Next Article in Journal
Research on Digital Twin Method for Spaceborne Along-Track Interferometric Synthetic Aperture Radar Velocity Inversion of Ocean Surface Currents
Previous Article in Journal
Estimating Vertical Distribution of Total Suspended Matter in Coastal Waters Using Remote-Sensing Approaches
 
 
Search for Articles:
Title / Keyword
Author / Affiliation / Email
Journal
Article Type
 
 
Advanced Search
 
Section
Special Issue
Volume
Issue
Number
Page
 
Journals
Remote Sensing
Volume 16
Issue 19
10.3390/rs16193738
remotesensing-logo
Submit to this Journal Review for this Journal Propose a Special Issue
► Article Menu

Article Menu

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

Remote Sensing Mapping and Analysis of Spatiotemporal Patterns of Land Use and Cover Change in the Helong Region of the Loess Plateau Region (1986–2020)

by
Jingyu Li
,
Yangbo Chen
*,
Yu Gu
,
Meiying Wang
and
Yanjun Zhao
School of Geography and Planning, Sun Yat-sen University, Guangzhou 510275, China
*
Author to whom correspondence should be addressed.
Remote Sens. 2024, 16(19), 3738; https://doi.org/10.3390/rs16193738
Submission received: 15 August 2024 / Revised: 5 October 2024 / Accepted: 7 October 2024 / Published: 8 October 2024
Browse Figures
Versions Notes

Abstract

:
Land use and cover change (LUCC) is directly linked to the sustainability of ecosystems and the long-term well-being of human society. The Helong Region in the Loess Plateau has become one of the areas most severely affected by soil and water erosion in China due to its unique geographical location and ecological environment. The long-term construction of terraces and orchards is one of the important measures for this region to combat soil erosion. Despite the important role that terraces and orchards play in this region, current studies on their extraction and understanding remain limited. For this reason, this study designed a land use classification system, including terraces and orchards, to reveal the patterns of LUCC and the effectiveness of ecological restoration projects in the area. Based on this system, this study utilized the Random Forest classification algorithm to create an annual land use and cover (LUC) dataset for the Helong Region that covers eight periods from 1986 to 2020, with a spatial resolution of 30 m. The validation results showed that the maps achieved an average overall accuracy of 87.54% and an average Kappa coefficient of 76.94%. This demonstrates the feasibility of the proposed design and land coverage mapping method in the study area. This study found that, from 1986 to 2020, there was a continuous increase in forest and grassland areas, a significant reduction in cropland and bare land areas, and a notable rise in impervious surface areas. We emphasized that the continuous growth of terraces and orchards was an important LUCC trend in the region. This growth was primarily attributed to the conversion of grasslands, croplands, and forests. This transformation not only reduced soil erosion but also enhanced economic efficiency. The products and insights provided in this study help us better understand the complexities of ecological recovery and land management.

