Introduction

Land use and land cover change (LUCC) significantly influences various temporal and spatial scales1,2. Recent studies underscore LUCC’s critical role in understanding environmental sustainability3,4, urban growth5, and ecosystem health6. Precise measurement of LUCC across scales is essential for tackling issues such as climate change, carbon storage, and biodiversity conservation7,8,9.

The spatio-temporal scale of LUCC serves as a crucial foundation for understanding its scale dependence and sensitivity expression10,11. Studies have demonstrated that changes in LUCC significantly influence ecosystem functions and processes across various temporal and spatial scales. For example, in the context of urbanization, the intricate interactions of LUCC drive the evolution of spatial patterns in ecological networks, impacting ecological connectivity and the efficiency of ecological process cycles4. In arid river basins, the combined effects of LUCC and climate change substantially modify the spatiotemporal distribution of blue and green water resources12. Most research has predominantly concentrated on multi-scenario simulation and prediction13, driving mechanisms and influencing factors6,14, trade-offs and synergies in ecosystem services15, as well as spatiotemporal distribution patterns and environmental regulation16. The field is progressively shifting from static analyses at single scales to dynamic simulations across multiple scales and scenarios.

At present, the emergence of Google Earth Engine (GEE) has provided a crucial tool for the large-scale interpretation of remote sensing images, expanding the possibilities for multi-scale analysis of LUCC dynamics17,18,19. The annual calculation of land use classification results for LUCC data leads to data redundancy and affects efficiency. Moreover, large-scale remote sensing classification using machine learning algorithms typically exhibits lower accuracy compared to supervised classification methods, primarily due to the reliance on unsupervised classification techniques. If the patterns of land use change can be analyzed using a large number of remote sensing images before interpretation, efficiency can be significantly improved, and storage space can be conserved. Currently, there is still a lack of methods to automatically identify the key time nodes of LUCC, which are essential for uncovering the periodic change patterns of land use over a specific period and providing a scientific basis for temporal-scale division. Previous studies on spatial-scale have primarily focused on identifying suitable spatial resolutions20,21. However, regional spatial differentiation characteristics can be described by designing a spatial grid (fishnet size) that retains high spatial resolution. This grid can serve as an independent geographical unit to present statistical analysis results at the plot scale, thereby addressing the challenge of determining the appropriate fishnet size. Consequently, developing a suitable spatio-temporal scale analysis framework for LUCC remains a highly valuable area of exploration.

Identifying the spatio-temporal scale is fundamental to understanding land use change characteristics. Determining the appropriate spatio-temporal scale LUCC effectively elucidates complex geographical patterns22. Studies have shown that LUCC’s evolutionary traits and spatial phenomena vary across different spatial and temporal scales23,24,25. These studies employ various indicators to describe LUCC’s spatio-temporal variations, including the quantitative structure of land use types, land use dynamics, land use conflict measurement26,27, land use multifunctionality28, regional differences, and ecological environment quality indices29,30. They attempt to demonstrate that different spatial and temporal scales influence LUCC characteristics, change processes, and influencing factors31,32. However, these efforts fall short of directly proving a spatio-temporal scaling effect in LUCC. Characteristics observed at a specific spatio-temporal scale cannot be extrapolated to other scales. Thus, there is an urgent need to automatically identify the optimal spatial and temporal scales for LUCC, a current challenge in LUCC research.

Previous studies on the spatial-scale division of LUCC predominantly rely on empirical formulas33,34, direct references to prior findings35, or administrative boundaries36. However, there remains a need to partition the spatial-scale into finer grids, such as geographical units, to accurately calculate or simulate land use information3,37,38. These spatial scales can serve as independent geographical units for spatializing LUCC and its index parameters. Nonetheless, existing methods exhibit significant limitations, including subjectivity, experimental errors, the necessity for specific assumptions, and complexity. Thus, developing an objective and automated method for identifying the spatial-scale of LUCC is imperative.

