Next Article in Journal
Importance of Forest Ecosystem within Important Plant Areas (IPAs) for the Development of Nature-Based Tourism—A Case Study of Fruška Gora National Park
Previous Article in Journal
Regional Scale Inversion of Chlorophyll Content of Dendrocalamus giganteus by Multi-Source Remote Sensing
 
 
Font Type:
Arial Georgia Verdana
Font Size:
Aa Aa Aa
Line Spacing:
Column Width:
Background:
Article

A New Paradigm for Assessing Detailed Dynamics of Forest Landscape Fragmentation

1
College of Art & Design, Putian University, Putian 351100, China
2
College of Transportation and Civil Engineering, Fujian Agriculture and Forestry University, Fuzhou 350002, China
*
Author to whom correspondence should be addressed.
Forests 2024, 15(7), 1212; https://doi.org/10.3390/f15071212
Submission received: 12 June 2024 / Revised: 23 June 2024 / Accepted: 11 July 2024 / Published: 12 July 2024
(This article belongs to the Section Forest Ecology and Management)

Abstract

:
There is an urgent need for a thorough assessment of forest landscape fragmentation to inform forest protection and restoration, and reforestation policies. However, there is currently a lack of an effective comprehensive index for forest landscape fragmentation, and detailed knowledge of the forest landscape fragmentation dynamics remains insufficient. Here, taking Putian City of Fujian Province in Southeastern China as a case, we employed a forest fragmentation comprehensive index (FFCI) to capture key features of forest landscape fragmentation, such as patch size, number, and distribution. Then, bivariate spatial autocorrelation analysis was employed to identify the spatial associations between the static forest landscape fragmentation (FFCI) and the dynamic forest landscape fragmentation (ΔFFCI), and the spatial coupling modes among the three individual components of FFCI (mean patch area, MPA; aggregation index, AI; patch density, PD) were identified to explore the detail process of forest landscape fragmentation. Finally, the random forest model was applied to observe the impact factors of forest landscape fragmentation dynamics. The findings showed that forest landscapes with different degrees of fragmentation exhibited more noticeable changes at both ends (i.e., either high or lower-level fragmentation), with the intermediate level remaining consistent from 2000 to 2020. Around 18.3% of forest landscapes experienced a decrease in fragmentation, particularly in the northern part of the study area, while approximately 81.7% of forest landscapes exhibited an increasing trend in fragmentation. The bivariate spatial autocorrelation analysis indicated that the proportion of Low–High-type grids was the highest at 17.3%, followed by the High–High type at 7.0%. We also identified eight forest landscape fragmentation modes, which indicate the most significant forest landscape fragmentation pattern is a decrease in MPA and an increase in PD. Moreover, the anthropogenic factors (e.g., population density and night light intensity) were found to dominate the FFCI dynamics during 2000–2020. This study offers an efficient research paradigm for the dynamics of forest landscape fragmentation. The outcomes are conducive to an in-depth comprehension of the detailed dynamic information of forest landscape fragmentation, and supply a scientific foundation for enhancing the overall ecological service function of the forest.

