Next Article in Journal
A Reconstructed Method of Acoustic Logging Data and Its Application in Seismic Lithological Inversion for Uranium Reservoir
Next Article in Special Issue
Extracting and Evaluating Urban Entities in China from 2000 to 2020 Based on SNPP-VIIRS-like Data
Previous Article in Journal
Smooth GIoU Loss for Oriented Object Detection in Remote Sensing Images
Previous Article in Special Issue
Automatic Extraction of Urban Impervious Surface Based on SAH-Unet
 
 
Search for Articles:
Title / Keyword
Author / Affiliation / Email
Journal
Article Type
 
 
Advanced Search
 
Section
Special Issue
Volume
Issue
Number
Page
 
Journals
Remote Sensing
Volume 15
Issue 5
10.3390/rs15051261
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

The Differences and Influence Factors in Extracting Urban Green Space from Various Resolutions of Data: The Perspective of Blocks

by
Xiao Wei
,
Mengjun Hu
and
Xiao-Jun Wang
*
Department of Landscape Architecture, School of Architecture, Southeast University, Nanjing 210096, China
*
Author to whom correspondence should be addressed.
Remote Sens. 2023, 15(5), 1261; https://doi.org/10.3390/rs15051261
Submission received: 30 December 2022 / Revised: 21 February 2023 / Accepted: 23 February 2023 / Published: 24 February 2023
(This article belongs to the Special Issue Urban Resilience with Remote Sensing - Observation, Measurement, Evaluation and Applications)
Browse Figures
Review Reports Versions Notes

Abstract

:
The appropriate resolution has been confirmed to be crucial to the extraction of urban green space and the related research on ecosystem services. However, the factors affecting the differences between various resolutions of data in certain application scenarios are lacking in attention. To fill the gap, this paper made an attempt to analyze the differences of various resolutions of data in green space extraction and to explore where the differences are reflected in the actual land unit, as well as the factors affecting the differences. Further, suggestions for reducing errors and application scenarios of different resolutions of data in related research are proposed. Taking a typical area of Nanjing as an example, data taken by DJI drone (0.1 m), GaoFen-1 (2 m) and Sentinel-2A (10 m) were selected for analysis. The results show that: (1) There were minimal differences in the green space ratio of the study area calculated by different resolutions of data on the whole, but when subdivided into each land use type and block, the differences were obvious; (2) The function, area and shape of the block, as well as the patch density and aggregation degree of the internal green space, had a certain impact on the differences. However, the specific impact varied when the block area was different; and (3) For the selection of the data source, the research purpose and application scenarios need to be comprehensively considered, including the function and attributes of the block, the distribution characteristics of green space, the allowable error limits and the budget. The present study highlighted the reasons of differences and hopefully it can provide a reference for the data selection of urban green space in the practical planning and design.

1. Introduction

Urban green space (UGS) is a vital part of urban ecosystems [1,2]. Green space can regulate the urban climate and mitigate the heat island effect through shading and transpiration of plants [3,4,5]. At the same time, it is able to substantially reduce air pollution while relieving related respiratory and cardiovascular diseases [6,7]. UGS can also promote the mental health of the public, reduce stress reactions, and improve social communication [8,9,10]. In order to achieve sustainable development of urban systems, rational planning of the UGS system is essential [11].
With the rapid development of technology, multiple data sources have become available for UGS research, including remote sensing data, aerial photography, mobile terrestrial sensors and LiDAR [12,13,14,15]. Using various data to extract and map vegetation has emerged as a fundamental research for related study [16,17,18,19]. The extraction mainly depends on the spectral, textural, geometric, contextual features and 3D features of vegetation [20,21,22,23], in which some features may require certain types of data to achieve accurate classification. For example, the textural features are more applicable with high-resolution data [24] and 3D features can be only obtained from LiDAR or terrestrial sensors like Google Street View [25]. Based on the distinguished features, supervised learning has been adopted as the mainstream method for classification, such as support vector machine, decision tree, random forest [26,27,28]. Recently, deep learning has emerged as a promising strategy to perform automated feature extraction in a supervised manner [17]. Additionally, multi-temporal images analysis combined with plant phenological characteristics has been extensively adopted, which can accurately distinguish vegetation types and even species [29,30,31]. However, the technical content of the aforementioned research is relatively high, and non-professionals will encounter difficulties during operation. Besides, available data sources may not always be satisfactory. Therefore, investigating how to reduce errors as much as possible and improving the accuracy of green space extraction under the condition of limited resources are critical issues that should be explored.
As revealed by numerous studies, there would be differences in UGS extracted from data with different resolutions, leading to contradictory results in relevant studies. Qian et al. found that the dynamic change of green space across years could only be observed in high-resolution images, because the percentage of green space would be greatly underestimated by medium resolution data [32]. Rioux et al. also concluded that the green space ratio was generally underestimated at coarser spatial resolutions, as well as the potential supply and flow of ecosystem services [33]. Holt et al. observed that the identification of key sites of urban ecosystem services depended on the spatial resolution of map units [1]. Zhou and Yan identified that the correlation between the spatial configuration metrics of green space and land surface temperature (LST) was stronger as the size of the analytical unit increased, but the influence of vegetation cover decreased [34,35]. Further, Li et al. indicated that the correlation at various resolutions differed not only in magnitude, but also in significance and direction [36]. In addition, complex relationships were also found between UGS and human health under different resolutions. As observed by Wood et al., smaller pocket parks that could not be identified by low-resolution data were significantly related to the improvement of mental health [37]. Reid et al. reported insignificant associations with individual health outcomes when using MODIS greenness data, but found significant associations for the same health outcomes with higher resolution data [38]. Browning et al. demonstrated that the academic performance of third-grade students had positive associations with low-resolution NDVI measures around schools and in neighborhoods [39].
Most of the aforementioned studies indicated that low-resolution data would ignore details, causing the absence or weakening of correlation. However, high spatial resolution data often come at the cost of low spectral resolution, which may cause the absence of important identification features [40,41]. Meanwhile, the complex background information in urban environment and high heterogeneity of vegetation structure further exacerbates the difficulty of green space extraction [42,43]. Therefore, in consideration of various research purposes, funding costs and other issues, different resolutions may have their own application scenarios. Lin et al. demonstrated that the spatial resolution of UGS maps significantly influenced habitat fragmentation and connectivity, in which high spatial resolution maps are appropriate for analyzing highly heterogeneous urban areas, and medium spatial resolution maps are more applicable to urban periphery and suburban areas with larger UGS patches and less fragmentation [44]. For the planning of UGS system, both overall large-scale planning and detailed fine-scale design are required [45], which is why the selection of appropriate data needs further research.
There are several studies comparing the UGS extraction from the functional perspective to quantify multiple ecosystem services [46,47], while few studies pay further attention to the block perspective. On the one hand, the block is the basic unit of various functional zones in urban planning and design [48]. On the other hand, it reflects the urban fabric and is closely related to the urban thermal environment [3,49]. Therefore, it is essential to investigate the differences of green space extraction with data of various resolutions and the influence factors from the perspective of blocks. Taking a typical area in Nanjing, China as an example, data from the DJI drone (0.1 m), GaoFen-1 (2 m) and Sentinel-2A (10 m) were selected to analyze the differences in extracting green space. Two kinds of division methods were used for comparison of the differences, fishnet in the same size and blocks classified by land use types, respectively. Through the use of the confusion matrix, the differences of various resolutions were analyzed to further investigate the influence factors. Based on the results, the strategy for reducing the error and the applicable scenarios of data with different resolutions were proposed, with the hope of providing a reference for data selection in UGS-related research.