The Box-Counting Dimension Method is a widely utilized technique in fractal geometry analysis, primarily employed to examine the spatial distribution characteristics and scale properties of complex systems39,40. By calculating the fractal dimension, this method quantitatively characterizes the spatial complexity and irregularity of geographical phenomena, including land use, vegetation distribution, and urban morphology41,42.The Box-Counting Dimension Method elucidates the self-similarity of geographical phenomena across various scales, suggesting that certain spatial distribution patterns exhibit analogous behaviors at different scales43. The magnitude of the fractal dimension serves as an indicator of the sparsity, clustering, or complexity of spatial distributions44.

Previous studies often selected periods and time nodes for LUCC analysis randomly45. Typically, the temporal scale of LUCC is determined by integrating remote sensing imagery with key economic and social development data46. Methods to establish temporal scales have included dividing cycle time based on practical experience in fields such as water management, climate science, hydrology, ecology, and agriculture47,48,49,50, employing hyperbolic models for hybrid microgrid systems51, and analyzing multi-temporal image data to assess land surface phenology changes52.The aforementioned methods can aid in determining temporal scales. However, LUCC data are typically derived from remote sensing image interpretation, where short-term changes are often negligible53. Due to the complex and labor-intensive nature of this process, annual interpretation is uncommon54. Consequently, temporal scales are often set arbitrarily, lacking scientific rigor. Currently, wavelet variance is recognized as an effective tool for both qualitative and quantitative temporal-scale analysis in climatology, hydrology, and geography31,55,56,57. Wavelet variance offers significant advantages over traditional methods, being non-destructive, requiring minimal data, and providing high precision. Nonetheless, its application in determining LUCC temporal scales remains limited.

This study examines the Shenyang Economic Zone using an enhanced fractal box-counting dimension model and wavelet analysis to investigate the spatial and temporal scales of LUCC. The research aims to introduce novel methods for identifying the spatio-temporal scales of LUCC. Furthermore, it seeks to validate these methods against established measurement indicators. Specifically, the objectives are: (1) to automatically identify the spatial scale of LUCC using the improved box-counting dimension model; (2) to determine the temporal scale of LUCC via the enhanced wavelet analysis model; and (3) to assess the feasibility of the spatio-temporal scale identification method using the land use dynamics degree model.

Materials and methods

Study area

The Shenyang Economic Zone, situated in southern Northeast China and central Liaoning Province, serves as the core of Northeast Asia (122° 10′ 56″ ~ 125° 47′ 05″ N, 40° 02′ 10″ ~ 43° 02′ 25″ E). Designated as the 8th National Comprehensive Reform Pilot Zone by the State Council in 2010, it is a key hub for heavy industry in China. The zone’s five major cities exemplify resource-based urban centers, characterized by high concentrations of energy-intensive and polluting industries. Consequently, the region faces significant land use challenges, including limited land resources, urban sprawl, a sharp reduction in farmland and ecological areas, low land use efficiency, and an imbalanced land use structure58. The conflict between ecological degradation and land use has become increasingly acute. Therefore, it is imperative to study the spatio-temporal dynamics of LUCC in this area. Such research could inform the identification of global heavy industrial zones on spatial and temporal scales (Fig. 1).

Fig. 1
Fig. 1The alternative text for this image may have been generated using AI.
Full size image

Location map of the study area. Notes: The process of creating Fig. 1 was as follows: ① Download the China base map with review number GS (2019) 1651 from the national geographic information public service platform’s standard map service system (https://www.tianditu.gov.cn/). ② Use ENVI 5.1 to create land use type data for the study area and export them as independent images, integrating with ArcGIS 10.2. ③ Arrange and synthesize these images in Adobe Photoshop 2019 to generate figure. (URL: https://pan.baidu.com/s/1SD7VFBXI-lX95SD7kLAH5Q?pwd=zh44 Extractioncode: zh44).

Data sources and processing

Considering the phenological characteristics of the study area and the clarity of remote sensing images, we selected images from June to September with less than 10% cloud coverage from 1980 to 2019. Landsat satellite data for this period were obtained from the United States Geological Survey (USGS) (http://Landsat.usgs.gov/). The data included images from 1980 to 1983 (Multi Spectral Scanner, MSS), 1984 to 2000 and 2002 to 2010 (Thematic Mapper, TM), 2011 to 2012 (Enhanced Thematic Mapper Plus, ETM +), and 2013 to 2019 (Operational Land Imager, OLI). In 2001, ETM images were utilized. The images were processed using geometric correction59, clipping60, data fusion61, and image enhancement62 (Bajpai K et al., 2017), among other techniques63,64.