1. Introduction

Forests, the lungs of our planet, play a pivotal role in maintaining ecological balance and biodiversity [1,2]. However, the phenomenon of forest landscape fragmentation, the process in which continuous forest cover is broken down into smaller, isolated patches, has emerged as a significant threat to the health and integrity of these vital ecosystems [3]. Forest landscape fragmentation poses a significant threat to ecosystems by reducing habitat quality and connectivity, making it difficult for species to find mates, disperse, and maintain genetic diversity [4]. Forest landscape fragmentation can disrupt important ecosystem processes, such as nutrient cycling, water regulation, and pollination, increasing the vulnerability of species to environmental changes [5]. Therefore, understanding the detailed dynamics of forest landscape fragmentation and identifying the driving forces behind this change is crucial for developing effective conservation strategies and mitigating its negative impacts.
Over the past few decades, research on forest landscape fragmentation has made significant strides in understanding the causes, consequences, and potential solutions of this pressing environmental issue [1,2,6]. For example, meta-analyses and synthesis studies have played a crucial role in synthesizing existing research findings on forest landscape fragmentation across different regions and ecosystems [7,8]. These studies help identify common patterns, trends, and drivers of fragmentation, providing valuable insights for conservation planning and management. Recently, researchers have increasingly focused on understanding the resilience of fragmented ecosystems to environmental stressors such as climate change, invasive species, and habitat loss [9]. Efforts in restoration ecology have explored innovative approaches to restoring connectivity in fragmented landscapes through habitat restoration, reforestation, corridor creation, and landscape planning. These initiatives aim to enhance habitat quality, promote species movement, and improve ecosystem functioning in fragmented areas [10].
As human activities expand and natural habitats become increasingly disrupted, an accurate understanding of the dynamics of forest landscape fragmentation is a prerequisite for effective conservation and sustainable management [2]. Advances in remote sensing technologies, such as satellite imagery and Light Detection and Ranging (LiDAR), have revolutionized the monitoring and assessment of forest landscape fragmentation at various spatial scales [11,12]. Researchers have developed a wide range of landscape metrics and indices to quantify the degree of forest landscape fragmentation within a landscape [13,14]. These metrics include measures of patch size, shape complexity, edge density, connectivity, and landscape configuration, providing valuable insights into the spatial structure of fragmented landscapes. Geographic Information Systems (GIS), spatial analysis and geostatistical analysis tools have become essential for measuring forest fragmentation by integrating landscape metric layers, assessing landscape connectivity, identifying critical habitat corridors, and prioritizing conservation efforts based on fragmentation patterns [15]. Forest landscape fragmentation is a manifestation of the comprehensive characteristics of forests, including various features such as forest patch size, shape, and distribution [2]. However, most previous studies have used individual landscape indices to characterize forest landscape fragmentation separately, and this cannot comprehensively depict the multidimensional features of forest landscape fragmentation [1]. Therefore, there is an urgent need to construct a forest fragmentation comprehensive index (FFCI) that can reflect various information such as forest patch size, shape, and spatial configuration. In addition, past research on local forest landscape fragmentation has mainly concentrated on evaluating fixed landscape patterns [1,6], neglecting the dynamic aspect of forest landscape fragmentation over time. Although the effects of forest landscape fragmentation can last for up to a century [16], they are usually most noticeable and prominent in the initial few decades after fragmentation occurs [17]. Therefore, a precise assessment of both the present patterns and dynamics of forest landscape fragmentation is crucial for averting future biodiversity decline and ecosystem deterioration. Nevertheless, the absence of recent decades’ quantification of local forest landscape fragmentation from a dynamic perspective underscores a notable gap in knowledge.
Forest landscape fragmentation dynamics are primarily driven by human activities such as deforestation, urbanization, agriculture expansion, infrastructure development, and logging [11,18]. These activities lead to the division of continuous forest areas into smaller patches, resulting in increased edge effects, habitat loss, and the isolation of wildlife populations. For example, hydrological facilities were found to have a significant impact on the distribution and quantities of forest landscapes [18]. Climate change can also play a role in forest landscape fragmentation by altering the distribution and composition of forests [19]. Additionally, natural factors such as wildfires, insect outbreaks, and disease can contribute to forest landscape fragmentation dynamics [20]. Several methods have been extensively used to study the driving forces of forest landscape fragmentation dynamics. These include the literature review method, spatial analysis method, and mathematical statistics method. The literature review method involves reviewing and analyzing many relevant research papers to summarize and categorize the main factors influencing forest landscape fragmentation [8]. The spatial analysis method quantitatively analyzes the spatial distribution and changes in forest landscape fragmentation using GIS and remote sensing technology [15]. Through spatial overlay, buffer analysis, correlation analysis, and other methods, the spatial relationships between influencing factors and forest landscape fragmentation can be revealed. The mathematical statistics method collects a large amount of sample data and uses statistical principles and methods to analyze the quantitative relationship between influencing factors and forest landscape fragmentation [6]. Common statistical methods include regression analysis, variance analysis, principal component analysis, etc., which help researchers determine the importance and direction of influencing factors. With technological advancements and continuous improvement in research methods, new methods and approaches will continue to emerge, providing more accurate and in-depth results for analyzing the influencing factors of forest landscape fragmentation. In recent years, deep learning techniques have emerged as powerful tools for analyzing complex spatial data and modeling landscape changes [21]. By leveraging the capabilities of deep learning algorithms, researchers can gain deeper insights into the drivers of forest landscape fragmentation at various scales [22,23].
Thus, research gaps need to be filled, such as a single index being unable to fully grasp the multi-dimensional characteristics of forest landscape fragmentation, and the knowledge on the detailed dynamic process of forest landscape fragmentation remaining limited. Hence, taking a fast urbanization area in Southeastern China—Putian City as a case, an FFCI was developed to depict the primary attributes of forest landscape fragmentation, encompassing patch size, shape, isolation, and aggregation information; then, the FFCIs in the years of 2000 and 2020, along with their variances (ΔFFCI) between 2000 and 2020, were utilized to ascertain the static and dynamic trends in forest landscape fragmentation. Through the integration of remote sensing data, GIS, and machine learning algorithms, this paper aims to (1) investigate spatial associations between the static and dynamic forest landscape fragmentation; (2) identify the spatial coupling modes of forest landscape fragmentation process; and (3) uncover the underlying factors contributing to forest landscape fragmentation dynamics. The findings of this research are expected to contribute valuable insights to conservation practitioners, policymakers, and land managers in developing strategies to mitigate the impacts of forest landscape fragmentation and promote sustainable land management practices.

2. Materials and Methods

2.1. Study Area

Putian City in Fujian Province is in the southeastern part of China (Figure 1). It is situated along the coast of the Taiwan Strait, bordering Quanzhou City to the north and Zhangzhou City to the south. It covers an area of approximately 4200 km2 and has a population of over 300 million residents. The city is known for its rich cultural heritage, including historical sites, traditional architecture, and local customs. The city has a diverse landscape that includes coastal areas, plains, hills, and rivers, providing a unique environment for various economic activities such as agriculture, fishing, and tourism. Economically, Putian City is a significant hub for industries such as footwear manufacturing, electronics production, and agricultural product processing. The city’s strategic location near major transportation routes and ports has contributed to its growth as a key economic center in the region. In terms of natural resources, Putian City boasts fertile farmland suitable for cultivating crops such as rice, tea, fruits, and vegetables. The city also has abundant marine resources due to its coastal location, supporting a thriving fishing industry. Overall, Putian City offers a blend of cultural heritage, economic opportunities, and natural beauty that makes it an intriguing area for study and exploration.

2.2. Data Sources and Processing

2.2.1. Land Use Data

The China Land Cover Dataset (CLCD) was an annual land cover dataset of China (1985–2022) created by Professor Huang from Wuhan University based on 335,709 Landsat scenes from Google Earth Engine using a random forest classifier. The overall accuracy of this dataset is 80%. CLCD includes nine land use/cover types, such as cropland, forest land, shrub, grassland, water body, snow/ice, barren, impervious, and wetland [24].

2.2.2. Driving Factor Data

Table 1 presents the driving factors considered in this study, encompassing geomorphic, natural, and socio-economic factors. The digital elevation model (DEM) utilized in this study was sourced from the ASTER GDEM V3 product, released jointly by National Aeronautics and Space Administration (NDAS) and Japan’s Ministry of Economy, Trad, and Industry (METI) in Tokyo, with a spatial resolution of 30 m, available at https://www.gscloud.cn/ (accessed on 6 December 2021). This DEM was employed to derive elevation and slope data using the Surface tool within the ArcGIS software version 10.3. The Normalized Difference Vegetation Index (NDVI) dataset with a spatial resolution of 30 m was obtained from Resource and Environmental Science, Chinese Academy of Sciences at http://www.resdc.cn/ (accessed on 6 December 2021). Population density data from the WorldPop dataset with a spatial resolution of 1 km was also included. Nighttime light intensity data with a spatial resolution of 1 km from the National Earth System Science Data Center at http://www.geodata.cn/ (accessed on 6 December 2021) was used to represent socio-economic information. The road network shapefiles (SHP) in vector formats were sourced from Open Street Map (OSM). Subsequently, the distance tool within ArcGIS software was utilized to calculate the distance to forest (DTF) and distance to road (DTR). All datasets mentioned above were transformed into the Albers coordinate system.

