Exploring Relationships between Spatial Pattern Change in Steel Plants and Land Cover Change in Tangshan City

Abstract

It is of great significance for the sustainable development of steel cities to explore the relationship between the spatial pattern change in steel plants and land cover change during the transformation of steel cities. To address the issue of unsatisfactory results for segmenting steel plants based on high-resolution remote sensing images, due to insufficient sample datasets and task complexity, we proposed a steel plant segmentation strategy that combines high-resolution remote sensing images, POI data, and OSM data. Additionally, we discussed the effect of POI data and OSM data on steel plant segmentation, analyzing the spatial pattern change in steel plants in Tangshan City during 2017–2022 and its relationship with land cover change. The results demonstrate that: (1) The proposed strategy can significantly improve the accuracy of steel plant segmentation. The introduction of POI data can significantly improve the precision of steel plant segmentation, however, it will to some extent reduce the recall of steel plant segmentation, and this phenomenon weakens as the distance threshold increases. The introduction of OSM data can effectively improve the effectiveness of steel plant segmentation, however, it has significant limitations. (2) During 2017–2022, the spatial distribution center of steel plants in Tangshan City moved obviously to the southeast, and the positive change in steel plants was mainly concentrated in the coastal regions of southern Tangshan City, while the negative change in steel plants was mainly concentrated in central Tangshan City. (3) There is a relatively strong spatial correlation between the positive change in steel plants and the transition from vegetation to built area, as well as the transition from cropland to built area.

1. Introduction

Steel products are widespread and prevalent in daily life and industrial manufacturing due to their malleability and low cost. Cities that specialize in steel production (hereinafter referred to as steel cities) played a crucial role in global industrialization [1]. However, long-term extensive economic development has resulted in problems such as severe air pollution, a fragile ecological environment, and an imbalanced industrial structure, impeding the sustainable development of steel cities [2,3]. China is the world’s largest steel producer, and the economic and environmental situations of Chinese steel cities, particularly Tangshan City, are facing significant challenges [4,5]. In response to this predicament, Tangshan has recently transformed its economic development and facilitated industrial transformation, attaining certain results in achieving coordinated economic and environmental development [6]. Industrial transformation and upgrading is often accompanied by the spatial reorganization of traditional industries [7,8], which can lead to substantial changes in land cover and subsequently affect climate [9], hydrology [10], surface temperature [11], and biodiversity [12]. Land use/cover change (LUCC) is an important topic in global environmental change and sustainable development [13,14], and exploring the relationships between spatial pattern change in steel plants and land cover change during Tangshan’s industrial transformation is essential for promoting sustainable development in the city.

Steel plants are intricate structures comprising various buildings, facilities, and other ground objects, and they vary in scale and characteristics. Remote sensing image (RSI) is widely considered to be objective and authentic data that can offer ample physical and spatial information. The availability of diverse remote sensing products provides strong data support for the rapid identification of steel plants. Energy-intensive plants, such as steel plants, release substantial amounts of heat during their production processes, and lighting is commonly required within these plants. Thermal anomaly products and night-time light data, derived from remote sensing images, provide a reliable means of identification for the energy-intensive plants [15]. However, it is challenging to conduct fine-grained studies using thermal anomaly products and night-time light data due to the limitation of spatial resolution. It has become feasible to identify steel plants from high-resolution images due to the rapid advancements in both high-resolution remote sensing technology and deep learning techniques. Lu et al. developed a deep learning target detection network for steel plant extraction from high-resolution remote sensing images based on deep learning target detection network SSD and conducted experiments on steel plants located in the Beijing-Tianjin-Hebei region using GF-1 data [16]. Nonetheless, the deep learning approach requires a large number of labeled samples for training. The total number of steel plants is small, and their distribution is relatively scattered. This presents challenges in data acquisition and model training. Due to the complex background of remote sensing images, insufficient training samples make it challenging for the model to learn the features that distinguish steel plants from other ground objects, leading to the identification of a significant number of confusing ground objects that are not steel plants. Therefore, in the task of steel plant segmentation based on high-resolution remote sensing images, it is crucial to combine other geospatial data for post-processing of the initial results in order to enhance the accuracy of the segmentation outcomes.

Point of interest (POI) data offers the advantage of providing highly precise geographical locations and rich socioeconomic information, while also benefiting from fast updating and easy acquisition. In recent years, scholars have put into practice the utilization of POI data for research related to industrial spatial patterns and gained productive outcomes [17,18,19]. Nevertheless, the limitations of POI data cannot be overlooked [20,21]. Firstly, the constraint on spatial attributes must be mentioned. POI data is abstracted from geographic objects, such as schools, parks, or lakes, and it lacks the specific area information associated with its corresponding geographical entity. As a result, the area represented by each POI is considered to be uniform. The other side is the limitation of data quality. POI data is of uneven quality, as it is susceptible to subjective influence from data production personnel, resulting in noise being commonly present in POI data, such as incorrect names or categories as well as multiple POIs corresponding to the same geographical entity. The limitations result in significant constraints when solely using POI data to investigate steel plant changes. In addition, the industry classification of POI data is not rigorous or standard enough. This limitation prevents the direct extraction of steel plant POIs from the raw POI dataset. Therefore, effectively acquiring the steel plant POIs is crucial in applying POI data to the study of steel plant changes.

Parcels are the basic spatial units for fine-scale urban modeling, urban studies, and urban spatial planning, however, due to the poorly developed digital infrastructures and the fact that parcel maps are tagged as confidential in existing Chinese planning agencies, it is difficult to obtain parcel maps of Chinese cities, especially small- and medium-sized Chinese cities [22]. OpenStreetMap (OSM) is one of the most successful crowdsourced GIS projects, providing rich and detailed street network data for many cities. In recent years, many scholars have used street network data provided by OSM (hereinafter referred to as OSM data) to segment urban parcels and used the parcels as the basic spatial units of urban research [23,24,25]. In addition, Wang et al. applied the OSM data to the identification of industrial land in the Beijing-Tianjin-Hebei region with satisfactory results [26]. Therefore, this study attempts to introduce the OSM data into the segmentation of steel plants.

To address the issue of unsatisfactory segmentation accuracy caused by both insufficient training data and the complex composition of steel plants, we proposed a steel plant segmentation strategy based on RSI, POI data, and OSM data. To rapidly and accurately acquire steel plant POIs, we matched the registration data of steel plants with the steel plant POIs. Using this strategy, we segmented the steel plants for 2017 and 2022 in Tangshan City and analyzed the spatial pattern change in the steel plants in Tangshan City during 2017–2022. We also combined land cover data to explore the relationships between the spatial pattern change in steel plants and land cover change in Tangshan City during 2017–2022. The results can provide support for further industrial layout planning and land use planning in Tangshan City and can provide a realistic reference for the transformation of other resource-based cities.

2. Materials and Methods

The framework of this study is shown in Figure 1, which involves three main components, including data pre-processing, steel plant segmentation, and spatial analysis.