2. Materials and Methods

2.1. Study Area

Nanjing (31°14′–32°37′N, 118°22′–119°14′E) is a megacity in the Yangtze River Delta Economic Development Zone of China with a history of over 1800 years. The total area of Nanjing is approximately 6587 km2 and the population of permanent residents is about 9.3 million (Nanjing Statistical Yearbook, 2020). Located in the subtropical monsoon climate zone with four distinct seasons and abundant rainfall, the mean temperature of Nanjing was 17.1 °C and the total precipitation was 1294.4 mm in 2020 (Nanjing Statistical Yearbook, 2020). In the present study, an area of 11.46 km2 was selected in the northwest of the city center for investigation (shown in Figure 1a). As a multifunctional area, residential, educational, municipal, commercial, recreational and historical land types were covered, which is representative for the study of green space in terms of different land use types.

2.2. Data Source and Preprocessing

Three resolutions of data were selected for the study, which were all obtained in sunny and cloudless weather from June to August 2022. The specific information of the selected data is shown in Table 1. GF1 and S2A are remote sensing images, and DJI is a visible light image taken by DJI M300 RTK, a UAV equipped with a high-resolution camera. Due to the limited flight altitude and shooting range of the UAV, the images of the study area were taken in three days from 10 a.m. to 4 p.m., and the final visible light image of the whole study area was obtained by combining images from the three days. The two kinds of remote sensing data were preprocessed, which included radiometric calibration, atmospheric correction, geometric correction and research area clipping.

2.3. Division of Land Units

In the present study, two methods were used to divide the land units. The first method involved using ArcGIS fishnet tools to divide the study area into multiple grids of same size. Subsequently, the size of the fishnet was altered from 10 m to 300 m to obtain different units. Then, the mean green space ratio (GR) in each grid from the three different resolutions of data were calculated. The second method involved dividing the study area into blocks according to the actual land use types classified by the code for classification of urban land use and planning standards of development land (GB50137-2011) in China. As a result, the study area was divided into 10 categories and 776 blocks, as shown in Figure 1b. The specific definition of each land use type, the number of blocks and corresponding area ratio in the total study area are shown in Table 2.

2.4. Extraction Method of UGS

Since the image taken by DJI only included visible light bands of red, green and blue, the UGS was extracted through the Green Leaf Index (GLI) [50]. The calculation formula was as follows (1):
GLI = (Green × 2 − Red − Blue)/(Green × 2 + Red + Blue)
where Red, Green and Blue correspond to the values of red, green and blue bands of the DJI images, respectively. For GF1 image, Red, Green and Blue correspond to the values of Band 3, 2 and 1, respectively; and for the S2A image, they correspond to Bands 4, 3 and 2, respectively.
Further, the vegetation coverage (VC) was calculated according to the pixel dichotomy [51]; the calculation formula was as follows (2):
VC = (GLI − GLImin)/(GLImax − GLImin)
where GLI is the actual GLI value of the grid; and GLImin and GLImax are the minimum and maximum values of GLI in the whole study area, respectively. In the present study, values with a cumulative probability of 5% and 95% were selected as the minimum and maximum values of GLI through the quick stat function in ENVI software.
In order to calculate the GR, the UGS inside the land unit should be extracted first based on a certain threshold, for which VC may not be the same as the GR in the actual situation. That threshold was selected by repeated comparison with the high-resolution image of DJI as well as the Baidu Map (https://map.baidu.com (accessed on 1 September 2022)) until the accuracy of the extracted UGS was high enough; and the evaluation of accuracy was to select 1000 points randomly and judge whether the extracted UGS was right or not until the Kappa Coefficient was higher than 85%. Finally, the thresholds were 60%, 70% and 73% in DJI, GF1, S2A, respectively, so the area where VC was higher than the threshold was identified as UGS.
Due to the GLI value of green space and water being close to each other for their similar colors, the rivers without vegetation covered in the study area were erased from the UGS maps to avoid error. The rivers were identified by the index of NDWI, which was calculated as follow (3):
NDWI = (Green − Nir)/(Green + Nir)
where Green and Nir correspond to the values of Band 2 and 4 in GF1 image, respectively, and for the S2A image, they correspond to Bands 3 and 8, respectively. The area that NDWI > 0.02 was identified as rivers. The rivers in the DJI image were identified based on the GF1 image along with manual calibration according to the high-resolution visible image of DJI. The selection of the threshold and the evaluation of accuracy were the same as UGS mentioned above.
The threshold may indeed not be the same when the images were collected in various conditions, such as sensor types, acquisition time, soil conditions and so on. Different thresholds may impact the result of UGS extraction, but the accuracy evaluation has been conducted to make sure the threshold was suitable in this context so it can still provide references for other studies.
Finally, the GR of each land unit was calculated as follows (4):
GR = Areag/Areal
where Areag is the UGS area in each land unit; and Areal is the area of the corresponding land unit.

2.5. Difference Analysis of Green Space Extraction between Data with Various Resolutions

In order to further compare the differences of GR extracted from images with different resolutions, the GR was divided into five grades from low ratio to high ratio according to a grade of 20%, as shown in Table 3. Through the use of the confusion matrix [52], taking GF1 as the benchmark, the differences in GR grades between GF1 and DJI and that between GF1 and S2A in different land units were compared. The rows and columns of the confusion matrix (5) corresponded to the GR grades extracted from GF1 and those extracted from DJI or S2A, respectively. The n in the matrix represented the number of the grades, which was 5 in the present study; Sij refers to the area that is the i-th grade in GF1 but divided into the j-th grade in DJI or S2A; and S is the total area of the study area.
S 11 ⋯ S 1 j ⋯ S 1 n ⋮ ⋱ ⋮ ⋱ ⋮ S i 1 ⋯ S ij ⋯ S in ⋮ ⋱ ⋮ ⋱ ⋮ S n 1 ⋯ S nj ⋯ S nn
The precision index was calculated to show the differences of each GR grade between different data according to (6):
Precision = Sij/(S1j + … + Sij +… + Snj)
in which i = j, namely, the area belonged to the i-th grade in both GF1 and DJI (or S2A) divided by all area that belonged to the i-th grade in DJI (or S2A). The degree of difference (DD) was the index to measure the overall differences between various resolutions of data. The calculation method was as follows (7):
DD = 1 − (S11 + …+ Sij + … + Snn)/S
To analyze whether the differences in GR were related to land use types, SPSS software was used to conduct Chi-Square test on the aforementioned two variations. Further, to investigate the influence factors of the GR difference between various resolutions, the area and shape index of each block, the patch density and aggregation degree of UGS inside each block (based on the UGS extracted from GF1) were selected as factors, of which the specific calculation method is shown in Table 4. Pearson correlation analysis was conducted between the factors and the absolute value of the GR differences between GF1 and DJI, as well as that between GF1 and S2A.

3. Results

The GR values of the study area extracted from DJI, GF1 and S2A data were 23.84%, 24.04% and 24.40% respectively, with the maximum difference of 0.56%, indicating that there were small differences in the GR of the total study area between different resolutions of data, which provided a guarantee for subsequent research. In the following sections, the analysis focuses on where the differences were reflected and the reasons for that.

