zhengzhou (22 February 2018) and delineation of the eastern/western suburbs and developing/built-up zones.
Remotesensing 14 03805 g001
Remotesensing 14 03805 g002 550
Figure 2. Field investigation pictures of (a) construction waste covered with dust-proof nets, (b) bare soil covered with dust-proof nets captured by drones and (c) the process of dumping bare soil with a shovel.
Figure 2. Field investigation pictures of (a) construction waste covered with dust-proof nets, (b) bare soil covered with dust-proof nets captured by drones and (c) the process of dumping bare soil with a shovel.
Remotesensing 14 03805 g002
Remotesensing 14 03805 g003 550
Figure 3. The workflow of this study.
Figure 3. The workflow of this study.
Remotesensing 14 03805 g003
Remotesensing 14 03805 g004 550
Figure 4. Example data of features from Google Earth. (a) Bare soil; (b) Green and blue dust-proof nets; (c) Regularly arranged temporary buildings with roofs similar in color to blue dust-proof nets, commonly found on construction sites.
Figure 4. Example data of features from Google Earth. (a) Bare soil; (b) Green and blue dust-proof nets; (c) Regularly arranged temporary buildings with roofs similar in color to blue dust-proof nets, commonly found on construction sites.
Remotesensing 14 03805 g004
Remotesensing 14 03805 g005 550
Figure 5. Two subsets of construction waste samples from western Zhengzhou in Sentinel-2 images. (a) The yellow and green parts represent bare soil and dust-proof nets, respectively (22 February 2018); (b) The yellow, blue, green and purple parts represent bare soil, blue dust-proof nets, green dust-proof nets and temporary buildings, respectively (7 July 2019).
Figure 5. Two subsets of construction waste samples from western Zhengzhou in Sentinel-2 images. (a) The yellow and green parts represent bare soil and dust-proof nets, respectively (22 February 2018); (b) The yellow, blue, green and purple parts represent bare soil, blue dust-proof nets, green dust-proof nets and temporary buildings, respectively (7 July 2019).
Remotesensing 14 03805 g005
Remotesensing 14 03805 g006 550
Figure 6. The reflectance plot showing the spectral characteristic of four classes derived from the Sentinel-2 image on 7 July 2019.
Figure 6. The reflectance plot showing the spectral characteristic of four classes derived from the Sentinel-2 image on 7 July 2019.
Remotesensing 14 03805 g006
Remotesensing 14 03805 g007 550
Figure 7. Construction waste and dust-proof nets map of Zhengzhou city from 2015 to 2020. (The green, blue and yellow parts represent green dust-proof nets, blue dust-proof nets and bare soil, respectively. The temporary buildings in the classification results are not displayed after post-processing).
Figure 7. Construction waste and dust-proof nets map of Zhengzhou city from 2015 to 2020. (The green, blue and yellow parts represent green dust-proof nets, blue dust-proof nets and bare soil, respectively. The temporary buildings in the classification results are not displayed after post-processing).
Remotesensing 14 03805 g007
Remotesensing 14 03805 g008 550
Figure 8. Comparison of ground truth and classification results. (A–D) Different regions of this study area. In each group, the left is the ground truth, and the right is the prediction result.
Figure 8. Comparison of ground truth and classification results. (A–D) Different regions of this study area. In each group, the left is the ground truth, and the right is the prediction result.
Remotesensing 14 03805 g008
Remotesensing 14 03805 g009 550
Figure 9. The area of construction waste in western suburbs, eastern suburbs and the whole area from 2015 to 2020.
Figure 9. The area of construction waste in western suburbs, eastern suburbs and the whole area from 2015 to 2020.
Remotesensing 14 03805 g009
Remotesensing 14 03805 g010 550
Figure 10. The variation of construction waste in the developing zone and built-up zone, which are represented by the yellow and red line, respectively. The green line named “Reference density” is the density of construction waste in the developing zone relative to the built-up zone. A comparison of the red and green lines (with the difference in the area of partitions removed) highlights the true disparity between the two regions.
Figure 10. The variation of construction waste in the developing zone and built-up zone, which are represented by the yellow and red line, respectively. The green line named “Reference density” is the density of construction waste in the developing zone relative to the built-up zone. A comparison of the red and green lines (with the difference in the area of partitions removed) highlights the true disparity between the two regions.
Remotesensing 14 03805 g010
Remotesensing 14 03805 g011 550
Figure 11. The coverage of dust-proof nets in the built-up zone and the whole area from 2015 to 2020.
Figure 11. The coverage of dust-proof nets in the built-up zone and the whole area from 2015 to 2020.
Remotesensing 14 03805 g011
Remotesensing 14 03805 g012 550