2.1. Study Area

As shown in Figure 2, Tangshan City is situated in the northeast region of Hebei Province, bordered by Beijing and Tianjin to the west, the Bohai Sea to the south, and the Yanshan Mountain to the north. The administrative area of Tangshan spans about 14,372 square kilometers, and as of 2021, it had a permanent population of 7.697 million. As a vital component of the Beijing-Tianjin-Hebei Economic Zone, Tangshan City has exhibited robust industrial prowess and rapid economic growth. In 2021, Tangshan’s total GDP was reported to be CNY 823.06 billion, ranking it first in Hebei Province, of which the GDP of the secondary industry accounted for over 55% with CNY 454.68 billion.

Tangshan City was built for coal and prospered with steel. It is endowed with abundant mineral resources and is one of the three major iron ore concentration areas in China.

Relying on rich mineral resources, Tangshan’s steelmaking industry has developed rapidly, making a huge contribution to economic growth. In 2021, the total production of crude steel in Tangshan City was 131 million tons, ranking first in China. However, in the process of urban development, a greater emphasis on economic output, while neglecting ecological benefits, has led to the degradation of the ecological environment in Tangshan City [27,28]. In recent years, ecological protection has received significant attention. To achieve sustainable development, Tangshan has proactively promoted industrial structure adjustment and upgrading, and expedited the relocation and integration of high energy consumption and high-pollution industries, such as steelmaking. As one of the first industrial transformation and upgrading demonstration zones for resource-based cities in China, Tangshan’s transformation experience holds substantial reference value.

2.2. Data Source and Pre-Processing

In this study, RSIs, POI data, and OSM data were used to segment steel plants. The RSIs with a spatial resolution of 1.18 m were obtained from Google Earth and downloaded through the Rivermap software (version 4.1). POI data was obtained from AMap via the application programming interface (API). Amap belongs to Alibaba Group, has a high-quality digital map database, and is a leading digital map content, navigation, and location service solution provider in China. Due to the original POI data on the classification of the industry being not detailed, it is difficult to directly extract the POIs of steel plants. In addition, AMap provides more than 70 million POIs across China, and it is obviously cumbersome and expensive to download all the POIs and then filter them. According to the “Provisions on Administration of Enterprise Name Registration” issued by the State Administration for Industry and Commerce of the People’s Republic of China (SAIC), the enterprise names are composed of administrative division name, business name, industry, and organizational form. We selected enterprises from the enterprise registration database that are engaged in the ferrous metal smelting and rolling processing industry, and whose business scope includes steelmaking. Subsequently, the names of the enterprises were segmented to extract the keywords, and the steel plant POIs were obtained based on the keywords. The enterprise registration data is downloaded from the official website of Aiqicha, which is an enterprise information query website based on Baidu artificial intelligence and big data technology. The keyword extraction process for the two typical steel enterprise names is shown in Figure 3. Finally, taking the steel plant POIs as the center, we built square regions with a side length of 6 km (the actual length and width of steel plants are generally 1–3 km [16]), and obtained the Google Earth images of these regions to build a steel plant dataset.

We downloaded the 2017 and 2022 annual road network data from OpenStreetMap geographic data platform [29].The pre-processing of OSM data, and the method of using OSM data to segment parcels, refer to the literature [24]. The steel plant POIs were used to extract the parcels related to steel plants, and the parcels intersecting with the steel plant POIs were considered as steel plant parcels.

In this study, the land cover data for 2017 and 2022, published by ESRI, were used to obtain the land cover change in Tangshan City. The data were generated by a deep learning land classification model using billions of training pixels, with an accuracy of over 80% [30]. The land cover data uses a classification system of 9 classes, including Water, Tree, Flooded Vegetation, Crops, Built Area, Bare Ground, Snow/Ice, Clouds, and Rangeland. According to the class definition of the land cover data [31], the steel plants belong to the Built Area class. In this study, we excluded two land cover classes of Snow/Ice and Clouds, which were less distributed in Tangshan City and had no significant contribution to the result. In addition, referring to the literature [32], we reclassified the land cover classes of Flooded Vegetation, Trees, and Rangeland into the Vegetation class.

2.3. Segmentation and Changing Region Identification of Steel Plants

The DeeplabV3+ model is currently the most representative semantic segmentation model, which performs well in the semantic segmentation of RSIs. In this study, we used DeeplabV3 + model to generate initial steel plant segmentation results from the RSIs. Subsequently, we performed post-processing on the initial segmentation results using steel plant POIs and steel plant parcels. The following are the detailed steps involved:

Step 1: Using the DeeplabV3 + model to segment steel plants from the RSIs and vectorize the segmentation results.

Step 2: Using steel plant POIs to filter the segmentation results outside the manually set distance threshold (DT).

Step 3: If the proportion of the area of the segmentation results within the steel plant parcel to the area of the steel plant parcel exceeds 50%, the steel plant parcel will be used as the segmentation result for this steel plant.

Spatial overlay analysis was used to obtain the change regions of the steel plant regions, then the area of the change regions was calculated and assigned to the geographical center of the change regions to obtain the point data with the changing area. This process is shown in Figure 4. The spatial changes of steel plants include appearance, disappearance, expansion, and contraction. In this study, appearance and expansion of the steel plants are regarded as positive change in the steel plants (hereinafter referred to as PC), while disappearance and contraction of the steel plants are regarded as negative change in the steel plants (hereinafter referred to as NC).

The DeeplabV3+ model has a complex structure and requires a large amount of computing resources in the training process [33]. The emergence of remote sensing computing cloud platforms with high-performance computing resources greatly relieves this problem [34]. In this study, the training and the prediction of the DeeplabV3+ model were completed through the PIE-Engine AI platform [35], which is an end-to-end full-stack remote sensing image intelligent interpretation development platform based on cloud elastic GPU resources.

The precision, recall, F1 score, and Intersection of Union (IoU) were used to evaluate the results of steel plant segmentation quantitatively. The calculation formula is as follows:

$$\mathit{precision}=\frac{TP}{TP+FP},\mathit{recall}=\frac{TP}{TP+FN}$$

(1)

$$\mathrm{F}1=2\times \frac{precision\times recall}{precision+recall},\mathrm{IoU}=\frac{TP}{TP+FP+FN}$$

(2)

where TP, FP, TN, and FN represent the true positive case, the true negative case, the false positive case, and the false negative case in the prediction results, respectively.

2.4. Spatial Analysis

2.4.1. Kernel Density Estimation

The kernel density estimation (KDE) is widely used in industrial agglomeration, spatial pattern, crime geography, and other studies, which considers the attenuation law of distance and can intuitively reflect the spatial distribution and agglomeration state of geographic elements [19]. The calculation formula is as follows:

$$f\left(s\right)=\frac{1}{n{h}^2}{\displaystyle \sum _{i=1}^{n}K\left(\frac{w_i\times d\left(s,x_i\right)}{h}\right)}$$

(3)

where h represents the bandwidth set artificially; K() represents the kernel function selected artificially; d(s, x_i) represents the distance from the steel plant i to the location s; ${\omega }_i$ represents the area of the steel plant i; and n is the number of steel plants.