2.3. Calculation of Forest Landscape Fragmentation Comprehensive Index

2.3.1. Selection of Landscape Metrics

The inspiration for the selection of landscape metrics in this study was drawn from the findings of the previous study [2,6]. We developed the FFCI to capture key features of forest landscape fragmentation, such as patch size, number, and distribution. In this regard, we utilized mean patch area (MPA) to represent the average size of all forest patches, and largest patch index (LPI) to indicate the proportion of the largest forest patch relative to the total area for patch size characteristics. For patch number characteristics, we employed patch density (PD) as the ratio of forest patch number to total area. To represent patch distribution characteristics, we selected aggregation index (AI) to measure the frequency of adjacent forest patches in the landscape, and DIVISION to assess the spatial proximity of forest patches. These landscape metrics were computed using FRAGSTATS 4.2 software with a moving window approach set at a size of 1 km.

2.3.2. Synthesis of Forest Landscape Fragmentation Comprehensive Index

Following the computation of landscape metrics mentioned above, our objective was to develop a composite index, known as FFCI, to provide a rapid and quantitative evaluation of forest landscape fragmentation in a region. The weighting process is crucial in the creation of FFCI. Various weighting methods exist, such as AHP and Delphi [25,26]. However, even with a simple weighting method, subjective biases may influence weight distribution in practice. To address this issue, we employed principal component analysis (PCA) to determine the relative importance of each variable. PCA is a multidimensional data compression technique that eliminates any collinearity effects among the five factors [27]. Importantly, this method automatically and objectively assigns weights to each factor based on their contribution to the principal components, thereby avoiding variations or errors in weight assignment due to individual biases [28].
Prior to conducting the PCA, all factors (MPA, LPI, PD, AI, and DIVISION) were standardized to a range of 0 to 1. Subsequently, the PCA was performed using the PCA Rotation tool in ENVI (version 5.1) software, resulting in the creation of a single-band image known as the FFCI image. Throughout the study period, all the percentage eigenvalues of PC1 exceeded 88% (88.1% for 2000 and 90.2% for 2020), indicating that this component encapsulated most characteristics of all factors. Therefore, PC1 was selected to represent the FFCI in this study. To enable temporal comparisons across different study periods, the FFCI values were re-normalized to a range of 0 to 1, with higher values denoting increased forest landscape fragmentation and lower values indicating reduced forest landscape fragmentation levels.

2.4. Bivariate Spatial Autocorrelation Analysis

The method of bivariate spatial autocorrelation analysis involves examining the spatial relationships between two different variables (i.e., FFCI in 2000 and ΔFFCI) within a geographic area. This analysis helps to identify patterns and associations between the two variables across space, providing insights into potential spatial dependencies and interactions [28]. In this study, the bivariate spatial autocorrelation analysis was conducted using the GeoDa software version 1.22 [29]. The program utilized the rook’s contiguity weighting method during execution. It was configured to conduct 999 permutation tests with a significance level of 0.05. Subsequently, a spatial cluster map was generated, classifying the associations between FFCI in 2000 and ΔFFCI into five categories: High–High (pixels with a high level of FFCI encircled by pixels of high ΔFFCI), Low–Low (pixels with a low level of FFCI encircled by pixels of low ΔFFCI), High–Low (pixels with a high level of FFCI encircled by pixels of low ΔFFCI), Low–High (pixels with low level of FFCI encircled by pixels of high ΔFFCI), and “not significant”.

2.5. Machine Learning Algorithms

The impact of geomorphic, natural, and socio-economic factors on FFCI dynamics between 2000 and 2020 was investigated with a machine learning algorithm (i.e., random forest model) in R version 4.3.1. The analysis utilized packages such as tidyverse, raster, caret, randomForest, and CAST for data processing, model development, evaluation, and result presentation [23]. Cross-validation of the model was conducted using 75% of the dataset for training and the remaining portion for validation in each resampling iteration. The model training involved the use of 500 trees (ntree value) and the sampling of 2 to 7 predictors (repeated 2 times with a total 12 model runs) at each node for splitting (mtry value) to minimize the root mean square error (RMSE). To assess the impact of individual variables on FFCI dynamics, the response curve was examined by observing how model predictions changed across variable ranges while holding other variables constant at the 0.25 and 0.75 quantiles. This method offers a descriptive interpretation of the random forest model output that can be easily visualized and understood. Due to the inherent randomness in random forest modeling, variable importance metrics and model predictions may vary with different seed numbers and across runs. A “variability assessment” was conducted through 12 model runs with adjusted tuning predictors based on caret to address this variability.

3. Results

3.1. Spatial Distribution in Static and Dynamic Forest Landscape Fragmentation

We calculated the static FFCI using PC1 of MPA, LPI, PD, AI, and DIVISION for 2000 and 2020 (Figure 2a,b). Forest landscapes with low static fragmentation (FFCI < 0.4) were mainly distributed in the northern part of the study area, while forests with high static fragmentation (FFCI > 0.7) were mainly located in the southern part of the study area, with an urban central area and proximity to the road networks. The proportion of forest landscapes with varying levels of fragmentation revealed more pronounced variations at both ends, while the middle level remained stable from 2000 to 2020 (Figure 2c). There was a slight decrease in the proportion of forest landscapes with low static fragmentation (FFCI < 0.4) from 2000 to 2020, dropping from 21.3% to 17.0%. Similarly, the percentage of forest landscapes with higher static fragmentation (0.8 < FFCI < 0.9) also saw a slight decline, decreasing from 33.6% in 2000 to 30.5% in 2020. In contrast, the proportion of forest landscapes with the highest static fragmentation (FFCI > 0.9) notably increased from 8.9% in 2000 to 15.4% in 2020.
However, the dynamic FFCI (ΔFFCI) for the period between 2000 and 2020 displayed a significantly different trend compared to the static FFCIs. Around 18.3% of forest landscapes experienced a decrease in fragmentation (ΔFFCI < 0), particularly in the northern part of the study area, while approximately 81.7% of forest landscapes exhibited an increasing trend in fragmentation (ΔFFCI > 0) (Figure 2d,e). In most study regions during 2000–2020, fragmentation remained relatively stable (−0.2 < ΔFFCI < 0.2).