Figure 12. The linear regression models of (a) air quality index, (b) concentration of PM2.5, (c) concentration of PM10 and the dust-proof net’s coverage. The dashed lines denote the best-fit lines from linear regression.
Figure 12. The linear regression models of (a) air quality index, (b) concentration of PM2.5, (c) concentration of PM10 and the dust-proof net’s coverage. The dashed lines denote the best-fit lines from linear regression.
Remotesensing 14 03805 g012
Table
Table 1. Main characteristics of the satellites and bands used in this study *.
Table 1. Main characteristics of the satellites and bands used in this study *.
SatelliteLandsat-8Sentinel-2
Spatial Resolution (m)3010/20
Available BandsBand 2 (Blue: 0.49–0.51  μ m)
Band 3 (Green: 0.53–0.59  μ m)
Band 4 (Red: 0.64–0.67  μ m)
Band 5 (NIR: 0.85–0.88   μ m)
Band 6 (SWIR: 1.57–1.65  μ m)
Band 7 (SWIR: 2.11–2.29  μ m)		Band 2 (Blue: 0.46–0.52  μ m)
Band 3 (Green: 0.54–0.58  μ m)
Band 4 (Red: 0.65–0.69  μ m)
Band 5 (Red Edge: 0.70–0.71  μ m)
Band 6 (Red Edge: 0.73–0.75  μ m)
Band 7 (Red Edge: 0.77–0.79  μ m)
Band 8 (NIR: 0.79–0.90  μ m)
Band 8A (Red Edge: 0.86–0.88  μ m)
Band 11 (SWIR: 1.57–1.66  μ m)
Band 12 (SWIR: 2.10–2.29  μ m)
Image Acquisition Time14 September 2015
2 October 2016		27 February 2017
22 February 2018
7 July 2019
6 July 2020
* The Available Bands are candidates for feature selection, which will be discussed in detail in Section 3.4.1 for the suitability for solid waste extraction. The Spatial Resolution indicates the resolution of the Available Bands.
Table
Table 2. Number of construction waste samples from 2015 to 2020.
Table 2. Number of construction waste samples from 2015 to 2020.
Image Acquisition TimeBare SoilGreen Dust-Proof NetsBlue Dust-Proof NetsTemporary Buildings
September 201512,7626570——
October 2016692114,238——
February 201768278728——
February 2018619314,173——
July 2019678415,490454511,558
July 20203143171541307852
Table
Table 3. Accuracy and Time-Consuming Comparison of Four Classifiers Derived from Sentinel-2 (7 July 2019).
Table 3. Accuracy and Time-Consuming Comparison of Four Classifiers Derived from Sentinel-2 (7 July 2019).
ClassifierOverall Accuracy (%)Kappa CoefficientTime Consumption (s)
Parallelepiped Classification86.830.818
Mahalanobis Distance Classification95.770.9426
Maximum Likelihood Classification97.150.9635
Support Vector Classification97.730.97558
Table
Table 4. Band standard deviation ranking derived from Sentinel-2 (7 July 2019) and Landsat-8 (2 October 2016), respectively.
Table 4. Band standard deviation ranking derived from Sentinel-2 (7 July 2019) and Landsat-8 (2 October 2016), respectively.
RankSentinel-2Landsat-8
1Band 12—SWIR-2B7—SWIR-2
2Band 11—SWIR-1B6—SWIR-1
3Band 8—NIRB5—NIR
4Band 8A—Red EdgeB4—Red
Table
Table 5. The overall accuracy and Kappa coefficient of MLC under different minimum similarity threshold.
Table 5. The overall accuracy and Kappa coefficient of MLC under different minimum similarity threshold.
Threshold of Minimum SimilarityOverall Accuracy ( %)Kappa Coefficient
0.0291.320.88
0.0192.630.90
0.00594.100.91
0.00495.060.93
0.00195.130.93
Table
Table 6. The integrated atmospheric data corresponding to the time of construction waste extraction *.
Table 6. The integrated atmospheric data corresponding to the time of construction waste extraction *.
Image Acquisition TimePM10 ( μ g/m 3 )PM2.5 ( μ g/m 3 )AQI
14 September 201579.6642.366.84
2 October 201697.1553.105.49
27 February 2017148.9689.288.79
22 February 2018135.8880.216.84
7 July 201961.4526.954.48
6 July 202066.4432.533.80
* The AQI is monthly data, while PM10 and PM2.5 concentrations are averaged over the thirty days following the time of image acquisition.
Table
Table 7. Classification Accuracy Assessment Results Based on the Validation Samples.
Table 7. Classification Accuracy Assessment Results Based on the Validation Samples.
Image Acquisition TimeClassProd. Acc.User Acc.CommissionOmissionOverall AccuracyKappa Coefficient
Bare soil99.23%100.00%0.000.77
14 September 2015Green dust-proof nets94.16%100.00%0.255.84 97.35% 0.94
Blue dust-proof nets————
Temporary buildings————
Bare soil97.67%100.00%0.002.33
2 October 2016Green dust-proof nets95.40%100.00%0.004.60 96.04% 0.91
Blue dust-proof nets————
Temporary buildings————
Bare soil98.68%99.72%0.281.32
27 February 2017Green dust-proof nets98.75%100.00%0.001.25 98.72% 0.97
Blue dust-proof nets————
Temporary buildings————
Bare soil99.51%100.00%0.000.49
22 February 2018Green dust-proof nets98.09%99.97%0.031.91 98.50% 0.96
Blue dust-proof nets————
Temporary buildings————
Bare soil90.70%100.00%0.009.30
7 July 2019Green dust-proof nets93.42%99.04%0.966.58 95.24% 0.93
Blue dust-proof nets98.64%85.20%14.801.36
Temporary buildings98.13%99.84%0.161.87
Bare soil91.78%100.00%0.008.22
6 July 2020Green dust-proof nets92.08%91.37%8.637.92 91.82% 0.89
Blue dust-proof nets88.71%99.69%0.3111.29
Temporary buildings94.96%98.34%1.665.04
Publisher’s Note: MDPI stays neutral with regard to jurisdictional claims in published maps and institutional affiliations.