The kernel function K and bandwidth h need to be set artificially. The selection of kernel function has little effect on the result; however, the selection of bandwidth has a great influence on the result [36]. In this study, the bandwidth selection method proposed by Silverman was used to preliminarily determine the bandwidth h; the calculation formula is as follows [37]:

$$h=0.9\times \min \left(SD,\sqrt{\frac{1}{ln2}}\times D_m\right)\times n^{-0.2}$$

(4)

$$SD=\sqrt{\frac{{\displaystyle \sum _{i=1}^n{\left(x_i-X\right)}^2}+{\displaystyle \sum _{i=1}^n{\left(y_i-Y\right)}^2}}{n}}$$

(5)

where D_m represents the median distance from each steel plant to the center point (the geographic center of all steel plants); x_i and y_i represent the latitude and longitude of the steel plant i, respectively; X and Y represent the latitude and longitude of the center point; and n is the number of steel plants.

2.4.2. Standard Deviational Ellipse

The standard deviational ellipse (SDE) can intuitively describe the directional, density, and dynamic changing trend of geographic element spatial distribution, and is widely used in spatial evaluation [38]. The coordinates of the center of the SDE, respectively denoted as SDE_x and SDE_y, can be calculated using the following formulas:

$${\mathrm{SDE}}_x=\sqrt{\frac{{\displaystyle \sum _{i=1}^n{\left(x_i-\overline{X}\right)}^2}}{n}},{\mathrm{SDE}}_y=\sqrt{\frac{{\displaystyle \sum _{i=1}^n{\left(y_i-\overline{Y}\right)}^2}}{n}}$$

(6)

where x_i and y_i represent the latitude and longitude of steel plant i, respectively; $\overline{X}$ and $\overline{Y}$ represent the center point of all steel plants; and n is the number of steel plants.

Then, the direction of the SDE, and the major and minor axis of the SDE, can be calculated by the following formulas:

$$\tan \theta =\frac{{\displaystyle \sum _{i=1}^n{\omega }_i{\tilde{x}}_i^2}-{\displaystyle \sum _{i=1}^n{\omega }_i{\tilde{y}}_i^2}+\sqrt{{\left({\displaystyle \sum _{i=1}^n{\omega }_i{\tilde{x}}_i^2}-{\displaystyle \sum _{i=1}^n{\omega }_i{\tilde{y}}_i^2}\right)}^2+4{\left({\displaystyle \sum _{i=1}^n{\omega }_i^2{\tilde{x}}_i{\tilde{y}}_i}\right)}^2}}{2{\displaystyle \sum _{i=1}^n{\tilde{x}}_i{\tilde{y}}_i}}$$

(7)

$${\sigma }_x=\sqrt{2}\times \sqrt{\frac{{{\displaystyle \sum _{i=1}^n\left({\omega }_i{\tilde{x}}_i\cos \theta -{\omega }_i{\tilde{y}}_i\sin \theta \right)}}^2}{{\displaystyle \sum _{i=1}^n{\omega }_i}}}$$

(8)

$${\sigma }_y=\sqrt{2}\times \sqrt{\frac{{{\displaystyle \sum _{i=1}^n\left({\omega }_i{\tilde{x}}_i\sin \theta +{\omega }_i{\tilde{y}}_i\cos \theta \right)}}^2}{{\displaystyle \sum _{i=1}^n{\omega }_i}}}$$

(9)

where θ is the rotation angle of the ellipse; ${\mathsf{\sigma }}_x$ and ${\sigma }_y$ represent the length of the major and minor axis respectively; ${\omega }_i$ represents the area of the steel plant i; and ${\tilde{x}}_i$ and ${\tilde{y}}_{\mathrm{i}}$ represent the difference of latitude and longitude between the steel plant i and the center point of all steel plants, respectively.

2.4.3. Bivariate Spatial Autocorrelation

The bivariate spatial autocorrelation analysis, proposed by Anselin, can effectively reflect the association and dependence characteristics of the spatial distribution of the two kinds of variables [39]. The bivariate spatial autocorrelation was divided into bivariate global spatial autocorrelation (also known as bivariate Moran’s I) and bivariate local spatial autocorrelation (also known as bivariate LISA); the bivariate Moran’s I can be calculated as follows:

$$\mathrm{I}=\frac{{\displaystyle \sum _{i=1}^n{\displaystyle \sum _{j=1}^n{\omega }_{ij}\left(x_i-\overline{x}\right)\left(y_j-\overline{y}\right)}}}{S^2{\displaystyle \sum _{i=1}^n{\displaystyle \sum _{j=1}^n{\omega }_{ij}}}}$$

(10)

where $\overline{x}$ and $\overline{y}$ represent the mean of the changing area of steel plants and the mean of changing area of land, respectively; ${\omega }_{ij}$ represents the spatial weight matrix; x_i and y_j represent the changing area of steel plants in the spatial unit i and the changing area of land in the spatial unit j, respectively; S² represents the variance of all samples; and n is the number of all spatial units.

The calculation method of bivariate LISA as follows:

$${\mathrm{I}}_i=\frac{x_i-\overline{x}}{S_x}{\displaystyle \sum _{j=1}^n{\omega }_{ij}\frac{y_j-\overline{y}}{S_y}}$$

(11)

where $S_x$, $S_y$ are the variance of the changing area of steel plants and the changing area of land, respectively, and the rest is the same as Equation (10).

In this study, the changing areas of steel plants and land cover were mapped to 500 m $\times$ 500 m grids covering Tangshan City, and queen adjacency was chosen to construct spatial weight matrices with orders of contiguity (OC) of 1, 3, 5, and 10, respectively. The spatial correlation between the two changes was explored by using bivariate global Moran’s I and bivariate LISA. By using Geoda version 1.20, spatial weight matrix was constructed and bivariate spatial autocorrelation analysis was performed.

3. Results

3.1. Results of Segmentation and Change Region Identification of Steel Plants

In this study, we built a steel plant dataset based on the obtained Google Earth images for 2017 and 2022, in which the images have 1024 $\times$ 1024 pixels. The dataset contained a training set and two test sets, Seg-SP17 and Seg-SP22. The Seg-SP17 contained 84 images of Tangshan City in 2017, the Seg-SP22 contained 111 images of Tangshan City in 2022, and the training set contained 2085 images of other cities. Before training the model, data augmentation methods, provided by the PIE-Engine AI platform, including color change, horizontal flipping, vertical flipping, and ninety degree rotation, were used for data augmentation.