1. Introduction

Land use and cover change (LUCC) is a critical component of global climate and environmental change research, and it has become one of the core issues in current global change studies [1]. LUCC is a direct manifestation of the interactions and mutual influences between human activities and the terrestrial natural environment. In recent years, due to intense human activities, there have been significant changes in the processes and patterns of LUCC at regional and even global scales, profoundly impacting the conditions of the Earth’s surface, water cycle processes, and biodiversity [2,3,4]. For example, urbanization has led to the encroachment of buildings and roads on cropland, while agricultural expansion and deforestation have altered the vegetation coverage of the land surface [5,6]. These changes have not only impacted the global ecological environment but also significantly affected economic development [7]. Therefore, studying the spatiotemporal patterns of LUCC is particularly important for guiding ecological conservation efforts and promoting high-quality development in China.
The Loess Plateau in the midstream of the Yellow River Basin, especially the section from Hekou Town to Longmen River (referred to as the Helong Region), has become one of the areas most severely affected by soil erosion in the world due to the complex terrain, loose soil, concentrated rainfall, and irrational human activities [8]. This poses a serious threat to the safety of people’s lives and property. To effectively alleviate the ecological pressure in this region, the government has implemented ecological restoration projects over the past few decades, among which the Grain for Green Project (GGP) has been a key measure for soil erosion control and ecological restoration on the Loess Plateau [9,10]. Meanwhile, when food security was threatened, policies emphasized measures that combined economic development with ecological restoration, such as transforming slope farmland to terraces [11] and the GGP. These measures significantly altered the spatiotemporal patterns of the region, promoted economic development, and mitigated the severity of soil erosion. Large-scale environmental destruction and ecological restoration activities have profoundly affected the land systems of the region. More importantly, the Helong Region has significant topographical variation and strong surface heterogeneity. Its complex terrain and land features affect the LUCC and the ecosystem’s response in this area. It is evident that, under the combined influence of environmental factors, ecological conservation projects, and socioeconomic conditions, the region’s vegetation coverage and LUC have undergone dramatic changes.
However, the long-term evolution of environmental patterns in the Helong Region is unclear, and the effectiveness of ecological conservation projects requires long-term monitoring and evaluation. Generating long-term LUCC information to reveal the processes and patterns of land use change in the Helong Region is highly desirable. In fact, many LUC products have already been widely applied in the Loess Plateau region. However, these products, such as FROM-GLC [12], GlobeLand30 [13], and GLC-FCS30 [14], mostly provide macroscale land cover information at the global or national level. This makes them insufficient for studying the impacts of land use changes in specific regions and specific fields. The Helong Region has a unique and significant type of land use: terraces and orchards. This type of land use occupies a considerable proportion of the Helong Region and has shown significant changes. Moreover, the construction of terraces and orchards is a key focus for controlling soil erosion and restoring the ecological environment on the Loess Plateau [15]. However, most current land use products do not consider terraces and orchards, and there are relatively few data on the spatiotemporal variability of this type. Additionally, most studies are short-term, with little information on long-term temporal variability [16]. Therefore, existing studies have struggled to accurately describe the transformation of unique land features in the region, and conducting assessments and optimizing land management and ecological restoration strategies based on LUCC has also been challenging. For regions with strong land heterogeneity, such as the Helong Region, there is an urgent need to establish a set of region-specific land use data.
To address the research gap, in this study, we utilized the Random Forest (RF) method, with the Google Earth Engine (GEE) platform and Landsat satellite images, to develop LUCC maps for eight periods from 1986 to 2020 in the study area, with a 5-year interval for each period. We specially designed a land use classification system that includes terraces and orchards to explore and analyze the characteristics and underlying causes of land use changes in the Helong Region. By introducing the category of terraces and orchards, this study not only reveals the effectiveness of ecological restoration projects but also offers a novel perspective for understanding the complexities of ecological recovery and land management.

2. Materials and Methods

2.1. Study Area

The Helong Region is located in the upper midstream of the Yellow River Basin, which is an important part of the Loess Plateau (Figure 1). The total area is about 1.3 × 105 km2, accounting for approximately 37.5% of the area of the midstream. The region is located in the western part of the Lvliang Mountain Range in Shanxi, bordering the Ordos Plateau in Inner Mongolia to the northwest and the Baiyu and Huanglong Mountains in Shanxi to the southwest [17]. The region is dominated by the hilly erosive landscape of the Loess Plateau. The terrain is low in the middle and high on the periphery, with a higher elevation in the north and a lower elevation in the south, making the landscape complex and fragmented. This region is located in the transitional zone between semi-arid and semi-humid regions, characterized by a temperate continental monsoon climate. The long-term average temperature is around 8 °C. The annual average rainfall is about 400 mm, with precipitation being concentrated and unevenly distributed spatially. Due to the special geographic and climatic conditions, the Helong Region has long faced serious soil erosion problems. To improve the situation, the region was actively engaged in the construction of a large number of terraces and orchards. Terraces and orchards have become an indispensable green asset between the River Dragon districts.

2.2. Data Source

The Landsat images were all obtained from the Google Earth Engine (GEE) platform. The GEE is a cloud platform specifically designed for processing and analyzing vast amounts of satellite imagery and other Earth observation data [18]. It integrates data from various sensors and platforms, including land satellites, MODIS, and Sentinel, as well as other datasets [18,19]. Table 1 shows the types of Landsat images used in different years. The Landsat project is a key Earth observation program jointly managed by the United States Geological Survey (USGS) and the National Aeronautics and Space Administration (NASA) [20]. Since 1972, the Landsat series of satellites has provided continuous surface imagery that is valuable for monitoring environmental changes and land use dynamics [21]. The spatial resolution and long-term consistency of Landsat data make it an important tool for regional-scale LUC analysis. In addition, the free and open access to Landsat data greatly facilitates scientific research and practical applications on a global scale [22].