Share and Cite

MDPI and ACS Style

Li, Z.; Guo, H.; Zhang, L.; Liang, D.; Zhu, Q.; Liu, X.; Zhou, H. Time-Series Monitoring of Dust-Proof Nets Covering Urban Construction Waste by Multispectral Images in Zhengzhou, China. Remote Sens. 2022, 14, 3805. https://doi.org/10.3390/rs14153805

AMA Style

Li Z, Guo H, Zhang L, Liang D, Zhu Q, Liu X, Zhou H. Time-Series Monitoring of Dust-Proof Nets Covering Urban Construction Waste by Multispectral Images in Zhengzhou, China. Remote Sensing. 2022; 14(15):3805. https://doi.org/10.3390/rs14153805

Chicago/Turabian Style

Li, Zilu, Huadong Guo, Lu Zhang, Dong Liang, Qi Zhu, Xvting Liu, and Heng Zhou. 2022. "Time-Series Monitoring of Dust-Proof Nets Covering Urban Construction Waste by Multispectral Images in Zhengzhou, China" Remote Sensing 14, no. 15: 3805. https://doi.org/10.3390/rs14153805

APA Style

Li, Z., Guo, H., Zhang, L., Liang, D., Zhu, Q., Liu, X., & Zhou, H. (2022). Time-Series Monitoring of Dust-Proof Nets Covering Urban Construction Waste by Multispectral Images in Zhengzhou, China. Remote Sensing, 14(15), 3805. https://doi.org/10.3390/rs14153805

Note that from the first issue of 2016, this journal uses article numbers instead of page numbers. See further details here.

Article Metrics

Back to TopTop