3.1. The Differences of GR in Fishnet with Various Sizes

3.1.1. The Precision of Each GR Grades in Fishnet with Various Sizes

The precision of each GR grade between GF1 and DJI as well as that between GF1 and S2A in fishnet with various sizes is shown in Figure 2. For the comparison between GF1 and DJI (Figure 2a), L_GR and SL_GR had higher precision, while SH_GR and H_GR had lower precision. With the increase of fishnet size, the precision of L_GR and M_GR was slightly increased, the precision of SL_GR was greatly increased, while the precision of SH_GR and H_GR were decreased. For the comparison between GF1 and S2A (Figure 2b), the precision of L_GR and H_GR were generally higher than other grades. With the increase of fishnet size, the precision of SL_GR, M_GR and SH_GR all performed as upward trend with a few turning after 100 m fishnet. It should be noticed that the precision was 0 in some cases, because there was no area divided into that grade, such as there was no H_GR in 300 m fishnet, which was reasonable for the large size of the fishnet.
In summary, the precision between GF1 and S2A was higher than that between GF1 and DJI. With the increase of fishnet size, the precision of all GR grades between GF1 and S2A was increased to a certain extent, while only the precision of L_GR, SL_GR and M_GR between GF1 and DJI presented an upward trend.

3.1.2. The DD in Fishnet with Various Sizes

The DD between three resolutions of data calculated by the confusion matrix is shown in Figure 3. With the increase of the fishnet size, the DD was generally decreasing, and the DD between GF1 and S2A was smaller than that between GF1 and DJI. The minimum DD between GF1 and DJI was in the fishnet of 300 m (20.83%), and the minimum DD between GF1 and S2A was in the fishnet of 210 m (11.06%).
Taking the confusion matrices of 150 m and 180 m fishnets (Figure 4) and the spatial distribution of GR grades in a typical region (Figure 5) as examples to show the difference in GR grades. The darker the color, the higher the accuracy of the two types of data in such a GR grade in Figure 4. The comparison between GF1 and DJI (Figure 4a,c) showed that most of the GR grades identified by DJI were lower than GF1, especially with the increase in the fishnet size, with such findings being more obvious in H_GR, while most of GF1 and S2A were in the same GR grade and the highest identification accuracy was in L_GR. The spatial distribution of GR grades was more intuitive (Figure 5). GR grades identified by DJI (Figure 5b,f) were lighter than GF1 (Figure 5c,g) and S2A (Figure 5d,h), which means the GR grades identified by DJI were lower than the other two, while the consistency of GR grades identified by GF1 and S2A were relatively higher.

3.2. The Differences of GR in Blocks with Different Land Use Types

3.2.1. The GR of Different Land Use Types

The mean GR of each land use type and the Chi-Square test therebetween is shown in Table 5. The results show that p < 0.001 for all three kinds of data, indicating that GR was considerably significant between different land use types.
Except for Type H and Type J, the GR of other land use types extracted from GF1 and S2A was basically the same or had a little difference, while GR from DJI was generally higher than the former two. In the park land (Type H), the GR from DJI was the same as that from GF1 but higher than S2A. In the water area (Type J), the GR values from DJI were significantly smaller than that from GF1 and S2A.

3.2.2. DD in Blocks with Different Land Use Types

The Chi-Square tests of the D_GF1_DJI and D_GF1_S2A between different land use types exhibited significant differences (p < 0.001), indicating that the differences between various resolutions were also remarkably influenced by land use types. Further, through the use of the confusion matrix, the DD between land use types was compared (Figure 6). In general, the DD between GF1 and DJI was higher than that between GF1 and S2A. Except for industrial land (Type D) and logistical land (Type E), the order of DD among other land use types of GF1 and DJI was basically the same with that of GF1 and S2A, among which the DDs in commercial (Type C) and other land (Type J) were the lowest, while the differences between park land (Type H) and water area (Type J) were the highest, indicating that the two types had more errors in the identification of UGS.
The park land (Type H) and water area (Type J) were taken as examples to observe the specific differences in GR grades from the confusion matrix, as shown in Figure 7. Comparing GF1 with DJI in park land (Figure 7a), in the L_GR, SL_GR and M_GR, the GR grades identified by DJI were higher, while in SH_GR and H_GR, the GR grades identified by DJI were lower. As for GF1 with S2A in park land (Figure 7b), the consistency of each grade was higher compared with GF1 and DJI, but there were still partially differentiated that a few GR grades of S2A were higher in L_GR, SL_GR, and lower in M_GR, SH_GR and H_GR. In the water area, GF1 was much lower than DJI (Figure 7c), except for the case in L_GR. Comparing GF1 with S2A (Figure 7d), the differences in SH_GR and H_GR were the greatest, where the GR grades of S2A were lower than GF1.

3.2.3. Influence Factors of the Differences between Various Resolutions

In order to further explore the reasons of the differences, four influencing factors related to the block and UGS were selected for analysis of the correlations therebetween. The areas of the 776 blocks considerably varied, the minimum being 71.84 m2 and the maximum being 238,813.59 m2. As such, all blocks were divided into four groups according to the area, namely, less than 1 ha (100 m × 100 m), 1~4 ha (200 m × 200 m), 4~9 ha (300 m × 300 m) and larger than 9 ha, so as to correspond to the above 10~300 m fishnet size, and in consideration of the fact that 100 m is the most pedestrian and humanized block size [53]. As a result, there were 510 blocks with an area less than 1 ha, with the total area accounting for 17.10% of the study area; 199 blocks with an area between 1~4 ha, with the total area accounting for 35.20%; 46 blocks with an area between 4~9 ha, with the total area accounting for 22.40%; and 21 blocks with an area of more than 9 ha, with the total area accounting for 25.31%. All blocks were grouped for correlation analysis, and the results are shown in Table 6.
When the block area was less than 1 ha, there were significant correlations between D_GF1_DJI and the four factors, which also existed between D_GF1_S2A and the factors of UGS. A negative correlation between the block area and the difference was found, indicating that the smaller the block area, the greater the difference. There was a significant positive correlation between the shape index and the difference value, which indicated that more elongated the shape of the block, the greater the difference value. At the same time, the significant positive correlation between the difference and patch density as well as the aggregation index of UGS in the block revealed that more UGS patches and more clustered distribution in the block, the greater the difference.
When the block area was between 1~4 ha, D_GF1_DJI had a significant positive correlation with the shape index of the block (r = 0.254) and the patch density of UGS (r = 0.182), while no significant correlation was found between D_GF1_S2A and all factors. Similarly, D_GF1_DJI and D_GF1_S2A had no significant correlation with all factors when the block area was in the range 4~9 ha. When the block area was larger than 9 ha, D_GF1_S2A had a significant positive correlation with the patch density (r = 0.445). D_GF1_DJI had a negative correlation with the aggregation index of UGS (r = −0.542), which was the opposite when the block area was smaller than 1 ha, indicating that when the block area reached a certain range (9 ha in the present study), the greater the aggregation degree of UGS within the block, the smaller the difference between GF1 and DJI.
In general, the characteristics of the block and the distribution of UGS had a significant impact on the difference of UGS identified by data with different resolutions, and the effect was distinguished when the block area was different. The area had a significant negative impact on the difference when the block area was less than 1 ha. The influence of the shape index of the block on the difference was mainly reflected in the case where the area was less than 4 ha. The patch density of green space was positively related with the differences when the block area was less than 4 ha and higher than 9 ha. The aggregation index of UGS had a positive impact when the block area was less than 1 ha, while it was negative when the block area was greater than 9 ha.