2.3. Methodology

To decipher terraces and orchards in more detail, this study constructed a targeted LUC classification framework. The entire process relied on the GEE platform, including data preprocessing, LUC data generation, and LUC change analysis. The data used in this study were based on selected and corrected Landsat imagery, which included six spectral bands of Landsat. Additionally, this study incorporated digital elevation model (DEM) data provided by the Shuttle Radar Topography Mission (SRTM) [23] to generate terrain factor data, which are indispensable for distinguishing terraces from other flat crop fields [24]. To further enhance classification accuracy, we used spectral indices computed from Landsat imagery in combination with texture features extracted from the gray-level covariance matrix (GLCM) to capture the unique texture patterns of terraces [25,26]. In the selection of the classification algorithm, this study employed the RF algorithm [27], which is recognized for its high accuracy in processing complex land cover types. To optimize the final map product, we applied a majority filtering method [28] to spatially smooth the mapping results, ensuring the continuity and consistency of the terraced areas. By employing this method, we successfully generated eight 30-m resolution LUCC maps for the Helong Region from 1986 to 2020, with intervals of five years (except for the period from 1986 to 1990, where the interval was four years). To conduct an in-depth analysis of the spatiotemporal characteristics of the terraces and orchards, we utilized tools such as the changes in the LUCC area, spatial grid area rate of change, and LUC transition matrices. A detailed overview of the entire workflow is shown in Figure 2.

2.3.1. Data Preprocessing

To obtain more high-quality and clear images, we filtered out Landsat images with high cloud coverage. Specifically, we used the filterMetadata() function to select images with less than 10% cloud cover. Additionally, we employed the custom functions “rmL8CloudNew” and “rmL457CloudNew” to remove clouds and cloud shadows further. From 1986 to 2020, we obtained a total of 1060 Landsat images (with a cloud cover percentage of less than 10%) of the study area (Figure 3). We found that the images used had more distribution at high latitudes (39°N–42°N) than at low latitudes (35°N–38°N). To repair the missing pixels in the Landsat scenes and to address the gaps in the Landsat 7 images, we applied the morphological mean filter [29] to an image with gaps, obtaining a blurred-filled image. Finally, we blended the original image with gaps onto the blurred-filled image, resulting in a repaired image where the gaps were filled with blurred data. This not only preserved the original data but also filled the gaps (Figure 4). To obtain a sufficient number of high-quality Landsat scenes for classification in the target year, we followed the approach of previous studies and selected a 12-month time window to synthesize the annual image for the target year. We explored five compositing methods in the GEE to study land use changes in the Helong Region. These methods were simple compositing, max value compositing, min value compositing, mean compositing, and median compositing. After testing each method, we found that median compositing performed the best; it was also more efficient in terms of computational resources and saving memory. Median compositing is a pixel-based image compositing technique that effectively eliminates extreme values caused by clouds and cloud shadows [30]. Additionally, this method reduced the edge effects between different Landsat image scenes, thereby improving the quality and consistency of the composite images [31]. Therefore, we selected the median compositing method to create LUCC image datasets from the filtered, cloud-removed, and data-repaired Landsat images.

2.3.2. Design of the Land Use and Cover Classification System

Based on the characteristics of the midstream and the quality of the image data, this study developed a classification system suitable for the study area (Table 2). The image displayed in Table 2 is a pseudo-color composite from Landsat 8, using the following RGB channel bands: Band 5, Band 4, and Band 3. The first is the identification of terraces and orchards. As an important measure for gully erosion control works and soil and water conservation projects, artificial terraces planted with fruit trees and crops are widely distributed in the Loess Plateau and play a significant role in reducing soil erosion and enhancing agricultural production [32,33]. Therefore, our classification system included a “terraces and orchards” category, which was mainly distributed on slopes and tablelands and showed significant signs of human activity. This categorization divided agricultural land into “cropland” and “terraces and orchards”. Moreover, since the end of the last century, the Loess Plateau region has implemented a series of ecological restoration projects, including GGP, as well as the Natural Forest Conservation Project (NFCP) [34]. Therefore, grassland, forest, and cropland are essential land types in this region. Thus, our classification system included a “terraces and orchards” category, which was mainly distributed on slopes and tablelands and showed significant signs of human activity. Based on familiarity with the basic characteristics of different land cover types and image interpretation signs in the study area, the LUC classification system contained seven categories: forest, grassland, impervious surface, water, bare land, cropland, and terraces and orchards.