From 1980 to 2019, principal component bands were derived as the original signal from multi-band remote sensing images using K-L transformations in ENVI 5.165. These bands accounted for 82% of the information in the images. The band data, characterized by matching DN values (gray levels), were employed to identify temporal-scale features of LUCC. To prevent spatial-scale mismatches, all principal component bands from 1980 to 2019 were resample ed to a 90 m × 90 m grid.

After identifying the temporal scale of LUCC in the study area, the years corresponding to different time nodes were determined. Using supervised classification, land use types were mapped into seven categories: dry land, paddy field, forest land, grassland, construction land, water area, and other land. This method ensured an interpretation accuracy exceeding 94%. These land use types served as the foundational data for verifying the spatio-temporal scale characteristics of LUCC.

Models

Establishment of spatial-scale identification model of LUCC

Previously, the fractal box-counting dimension model primarily analyzed the fractal characteristics of population distribution in spatial occupation66. New algorithms using second derivatives were developed and integrated into an enhanced fractal box-counting dimension model to determine the optimal spatial scale for LUCC. This development involved: ① a continuous point sequence selection algorithm, ② an enclosed area optimization algorithm, and ③ an improved scale-free area selection algorithm. When the fractal box-counting dimension could ascertain the fractal structure of LUCC, it was employed as the optimal spatial scale for measuring LUCC. The model development and refinement process is detailed as follows:

  1. (1)

    Acquisition of non-empty box-counting in LUCC

    Assuming the two-dimensional LUCC area is a square with area M, it can be entirely covered by an L × L square. Initially, the L × L square serves as the ruler with scale (s1) is L, and the number of non-empty boxes, (N(s1)), covering M is 1. Subsequently, the size of the nth square ruler is reduced to L/2n-1 × L/2n-1, maintaining an equal division of length and width in the square space M. The scale of the nth square ruler and the number of non-empty boxes in M are denoted as sn (sn = L/2n-1) and N(sn), respectively.

  1. (2)

    Relationship between the linear scale of LUCC and the box-counting number

    In estimating the fractal dimension of LUCC, the linear scale and box-counting number exhibit a negative power law relationship, as described by Eq. (1):

    $$N\left( {\text{s}} \right){ } \propto {\text{ s}}^{{{\text{ - D}}_{{1}} }}$$
    (1)

    where D1 represents the box-counting dimension, s denotes the scale, and N(s) is the count of non-empty boxes. In two-dimensional space, square rulers must maintain a consistent aspect ratio to establish a unified scale standard. The logarithm is applied to both sides of Eq. (1) simultaneously, yielding the expression in Eq. (2):

    $${\text{logN}}\left( {\text{s}} \right){\text{ = - D}}_{{1}} {\text{logs + C}}$$
    (2)

    where C was constant. In the fractal scenario, s and N(s) exhibited a linear relationship in a double logarithmic plot. The least squares method was applied for linear regression between s and N(s). The absolute value of the regression line’s slope (-D1) represented the box-counting dimension.

  1. (3)

    The improvement of box-counting dimension model and the process of automatic identification of spatial-scale of LUCC

    An automatic identification method utilizing the second derivative of the lns-lnN(s) curve was developed using Matlab R2013a to enhance the objectivity and precision of scale-free area identification. In a specific region, the lns-lnN(s) curve approximates a straight line, indicating a scale-free area67.

    In the scale-free region, the local slope of the lns-lnN(s) curve exhibited minor fluctuations around a constant value. The first derivative at a specific point (i) on the double logarithmic curve represented the local slope of the curve, as described by Eq. (3).