The segmentation results of steel plant using only RSIs (marked as initial results), and the segmentation results of steel plant using the strategy proposed in this study with the DT of 500 m (marked as final results), are shown in Figure 5. In the Seg-SP17 test set, the precision, recall, F1 score, and IoU of the initial results were 50.09%, 53.71%, 51.84%, and 34.99%, respectively. The precision, recall, F1 score, and IoU of the final results were 82.35%, 55.30%, 66.17%, and 49.44%, respectively. In the Seg-SP22 test set, the precision, recall, F1 score and IoU of the initial results were 59.89%, 63.75%, 61.76%, and 44.68%, respectively. The precision, recall, F1 score, and IoU of the final results were 84.77%, 70.11%, 76.74%, and 62.26%, respectively.

3.2. Analysis of Spatial Pattern Change in Steel Plants in Tangshan City

To improve the accuracy of subsequent analysis results as much as possible, the segmentation results of the steel plants were corrected manually before conducting subsequent analysis. Following that, the changing regions of steel plants were identified, and the corresponding changing points with changing area were obtained.

The area change in steel plants in all counties of Tangshan City from 2017 to 2022 is shown in Figure 6. It can be seen that the steel plants in Tangshan City were mainly concentrated in Fengnan, Laoting, Caofeidian, Luanzhou, and Qian’an, and that the area of the steel plants in these counties had increased to varying degrees from 2017 to 2022. Among them, the area of the steel plants in Laoting and Caofeidian had increased significantly, with an area increase of over 1000 hm² and a growth rate of over 100%. In 2022, the area of the steel plants in Laoting and Caofeidian exceeded that in Qian’an, which ranked first in Tangshan City in terms of steel plant area in 2017. The area of the steel plants in Fengrun, Luannan, and Yutian had significantly decreased, among which Fengrun has the largest reduction of 180 hm², and the area of the steel plants in Yutian decreased to 0 hm² in 2022.

SDE and KDE, with the area of steel plants as the weight parameter, were used to depict the spatial distribution changes of steel plants in Tangshan City during 2017–2022. The results of KDE were classified into three categories of low-density regions, medium-density regions, and high-density regions using Jenks Natural Breaks Algorithm (JNBA) [40], the same below. The KDE results are shown in Figure 7. Figure 7a,b show that the high-density regions were mainly concentrated in southwestern Qian’an, southern Caofeidian, and southern Laoting in both 2017 and 2022. However, compared to that in 2017, the high-density region in Qian’an contracted in 2022, while the high-density region in Laoting significantly expanded. Additionally, the medium-density regions in the central Tangshan also experienced contraction, and new medium-density regions appeared in southwestern Fengnan and central Caofeidian. From Figure 7c,d, it can be seen that obvious NC occurred in northern Fengnan, central Fengrun, and obvious PC occurred in southern Fengnan, southern Caofeidian, and southern Laoting.

The results of SDE are shown in Figure 8. It can be seen that the spatial distribution of steel plants in Tangshan City was in the northwest–southeast direction in 2017 and the same goes for 2022, without obvious change. However, the spatial distribution center of steel plants had obviously moved to the southeast. In 2022, both the major and minor axes of SDE were longer than that of 2017, indicating that the directionality of the spatial distribution of steel plants in Tangshan City was more obvious, and the centripetal degree of the spatial distribution of the steel plants in Tangshan City was weakened.

3.3. Relationships between Steel Plant Spatial Pattern Change and Land Cover Change in Tangshan City

The single land use dynamic degree (SLUDD) and the land use transfer matrix (LUTM) were used to quantitatively analyze land cover change in Tangshan City.

The SLUDD can quantify the dramatic degree of land use change over a period of time and has an important role in studies related to land cover change [41]. The calculation formula is as follows:

$$SLUDD\left(i\right)=\frac{C\left(i,t_2\right)-C\left(i,t_1\right)}{\Delta tC\left(i,t_1\right)}\times 100\%$$

(12)

where SLUDD(i) is the single land use dynamic degree of land cover type i; C(i, t₁) and C(i, t₂) represent the areas of land cover type i for the start year t₁ and the end year t₂, respectively; and Δt equals t₂ − t₁.

The LUTM is currently the most widely used method in studies of land cover change. LUTM is the application of the Markov model in land use, which can reflect the changes in land use structure comprehensively and concretely [42]. Its formula is as follows:

$$S_{ij}=\left(\begin{array}{ccc}{S}_{11}& \dots & S_{1n}\\ ⋮& \ddots & ⋮\\ S_{n1}& \cdots & S_{\mathrm{n}n}\end{array}\right)$$

(13)

where i and j represent the land cover type at the start year and the end year, respectively; S_ij represents the area converted from i to j; and n is the number of land cover types.

The land cover data for 2017 and 2022 were used to calculate the LUTM and the SLUDD of Tangshan City from 2017 to 2022, and the calculation results are shown in

Table 1. Results of LUTM and SLUDD in Tangshan City during 2017–2022, the values of changing area exceeding 100 hm² are in bold.

	W * (hm2)	V * (hm2)	C * (hm2)	BA * (hm2)	BG * (hm2)	Total (hm2)	LUDD (%/Year)
W (hm2)	1303.612	24.768	41.532	47.112	9.907	1426.931	0.109
V (hm2)	21.124	2151.671	401.361	167.402	11.035	2752.593	−1.774
C (hm2)	81.110	242.771	5874.815	328.411	3.822	6530.929	−0.257
BA (hm2)	10.675	54.678	117.097	3113.839	4.360	3300.649	2.299
BG (hm2)	8.155	34.618	12.071	33.262	23.770	111.876	−10.544
Total (hm2)	1424.676	2508.506	6446.876	3690.026	52.894	-	-

* The W represents the Water class, the V represents the Vegetation class, the C represents the Cropland class, the BA represents the Built Area class, and the BG represents the Bare Ground class.

.

For the convenience of description, A-B is used in the following text to represent the transition from class A to class B. As can be seen from Table 1, V, C, and BA were the main land cover classes in Tangshan City, and the total area of C was the largest (6530.929 hm² in 2017 and 6446.876 hm² in 2022). The land cover change in Tangshan City was relatively dramatic. From the perspective of change area, V-C, V-BA, C-V, C-BA, and BA-C were the main change types, with change area over 100 hm². From the perspective of SLUDD, the class with the most drastic change was BG (−10.544%/year), and the class with the most stable change was W (0.109%/year). Among the three major land cover classes in Tangshan City, BA (2.299%/year) underwent the most drastic change and showed an increase trend, and the main land cover classes that transferred in BA were C with an area of 328.411 hm² and V with an area of 167.402 hm². V (−1.774%/year) also underwent a relatively drastic change and showed a decrease trend, and mainly transferred out as C and BA with area of 401.361 hm² and 167.402 hm², respectively. C (−0.257%/year) underwent a relatively stable change and the total area changed slightly. However, the area of transitions from other land cover classes to C was 518.458 hm², and the area of transitions from C to other land cover classes was 571.182 hm², which shows that C changed actively.