3.2. Spatial Coupling Modes of Forest Landscape Fragmentation Processes

We examined the bivariate spatial autocorrelation between the static FFCI and dynamic FFCI, with the Moran’s I 0.689 (p < 0.01) indicating an obvious spatial aggregation effect between the FFCI and ΔFFCI (Figure 3). Except for the non-significant and neighborless grids, the proportion of Low–High-type grids was the highest, at 17.3%, followed by the High–High type at 7.0%. The Low–High type was mainly distributed in the central region of the forest, while the High–High type was primarily found at the boundary between forest and non-forest areas. Other types, such as Low–Low and High–Low, had relatively small proportions.
Based on the potential combinations of alterations (either increases or decreases) in the three individual constituents of the FFCI (MPA, AI, and PD), we determined eight modes of forest landscape fragmentation and calculated the respective area composition ratios of these eight patterns (Figure 4). The mode MPA↓AI↑PD- was one of the most prevalent and widely distributed patterns across the study area, accounting for 33.4% of the total area. It was followed by the mode MPA↓AI↑PD↑, which represented 32.6% of the area and was primarily found at the non-forest edges. Other modes, such as MPA↑AI↑PD↓ and MPA↑AI↑PD-, accounted for 17.1% and 15.9% of the area, respectively, and were predominantly concentrated in the central forest region.

3.3. Impact Factors of Forest Landscape Fragmentation Processes

Figure 5 revealed that the anthropogenic factors (e.g., POD and NTL) dominated the FFCI dynamics during 2000–2020. The POD had the greatest impact on the dynamics of forest landscape fragmentation, with an average standardized importance value of 91.0%, followed by NDVI and NTL, with average standardized importance values of 66.0% and 65.8%, respectively. However, the impact of terrain factors (e.g., Elevation and Slope) on the dynamics of forest landscape fragmentation was relatively small. In different regression models, POD consistently maintained a high level of impact on forest landscape fragmentation dynamics, with relatively low variability in its influence (Std. = 16.5). However, the impact of DTR and DTF on forest landscape fragmentation dynamics varied significantly across different models, with the Std. of 29.8 and 27.5, respectively.

4. Discussion

4.1. Forest Landscape Fragmentation Process

We created an integrated FFCI and assessed the static and dynamic forest landscape fragmentation patterns from 2000 to 2020. The findings highlighted the forests with the highest fragmentation levels and those that underwent the most significant fragmentation. In line with prior research [6,30], our findings indicated that the forest landscapes in the northern mountains of the study area remained largely undisturbed, while the forest landscapes in the southern urban area of the study area were in a highly fragmented state. It is worth noting that the forest landscapes at both extreme fragmentation levels (e.g., FFCI < 0.4 and FFCI > 0.8) also experienced significant variation during the study period. In particular, the highly fragmented forest landscapes with an FFCI greater than 0.9 experienced the most severe forest landscape fragmentation over the last two decades, according to an increase in their area by more than 1.7 times (Figure 2c). Moreover, consistent with previous studies [2], we have observed that forest landscapes that were relatively intact have also undergone significant forest landscape fragmentation in the past two decades. These results are robust and may be further confirmed in the bivariate spatial autocorrelation analysis (Figure 3). The bivariate spatial autocorrelation analysis revealed that as much as 17.3% of the study area exhibited a Low–High (i.e., low-level FFCI with high-level ΔFFCI) clustering pattern, and 7% of the study area exhibited a High–High (i.e., high-level FFCI with high-level ΔFFCI) clustering pattern. For example, the boundary areas between forest and non-forest encountered strong interference, leading to the transformation of intact forest into fragmented forests [30,31]. These findings suggest that using both the static FFCI and the dynamic FFCI (ΔFFCI) together is more appropriate for assessing the status and trends in forest landscape fragmentation. The static FFCI primarily depicts forest distribution patterns that result from long-term factors, such as artificial, climate, and topography patterns since the beginning of the Anthropocene era, while the dynamic FFCI provides a more accurate representation of forest landscape fragmentation processes that result from short-term factors such as anthropogenic interference [32,33].
Enhancing our understanding of forest landscape fragmentation processes and drivers involves identifying the distribution and composition of forest landscape fragmentation patterns. We found that the most typical FFCI decrease modes (MPA↑AI↑PD↓ and MPA↑AI↑PD-) accounted for 17.1% and 15.9% of the area and accounted for 33.0% of the total forest landscapes, which indicates that patch size, aggregation, and patch numbers changed synergistically with forest cover change in these areas. However, the most typical FFCI increase modes (MPA↓AI↑PD- and MPA↓AI↑PD↑) accounted for 66.0% of the total forest landscapes, which indicated that forest landscape fragmentation was primarily characterized by the division of forests into smaller fragments. Fortunately, the aggregation of these forest landscapes increased during the study period. Hence, it is essential to tailor efforts towards an understanding of the in situ mechanisms driving changes in forest landscape fragmentation, emphasizing the correlation between forest landscape patterns and their explanatory factors.