    $${\text{ln}}{\prime} {\text{N }}\left( {{\text{s}}_{{\text{i}}} } \right){ = }\frac{{{\text{d}}\left[ {{\text{lnN}}\left( {{\text{s}}_{{\text{i}}} } \right)} \right]}}{{{\text{d}}\left( {{\text{lns}}_{{\text{i}}} } \right)}}{ = }\frac{{{\text{lnN}}\left( {{\text{S}}_{{\text{i + 1}}} } \right){\text{ - lnN}}\left( {{\text{s}}_{{\text{i}}} } \right)}}{{{\text{lns}}_{{\text{i + 1}}} {\text{ - lns}}_{{\text{i}}} }}{ }\left( {{\text{i = 1, 2, 3, \ldots , K - 1}}} \right)$$
    (3)

    where ln’N(si) was the local slope of a certain point (i) on the lns-lnN(s) curve; k was the number of points on the lns-lnN(s) curve. The local slope of ln’N(si) was derivated again, that was, the second derivative of a certain point (i) on the lns-lnN(s) curve was found Eq. (4):

    $${\text{ln}}^{^{\prime\prime}} {\text{N }}\left( {{\text{S}}_{{\text{i}}} } \right) = \frac{{{\text{d}}\left[ {{\text{lnN}}\left( {{\text{S}}_{{\text{i}}} } \right)} \right]}}{{{\text{d}}\left( {{\text{lnS}}_{{\text{i}}} } \right)}}{ = }\frac{{{\text{ln}}{\prime} {\text{N}}\left( {{\text{S}}_{{\text{i + 1}}} } \right){\text{ - ln}}{\prime} {\text{N}}\left( {{\text{S}}_{{\text{i}}} } \right)}}{{{\text{lnS}}_{{\text{i + 1}}} {\text{ - lnS}}_{{\text{i}}} }} \left( {{\text{i = 2, 3, 4, \ldots , K - 2}}} \right)$$
    (4)

The analysis indicates that identifying the scale-free area involves locating a region where the second derivative curve approaching 0. A greater number of fitting points within this continuous region enhances the accuracy of the slope of the fitted line. The specific method for identifying the scale-free area includes: (1) Selection of continuous point sequence: On the second derivative graph of lns-lnN(s) curve, select continuous point sequence with value approaching 0 as candidate region, length P, satisfying 2 ≤ P ≤ K-2. That is to say, the fluctuation of the second derivative value approaching 0 is the key, and the larger P, the more fitting points, the more accurate slope, which is helpful to identify the scale-free region. (2) calculating the area S of the enclosed by points P along the x-axis to optimize the algorithm for determining the enclosed area (Eq. 5). For the same number of points P, a smaller S indicates a reduced fluctuation range near zero. Calculate the area S enclosed by P continuous point series and x axis, and measure the fluctuation amplitude of point series near 0. Under the same P, the smaller S means the smaller fluctuation, which is more in line with the requirements of scale-free region. Thus, identifying the scale-free area involves minimizing S and maximizing P, ultimately enhancing the spatial-scale model for automatic LUCC identification.

$$S = \mathop \sum \limits_{i = j}^{j + P - 1} \left| {{\text{ln}}^{\prime \prime } N\left( {S_{i} } \right)} \right| \times \Delta \ln S{ 2 } \le {\text{ P }} \le {\text{ K - 4, 2 }} \le {\text{ j }} \le {\text{ K - P - 1}}$$
(5)

Initially, The S needed to set a threshold μ, and the size of the threshold μ determines the linearity degree of the selected scale-free areas on the lns-lnN(s) curve. The smaller the μ was, the higher the requirement for linearity was. Subsequently, When S ≤ μ (set μ = 0.5), the slopes fluctuation amplitude of the points was considered to meet the requirement of scale-free area. Among these candidates, the sequence with the highest P and lowest S is preferentially chosen to define the scale-free region. And then, the points were selected as the candidate points sequence of scale-free area.

Establishment of temporal-scale identification model of LUCC

The improvements had been made in the original signal input data processing and wavelet analysis model construction of temporal-scale of LUCC, and the detail improvements contained the improvement of the extraction method of the original signal input data; the improvement of the spatialization of the time domain data; and the establishment of most important time span extraction method based on Fast Fourier Transform (FFT) (i.e. fundamental frequency). The model building and improvement process was shown as follows:

  1. (1)

    Continuous wavelet transform of single-band remote sensing images.