Using bivariate spatial autocorrelation to explore the spatial correlation between the spatial pattern change in steel plants and land cover change in Tangshan City from 2017 to 2022. The calculation results of the bivariate Moran’s I are shown in

Table 2. Results of bivariate Moran’s I, the two largest values are in bold.

		V-C	V-BA	C-V	C-BA	BA-C
OC = 1	PC	−0.014	0.212	0.006	0.137	−0.016
	NC	−0.003	0.002	−0.003	−0.005	0.007
OC = 3	PC	−0.010	0.122	0.004	0.063	−0.012
	NC	−0.003	0.002	−0.004	−0.001	0.006
OC = 5	PC	−0.009	0.060	0.002	0.033	−0.011
	NC	0.000	0.001	−0.003	0.002	0.003
OC = 10	PC	−0.010	0.033	−0.004	0.009	−0.011
	NC	−0.001	0.000	−0.003	0.007	0.004

. It can be seen that PC was positively correlated with C-BA, V-BA, and C-V, while negatively correlated with BA-C and V-C. Among them, a strong spatial correlation is observed between PC and V-BA, as well as PC and C-BA, and the spatial correlations weakened with the increase in OC. When OC = 1, the bivariate Moran’s I of PC with C-BA was 0.137, and that of PC with R-BA was 0.212. The spatial correlations between NC and the five major land cover change types were not strong.

Group 1 (PC and C-BA at OC = 1) and group 2 (PC and V-BA at OC = 1), with relatively strong spatial correlation, were selected to draw bivariate LISA clustering maps. The results are shown in Figure 9. It can be seen that the H-H clustering regions of group 1 were distributed in southeastern Laoting, central Caofeidian, southwestern and central Fengnan, southwestern and central Luanzhou, and southwestern Qian’an. And the H-H clustering regions of the group 2 were distributed in southeastern Laoting, central and southern Caofeidian, western Qianxi, and southwestern Qian’an.

Unexpectedly, as shown in Figure 7d and Figure 9a, central Fengnan and southwestern Luanzhou were classified as low-density regions in the KDE result of PC, while they were clustered as H-H clustering regions in the bivariate LISA clustering map of PC and C-BA, which appears to be contradictory. To validate these findings, satellite images were employed to examine the two regions. As shown in Figure 10, the spatial pattern change in steel plants and the land cover change in central Fengnan and southwestern Luanzhou, were consistent with the bivariate LISA clustering map of PC and C-BA, that is, PC and the transition of C to BA both occurred in the two regions. The contradiction may arise because the steel plants in the two regions underwent small-scale expansion of the original steel plants. The changing areas are relatively small compared to that in the coastal regions, and without PC nearby, resulting in a smaller KDE value in the two regions, which were classified as low-density regions in the KDE results.

4. Discussion

4.1. Effect of POI Data and OSM Data on Segmentating Steel Plants

To investigate the impact of different DTs on the post-processing of the initial steel plant segmentation results using steel plant POIs, we selected three DTs of 500 m, 1000 m, and 1500 m. In addition, in order to evaluate the effect of POI data and OSM data on segmentating steel plants, we set up the following steel plant segmentation strategies: strategy using RSIs (marked as RSI), strategy using RSIs and OSM data (marked as RSI + OSM), strategies using RSIs and POI data with different DTs (respectively marked as RSI + POI_500 m, RSI + POI_1000 m, and RSI + POI_1500 m), and strategies using the three types of data with different DTs (respectively marked as RSI + OSM + POI_500 m, RSI + OSM + POI_1000 m, and RSI + OSM + POI_1500 m). The experiments were conducted on the Seg-SP17 test set and the Seg-SP22 test set, respectively. The evaluation results of different strategies are shown in

Table 3. Evaluation of different steel plant segmentation strategies.

Test Set	Strategy	Precision (%)	Recall (%)	F1 Score (%)	IoU (%)
Seg-SP17	RSI	50.09	53.71	51.84	34.99
	RSI + OSM	54.80	56.51	55.64	38.54
	RSI + POI_500 m	73.70	46.10	56.72	39.58
	RSI + POI_1000 m	61.04	52.92	56.69	39.56
	RSI + POI_1500 m	55.31	53.68	54.48	37.44
	RSI + OSM + POI_500 m	82.35	55.30	66.17	49.44
	RSI + OSM + POI_1000 m	67.48	56.08	61.25	44.15
	RSI + OSM + POI_1500 m	60.80	56.81	58.74	41.58
Seg-SP22	RSI	59.89	63.75	61.76	44.68
	RSI + OSM	65.42	72.20	68.64	52.26
	RSI + POI_500 m	78.84	52.64	63.13	46.12
	RSI + POI_1000 m	71.13	63.68	67.20	50.60
	RSI + POI_1500 m	64.95	63.75	64.34	47.43
	RSI + OSM + POI_500 m	84.77	70.11	76.74	62.26
	RSI + OSM + POI_1000 m	77.12	72.16	74.56	59.44
	RSI + OSM + POI_1500 m	70.70	72.18	71.43	55.56

, and the best values for each metric are shown in bold. From Table 3, it can be seen that:

(1)

The RSI + OSM strategy effectively improves the effectiveness of steel plant segmentation. Compared to the RSI strategy, the experiment in the Seg-SP17 test set showed improvements of 4.17% in precision, 2.90% in recall, 3.84% in F1 score, and 3.55% in IoU, and that, in the Seg-SP22 test set, these values were 5.53%, 8.45%, 6.88%, and 7.58%, respectively. It can be observed that the RSI + OSM strategy showed greater improvements in the Seg-SP22 test set compared to the Seg-SP17 test set. This is because the steel plant parcels segmented by OSM data in the coastal regions of Tangshan, particularly in southeastern Laoting and southwestern Fengnan, were better matched with the actual situation in 2022 compared to that in 2017. In other words, there is a stronger correlation between the actual steel plants and the steel plant parcels in the Seg-SP22 test set. By counting the number of steel plant parcels that meet the judgment criteria in the two test sets, it can be seen that there are 15 steel plant parcels that meet the criteria in Seg-SP22 test set, which is approximately twice that in Seg-SP17 test set (7 parcels).

(2)

Compared to the RSI strategy, the strategies of using RSIs and POI data showed significant improvements in precision in both the Seg-SP17 and Seg-SP22 test sets. However, these strategies have a negative impact on recall. The varying degrees of improvements and reductions depended on the selection of the DT. In the Seg-SP17 test set, the improvements in precision and recall were 23.61% and −7.61% at DT = 500 m, 6.24% and −3.69% at DT = 1000 m, and 5.22% and −0.03% at DT = 1500 m, respectively. While that in the Seg-SP22 test set were 18.95% and −11.11% at DT = 500 m, 11.24% and −0.07% at DT = 1000 m, and 5.06% and −0.00% at DT = 1500 m, respectively. It can be observed that the strategies of using RSIs and POI data have a trade-off effect between recall and precision, with a lower DT resulting in a more pronounced improvement in precision and reduction in recall. Additionally, the strategies also showed improvements in F1 score and IoU, where the best values for F1 score and IoU were achieved at DT = 500 m in Seg-SP17 and DT = 1000 m inSeg-SP22, respectively.