2.3.3. Classification Method and Strategy

To determine the most appropriate LUC classification method for the study area, we conducted tests using 2010 as an example. The 1151 reference polygon samples were generated through visual interpretation of high-resolution imagery on the Google Earth Engine (GEE). Using Google Earth high-resolution imagery, we identified and delineated different land cover types to create polygons that accurately represented these features. The test feature set consisted of data in six spectral bands from Landsat 7 and a sample of 1151 reference polygons in the region. To evaluate the accuracy of the classifier, 70% of the reference samples were used for training, and the remaining 30% were used for validation. The classification methods involved in testing can be used directly in the GEE. By comparing the accuracy of different classification methods, we found that the overall accuracy of the RF classifier was 90.16%, with a Kappa coefficient of 86.37%, both of which were higher than those of the other classifiers (Table 3). This indicates that the RF classification method is reliable for land use classification in regions with complex geomorphology and high land heterogeneity [29]. Moreover, Sheykhmousa et al. [35] examined and compared RF and SVM, which are both very successful ML classifiers, and they found that the SVM classifier is more effective when fewer LUCC classes are used (less than 5.5 classes) and that RF one is more effective when the number of is higher. This can be seen in this study, in which the number of classes was set to 7 in order to justify the use of RF over SVM. Therefore, we proceeded with our research based on the RF classification algorithm.
To enhance the classification accuracy and reliability of specific land cover types, such as terraces and orchards, this study introduced terrain factor features, texture features, and spectral features. Firstly, terrain factor features were introduced. Terrain factor features are key to distinguishing slopes from flat terrain, and they are particularly important for terraces, which are usually located in areas with larger slopes [24]. We utilized SRTM DEM data (30 m spatial resolution), from which we extracted terrain factor features such as elevation, slope, aspect, and topographic diversity. The elevation data helped us to identify the distribution of terraces and orchards at different altitudes. Furthermore, the slope information directly reflected the degree of the inclination of the terraces. The aspect data provided indications of the lighting conditions for terraces and orchards, and the topographic diversity index revealed the complexity of the terrain where the terraces and orchards were located. Texture features were considered. Terraces and orchards are fields that have been excavated and constructed along contour lines on the hillsides of mountainous and hilly areas, mainly distributed on the plateau surfaces, and they possess unique textural patterns. The gray level cooccurrence matrix (GLCM) [36] helps capture these textural characteristics [26]. We calculated six classic GLCM texture features based on two bands. These texture features reflected the surface roughness, structural directionality, and complexity of the terraces and orchards, providing auxiliary information for the identification of terraces. During the classification process, we introduced canny edge detection technology to assist in recognizing land class boundaries [37]. Finally, various spectral features could enhance the accuracy of land-type mapping [38]. Certain vegetation indices could assist the classifiers in distinguishing vegetation types from a phenological perspective. Commonly used indices such as NDVI [39], RVI [38], and EVI [40] were employed. In addition, spectral indices for extracting water (NDWI) [41], impervious surfaces (NDBI) [42], and bare soil features (BSI) [43] were utilized. These spectral indices improve the identification accuracy of non-terrace areas, thereby indirectly improving the recognition precision of terraces and orchards with unique spectral characteristics [24]. By integrating these factors, the application of these data in mountainous areas with complex terrains improved the precision and reliability of terrace identification. In summary, we selected a total of 29 features (Table 4). Within the study area, we evenly mapped the reference samples for each period using the GEE. During the plotting process, we combined Landsat’s RGB false-color composite images and high-resolution imagery from Google Earth for a visual interpretation to ensure the accuracy of the samples. The distribution of these sample polygons in time and category is shown in Figure 5. In the post-processing classification, due to the fragmentation of the land surface in the study area and the poor image quality in some years, there were still some debris patches and the “pretzel phenomenon” in the classification results after using RFs for LUC classification. To improve this phenomenon, this study used the majority filter method [32] to smooth the images. We used the majority parameter, eight-neighborhood, filter, and optimal smoothing window (kernel size = 5) in the majority filter to process the land cover classification results through two iterations. Through several adjustments, we found that the majority filter with these parameter settings could effectively reduce the number of small-area anomalous patches generated during classification while preserving the edge features of the terrain.