4.2. Driving Patterns of Forest Landscape Fragmentation Process

The driving patterns of forest landscape fragmentation processes are a complex and critical topic that requires thorough examination and understanding. Forest landscape fragmentation, the division of continuous forest areas into smaller and more isolated patches, is primarily caused by human activities such as urbanization, agricultural expansion, infrastructure development, and logging [2]. These activities lead to the loss of forest cover, habitat destruction, and the disruption of ecological processes [8].
We examined ΔFFCI and its correlations with human-induced and natural factors, uncovering potential driving forces for the dynamics of forest landscape fragmentation in the study area. We revealed that the increase in FFCI was mainly attributed to human disturbances (i.e., POD and NTL) and changes in NDVI, respectively (Figure 5). Our research is consistent with the findings of previous studies [34,35,36], which have reported that one key driving pattern in forest landscape fragmentation is land use change. As human populations grow and urban areas expand, forests are often cleared to make way for agriculture, settlements, and infrastructure [36]. These factors led to an increase in small forest patches and patch numbers, directly contributing to forest landscape fragmentation from 2000 to 2020 (Figure 4). This conversion of forested land into fragmented patches not only reduces the overall forest cover but also disrupts wildlife habitats and migration corridors [10,37].
Another important driving pattern is the construction of roads and other linear infrastructure [38]. Roads can act as barriers to wildlife movement, fragmenting habitats and isolating populations. They also facilitate access to previously remote areas, leading to increased deforestation and degradation. Our study also confirms this finding, showing that forest landscape fragmentation was further increased during the study period near roads (Figure 5). However, the influence of roads on forest landscape fragmentation dynamics varies greatly with the change in sample set in terms of different model runs. This shows that the impact of roads on forest landscape fragmentation is characterized by spatial instability [39,40]. This is closely associated with the implementation of ecological restoration projects amid the rapid economic development in China [41].
There are other influencing factors (e.g., forest diseases and pests) that may lead to forest landscape fragmentation, but due to the difficulty of data acquisition, this study did not consider them. For example, the Masson pine occupies a certain proportion of the study area, and pine wood nematode disease sometimes causes the death of the plantation, which leads to the fragmentation of the forest landscape. Hence, the subsequent research on the dynamic attribution of forest landscape fragmentation should include more impact factors (e.g., forest diseases and pests) for a more comprehensive and systematic examination.
Therefore, to effectively address the driving patterns of the forest landscape fragmentation process, it is essential to implement sustainable land use planning practices that prioritize conservation and restoration efforts. This includes promoting landscape connectivity through the creation of wildlife corridors, protected areas, and sustainable land management practices [2,10].

4.3. Variability in Random Forest Models

Because of the inherent stochasticity of random forest modeling, the measurements of variable importance and the predictions derived from the trained models can be susceptible to random alterations in seed numbers, and random forest models might present differences among runs [23]. The seed number is used to initialize the random number generator in machine learning algorithms like random forests. Different seed numbers can lead to different random samples being generated during model training. Due to this sensitivity to seed numbers, small changes in the seed value can result in variations in the composition of training data for individual trees within the random forest [42]. These variations can impact the importance assigned to different variables during model training and influence the final predictions made by the ensemble of trees [43]. In practice, running a random forest model multiple times with different seed numbers or under slightly varied conditions can lead to variations in variable importance rankings and prediction outcomes [23]. These differences between runs highlight that while random forests provide robust and reliable predictions on average, individual model instances may vary due to their stochastic nature. To address the variability introduced by stochasticity in random forest modeling, researchers often perform multiple models runs with different seeds [44]. Under the guidance of these papers, we used a “variability assessment” set of 12 model runs specified with the adjusted tuning parameters and random seeds based on caret. Such leveraging ensemble techniques and robust evaluation methods can enhance the reliability and generalization capabilities of random forest models [45].

5. Conclusions

In this study, we developed the FFCI to capture key features of forest landscape fragmentation, such as patch size (MPA, LPI), number (PD), and distribution (AI and DIVISION). Then, taking Putian City of Fujian Province in Southeastern China as a case, we analyzed the spatial distribution in the static and dynamic forest landscape fragmentation, identified the coupling modes of forest landscape fragmentation, and explored the impactors of forest landscape fragmentation dynamics for the study area. The findings showed that:
(1)
The forest landscapes with different degrees of fragmentation exhibited more noticeable changes at both ends, with the intermediate level remaining consistent from 2000 to 2020. Around 18.3% of forest landscapes experienced a decrease in fragmentation, particularly in the northern part of the study area, while approximately 81.7% of forest landscapes exhibited an increasing trend in fragmentation. In most study regions during 2000–2020, fragmentation remained relatively stable (−0.2 < ΔFFCI < 0.2).
(2)
The bivariate spatial autocorrelation analysis indicated that the proportion of Low–High-type grids was the highest, at 17.3%, followed by the High–High type at 7.0%.
(3)
We also identified eight modes of fragmentation that indicate that the most significant forest landscape fragmentation pattern is a decrease in MPA and an increase in PD. The mode MPA↓AI↑PD- was one of the most prevalent and widely distributed patterns across the study area, accounting for 33.4% of the total area. It was followed by the mode MPA↓AI↑PD↑, which represented 32.6% of the area and was primarily found at the non-forest edges. Other modes, such as MPA↑AI↑PD↓ and MPA↑AI↑PD-, accounted for 17.1% and 15.9% of the area, respectively, and were predominantly concentrated in the central forest region.
(4)
The anthropogenic factors (e.g., population density and night light intensity) were found to dominate the FFCI dynamics during 2000–2020. However, the impact of DTR and DTF on forest landscape fragmentation dynamics varied significantly across different models.

Author Contributions

Conceptualization, X.H.; methodology, X.H. and X.L.; software, Q.Z. and S.Z.; data curation, Q.Z. and S.Z.; writing—original draft preparation, X.L.; writing—review and editing, X.L. and X.H.; visualization, S.Z.; supervision, X.H. All authors have read and agreed to the published version of the manuscript.

Funding

This research was funded by the National Natural Science Foundation of China (No. 31971639), the Natural Science Foundation of Fujian Province (No. 2023J01477), and the General Project of Education and Scientific Research for Young and Middle-aged Teachers of Fujian Province (No. JAS22135).

Data Availability Statement

The data are contained within the article, and all data sources are mentioned.

Conflicts of Interest

The authors declare no conflicts of interest.