(3)

Compared to the RSI strategy, the strategies of using RSIs, OSM data, and POI data can significantly enhance the effectiveness of steel plant segmentation. In the Seg-SP17 test set, the improvements were observed as: 32.26% in precision, 1.59% in recall, 14.33% in F1 score, and 14.45% in IoU at DT = 500 m; 17.39% in precision, 2.37% in recall, 9.41% in F1 score, and 9.16% in IoU at DT = 1000 m; and 10.71% in precision, 3.10% in recall, 6.90% in F1 score, and 6.59% in IoU at DT = 1500 m. While that in the Seg-SP22 were: 24.88%, 6.36%, 14.98%, and 17.58% at DT = 500 m; 17.23%, 8.41%, 12.80%, and 14.76% at DT = 1000 m; and 10.71%, 3.10%, 6.90%, and 6.59% at DT = 1500 m. Meanwhile, compared to the strategy that combines RSIs and OSM data, as well as the strategies that combine RSIs and POI data with different DT, these strategies can significantly improve the precision of steel plant segmentation while effectively controlling the reduction in recall. Among all steel plant segmentation strategies, the best F1 score and IoU for both Seg-SP17 and Seg-SP22 appeared when using this strategy with a DT of 500 m.

The visual comparisons between the six different strategies of steel plant segmentation are shown in Figure 11. As can be seen from these figures, if the steel plant parcels segmented using OSM data were more in line with actual planning, the RSI + OSM strategy can effectively improve the TP cases, thereby improving the steel plant segmentation accuracy (Figure 11d1). However, if the steel plant parcels are significantly different from the actual planning, that is, if the steel plant parcels contain more other ground objects besides the steel plant, this strategy has no effect on optimizing the results (Figure 11d3). Surprisingly, in the results of the Seg-SP22 test set, this strategy unexpectedly led to an increase in FP cases (Figure 11d2), which led to the reduction in precision, F1 score, and IoU. Fortunately, these occurrences are extremely rare. The strategies employing RSIs and POI data effectively reduce FP cases, thereby significantly improving the precision of the results, and, the smaller the DT, the more obvious the effect. However, a too-small DT may also filter out a portion of TP cases, leading to an increase in FN cases (Figure 11e3). The decrease in TP cases and the increase in FN cases together contribute to the decrease in recall.

4.2. Effect of the Spatial Pattern Change in Steel Plants on Land Cover Change in Tangshan City during the Study Period

Cropland is a natural resource essential to human survival, and strict protection of arable land is crucial for food security, social stability, and economic stability [43]. In 2004, China set a red line of 180 million mu (120 million hectares) for cropland protection, which is still in use now [44]. PC was positively correlated with C-BA spatially in Tangshan City during 2017–2022, and the bivariate LISA clustering map of PC and C-BA shows that the effect of PC on C-BA was concentrated in southeastern Laoting, central Caofeidian, southwestern and central Fengnan, southwestern and central Luanzhou, and southwestern Qian’an. Meanwhile, the transitions of C were relatively active in Tangshan City from 2017 to 2022, with a small decrease in the total area (52.724 hm²). This indicates that Tangshan City paid more attention to cropland protection in the process of the spatial relocation of steel plants.

Vegetation has an important role in preventing soil erosion, maintaining biodiversity, preventing climate change, and increasing soil carbon storage [45]. Among the three major land cover types in Tangshan City, V changed the most drastically, with a significant decrease in a total area of 244.087 hm² and LUDD of −1.744%/year. The bivariate Moran’s I of PC and V-BA at OC = 1 is 0.212, indicated a relatively strong spatial correlation. The bivariate LISA clustering map of PC and V-BA at OC = 1 shows that the H-H clustering regions were mainly concentrated in the coastal regions of southern Tangshan City, indicating a strong connection between PC and the transition of V-BA in the regions. In resource-based cities such as Tangshan City, the long-term exploitation of natural resources and the disorderly expansion of resource-based industries seriously damage the ecological environment and weaken the connections between ecosystems [46]. The degradation of the ecological environment will have a negative impact on urban production and life. When the ecological environment deteriorates to the point where it seriously threatens human health, the government is required to allocate substantial funds and change development models to enhance the urban ecological environment [47]. Existing studies show that the interrelationship between urban development and the ecological environment is lagging, and optimizing urban development models can not immediately improve the ecological environment [48]. This means that the restoration of the ecological environment requires not only a large amount of funds but also a long restoration period, and the development of the urban economy may be constrained. In addition, Tangshan City is currently in the stage of urbanization development, and studies have shown that occupying original ecological land is the driving force for further urbanization [27]. Therefore, in the subsequent industrial spatial restructuring and urbanization development process of Tangshan City, the protection of the ecological environment and the layout of ecological space should be given utmost attention by the authorities.

4.3. Driving Mechanism and Possible Impact of the Spatial Pattern Change in Steel Plants in the Context of Resource-Based City Transformation

In the context of industrial transformation, the spatial pattern of steel plants has changed drastically in Tangshan City during 2017–2022, with the spatial distribution center obviously migrating to the southeast. The areas of steel plants in central Fengrun and northern Fengnan have decreased, while that in the southern coastal regions (including southwestern Fengnan, southern Caofeidian, and southern Laoting) have increased significantly. These changes were the result of the combined effect of multiple factors.

In terms of the policy factor, Tangshan has introduced a series of measures and policies targeting the relocation of steel plants, including relocation compensation, tax incentives, industrial land approval, and favorable water and electricity charges. These incentive measures and subsidy policies greatly alleviate the financial pressure faced by steel enterprises during the relocation process, as it involves production stagnation, reduced revenue, and the need for significant capital investment in the reconstruction of plants.

In terms of the market factor, the distributional disparity of iron ore resources in China results in the characteristic of stronger steel production capacity in the north and a weaker capacity in the south. North-to-south steel transportation has become the primary means of regional adjustment. Relocating steel plants from inland regions to coastal regions is beneficial for reducing logistics costs for steel enterprises, including the transportation costs of imported iron ore from the ports to the plants, as well as the transportation costs of steel products from the plants to the ports, which helps enhance the competitiveness and market share of steel products.

In terms of the ecological factor, the resources and environmental carrying capacity of urban regions is limited. Steel plants consume significant amounts of resources during the production process while also generating substantial pollution, which negatively impacts the surrounding environment and the health of residents. Relocating steel plants away from the urban center regions is beneficial for alleviating the pressure on resources and environment in the region, thereby improving both the ecological environment and the living environment for the region.

In terms of the land factor, the rapid urban development and population concentration in Tangshan City have resulted in steel plants, which were originally located far from urban residential regions, now being located nearby or even surrounded by the urban residential regions. This has led to a fragmented landscape pattern and disordered functional distribution in the central region of Tangshan City. Relocating steel plants away from the urban center regions is advantageous for optimizing the landscape pattern and restoring the order of functional distribution in the region.