4. Discussion

4.1. Differences in UGS Extraction from Different Resolutions of Data in Terms of Blocks

The requirements of data resolution for UGS research depend on the purpose of the research and the subsequent application scenarios. For reasonable UGS system planning, the focus needs to be placed not only on the large-scale distribution characteristics, but also on how to interact and communicate with water systems, roads, buildings and neighborhoods at a fine scale [54], which requires data with a certain resolution to identify the relationship between green space and the surrounding environment [32]. The focus of previous studies on the differences in UGS extraction from different resolutions of data were mostly on comparing the total GR or the correlation with other factors [32,35,38], and there was a scarcity of research on where the differences were reflected. Based on practical situations, in the present study, attempts were made to analyze the differences between different resolutions in UGS extraction in each block and the reasons for the differences. In order to confirm the reliability of the results and for the detailed comparison, both fishnet in the same size and the blocks classified by land use types were used for the division of land unit. The fishnet is a common division method used in related research [55,56]. Changing the size of the fishnet is equivalent to only taking the area size as the variable and excluding the interference of other factors as much as possible, while division according to the land use types takes the function, area, shape of each block into account, and is more applicable for the practice of landscape planning and design.
By varying the size of the fishnet, the differences among the data were found to decrease gradually with the increase of the fishnet size. The DD decreased with the increase of the fishnet size until 120 m when a fluctuation appeared. To a certain extent, such a variation pattern was consistent with the analysis of the correlation between block area and difference value in Section 3.2.3. When the block area was less than 1 ha, corresponding to the situation when the fishnet was less than 100 m, there was a significant negative correlation between block area and difference value. When the block area was between 1 and 9 ha, corresponding to the fishnet size being between 100 and 300 m, the variation of the DD was relatively complex and had no significant correlation (shown in Figure 3). Such results indicated that analysis of the differences in UGS extraction by block is reasonable. Further, because the study area is representative and covers different land use types and block sizes, the conclusions have a certain value and are expected to provide a reference for city-wide studies.

4.2. Factors Affecting the Differences in UGS Extraction with Different Resolutions of Data

The differences in green space extraction with different resolutions were shown to be correlated with the functional attributes of the blocks, and the distribution characteristics of UGS. Firstly, the Chi-Square test demonstrated that there was a significant difference between the GR of different land use types, as well as the GR differences between different resolutions of data, which is rarely mentioned in previous research.
The differences were not only reflected in the function, but also in the size and shape of the block and the distribution characteristics of the UGS inside the block. Such results indicate that the smaller the block size, the higher the differences, which was more pronounced in the case of blocks smaller than 1 ha. On the one hand, it indicates that the differences of GR in small blocks are larger. On the other hand, there may be a certain tolerance of block area in the calculation of GR. When the area reached a certain level, the differences between various resolutions would be weakened. Such findings also correspond to the results that the total GR extracted from the data of the three resolutions did not significantly differ (maximum difference of 0.58%), but the differences were more obvious when subdivided into each type of block (Figure 6 and Figure 7). In addition, the difference of GR in the total study area was relatively low compared with previous research. Qian et al. calculated that the coverage of the green space in Dongcheng District of Beijing was 7.19 % from low-resolution data (30 m) and 30.19% from high-resolution data (2.5 m) [32]. Such a difference may be attributed to the fact that the two kinds of resolution used in Qian’s research differ too much in magnitude compared with this study.
As for the shape of the block, the more elongated the shape, the greater the difference when the block area was less than 4 ha. Combined with the actual situation, there was a section of Ming City Wall in the study area, which is a linear heritage conservation land that can be well identified by high-resolution data, but the afforestation on both sides of the Wall is relatively good so that GF1 and S2A cannot identify the Wall separately because it is too narrow. Instead, the UGS on both sides of the Wall were identified as connected along with the Wall area, causing the difference. Similarly, it could also explain why the mean GR of water area (Type J) extracted from GF1 and S2A was much higher than that from DJI (see Table 5), as most of the water bodies in the study area are narrow rivers with abundant tree cover along the riverside.
Assuming that the allowable error limit between different resolutions is 0.5%, according to the linear regression equation between the area and shape of the block and the difference value, a conclusion can be drawn that the area of the block should be larger than 0.42 ha and the shape index should be less than 2.11. If there are blocks with a small area or elongated shape, specific ways for separate identification should be adopted, such as using high-resolution data, or confirming by means of field research to ensure the accuracy of the results.
In regard to the distribution characteristics of UGS inside the blocks, both the patch density and the degree of aggregation had significant effects on the differences, but the effects were varied at different block areas. In summary, the difference between GF1 and DJI and that between GF1 and S2A were higher when the UGS was densely distributed and aggregated in the relatively small block. However, when the block area was more than 9 ha, the difference between GF1 and DJI would be lower when the UGS was more aggregated, and the difference between GF1 and S2A would be higher when the UGS was more densely distributed. It was indicated that high-resolution data were more convincible and suitable when the UGS was aggregated while the coarse resolution data may cause more errors. The explanation might be that the high-resolution data could clearly identify the boundaries of individual green space patches and the gaps therebetween, while the low-resolution data could identify multiple clustered patches together and the gaps therebetween were ignored, with smaller green space patches not being identified (as shown in Figure 8).

4.3. Different Resolutions of Data Have Their Own Application Scenarios

In the present study, three kinds of data were selected to compare the difference in UGS extraction. It turned out that the differences between GF1 and S2A were relatively small. The advantage of those kinds of remote sensing data is the wide coverage, but such data are limited by time and weather conditions [57]. The high-resolution data taken by DJI drone are limited by the shooting height, resulting in the flight time being too long for large sites, leading to a part of UGS under the shadow of certain high-rise buildings not being identified. However, the advantages of high-resolution data are obvious, as the border of UGS and even the edge of individual trees, shrubs and forest canopy gaps can be accurately identified. Moreover, such data are more flexible in time and there is a certain time continuity.
Since data with different resolutions have pros and cons, the research purposes and application scenarios need to be comprehensively considered when selecting data, including the function, area and shape of the site, the distribution characteristics of the internal UGS, the allowable error limits, and the budget. For small sites or elongated sites, the parkland or sites with large numbers of green space densely distributed, or sites requiring fine design, the use of high-resolution data is more appropriate. For example, pocket parks and informal green space are important strategies in urban micro-regeneration, but the size and shape may be limited by the site, and thus, high-resolution data are needed in such studies [16,37,58]. If the site is larger and regular, or requires overall planning, the use of lower resolution data is more appropriate within the range of error permissibility.