3. Results

3.1. Data Validation and Evaluation

This study produced the annual land use classification product HL-LUC for the Helong Region, and it covers eight periods from 1986 to 2020. In this study, we used a stratified random sampling method to collect 1236 validation sample points on Google Earth for accuracy assessment (Figure 6). Due to the limitations imposed by the advent of the commercial era of high spatial resolution satellites (around 2000), it was challenging to obtain high-resolution remote sensing images from before 2001. Therefore, we conducted random sampling validation on the land use and cover maps from the last four periods. This study computed evaluation metrics, including the overall accuracy (OA), Kappa coefficient, and F1 score (weighted harmonic mean of producer’s and user’s accuracy), from the confusion matrix. The evaluation results are shown in Table 5. The validation results showed that the average total accuracy of the dataset was 87.54% and that the average Kappa coefficient was 76.94%. The confusion matrix analysis indicated that the average F1 scores such as forest, grassland, bare land, water, impervious surfaces cropland, and terraces and orchards were 0.91, 0.93, 0.85, 0.77, 0.68, and 0.70, respectively. There were some misclassifications between impervious surfaces and terraces and orchards due to the vulnerability of these types to human activities. In addition, some terraces and orchards were easily classified as cropland, possibly because most sloping cropland is located on slopes along with terraces and orchards. However, the overall accuracy of the images for all periods was above 80%. Therefore, terraces and orchards, as well as other land use types, have a high classification accuracy. Additionally, as shown in Table 5, the F1 scores for each land type in 2020 and 2015 were generally higher than those in the other years, with relatively lower misclassification rates. This was mainly due to the image quality of Landsat 8 being better than that of Landsat 7 and 5. Overall, the HL-LUC products exhibited high precision and reliability.

3.2. Land Use and Land Cover Change

By calculating the LUC area change, LUC transition matrix, and spatial distribution of the area change rates for each LUC type (Figure 7 and Figure 8 and Table 6), this study conducted an in-depth analysis of the spatiotemporal evolution patterns of various LUC types. The results indicated that from 1986 to 2020, the forest area increased from 12,403.05 km2 to 14,608.07 km2, mainly contributed to by grasslands (2631.84 km2) and croplands (550.8 km2). Forest growth was primarily concentrated in natural forest areas such as the northern part of the Huanglong Mountains and the Lvliang Mountains region. The grassland area also increased from 69,174.46 km2 to 87,828.04 km2, mainly contributed to by croplands (13,469.17 km2). Grasslands were mainly distributed in areas with a low terrain and flat topography, as well as in the sandy grassland areas in the northwest. The increase in forest and grassland areas demonstrates the significant benefits of national ecosystem protection policies such as the GGP and the NFCP [44].
Meanwhile, the growth of impervious surfaces was particularly significant, surging from 246.91 km2 in 1986 to 2285.10 km2 in 2020, being primarily converted from croplands (857.03 km2). The rapid growth of impervious surfaces was mainly concentrated in the northwest of Ordos City, Inner Mongolia, which was related to the accelerated urbanization process and the expansion of infrastructure construction in this region [45]. The cropland area decreased significantly from 27,524 km2 in 1986 to 13,990.84 km2 in 2020, being primarily converted to grasslands (13,469.17 km2), terraces and orchards (1001.67 km2), and impervious surfaces (857.03 km2). The expansion of urban areas in this region has benefited from national macroeconomic policies such as the Western Development Strategy and the Rise of Central China policy [46]. However, the reduction in cropland resources cannot be separated from the encroachment of construction land, which poses a challenge to agricultural production and food security [47]. Therefore, to prevent the excessive erosion of cropland by urbanization, the red line policy for cropland must be adhered to during the urbanization process. Basic cropland should be strictly protected, and the pace of urbanization should not be pursued blindly so as to avoid the conversion of basic cropland to non-food and non-agricultural uses [48]. Additionally, the area of bare land decreased significantly from 12,507.82 km2 in 1986 to 2661.41 km2 in 2020, being primarily converted to grasslands (6750.35 km2) and croplands (1613.48 km2). In the Mu Us Desert of Yulin City, there has been a notable increase in the grassland area, demonstrating the positive results of vegetation restoration measures [49]. The water area has shown fluctuating declines, decreasing from 655.13 km2 in 1986 to 602.03 km2 in 2020. The minor fluctuations during this period were primarily attributed to increased human activities (such as overgrazing) and decreased water levels due to permafrost thawing caused by rising temperatures [50,51]. The area of terraces and orchards has been continuously increasing, from 4267.11 km2 in 1986 to 5601.25 km2 in 2020, mainly concentrated in the central regions of Yulin and Lvliang cities. The spatiotemporal changes are discussed in detail in the next subsection.