The driving mechanisms and possible impacts of the spatial pattern changes of steel plants in Tangshan can be summarized as follows:

(1)

National policies drive Tangshan’s industrial transformation. The long-term extensive economic development model, dominated by mining and heavy industries, has led to prominent issues in resource-based cities, such as the imbalance of industrial structure, heavy reliance on resources, environmental degradation, and low land use efficiency. These issues have gradually reduced the investment attractiveness of resource-based cities and hindered their sustainable development. Industrial transformation and upgrading are crucial for modern societal development and the successful transformation of resource-based cities. To promote the transformation and sustainable development of resource-based cities in China, the State Council issued the “National Sustainable Development Plan for Resource-Based Cities (2013–2020)” in 2013 [49]. This plan outlines the overall strategic planning for the transformation and development of resource-based cities in China and promotes their industrial transformation. In 2017, the National Development and Reform Commission (NDRC) issued a notice titled “Supporting the Construction of Demonstration Zones for Industrial Transformation and Upgrading in the First Batch of Old Industrial Cities and Resource-Based Cities”, in which Tangshan City was designated as one of the first demonstration zones for industrial transformation and upgrading [50]. The notice requires the local governments in the demonstration zones to actively explore suitable paths and models for industrial transformation and upgrading based on their local conditions and characteristics. The following year, in order to improve air quality and address deteriorating living environments, the State Council issued a notice titled “Three-Year Action Plan for Winning the Battle for Blue Skies” [51], which designated the Beijing-Tianjin-Hebei region, including Tangshan City, as key regions. The notice emphasizes the adjustment and optimization of industrial structure as one of the important work objectives. These policies have accelerated the pace of industrial transformation in Tangshan City.

(2)

Tangshan’s industrial transformation led to the spatial pattern change in steel plants. The process of industrial transformation and the changes in industrial spatial patterns are interconnected in resource-based cities. On one hand, the industrial transformation and upgrading of resource-based cities often involve an adjustment of industrial structure, which is an important influencing factor in the changes of industrial spatial patterns [7,8]. On the other hand, by optimizing the industrial spatial pattern, it is possible to enhance the spatial competitiveness of enterprises, reduce production costs, and promote technological innovation, thereby driving the industries towards higher added value and technological content and achieving the transformation and upgrading of the industries [52,53]. Due to long-term development and renewal, the central region of Tangshan City has a relatively well-developed infrastructure and supporting facilities, serving as the commercial and residential center of the city. Steel plants occupy a significant amount of land, and the flow of people and logistics associated with steel plants can contribute to congestion and environmental issues in the city, limiting the development potential of the central region. Relocating the steel plants away from the region can free up land resources, alleviate the pressures on traffic and population, and attract high-value-adding industries, such as technology-intensive and knowledge-intensive industries, optimizing the city’s industrial structure. Additionally, the spatial restructuring of plants involves various aspects such as equipment relocation and reconstruction, factory region redevelopment, and employee resettlement. For the steelmaking industry, due to the uniqueness of steelmaking equipment, there are limited facilities that can be relocated and utilized, making it necessary to establish new facilities in a new location. This opportunity allows steel plants to phase out outdated equipment and redesign production capacity, which has significant positive implications for the development of the steelmaking industry and the reduction in excess capacity.

(3)

Tangshan’s future development plan decides the changing direction of the spatial pattern of the steel plants. When city policymakers engage in urban planning, they need to consider the current challenges and future development directions, ensuring that urban planning is both realistic and forward-looking. The “overall land-use plan for Tangshan City (2021–2035)” [54], published by the Tangshan Bureau of Natural Resources and Planning, proposes four major transformations including working towards an ecological civilization demonstration zone, a modern coastal city with land–sea linkage, a strong innovation-driven industrial city, and a prosperous, comfortable, beautiful, and livable city. Three major target, including serving as a window city for economic cooperation in Northeast Asia, a new industrialization base in the Bohai-Rim region, and an important hub in the Capital Economic Circle, were also proposed in the plan. These indicate that, in its future development, Tangshan will adjust its economic development model, promote ongoing industrial transformation, emphasize the protection of the ecological and living environment, leverage its geographical location and industrial foundation advantages, promote industrial agglomeration and develop industrial parks, strengthen exchanges and cooperation with Northeast Asian regions, and enhance and coordinate in the division of labor and collaboration with other cities within the Capital Economic Circle. In addition, the plan also mentions the “One Port, Two Cities” spatial development pattern, which signifies the establishment of a new urban center in the Caofeidian District, in addition to the existing central urban area of Tangshan City. These plans decided the spatial change direction for the steel plants of relocating from the central region towards coastal regions. The relocation of steel plants from the central region to coastal regions can alleviate environmental pollution and improve air quality in the central region, contributing to Tangshan’s transformation into an ecological civilization demonstration zone and a prosperous, comfortable, beautiful, and livable city. An efficient transportation network connecting the port and the inland is crucial for achieving the integration of land and sea. The relocation of steel plants will bring a significant population influx to the coastal regions, leading to an improvement in the surrounding infrastructure, including roads and railways. This development will contribute to Tangshan’s transformation into a modern coastal city with land–sea linkage. Additionally, the population influx and the improvement in infrastructure will further drive the development of the coastal regions, supporting the implementation of the “One Port, Two Cities” strategy, which can alleviate environmental and resource pressures in the central region. The relocation of the steel plants to the coastal regions can also leverage the influence of the steel industry as a dominant industry, attracting upstream and downstream industries, as well as other industries, to gather in the coastal regions, promoting the formation and development of the new Bohai-Rim industrialization base. The Bohai Sea serves as a gateway for communication between Tangshan and Northeast Asian countries. Its strategic geographical location makes it an important hub for exchanges between China and Northeast Asian countries. The formation and development of the new Bohai-Rim industrialization base will strengthen the connection and interaction between Tangshan and Northeast Asian countries. The strengthening of the connection and the interaction will promote economic cooperation and exchange, facilitate resource sharing, industrial collaboration, and market openness in the region, which is of great significance for promoting the formation of a development pattern that benefits all parties and achieves regional sustainable development.

4.4. Limitations and Prospects

One of the main aspects of our work is to detect the changes in steel plants based on high-resolution remote sensing images. However, unlike most existing change detection methods that directly detect changes from multi-temporal remote sensing images, we, respectively, segmented the steel plants from the individual remote sensing images for 2017 and 2022, then performed post-processing using POI data and OSM data. Finally, we obtained the changes of steel plants based on the segmentation results. One of the primary reasons for adopting this approach is the relatively low number of steel plants, and the fact that even fewer of them underwent changes. This poses challenges in constructing sufficient sample datasets for steel plant change detection, thereby hindering the effectiveness of supervised change detection methods.