4.4. Shortcomings and Prospects

The present study also has several shortcomings. When classifying the blocks, the differences in building height and density in the same land use type were not considered, which also have an impact on the distribution characteristics of UGS [59,60,61]. The selection of flight time was too long and the impact of building shadows was not taken into account, resulting in some UGS not being identified due to shadows. When analyzing the factors influencing the differences, the block was divided according to the thresholds of 1, 4 and 9 ha, which may not be detailed enough, and the reasons why certain influencing factors were not significantly correlated at medium area sizes need to be further analyzed. Compared with the commonly used NDVI index, the GLI index has a few deficiencies, such as that it may misclassify the green roofs, water and other surfaces with a similar color as UGS; also, the lawn with lighter color may not be identified as UGS. It is undeniable that the correlation and regression results may be different because of the sample quantity; but still, we believe that the conclusions have a certain reference value, as the selected area of this study is representative and large enough to cover all types of land use and blocks, and it is revealed that there are factors affecting the difference that should be noticed in the practical application.
In future research, the application scenarios of the methods and conclusions will continue to be explored, such as whether the GR calculated by high-resolution data can be used as a benchmark to invert the thresholds for UGS extraction by means of low-resolution data, and whether the conclusion is applicable in other regions and when the extraction index is different. At the same time, an in-depth study on the factors affecting the differences would be conducted, such as the height and density of buildings in the site and the relationship between buildings and UGS. Meanwhile, the difference of detailed classification for grass, shrub and tree in the fine scale will be further investigated.

5. Conclusions

In this paper, different resolutions of data were used to analyze the differences in UGS extraction and their influencing factors. The results show that the differences analysis by blocks based on the actual situation is essential to UGS planning. The present study indicates that the impact of the factors on the differences are distinguished in blocks with various sizes and when compared with different resolutions of data. The smaller the block area is, the larger the difference is when the block area is less than 1 ha. When the block area is less than 4 ha, the more elongated the shape, the larger the difference. When the block area is less than 1 ha, the larger the aggregation of UGS, the greater the difference, while when the area is more than 9 ha, the result is the opposite. When the block area is less than 4 ha and larger than 9 ha, the higher the patch density of UGS within the block, the higher the difference. Considering the actual planning and design, the appropriate data need to be selected by comprehensive judgment, including the purpose of the study, the application scenarios, the function, shape and area of the site, as well as the distribution characteristics of the internal green space within error permissibility. Taking 0.5% error allowance as an example, the use of high-resolution data is more appropriate for sites with small area or elongated shape, or sites with a large number of UGS and relatively aggregated distribution, or places that need precise design. If the site is larger and the shape is regular, or just overall planning is needed, low-resolution data can be used. The present study can provide a scientific reference for the data selection of UGS planning and design.

Author Contributions

Conceptualization, X.W. and X.-J.W.; Data curation, X.W. and M.H.; Formal analysis, X.W.; Investigation, X.W.; Methodology, X.W.; Resources, X.-J.W.; Software, X.W.; Supervision, X.-J.W.; Validation, X.W. and M.H.; Visualization, X.W. and M.H.; Writing—original draft, X.W.; Writing—review & editing, X.W., M.H. and X.-J.W. All authors have read and agreed to the published version of the manuscript.

Funding

This research was funded by the National Natural Science Foundation of China (50978054 and 51878144).

Data Availability Statement

The Sentinel-2A data are provided by ESA at: https://scihub.copernicus.eu/dhus/#/home (accessed on 30 August 2022). The other data are available from the corresponding author upon reasonable request.

Acknowledgments

The authors thank Quansheng Zha and Li Tao in Nanjing Fortuneait Intelligent Technology Co., Ltd., Nanjing, China for the support of UVA operation and Rui Zheng in Shanxi University of Finance and Economics for the instruction of data analysis. The authors also thank all the reviewers and editors for their constructive comments.

Conflicts of Interest

The authors declare no conflict of interest.