    A wavelet analysis model for identifying temporal-scale feature of LUCC was established68. Complex Morlet wavelet was used as the mother wavelet to carry out one-dimensional continuous wavelet transform for K-L transform single-band image DN value (gray level) from 1980 to 2019. The complex Morlet wavelet function had significant advantages in time domain localization of data (Eq. 6).

    $$W_{s} {(}a{\text{, b)}} = \left| a \right|^{ - 1/2} \mathop \smallint \limits_{{{ - }\infty }}^{{{ + }\infty }} {\text{s }}\left( {\text{t }} \right)\overline{\varphi } \left( {\frac{{\text{t - b}}}{a}} \right){\text{dt}}$$
    (6)

    where Ws(a, b) was the wavelet transform coefficient; s(t) was the analysis wavelet, which meant the original signal data function in the time domain; a was the scale parameter; b was the displacement parameter, which reflected the displacement in the time domain; φ(t) was the basis Wavelet function.

    The original signal data of remote sensing images in the time domain were discrete sequence, and the continuous wavelet transform could change the calculation form of Eq. (6) to Eq. (7):

    $$W_{s} (a, b) = \left| a \right|^{ - 1/2} \Delta t\mathop \sum \limits_{k = 1}^{n} s\left( {k \cdot \Delta t} \right)\overline{\varphi } \left( {\frac{k \cdot \Delta t - b}{a}} \right), k = 1, 2, 3, \ldots ,n;n = 50$$
    (7)

    where Δt was the sampling interval. The wavelet transform would obtain the low or high frequency information of the original signal data by increasing or decreasing the scaling scale, where the low frequency information was the whole change trend of the original signal data, and the high frequency information reflected the details of the original signal data at a specific time point.

  1. (2)

    Establishment of wavelet variance model for single-band remote sensing images.

    The wavelet variance was obtained by integrating the squared values of all wavelet coefficients in b (Eq. 8), which could reflect the evolution characteristics of the original signal energy fluctuation with the scale parameter a. The wavelet variance determined the relative strength of LUCC at different temporal-scales in the original signal. The temporal-scale corresponding to the peak of the wavelet variance map was the feasible temporal-scale that played a major role in this time domain.

    $$Var\left( a \right) = \mathop \smallint \limits_{{{ - }\infty }}^{{{ + }\infty }} \left| {W_{s} \left( {a{\text{,b}}} \right)} \right|^{{2}} {\text{db}}$$
    (8)

    The bandwidth and center frequency of the band-pass filter with complex Morlet mother wavelet were determined by the scale parameter a. The larger the bandwidth and center frequency of Morlet wavelet was, the better the time–frequency aggregation of Morlet wavelet was. The periodic signals of LUCC were expressed as a linear combination of sinusoidal functions by using FFT. The fundamental frequency corresponding to the maximum amplitude was the feasible temporal-scale of all study areas69.

The establishment of land use dynamics degree model

The land use dynamics degree quantitatively represents changes in land use types over time and can effectively predict future trends. It is categorized into positive and negative values, with a positive value indicating an increase in a specific land type area and a negative value indicating a decrease. The absolute value of the land use dynamics degree is used to verify the accuracy of spatio-temporal scale division in LUCC. A larger absolute value signifies greater transformation in land use type, suggesting significant changes, whereas smaller values indicate relative stability.

The model of the land use dynamics degree was employed to examine the spatio-temporal fluctuations of LUCC in the study area from 1980 to 2019. The expression is as follows:

$$S_{ij} = \left\{ {\mathop \sum \limits_{ij}^{n} \frac{{\Delta S_{i - j} }}{{S_{i} }}} \right\} \times \frac{1}{T} \times 100\%$$
(9)

where Sij was the dynamics degree of land use; i was the number of land use types; T was the time period of land use changes; Si was the area of the land use type i; ∆Si-j referred that the total area of land use type i converted to land use type j from the beginning year to the end year.

Results

The spatial-scale identification results of LUCC

Within the study area, a negative power law relationship was observed between different spatial scales of LUCC and the corresponding counts of non-empty boxes N(s). The study area was divided into a grid of 3310 rows by 3764 columns, each 90 m × 90 m. The spatial scale s was defined as s = (90 × n) m × (90 × n) m = 8100 × n2 m2, increasing by integer multiples of the minimum spatial scale. The parameter n, set at 50, was determined by the dataset’s maximum row and column count. As the spatial scale s increased, the number of non-empty boxes N(s) decreased significantly, exhibiting a typical negative power-law relationship. This indicates that at larger spatial scales, the LUCC features in the study area tend to become more aggregated.