References

  1. Taubert, F.; Fischer, R.; Groeneveld, J.; Lehmann, S.; Müller, M.S.; Rödig, E.; Wiegand, T.; Huth, A. Global patterns of tropical forest fragmentation. Nature 2018, 554, 519–522. [Google Scholar] [CrossRef]
  2. Ma, J.; Li, J.; Wu, W.; Liu, J. Global forest fragmentation change from 2000 to 2020. Nat. Commun. 2023, 14, 3752. [Google Scholar] [CrossRef] [PubMed]
  3. Arroyo-Rodríguez, V.; Melo, F.P.L.; Martínez-Ramos, M.; Bongers, F.; Chazdon, R.L.; Meave, J.A.; Norden, N.; Santos, B.A.; Leal, I.R.; Tabarelli, M. Multiple successional pathways in human-modified tropical landscapes: New insights from forest succession, forest fragmentation and landscape ecology research. Biol. Rev. 2017, 92, 326–340. [Google Scholar] [CrossRef] [PubMed]
  4. Chen, S.; Wu, S.; Ma, M. Ecological restoration programs reduced forest fragmentation by stimulating forest expansion. Ecol. Indic. 2023, 154, 110855. [Google Scholar] [CrossRef]
  5. Taylor, C.; Lindenmayer, D.B. Temporal fragmentation of a critically endangered forest ecosystem. Austral. Ecol. 2020, 45, 340–354. [Google Scholar] [CrossRef]
  6. Zhen, S.; Zhao, Q.; Liu, S.; Wu, Z.; Lin, S.; Li, J.; Hu, X. Detecting spatiotemporal dynamics and driving patterns in forest fragmentation with a forest fragmentation comprehensive index (FFCI): Taking an area with active forest cover change as a case study. Forests 2023, 14, 1135. [Google Scholar] [CrossRef]
  7. Magrach, A.; Laurance, W.F.; Larrinaga, A.R.; Santamaria, L. Meta-analysis of the effects of forest fragmentation on interspecific interactions. Conserv. Biol. 2014, 28, 1342–1348. [Google Scholar] [CrossRef] [PubMed]
  8. Siegel, T.; Magrach, A.; Laurance, W.F.; Luther, D. A global meta-analysis of the impacts of forest fragmentation on biotic mutualisms and antagonisms. Conserv. Biol. 2023, 38, e14206. [Google Scholar] [CrossRef] [PubMed]
  9. de Santana, N.S.; Santos, A.S.; Borges, D.B.; França, D.d.S.; Reis, J.B.L.; de Oliveira, F.A.; Barreto, M.A.; Corrêa, R.X.; Zucchi, M.I.; Martins, K.; et al. Genetic resilience of Atlantic forest trees to impacts of biome loss and fragmentation. Eur. J. For. Res. 2023, 142, 161–174. [Google Scholar] [CrossRef]
  10. Brennan, A.; Naidoo, R.; Greenstreet, L.; Mehrabi, Z.; Ramankutty, N.; Kremen, C. Functional connectivity of the world’s protected areas. Science 2022, 376, 1101–1104. [Google Scholar] [CrossRef]
  11. Gong, C.; Yu, S.; Joesting, H.; Chen, J. Determining socioeconomic drivers of urban forest fragmentation with historical remote sensing images. Landsc. Urban Plan. 2013, 117, 57–65. [Google Scholar] [CrossRef]
  12. Valle, D.; Silva, C.A.; Longo, M.; Brando, P. The Latent Dirichlet Allocation model applied to airborne LiDAR data: A case study on mapping forest degradation associated with fragmentation and fire in the Amazon region. Methods Ecol. Evol. 2022, 13, 1329–1342. [Google Scholar] [CrossRef]
  13. Frazier, A.E.; Kedron, P. Comparing forest fragmentation in Eastern, U.S. forests using patch-mosaic and gradient surface models. Ecol. Inform. 2017, 41, 108–115. [Google Scholar] [CrossRef]
  14. González-Ávila, S.; Ortega, E.; Martín, B. Multiscale fragmentation of forest types in Spain. For. Ecol. Manag. 2023, 546, 121317. [Google Scholar] [CrossRef]
  15. Jha, C.; Goparaju, L.; Tripathi, A.; Gharai, B.; Raghubanshi, A.; Singh, J. Forest fragmentation and its impact on species diversity: An analysis using remote sensing and GIS. Biodivers. Conserv. 2005, 14, 1681–1698. [Google Scholar] [CrossRef]
  16. Vellend, M.; Verheyen, K.; Jacquemyn, H.; Kolb, A.; Van Calster, H.; Peterken, G.; Hermy, M. Extinction debt of forest plants persists for more than a century following habitat fragmentation. Ecology 2006, 87, 542–548. [Google Scholar] [CrossRef] [PubMed]
  17. Gibson, L.; Lynam, A.J.; Bradshaw, C.J.A.; He, F.; Bickford, D.P.; Woodruff, D.S.; Bumrungsri, S.; Laurance, W.F. Near-complete extinction of native small mammal fauna 25 years after forest fragmentation. Science 2013, 341, 1508–1510. [Google Scholar] [CrossRef]
  18. Keken, Z.; Panagiotidis, D.; Skaloš, J. The influence of damming on landscape structure change in the vicinity of flooded areas: Case studies in Greece and the Czech Republic. Ecol. Eng. 2015, 74, 448–457. [Google Scholar] [CrossRef]
  19. Guo, H.; Duan, D.; Lei, H.; Chen, Y.; Li, J.; Albasher, G.; Li, X. Environmental Drivers of Landscape Fragmentation Influence Intraspecific Leaf Traits in Forest Ecosystem. Forests 2023, 14, 1875. [Google Scholar] [CrossRef]
  20. Sulieman, H.M. Exploring Drivers of Forest Degradation and Fragmentation in Sudan: The Case of Erawashda Forest and its Surrounding Community. Sci. Total. Environ. 2018, 621, 895–904. [Google Scholar] [CrossRef]
  21. Çalışkan, E.; Sevim, Y. Forest road detection using deep learning models. Geocarto Int. 2022, 37, 5875–5890. [Google Scholar] [CrossRef]
  22. Nasejje, J.B.; Mbuvha, R.; Mwambi, H. Use of a deep learning and random forest approach to track changes in the predictive nature of socioeconomic drivers of under-5 mortality rates in sub-Saharan Africa. BMJ Open 2022, 12, e049786. [Google Scholar] [CrossRef] [PubMed]
  23. Patoine, G.; Eisenhauer, N.; Cesarz, S.; Phillips, H.R.P.; Xu, X.; Zhang, L.; Guerra, C.A. Drivers and trends of global soil microbial carbon over two decades. Nat. Commun. 2022, 13, 4195. [Google Scholar] [CrossRef] [PubMed]