References

  1. Holt, A.R.; Mears, M.; Maltby, L.; Warren, P. Understanding spatial patterns in the production of multiple urban ecosystem services. Ecosyst. Serv. 2015, 16, 33–46. [Google Scholar] [CrossRef] [Green Version]
  2. Liu, O.Y.; Russo, A. Assessing the contribution of urban green spaces in green infrastructure strategy planning for urban ecosystem conditions and services. Sustain. Cities Soc. 2021, 68, 102772. [Google Scholar] [CrossRef]
  3. Voogt, J.A.; Oke, T.R. Effects of urban surface geometry on remotely-sensed surface temperature. Int. J. Remote Sens. 1998, 19, 895–920. [Google Scholar] [CrossRef]
  4. Ng, E.; Chen, L.; Wang, Y.N.; Yuan, C. A study on the cooling effects of greening in a high-density city: An experience from Hong Kong. Build. Environ. 2012, 47, 256–271. [Google Scholar] [CrossRef]
  5. Huang, Q.P.; Huang, J.J.; Yang, X.N.; Fang, C.L.; Liang, Y.J. Quantifying the seasonal contribution of coupling urban land use types on Urban Heat Island using Land Contribution Index: A case study in Wuhan, China. Sustain. Cities Soc. 2019, 44, 666–675. [Google Scholar] [CrossRef]
  6. Jennings, V.; Gragg, R.S.; Brown, C.P.; Hartel, D.; Kuehler, E.; Sinykin, A.; Johnson, E.; Kondo, M. Structural Characteristics of Tree Cover and the Association with Cardiovascular and Respiratory Health in Tampa, FL. J. Urban Health 2019, 96, 669–681. [Google Scholar] [CrossRef] [PubMed]
  7. Jim, C.Y.; Chen, W.Y. Assessing the ecosystem service of air pollutant removal by urban trees in Guangzhou (China). J. Environ. Manag. 2008, 88, 665–676. [Google Scholar] [CrossRef] [PubMed]
  8. Bratman, G.N.; Hamilton, J.P.; Daily, G.C. The impacts of nature experience on human cognitive function and mental health. Ann. N. Y. Acad. Sci. 2012, 1249, 118–136. [Google Scholar] [CrossRef]
  9. Markevych, I.; Schoierer, J.; Hartig, T.; Chudnovsky, A.; Hystad, P.; Dzhambov, A.M.; de Vries, S.; Triguero-Mas, M.; Brauer, M.; Nieuwenhuijsen, M.J.; et al. Exploring pathways linking greenspace to health: Theoretical and methodological guidance. Environ. Res. 2017, 158, 301–317. [Google Scholar] [CrossRef]
  10. van den Berg, A.E.; Maas, J.; Verheij, R.A.; Groenewegen, P.P. Green space as a buffer between stressful life events and health. Soc. Sci. Med. 2010, 70, 1203–1210. [Google Scholar] [CrossRef] [Green Version]
  11. Haaland, C.; van den Bosch, C.K. Challenges and strategies for urban green-space planning in cities undergoing densification: A review. Urban For. Urban Green. 2015, 14, 760–771. [Google Scholar] [CrossRef]
  12. Alonzo, M.; Bookhagen, B.; Roberts, D.A. Urban tree species mapping using hyperspectral and lidar data fusion. Remote Sens. Environ. 2014, 148, 70–83. [Google Scholar] [CrossRef]
  13. Brell, M.; Segl, K.; Guanter, L.; Bookhagen, B. 3D hyperspectral point cloud generation: Fusing airborne laser scanning and hyperspectral imaging sensors for improved object-based information extraction. ISPRS J. Photogramm. Remote Sens. 2019, 149, 200–214. [Google Scholar] [CrossRef]
  14. Burr, A.; Schaeg, N.; Hall, D.M. Assessing residential front yards using Google Street View and geospatial video: A virtual survey approach for urban pollinator conservation. Appl. Geogr. 2018, 92, 12–20. [Google Scholar] [CrossRef]
  15. Chance, C.M.; Coops, N.C.; Plowright, A.A.; Tooke, T.R.; Christen, A.; Aven, N. Invasive Shrub Mapping in an Urban Environment from Hyperspectral and LiDAR-Derived Attributes. Front. Plant Sci. 2016, 7, 1528. [Google Scholar] [CrossRef] [PubMed] [Green Version]
  16. Mathieu, R.; Freeman, C.; Aryal, J. Mapping private gardens in urban areas using object-oriented techniques and very high-resolution satellite imagery. Landsc. Urban Plan. 2007, 81, 179–192. [Google Scholar] [CrossRef]
  17. Neyns, R.; Canters, F. Mapping of Urban Vegetation with High-Resolution Remote Sensing: A Review. Remote Sens. 2022, 14, 1031. [Google Scholar] [CrossRef]
  18. Zhang, C.H.; Kovacs, J.M. The application of small unmanned aerial systems for precision agriculture: A review. Precis. Agric. 2012, 13, 693–712. [Google Scholar] [CrossRef]
  19. Jiang, L.L.; Jiapaer, G.; Bao, A.M.; Guo, H.; Ndayisaba, F. Vegetation dynamics and responses to climate change and human activities in Central Asia. Sci. Total Environ. 2017, 599, 967–980. [Google Scholar] [CrossRef] [PubMed]
  20. Blanco, S.R.; Heras, D.B.; Arguello, F. Texture Extraction Techniques for the Classification of Vegetation Species in Hyperspectral Imagery: Bag of Words Approach Based on Superpixels. Remote Sens. 2020, 12, 2633. [Google Scholar] [CrossRef]
  21. Li, C.K.; Dong, X.J.; Zhang, Q. Multi-scale Object-oriented Building Extraction Method of Tai’an City from High Resolution Image. In Proceedings of the 3rd International Workshop on Earth Observation and Remote Sensing Applications (EORSA), Changsha, China, 11–14 June 2014. [Google Scholar]
  22. Liu, C.; Zhang, J.; Yu, S.; Yang, A.M. Extraction of Urban Green Space Information Based on ZY-3 Remote Sensing Images. In Proceedings of the International Conference on Information Technology and Management Engineering (ITME), Hong Kong, China, 26–27 April 2014. [Google Scholar]
  23. Liu, Z.J.; Yang, L.; Liu, Z.H.; Duan, L.L.; Qiao, X.X.; Gong, J.J. Method of Vegetation Extraction Based on Deep Belief Network and Optimal Scale. Laser Optoelectron. Prog. 2018, 55, 021001. [Google Scholar] [CrossRef]
  24. Zhang, K.; Hu, B. Individual Urban Tree Species Classification Using Very High Spatial Resolution Airborne Multi-Spectral Imagery Using Longitudinal Profiles. Remote Sens. 2012, 4, 1741–1757. [Google Scholar] [CrossRef] [Green Version]
  25. Richards, D.R.; Edwards, P.J. Quantifying street tree regulating ecosystem services using Google Street View. Ecol. Indic. 2017, 77, 31–40. [Google Scholar] [CrossRef]
  26. Hoang, N.D.; Tran, X.L. Remote Sensing-Based Urban Green Space Detection Using Marine Predators Algorithm Optimized Machine Learning Approach. Math. Probl. Eng. 2021, 2021, 5586913. [Google Scholar] [CrossRef]
  27. Kuang, W.H.; Zhang, S.; Li, X.Y.; Lu, D.S. A 30 m resolution dataset of China’s urban impervious surface area and green space, 2000-2018. Earth Syst. Sci. Data 2021, 13, 63–82. [Google Scholar] [CrossRef]
  28. Zhou, X.S.; Li, L.; Chen, L.Q.; Liu, Y.Q.; Cui, Y.F.; Zhang, Y.; Zhang, T. Discriminating Urban Forest Types from Sentinel-2A Image Data through Linear Spectral Mixture Analysis: A Case Study of Xuzhou, East China. Forests 2019, 10, 478. [Google Scholar] [CrossRef] [Green Version]
  29. An, C.C.; Dong, Z.; Li, H.L.; Zhao, W.T.; Chen, H.L. Assessment of Vegetation Phenological Extractions Derived From Three Satellite-Derived Vegetation Indices Based on Different Extraction Algorithms Over the Tibetan Plateau. Front. Environ. Sci. 2021, 9, 794189. [Google Scholar] [CrossRef]
  30. Xu, Z.Y.; Zhou, Y.; Wang, S.X.; Wang, L.T.; Li, F.; Wang, S.C.; Wang, Z.Q. A Novel Intelligent Classification Method for Urban Green Space Based on High-Resolution Remote Sensing Images. Remote Sens. 2020, 12, 3845. [Google Scholar] [CrossRef]
  31. Yan, J.; Chen, Y.Y.; Zheng, J.Z.; Guo, L.; Zheng, S.Q.; Zhang, R.C. Multi-Source Time Series Remote Sensing Feature Selection and Urban Forest Extraction Based on Improved Artificial Bee Colony. Remote Sens. 2022, 14, 4859. [Google Scholar] [CrossRef]