The curve of lns-lnN(s) and its first and second derivative curve was used to determine the box-counting dimension and automatically identify the spatial scale of LUCC, a continuous points sequence P = 4 + 5 + … + K-4 = K × (K-7)/2 was obtained for K-7 points on the lns-ln″N(s) curve, and calculated the area S of the x-axis encircled by the continuous points sequence P. When K and n were both 50, K-7 and P were 43 points and 1075 continuous points sequence, respectively, and the minimum area of S was 0.0241. The spatial-scale of LUCC in the study area was 4500 m × 4500 m. Based on the spatial-scale, the number of samples covering the study area was 2516.

Then, the least square method was used to linearly fit the logarithm of the continuous points sequence in the scale-free area of study area (Eq. 10). The box-counting dimension D1 was 1.9987 in the spatial-scale of LUCC (4500 m × 4500 m), indicating that in this spatial-scale, the LUCC in the study area had obvious fractal characteristics.

$${\text{logN}}\left( {\text{s}} \right){ = - 1}{\text{.9987logs + 15}}{.0056}$$
(10)

The test results showed that the coefficient of linear fitting was 0.98, which indicated that the model could automatically identify the optimal spatial-scale of LUCC effectively. The results showed that the box dimension of LUCC was close to 2, and the LUCC had obvious fractal characteristics.

The temporal-scale identification results of LUCC

Figure 2 showed the existence of different temporal-scales of principal component bands, which could be reflected by wavelet coefficients of real part. The volatility characteristics in 10–20 annus(a), 30–40a and 40–50a temporal-scales were quasi-two shocks, quasi-three shocks and quasi-three shocks, respectively, during the land use and land change (LUCC) evolution in the study area. It could be seen that the changes of wavelet coefficients of real part in 30–40a and 40–50a temporal-scales were stable and global throughout the analysis period, and the changes in 10–20a temporal-scales were stable between 2006 and 2019. The order of energy density distribution obvious degree in the time frequency domain was the temporal-scale in 10–20a, 30–40a and 40–50a, respectively.

Fig. 2
Fig. 2The alternative text for this image may have been generated using AI.
Full size image

The LUCC wavelet coefficients of real part in the study area.

Figure 3 showed the wavelet variance of LUCC throughout the time domain in the study area. The wavelet variance of the inter-annual changes of LUCC had six local maximum, which corresponded to the temporal-scales of 5a, 13a, 18a, 23a, 33a and 40a in turn. The maximum peak corresponded to the temporal-scale of 40a, indicating that the 40a temporal-scale volatility was the strongest and it was the main temporal-scale of LUCC. The complete waveform was formed on the wavelet variance from 1980 to 2019, therefore, 40a was the temporal-scale of LUCC. The wavelet variance map was the change process of the wavelet variance with the temporal-scale, and it could reflect the energy distribution of the signal fluctuation under different temporal-scale. Therefore, the wavelet variance map could be used to determine the main period of the signal under study. Wavelet variance method was effective and adaptive in multi-scale feature recognition, and it could be applied to the recognition of multi-scale spatial structure and feature scale of LUCC50,56.

Fig. 3
Fig. 3The alternative text for this image may have been generated using AI.
Full size image

Wavelet variance of LUCC in the study area.

In the temporal-scale of 40a, the temporal-scale characteristics of LUCC were global in the time domain, and the LUCC had a strong fluctuation energy. The average fluctuation period of LUCC was about 13 years, and the important time nodes of LUCC in the case study area were 1980, 1993, 2006 and 2019 (Fig. 4).