  24. Yang, J.; Huang, X. The 30 m annual land cover datasets and its dynamics in China from 1985 to 2022. Earth Syst. Sci. Data 2023, 13, 3907–3925. [Google Scholar] [CrossRef]
  25. Cinelli, M.; Coles, S.R.; Kirwan, K. Analysis of the potentials of multi criteria decision analysis methods to conduct sustainability assessment. Ecol. Indic. 2014, 46, 138–148. [Google Scholar] [CrossRef]
  26. Norouzian-Maleki, S.; Bell, S.; Hosseini, S.-B.; Faizi, M. Developing and testing a framework for the assessment of neighbourhood liveability in two contrasting countries: Iran and Estonia. Ecol. Indic. 2015, 48, 263–271. [Google Scholar] [CrossRef]
  27. Seddon, A.W.R.; Macias-Fauria, M.; Long, P.R.; Benz, D.; Willis, K.J. Sensitivity of global terrestrial ecosystems to climate variability. Nature 2016, 531, 229–232. [Google Scholar] [CrossRef] [PubMed]
  28. Hu, X.; Xu, H. A new remote sensing index for assessing the spatial heterogeneity in urban ecological quality: A case from Fuzhou City, China. Ecol. Indic. 2018, 89, 11–21. [Google Scholar] [CrossRef]
  29. Anselin, L. GeoDa 0.9 User’s Guide; Spatial Analysis Laboratory (SAL), University of Illinois, Urbana-Champaign: Urbana, IL, USA, 2003. [Google Scholar]
  30. Chen, J.; Xu, C.; Lin, S.; Wu, Z.; Qiu, R.; Hu, X. Is there spatial dependence or spatial heterogeneity in the distribution of vegetation greening and browning in Southeastern China? Forests 2022, 13, 840. [Google Scholar] [CrossRef]
  31. Hu, X.; Xu, C.; Chen, J.; Lin, Y.; Lin, S.; Wu, Z.; Qiu, R. A synthetic landscape metric to evaluate urban vegetation quality: A case of Fuzhou City in China. Forests 2022, 13, 1002. [Google Scholar] [CrossRef]
  32. Laurance, W.F.; Camargo, J.L.C.; Fearnside, P.M.; Lovejoy, T.E.; Williamson, G.B.; Mesquita, R.C.G.; Meyer, C.F.J.; Bobrowiec, P.E.D.; Laurance, S.G.W. An Amazonian rainforest and its fragments as a laboratory of global change. Biol. Rev. 2018, 93, 223–247. [Google Scholar] [CrossRef] [PubMed]
  33. Hansen, M.C.; Wang, L.; Song, X.-P.; Tyukavina, A.; Turubanova, S.; Potapov, P.V.; Stehman, S.V. The fate of tropical forest fragments. Sci. Adv. 2020, 6, eaax8574. [Google Scholar] [CrossRef] [PubMed]
  34. Zhou, W.; Zhang, S.; Yu, W.; Wang, J.; Wang, W. Effects of urban expansion on forest loss and fragmentation in six megaregions, China. Remote Sens. 2017, 9, 991. [Google Scholar] [CrossRef]
  35. Lin, Y.; Hu, X.; Zheng, X.; Hou, X.; Zhang, Z.; Zhou, X.; Qiu, R.; Lin, J. Spatial variations in the relationships between road network and landscape ecological risks in the highest forest coverage region of China. Ecol. Indic. 2019, 96, 392–403. [Google Scholar] [CrossRef]
  36. Engert, J.E.; Campbell, M.J.; Cinner, J.E.; Ishida, Y.; Sloan, S.; Supriatna, J.; Alamgir, M.; Cislowski, J.; Laurance, W.F. Ghost roads and the destruction of Asia-Pacific tropical forests. Nature 2024, 629, 370–375. [Google Scholar] [CrossRef] [PubMed]
  37. Fischer, R.; Taubert, F.; Müller, M.S.; Groeneveld, J.; Lehmann, S.; Wiegand, T.; Huth, A. Accelerated forest fragmentation leads to critical increase in tropical forest edge area. Sci. Adv. 2021, 7, eabg7012. [Google Scholar] [CrossRef] [PubMed]
  38. Hu, X.; Wu, Z.; Wu, C.; Ye, L.; Lan, C.; Tang, K.; Xu, L.; Qiu, R. Effects of road network on diversiform forest cover changes in the highest coverage region in China: An analysis of sampling strategies. Sci. Total. Environ. 2016, 565, 28–39. [Google Scholar] [CrossRef] [PubMed]
  39. Hu, X.; Zhang, L.; Ye, L.; Lin, Y.; Qiu, R. Locating spatial variation in the association between road network and forest biomass carbon accumulation. Ecol. Indic. 2017, 73, 214–223. [Google Scholar] [CrossRef]
  40. Lin, Y.; Qiu, R.; Yao, J.; Hu, X.; Lin, J. The effects of urbanization on China’s forest loss from 2000 to 2012: Evidence from a panel analysis. J. Clean. Prod. 2019, 214, 270–278. [Google Scholar] [CrossRef]
  41. Bryan, B.A.; Gao, L.; Ye, Y.; Sun, X.; Connor, J.D.; Crossman, N.D.; Stafford-Smith, M.; Wu, J.; He, C.; Yu, D.; et al. China’s response to a national land-system sustainability emergency. Nature 2019, 559, 193–204. [Google Scholar] [CrossRef]
  42. Wang, Y.; Chen, X.; Gao, M.; Dong, J. The use of random forest to identify climate and human interference on vegetation coverage changes in southwest China. Ecol. Indic. 2022, 144, 109463. [Google Scholar] [CrossRef]
  43. Leroux, L.; Bégué, A.; Seen, D.L.; Jolivot, A.; Kayitakire, F. Driving forces of recent vegetation changes in the Sahel: Lessons learned from regional and local level analyses. Remote Sens. Environ. 2017, 191, 38–54. [Google Scholar] [CrossRef]
  44. Kafy, A.-A.; Saha, M.; Fattah, A.; Rahman, M.T.; Duti, B.M.; Rahaman, Z.A.; Bakshi, A.; Kalaivani, S.; Rahaman, S.N.; Sattar, G.S. Integrating forest cover change and carbon storage dynamics: Leveraging Google Earth Engine and InVEST model to inform conservation in hilly regions. Ecol. Indic. 2023, 152, 110374. [Google Scholar] [CrossRef]
  45. Augusto, L.; Boča, A. Tree functional traits, forest biomass, and tree species diversity interact with site properties to drive forest soil carbon. Nat. Commun. 2022, 13, 1097. [Google Scholar] [CrossRef]
Figure 1. Location of study area. DEM: digital elevation model.
Figure 1. Location of study area. DEM: digital elevation model.
Forests 15 01212 g001