  32. Qian, Y.; Zhou, W.; Yu, W.; Pickett, S.T.A. Quantifying spatiotemporal pattern of urban greenspace: New insights from high resolution data. Landsc. Ecol. 2015, 30, 1165–1173. [Google Scholar] [CrossRef]
  33. Rioux, J.-F.; Cimon-Morin, J.; Pellerin, S.; Alard, D.; Poulin, M. How Land Cover Spatial Resolution Affects Mapping of Urban Ecosystem Service Flows. Front. Environ. Sci. 2019, 7, 93. [Google Scholar] [CrossRef] [Green Version]
  34. Yan, J.; Zhou, W.; Jenerette, G.D. Testing an energy exchange and microclimate cooling hypothesis for the effect of vegetation configuration on urban heat. Agric. For. Meteorol. 2019, 279, 107666. [Google Scholar] [CrossRef]
  35. Zhou, W.; Wang, J.; Cadenasso, M.L. Effects of the spatial configuration of trees on urban heat mitigation: A comparative study. Remote Sens. Environ. 2017, 195, 1–12. [Google Scholar] [CrossRef]
  36. Li, X.; Zhou, W.; Ouyang, Z. Relationship between land surface temperature and spatial pattern of greenspace: What are the effects of spatial resolution? Landsc. Urban Plan. 2013, 114, 1–8. [Google Scholar] [CrossRef]
  37. Wood, L.; Hooper, P.; Foster, S.; Bull, F. Public green spaces and positive mental health—Investigating the relationship between access, quantity and types of parks and mental wellbeing. Health Place 2017, 48, 63–71. [Google Scholar] [CrossRef]
  38. Reid, C.E.; Kubzansky, L.D.; Li, J.Y.; Shmool, J.L.; Clougherty, J.E. It’s not easy assessing greenness: A comparison of NDVI datasets and neighborhood types and their associations with self-rated health in New York City. Health Place 2018, 54, 92–101. [Google Scholar] [CrossRef] [PubMed]
  39. Browning, M.H.E.M.; Locke, D.H. The greenspace-academic performance link varies by remote sensing measure and urbanicity around Maryland public schools. Landsc. Urban Plan. 2020, 195, 103706. [Google Scholar] [CrossRef]
  40. Mozgeris, G.; Juodkiene, V.; Jonikavicius, D.; Straigyte, L.; Gadal, S.; Ouerghemmi, W. Ultra-Light Aircraft-Based Hyperspectral and Colour-Infrared Imaging to Identify Deciduous Tree Species in an Urban Environment. Remote Sens. 2018, 10, 1668. [Google Scholar] [CrossRef] [Green Version]
  41. Degerickx, J.; Hermy, M.; Somers, B. Mapping Functional Urban Green Types Using High Resolution Remote Sensing Data. Sustainability 2020, 12, 2144. [Google Scholar] [CrossRef] [Green Version]
  42. Pu, R.; Landry, S. A comparative analysis of high spatial resolution IKONOS and WorldView-2 imagery for mapping urban tree species. Remote Sens. Environ. 2012, 124, 516–533. [Google Scholar] [CrossRef]
  43. Jensen, R.R.; Hardin, P.J.; Hardin, A.J. Classification of urban tree species using hyperspectral imagery. Geocarto Int. 2012, 27, 443–458. [Google Scholar] [CrossRef]
  44. Lin, Y.; An, W.Z.; Gan, M.Y.; Shahtahmassebi, A.; Ye, Z.R.; Huang, L.Y.; Zhu, C.M.; Huang, L.; Zhang, J.; Wang, K. Spatial Grain Effects of Urban Green Space Cover Maps on Assessing Habitat Fragmentation and Connectivity. Land 2021, 10, 1065. [Google Scholar] [CrossRef]
  45. Chang, Q.; Liu, X.W.; Wu, J.S.; He, P. MSPA-Based Urban Green Infrastructure Planning and Management Approach for Urban Sustainability: Case Study of Longgang in China. J. Urban Plan. Dev 2015, 141, A5014006. [Google Scholar] [CrossRef]
  46. Weber, C.; Petropoulou, C.; Hirsch, J. Urban development in the Athens metropolitan area using remote sensing data with supervised analysis and GIS. Int. J. Remote Sens. 2005, 26, 785–796. [Google Scholar] [CrossRef]
  47. Freeman, C.; Buck, O. Development of an ecological mapping methodology for urban areas in New Zealand. Landsc. Urban Plan. 2003, 63, 161–173. [Google Scholar] [CrossRef]
  48. Liu, B.H.; Deng, Y.B.; Li, M.; Yang, J.; Liu, T. Classification Schemes and Identification Methods for Urban Functional Zone: A Review of Recent Papers. Appl. Sci. 2021, 11, 9968. [Google Scholar] [CrossRef]
  49. Chen, H.; Ooka, R.; Huang, H.; Tsuchiya, T. Study on mitigation measures for outdoor thermal environment on present urban blocks in Tokyo using coupled simulation. Build. Environ. 2009, 44, 2290–2299. [Google Scholar] [CrossRef]
  50. Agapiou, A. Vegetation Extraction Using Visible-Bands from Openly Licensed Unmanned Aerial Vehicle Imagery. Drones 2020, 4, 27. [Google Scholar] [CrossRef]
  51. Li, F.; Chen, W.; Zeng, Y.; Zhao, Q.J.; Wu, B.F. Improving Estimates of Grassland Fractional Vegetation Cover Based on a Pixel Dichotomy Model: A Case Study in Inner Mongolia, China. Remote Sens. 2014, 6, 4705–4722. [Google Scholar] [CrossRef] [Green Version]
  52. Gao, Y.; Liu, L.Y.; Zhang, X.; Chen, X.D.; Mi, J.; Xie, S. Consistency Analysis and Accuracy Assessment of Three Global 30-m Land-Cover Products over the European Union using the LUCAS Dataset. Remote Sens. 2020, 12, 3479. [Google Scholar] [CrossRef]
  53. Turner, T. Urban design: Green dimensions—Moughtin, C. Cities 1997, 14, 244–245. [Google Scholar] [CrossRef]
  54. Semeraro, T.; Scarano, A.; Buccolieri, R.; Santino, A.; Aarrevaara, E. Planning of Urban Green Spaces: An Ecological Perspective on Human Benefits. Land 2021, 10, 105. [Google Scholar] [CrossRef]
  55. Li, Y.Y.; Ren, B.T.; Chen, Y.S.; Huang, L.C.; Sun, C.G. Multiscale spatiotemporal dynamics analysis of urban green space: Implications for green space planning in the rapid urbanizing Hefei City, China. Front. Ecol. Evol. 2022, 10, 998111. [Google Scholar] [CrossRef]
  56. Xu, Y.Q.; Wang, L.; Fu, C.; Kosmyna, T. A fishnet-constrained land use mix index derived from remotely sensed data. Ann. GIS 2017, 23, 303–313. [Google Scholar] [CrossRef] [Green Version]
  57. Joshi, N.; Baumann, M.; Ehammer, A.; Fensholt, R.; Grogan, K.; Hostert, P.; Jepsen, M.R.; Kuemmerle, T.; Meyfroidt, P.; Mitchard, E.T.A.; et al. A Review of the Application of Optical and Radar Remote Sensing Data Fusion to Land Use Mapping and Monitoring. Remote Sens. 2016, 8, 70. [Google Scholar] [CrossRef] [Green Version]
  58. Baker, F.; Smith, C.L.; Cavan, G. A Combined Approach to Classifying Land Surface Cover of Urban Domestic Gardens Using Citizen Science Data and High Resolution Image Analysis. Remote Sens. 2018, 10, 537. [Google Scholar] [CrossRef] [Green Version]
  59. Li, X.J.; Li, W.D.; Meng, Q.Y.; Zhang, C.R.; Jancso, T.; Wu, K.L. Modelling building proximity to greenery in a three-dimensional perspective using multi-source remotely sensed data. J. Spat. Sci. 2016, 61, 389–403. [Google Scholar] [CrossRef]
  60. Yuan, B.; Zhou, L.; Dang, X.W.; Sun, D.Q.; Hu, F.N.; Mu, H.W. Separate and combined effects of 3D building features and urban green space on land surface temperature. J. Environ. Manag. 2021, 295, 113116. [Google Scholar] [CrossRef]
  61. Zhu, S.J.; Yang, Y.; Yan, Y.; Causone, F.; Jin, X.; Zhou, X.; Shi, X. An evidence-based framework for designing urban green infrastructure morphology to reduce urban building energy use in a hot-humid climate. Build. Environ. 2022, 219, 109181. [Google Scholar] [CrossRef]
Remotesensing 15 01261 g001 550
Figure 1. The location of the study area: (a) the visible light image taken by DJI Mavic 2; (b) the land use types of the study area.
Figure 1. The location of the study area: (a) the visible light image taken by DJI Mavic 2; (b) the land use types of the study area.