Fig. 4
Fig. 4The alternative text for this image may have been generated using AI.
Full size image

The characteristic curve of LUCC in the study area from 1980 to 2019. Notes: The process of creating Fig. 5 was as follows: ① The Chinese base map was downloaded from the Standard Map Service System of the National Geographic Information Government Service Platform (https://www.tianditu.gov.cn/), and the base map revision number is GS (2019) 1651. ② The Landsat remote sensing data of Shenyang Economic Zone were obtained from the website of the United States Geological Survey (USGS) (http://Landsat.usgs.gov/). ③ The obtained remote sensing data were processed using ENVI 5.1 software to create the land use type classification data of the study area, and the classification results were exported as independent geospatial image files. ④ The geostatistical analysis module of ArcGIS 10.2 software was used to integrate and analyze the land use classification data of multiple time series to create a single dynamic visual land use map of the Shenyang Economic Zone from 1980 to 2019, after which the map was exported. ⑤ Organize, arrange and synthesize the classified images using Adobe Photoshop 2019 to obtain the mapping results as shown in Fig. 5. (URL: https://pan.baidu.com/s/1SD7VFBXI-lX95SD7kLAH5Q?pwd=zh44 Extractioncode: zh44).

land use dynamics degree analysis results

Table 1 presents the changes in land use areas within the study region across three intervals: 1980–1993, 1993–2006, and 2006–2019. Overall, dry land and grassland areas declined, particularly from 1993 to 2006, with grassland experiencing the most significant reduction from 2006 to 2019. Conversely, the area of built-up land expanded rapidly, peaking in growth between 2006 and 2019. Additionally, the areas of rice land, water, and other land uses exhibited fluctuations, with varying decreases or increases during the initial period and subsequent changes in later periods.

Table 1 The area variation of different land use types in the study area from 1980 to 2019.
Full size table

The overall land use dynamics degree was significantly different during the study period in Shenyang Economic Zone. Figure 5a indicated the overall spatial distribution trend of active LUCC in the study area gradually weakened from the northwest to the southeast from 1980 to 1993. High land use dynamics were primarily observed around Shenyang, Fushun, and Liaoyang, while other regions exhibited smaller changes and lower dynamics. Figure 5b indicates that areas with frequent LUCC were concentrated in the northern, southern, and western to central parts of the study area, whereas the southeastern region experienced minimal and stable LUCC. The overall dynamics were lower compared to stage (a), with a deceleration in land use change intensity. Figure 5c illustrates that LUCC momentum in the study area significantly weakened compared to the previous period, with a dispersion of LUCC momentum from 2006 to 2019. Overall, the trend in land-use dynamics showed a gradual decline over time, with higher LUCC activity primarily concentrated in regions connecting other cities to Shenyang City.

Fig. 5
Fig. 5The alternative text for this image may have been generated using AI.
Full size image

Spatial distribution characteristics of single land use dynamics degree in the study area from 1980 to 2019.

Discussion

The fractal box-counting dimension model and wavelet analysis model improved by this study implemented the process of determining the spatio-temporal scale of land use and land cover change (LUCC) evolution, and the spatial quantitative model analyzed the spatio-temporal scale evolution characteristics of LUCC in Shenyang Economic Zone from 1980 to 2019, which helped to reveal the evolution pattern of LUCC.

Firstly, in the process of determining the space grid, the grid size was usually obtained on the basis of previous researches or repeated debugging70, and most studies on grid division methods mainly focused on land use changes and landscape pattern changes71,72.

The determination of the grid size was similar to the identification process of the scale-free area. Generally, a section with a better linear relationship in the fitting curve was found as a scale-free area through the manual judgment or the empirical judgment. However, there were some differences in the scale-free area division of the same data among different observers. This method lacked of theoretical basis and had the shortcomings of strong subjectivity and large errors. On the spatial-scale of LUCC, based on the fractal theory, the fractal box-counting dimension model was applied in the field of LUCC, and the second derivative was used to improve the model. It was realized the innovation of the traditional method from manual identification to objectively automatic identification of the spatial-scale of LUCC. In addition, space grid could reflect the difference of LUCC analysis scale more accurately and make up for the deficiency of micro scale universality73,74.

On the space grid of 1, 2, 4, 6, 8, 10, 20, 40, 60 m, and 1 km, the changes of micro-geomorphology index, the local dynamics of urban growth, the spatio-temporal changes of Production-Living-Ecological space and the indexes of land use function were quantitatively analyzed, which provided a practical method for soil erosion, urban expansion and sustainable development of regional land use73,75,76,77. It not only increased the amount of data processing, but also the information couldn’t be well described. If the grid scale was too large, it would show high uncertainty, the overall accuracy decreased. Therefore, in order to reduce the computational burden and try to accurately capture the space grids heterogeneity, it was particularly important to identify moderate space grid scales78, and the proposed research approach precisely addresses this issue. The improved fractal box-counting dimension model survive a new index of fractal to explore the spatial-scale of LUCC.