Figure 2. Spatial distribution of the static forest landscape fragmentation (FFCI) and the dynamic forest landscape fragmentation (ΔFFCI) during 2000 and 2020. (a,b) Static FFCI in 2000 and 2020; (c) the area proportion of each FFCI class in 2000 and 2020; (d) dynamic forest landscape fragmentation (ΔFFCI) from 2000 to 2020; (e) the histogram of the numbers of forest pixels (count) of ΔFFCI, with the green indicates decreasing FFCI and the pink indicates increasing FFCI.
Figure 2. Spatial distribution of the static forest landscape fragmentation (FFCI) and the dynamic forest landscape fragmentation (ΔFFCI) during 2000 and 2020. (a,b) Static FFCI in 2000 and 2020; (c) the area proportion of each FFCI class in 2000 and 2020; (d) dynamic forest landscape fragmentation (ΔFFCI) from 2000 to 2020; (e) the histogram of the numbers of forest pixels (count) of ΔFFCI, with the green indicates decreasing FFCI and the pink indicates increasing FFCI.
Forests 15 01212 g002
Figure 3. Bivariate spatial autocorrelation analysis. (a) Spatial distribution of cluster types in terms of bivariate spatial autocorrelation between static FFCI and dynamic FFCI, with others including not significant and neighborless grids; (b) proportions of each cluster type.
Figure 3. Bivariate spatial autocorrelation analysis. (a) Spatial distribution of cluster types in terms of bivariate spatial autocorrelation between static FFCI and dynamic FFCI, with others including not significant and neighborless grids; (b) proportions of each cluster type.
Forests 15 01212 g003
Figure 4. Spatial distributions of forest landscape fragmentation process modes (a) and their proportions (b). 1: MPA↑AI↑PD↑; 2: MPA↑AI↑PD↓; 3: MPA↓AI↓PD↓; 4: MPA↓AI↑PD↑; 5: MPA↓AI↑PD↓; 6: MPA↑AI↑PD-; 7: MPA↓AI↑PD-; 8: MPA-AI↑PD-. No following modes: MPA↑AI↓PD↑, MPA↑AI↓PD↓, MPA↓AI↓PD↑. In which the sign “↑”, ”↓” and “-” mean increase, decrease and no change, respectively.
Figure 4. Spatial distributions of forest landscape fragmentation process modes (a) and their proportions (b). 1: MPA↑AI↑PD↑; 2: MPA↑AI↑PD↓; 3: MPA↓AI↓PD↓; 4: MPA↓AI↑PD↑; 5: MPA↓AI↑PD↓; 6: MPA↑AI↑PD-; 7: MPA↓AI↑PD-; 8: MPA-AI↑PD-. No following modes: MPA↑AI↓PD↑, MPA↑AI↓PD↓, MPA↓AI↓PD↑. In which the sign “↑”, ”↓” and “-” mean increase, decrease and no change, respectively.
Forests 15 01212 g004
Figure 5. Box plot and normal curve of the standardized importance of each variable to the dynamic forest landscape fragmentation (ΔFFCI) in the study area (n = 2760). The solid stars represent standardized importance estimates for each variable from 12 random forest model runs. The curves indicate the normal curve of the standardized importance of each variable. The root mean square error (RMSE) and R squared (R2) ranged between 69.9 and 75.0 and between 0.723 and 0.837, respectively. NTL: night light, POD: population density, NDVI: normalized difference vegetation index, DTF: distance to forest edge, DTR: distance to road.
Figure 5. Box plot and normal curve of the standardized importance of each variable to the dynamic forest landscape fragmentation (ΔFFCI) in the study area (n = 2760). The solid stars represent standardized importance estimates for each variable from 12 random forest model runs. The curves indicate the normal curve of the standardized importance of each variable. The root mean square error (RMSE) and R squared (R2) ranged between 69.9 and 75.0 and between 0.723 and 0.837, respectively. NTL: night light, POD: population density, NDVI: normalized difference vegetation index, DTF: distance to forest edge, DTR: distance to road.
Forests 15 01212 g005
Table 1. List of driving factors.
Table 1. List of driving factors.
TypeDataVariableSources
Geomorphic factorsDigital Elevation Model (DEM)Elevationhttps://www.gscloud.cn/ (accessed on 6 December 2021)
Slope
Natural factorsNormalized Difference Vegetation Index (NDVI)NDVIhttp://www.resdc.cn/ (accessed on 6 December 2023)
China Land Cover Dataset (CLCD)Distance to Forest (DTF)Google Earth Engine
Socio-economic factorsPopulation Density (POD)PODWorldPop dataset (accessed on 12 December 2021)
Nighttime-Light Intensity (NTL)NTLhttp://www.geodata.cn/ (accessed on 12 December 2021)
Open Street Map (OSM)Distance to Road (DTR)Open Street Map (accessed on 12 December 2022)
Disclaimer/Publisher’s Note: The statements, opinions and data contained in all publications are solely those of the individual author(s) and contributor(s) and not of MDPI and/or the editor(s). MDPI and/or the editor(s) disclaim responsibility for any injury to people or property resulting from any ideas, methods, instructions or products referred to in the content.

Share and Cite

MDPI and ACS Style

Lin, X.; Zhen, S.; Zhao, Q.; Hu, X. A New Paradigm for Assessing Detailed Dynamics of Forest Landscape Fragmentation. Forests 2024, 15, 1212. https://doi.org/10.3390/f15071212

AMA Style

Lin X, Zhen S, Zhao Q, Hu X. A New Paradigm for Assessing Detailed Dynamics of Forest Landscape Fragmentation. Forests. 2024; 15(7):1212. https://doi.org/10.3390/f15071212

Chicago/Turabian Style

Lin, Xin, Shiyong Zhen, Qing Zhao, and Xisheng Hu. 2024. "A New Paradigm for Assessing Detailed Dynamics of Forest Landscape Fragmentation" Forests 15, no. 7: 1212. https://doi.org/10.3390/f15071212

APA Style

Lin, X., Zhen, S., Zhao, Q., & Hu, X. (2024). A New Paradigm for Assessing Detailed Dynamics of Forest Landscape Fragmentation. Forests, 15(7), 1212. https://doi.org/10.3390/f15071212

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

Article Metrics

Back to TopTop