Remotesensing 15 01261 g001
Remotesensing 15 01261 g002 550
Figure 2. The precision of GR grades between GF1 and DJI (a) and that between GF1 and S2A (b).
Figure 2. The precision of GR grades between GF1 and DJI (a) and that between GF1 and S2A (b).
Remotesensing 15 01261 g002
Remotesensing 15 01261 g003 550
Figure 3. The DD between different resolutions of data in fishnet with various sizes.
Figure 3. The DD between different resolutions of data in fishnet with various sizes.
Remotesensing 15 01261 g003
Remotesensing 15 01261 g004 550
Figure 4. The confusion matrices of 150 m and 180 m fishnets: (a) comparison of GF1 and DJI in the 150 m fishnet; (b) comparison of GF1 and S2A in the 150 m fishnet; (c) comparison of GF1 and DJI in the 180 m fishnet; (d) comparison of GF1 and S2A in the 180 m fishnet.
Figure 4. The confusion matrices of 150 m and 180 m fishnets: (a) comparison of GF1 and DJI in the 150 m fishnet; (b) comparison of GF1 and S2A in the 150 m fishnet; (c) comparison of GF1 and DJI in the 180 m fishnet; (d) comparison of GF1 and S2A in the 180 m fishnet.
Remotesensing 15 01261 g004
Remotesensing 15 01261 g005 550
Figure 5. The spatial distribution of GR grades divided by 150 m and 180 m fishnets in a typical region: (a–d) the visible light image divided by 150 m fishnet and GR grades identified by DJI, GF1, S2A, respectively; (e–h) the visible light image divided by 180 m fishnet and GR grades identified by DJI, GF1, S2A, respectively.
Figure 5. The spatial distribution of GR grades divided by 150 m and 180 m fishnets in a typical region: (a–d) the visible light image divided by 150 m fishnet and GR grades identified by DJI, GF1, S2A, respectively; (e–h) the visible light image divided by 180 m fishnet and GR grades identified by DJI, GF1, S2A, respectively.
Remotesensing 15 01261 g005
Remotesensing 15 01261 g006 550
Figure 6. The DD between different resolutions of data in blocks with different land use types.
Figure 6. The DD between different resolutions of data in blocks with different land use types.
Remotesensing 15 01261 g006
Remotesensing 15 01261 g007 550
Figure 7. The confusion matrices of park land and water area: (a) comparison of GF1 and DJI in park land; (b) comparison of GF1 and S2A in park land; (c) comparison of GF1 and DJI in water area; (d) comparison of GF1 and S2A in water area.
Figure 7. The confusion matrices of park land and water area: (a) comparison of GF1 and DJI in park land; (b) comparison of GF1 and S2A in park land; (c) comparison of GF1 and DJI in water area; (d) comparison of GF1 and S2A in water area.
Remotesensing 15 01261 g007
Remotesensing 15 01261 g008 550
Figure 8. The comparison of extracting UGS from data with different resolutions (a) the visible light image taken by DJI; (b) UGS extracted from DJI; (c) UGS extracted from GF1; (d) UGS extracted from S2A.
Figure 8. The comparison of extracting UGS from data with different resolutions (a) the visible light image taken by DJI; (b) UGS extracted from DJI; (c) UGS extracted from GF1; (d) UGS extracted from S2A.
Remotesensing 15 01261 g008
Table
Table 1. Details of the data used in the study.
Table 1. Details of the data used in the study.
Data SourceDate of ImagesResolution Data Type
DJI M300 RTK (DJI)2022-8-3~50.1 mVisible spectral, including bands of Red, Green and Blue
Gaofen-1 (GF1)2022-6-152 mMultispectral, including bands of Red, Green, Blue and Near infrared
Sentient-2A (S2A)2022-8-1410 mThe same as GF1
Table
Table 2. The land use types of the study area.
Table 2. The land use types of the study area.
CodeLand Use TypesNumber of BlocksArea Ratio
AResidential land18131.38%
BAdministration and public services11211.65%
CCommercial and business facilities1097.85%
DIndustrial land40.83%
ELogistics and warehouse 40.38%
FRoad, street and transportation15418.29%
GMunicipal utilities231.94%
HPark, green space and square8215.51%
IOther types, including land that are abandoned and under construction706.74%
JWater375.62%
Table
Table 3. The grades of GR.
Table 3. The grades of GR.
Name of the GradesAbbreviationGR
Low GRL_GR0–20%
Sub-low GRSL_GR20–40%
Medium GRM_GR40–60%
Sub-high GRSH_GR60–80%
High GRH_GR80–100%
Table
Table 4. The calculation methods of influence factors.
Table 4. The calculation methods of influence factors.
Influence FactorsCalculation Methods
L_AreaArea of the land unit
L_SIShape index of the land unit, L_SI = 0.25P/ A , in which P is the perimeter of the land unit, and A is the area of the land unit
G_PDPatch density of the green space inside each land unit, G_PD = Ng/A, in which Ng is the number of the green space patches and A is the area of the land unit
G_AIAggregation index of the green space inside each land unit, G_AI = (gii /max → gii), gii is the number of like adjacencies involving the corresponding class, max → gii is the maximum possible number of like adjacencies involving the corresponding class.
Table
Table 5. The mean GR of each land use type and results of Chi-Square test.
Table 5. The mean GR of each land use type and results of Chi-Square test.
Data
Source		Mean GR of Each Land Use TypePearson
Chi-Square		p Value
ABCDEFGHIJ
DJI0.220.270.160.290.200.220.220.630.140.24213.712<0.001
GF10.100.220.090.220.040.140.160.630.080.54488.558<0.001
S2A0.140.220.110.230.040.160.170.580.090.49446.377<0.001
Note: A–J means different land use types shown in Table 2.
Table
Table 6. The correlations between the influence factors and the differences of GR from data with various resolutions.
Table 6. The correlations between the influence factors and the differences of GR from data with various resolutions.
AREA ≤ 1 ha1 ha < AREA ≤ 4 ha4 ha < AREA ≤ 9 haAREA > 9 ha
D_GF1_
DJI		D_GF1_
S2A		D_GF1_
DJI		D_GF1_
S2A		D_GF1_
DJI		D_GF1_
S2A		D_GF1_
DJI		D_GF1_
S2A
L_
AREA		−0.124 **−0.041−0.046−0.0120.1810.2310.304−0.143.
L_SI0.105 *0.0200.254 **0.0190.163−0.1310.2500.128
G_PD0.177 **0.231 **0.182 *0.0990.0560.113−0.2410.445 *
G_AI0.288 **0.137 **0.0530.1130.2470.105−0.542 *−0.230
Note: ** means p < 0.01, * means p < 0.05.
Disclaimer/Publisher’s Note: The statements, opinions and data contained in all publications are solely those of the individual author(s) and contributor(s) and not of MDPI and/or the editor(s). MDPI and/or the editor(s) disclaim responsibility for any injury to people or property resulting from any ideas, methods, instructions or products referred to in the content.

Share and Cite

MDPI and ACS Style

Wei, X.; Hu, M.; Wang, X.-J. The Differences and Influence Factors in Extracting Urban Green Space from Various Resolutions of Data: The Perspective of Blocks. Remote Sens. 2023, 15, 1261. https://doi.org/10.3390/rs15051261

AMA Style

Wei X, Hu M, Wang X-J. The Differences and Influence Factors in Extracting Urban Green Space from Various Resolutions of Data: The Perspective of Blocks. Remote Sensing. 2023; 15(5):1261. https://doi.org/10.3390/rs15051261

Chicago/Turabian Style

Wei, Xiao, Mengjun Hu, and Xiao-Jun Wang. 2023. "The Differences and Influence Factors in Extracting Urban Green Space from Various Resolutions of Data: The Perspective of Blocks" Remote Sensing 15, no. 5: 1261. https://doi.org/10.3390/rs15051261

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