Compared to supervised methods that require labeled data, unsupervised change detection methods can effectively extract change regions from multi-temporal remote sensing images without needing labeled data [55,56]. The goal of unsupervised change detection methods is to detect all potential changes without any labeled data; however, due to the lack of category information provided by labeled data, unsupervised change detection methods can not determine the types of changes, which is one of their main limitations [57]. This limitation makes it challenging to apply unsupervised change detection methods to the tasks that require specific attention to certain types of changes, such as focusing solely on the changes of steel plants in this study.

It is noteworthy that existing research has shown that Point of Interest (POI) data can effectively reflect human economic activities [58]. That is to say, if the research subject is appropriate, POI data can help indicate the content of detected changes. Therefore, in order to selectively identify changes of interest, or exclude changes of less interest, applying POI data to unsupervised change detection, and utilizing it to constrain the geographical locations of detected changes, may be a promising research direction.

It is important to clarify that our work did not propose a new network model or method specifically designed for segmenting steel plants. Instead, we focused on a post-processing strategy, in which we employed POI data and OSM data in order to optimize the initial results obtained from a deep learning network model (the DeepLabV3+ model was selected as the deep learning model in our work) and make the final results align as closely as possible with the ground truth (GT). Our goal is to address the issues caused by insufficient labeled data and the complexity of the segmentation task, such as the misidentification of non-steel plant objects as steel plants and incomplete segmentation of certain steel plants. As a result, we did not compare the proposed strategy with other change detection or semantic segmentation methods. Instead, we evaluated the effectiveness of our proposed strategy by comparing the accuracy of the results with/without its implementation.

From the current study results, the spatial correlation between NC and the five main types of land cover changes under the current land classification system is relatively weak. However, under a more fine-grained land use classification system, another phenomenon may appear (such as the conversion of industrial and mining land to residential land or industrial land to service industry land, etc.). Therefore, we believe that analyzing the changes in steel plants, and their spatial relationship with land use based on a more fine-grained land use classification system, is also a meaningful research direction.

In this study, we used bivariate spatial autocorrelation analysis to analyze the relationships between spatial pattern change in steel plants and land cover changes, even if the conclusions seem to be obvious even without the bivariate spatial autocorrelation analysis. This is because existing research suggests that the relocation of large plants has significant impacts on land use, and the impacts are significant and lagged [59]. The most intuitive impact is the conversion of other land use types into industrial land for the construction of plants, warehouses, and other industrial facilities. Additionally, the influx of a large labor force accompanying the relocation of large plants may also lead to the establishment of residential, commercial, and service facilities in the surrounding regions. Furthermore, the normal operation of large plants requires infrastructure support such as roads, electricity, water supply, and drainage, which can be factors contributing to local LUCC.

Bivariate spatial autocorrelation analysis can effectively analyze the spatial correlation between two variables, and many scholars have applied it in LUCC-related research [60,61]. In our experimental analysis using bivariate spatial autocorrelation, we set the changing area of steel plants as the independent variable and the changing area of land as the dependent variable to explore the relationship between spatial pattern changes of steel plants and land cover changes. Additionally, we set different orders of contiguity (referred to as OC in this paper), where increasing the OC included more observations as neighbors in the analysis, which allowed us to explore the influence range of the spatial pattern changes of steel plants on the land cover changes. Unfortunately, the experimental results show that only at OC = 1 is there a strong correlation between spatial pattern changes of steel plants and some types of land use changes, indicating that the spatial pattern changes of steel plants are only spatially correlated with land cover changes in a small region nearby. This may be due to the relatively short time interval between the research and the completion of steel plant relocations, resulting in the impacts having not yet spread. We believe that the continuous monitoring of the impact of steel plant relocation on the land cover pattern in the coastal regions of Tangshan City is also a meaningful research direction, which can provide valuable practical references for the formulation of future land use planning in Tangshan City and the transformation of other resource-based cities.

5. Conclusions

Understanding the spatial pattern change in steel plants and their effect on land cover in the process of industrial transformation of steel cities, is crucial to the formulation of transformation planning for resource-based cities. In this study, we proposed a steel plant segmentation strategy that combines RSIs, POI data, and OSM data, and analyzed the effect of POI data and OSM data on steel plant segmentation. Finally, the strategy was applied to the segmentation of steel plants in Tangshan City for 2017 and 2022. Based on the segmentation results, further exploration and analysis were conducted on the spatial pattern change in steel plants in Tangshan City during 2017–2022 and its relationship with land cover. The research results indicate that:

(1)

Compared to using only RSIs, the steel plant segmentation strategy combining RSIs and POI data can significantly improve the precision of steel plant segmentation; however, it will, to some extent, reduce the recall of steel plant segmentation. This phenomenon weakens as DT increases. For this strategy, the best F1 score and IoU appeared at DT = 500 m in the Seg-SP17, while, in the Seg-SP22 test set, the best F1 score and IoU appeared at DT = 1000 m. The steel plant segmentation strategy combining RSIs and OSM data can effectively improve the effectiveness of steel plant segmentation; however, it has significant limitations. The segmentation results of combining RSIs, POI data, and OSM data were the most excellent, and the best F1 score and IoU in the Seg-SP17 and Seg-SP22 test sets both appear at DT = 500 m.

(2)

The area of steel plants increased significantly with a total area increase of 2714 hm², while the spatial distribution center of steel plants moved to the southeast significantly. The PC was mainly distributed in the southern coastal regions, including southeastern Laoting, southern Caofeidian, and southwestern Fengnan, and the NC was mainly distributed in the central area, including central Fengrun and northern Fengnan.

(3)

The V-BA had the strongest spatial correlation with the PC among the five main types of land cover change, with the bivariate Moran’s I of 0.212 (OC = 1), 0.122 (OC = 3), 0.060 (OC = 5), and 0.033 (OC = 10). The impact of the PC on the V-BA was concentrated in southeastern Laoting, central and southern Caofeidian, southwestern Fengnan, western Qianxi, and southwestern Qian’an. The C-BA also had a strong spatial correlation with the PC, with the bivariate Moran’s I of 0.137 (OC = 1), 0.063 (OC = 3), 0.033 (OC = 5), and 0.009 (OC = 10). The impact of the PC on the C-BA was concentrated in southeastern Laoting, central Caofeidian, southwestern and central Fengnan, southwestern and central Luan-zhou, and southwestern Qian’an. This indicates that the PC led to C and V transfer to BA during the spatial reorganization process of steel plants in Tangshan City during 2017–2022. Considering the area change in C and V, the protection of cropland in the transformation process of Tangshan City is relatively outstanding. However, in the subsequent spatial restructuring of steel plants and the development of urbanization, attention needs to be paid to the protection of the ecological environment.

The strategy proposed in this study effectively addresses the issue of unsatisfactory results for segmenting steel plants based on high-resolution remote sensing images. The research findings in this study can provide support for further industrial layout planning and land use planning in Tangshan City and can provide a realistic reference for the transformation of other resource-based cities.