Secondly, as a new method of temporal-scale feature analysis, wavelet analysis had been widely applied in the analysis of time-series data, some studies focused on the periodic regularity of geographic phenomena based on the availability of data and the convenience of calculation79. The original data of time series was mostly the statistical data of continuous time series80.

Previous studies had shown that LUCC was generally no cyclical variation, and its changes were usually guided by policies81. However, it was found that LUCC had a 13a periodic change in the 40-year time span by using wavelet variance analysis in the study area. This study also found that although the area of regional cultivated land continued to decrease, the area of construction land had been increasing, which were the same as the results of other areas with rapid urbanization58. This may be that the change rate of LUCC was closely related to the periodic change of some factors. The next step is to find the key influencing factors of LUCC for verification.

Thirdly, the results showed that LUCC in the study area from 1980 to 2019 were consistent with the stage characteristics of land use dynamics degree. Dry land, grassland and construction land were the main forms of LUCC in Shenyang Economic Zone. Previous studies focused on expressing the results of LUCC with statistical data26,82, but lacked of spatial visual representation for LUCC characteristics. Our exploration results were supported by our model accuracy and previous studies of LUCC83. This study used the calculated spatial-scale and temporal-scale to measure land use dynamics degree, which was expressed quantitatively the changes of land use types in Shenyang Economic Zone during the study period. As shown in Fig. 5, the characteristics of LUCC during each period in this study had certain periodic rules, indicating that the spatio-temporal scales identified in this study were more applicable.

Future research should integrate diverse remote sensing data with high-resolution spatio-temporal datasets to enhance model accuracy and applicability. Investigating nonlinear characteristics in LUCC dynamics can be advanced through machine learning or deep learning algorithms. This methodology could be applied across various geographical regions or land use categories to assess its broader relevance. Additionally, a detailed analysis of the drivers of LUCC, including the interplay of policy, economic, and environmental factors, is essential for a comprehensive understanding of its evolutionary trends.

Conclusion

Future research should integrate diverse remote sensing data with high-resolution spatio-temporal datasets to enhance model accuracy and applicability. Investigating nonlinear characteristics in land use and land cover change (LUCC) dynamics can be advanced through machine learning or deep learning algorithms. This methodology could be applied across various geographical regions or land use categories to assess its broader relevance. Additionally, a detailed analysis of the drivers of LUCC, including the interplay of policy, economic, and environmental factors, is essential for a comprehensive understanding of its evolutionary trends.

  1. (1)

    The enhanced fractal box dimension model objectively and automatically identifies the LUCC spatial scale by integrating the second-order derivative and an automated calibration process, thus mitigating the subjectivity and experimental errors of traditional segmentation methods. The results demonstrate that a spatial grid division of 4500 m × 4500 m in LUCC analysis effectively reveals regional fractal characteristics, providing a scientific basis for future land use pattern studies.

  2. (2)

    Wavelet analysis reveals distinct fluctuation characteristics of land use and cover change (LUCC) in the study area across 10–20, 30–40, and 40–50-year time scales. Notably, the 40–50-year scale exhibits the most pronounced fluctuations, whereas the 10–20-year scale shows the least.

  3. (3)

    Wavelet variance emerges as a novel parameter for characterizing development patterns in LUCC time series. In the study area, LUCC displays a cyclic pattern with significant fluctuations approximately every 13 years. This finding provides a vital reference for further exploration of LUCC’s dynamic mechanisms and addresses the limitations of existing time-series analysis methods in detecting periodicity.

  4. (4)

    Land use dynamics were documented over three periods at a spatial scale of 4500 m × 4500 m, with values of 86.11 from 1980 to 1993, 65.06 from 1993 to 2006, and 60.53 from 2006 to 2019. These results suggest that fractal characteristics and wavelet variance are effective in capturing LUCC information, underscoring the robust applicability of the LUCC spatio-temporal scale identification method.