3.3. Transition From Multiple Land Use and Cover Types to Terraces and Orchards

During this study, we observed a significant increase in the total area of terraces and orchards, which expanded from 4267.11 km2 in 1986 to 5601.25 km2 in 2020 due to various LUC changes. This increase was primarily attributed to the conversion of forests (13.68 km2), grasslands (2263.94 km2), and croplands (1001.67 km2) to terraces and orchards. To visually represent the spatiotemporal distribution of these land use types, we created spatial distribution maps for the transition of different land types to terraces and orchards from 1986 to 2000, from 2000 to 2010, and from 2010 to 2020 (Figure 9). In these distribution maps, green represents the spatiotemporal distribution of forests transformed into terraces and orchards, blue indicates the spatiotemporal distribution of cropland transformed into terraces and orchards, and red represents the spatiotemporal distribution of grassland transformed into terraces and orchards. In addition, we further obtained land use transfer results between forests, grasslands, croplands, terraces, and orchards in adjacent periods (Figure 10). The results showed that the transition from forests to terraces and orchards mainly occurred at the eastern edge of the Lvliang Mountain area, indicating that human activities have encroached on natural forest areas. Meanwhile, the transformation of grassland and cropland to terraces and orchards occurred mainly on the slopes bordering the plains and mountains in the central and western parts of the study area. The area of terraces and orchards has seen a continuous increase, with notable expansions during the periods from 1986 to 1990 and from 2015 to 2020. In 1986–1990, terraces and orchards were mainly converted from grasslands (997.15 km2) and croplands (516.89 km2). During 2015–2020, terraces and orchards were mainly converted from grasslands (1883.27 km2), croplands (981.74 km2), and forests (21.79 km2). The significant transfer of land to orchards and terraces demonstrates the remarkable effectiveness of ecological restoration policies. Specifically, from the late 1970s to the late 1990s, the Loess Plateau implemented gentle slope terracing and developed the fruit industry, achieving good ecological governance results [52]. The significant increase in terraces and orchards from 1986 to 1990 reflects this policy. Compared with the first round of GGP (1990–2007), the second round (launched in 2014) of this project encouraged farmers to develop economic forestry, such as planting fruit trees and medicinal herbs [53]. However, this could also lead to some forest encroachment, as evidenced by the conversion of forest land to terraces and orchards from 2015 to 2020. A study showed that the apple coverage and the production of the Loess Plateau in 2016 were 1.3 million hectares and 23 million tons, accounting for 25.2% and 26.3% of the global land coverage and apple production, respectively [54]. From a sustainable perspective, the expansion of terraces and orchards has not only significantly reduced the rate of soil erosion and improved the ecological environment but has also improved economic efficiency and provided diversified livelihood options for rural households [55].

4. Discussion

4.1. Evaluation of the Accuracy of the HL-LUC Product in Relation to that of Other Products

To further evaluate the reliability of the LUC data in this study, we selected four areas within the study region and compared our LUC data with three other major existing LUC products under the first-level classification (Table 7). These first-level classifications included forests, grasslands, croplands, terraces and orchards, impervious surfaces, water, and bare lands. To better reflect the quality of the HL-LUC product, we compared the classification accuracy of the HL-LUC (2015) product with that of the other three products. Based on the validation sample points obtained from the historical imagery of the Helong Region in 2015 on Google Earth, we conducted an accuracy assessment of multiple source products (Figure 11). In terms of OA, HL-LUC was the highest (77.94%), and, in descending order of OA, the other data ranked as follows: FROM_GLC (72.49%), GLC_FCS30 (71.20%), and ESA_CCI-LC (50.81%). The Kappa coefficient for products with a spatial resolution of 30 m was above 70%, with HL-LUC (89.10%) and GLC_FCS30 (80.15%) both exceeding 80%. However, as the resolution of the products decreased, the overall classification accuracy also showed a downward trend. Therefore, HL-LUC achieved the highest accuracy results, primarily because the data were scaled using the Helong Region, which provided more samples in the study area than the other datasets. Additionally, the classification system of HL-LUC belonged to the first level of classification, which increased the accuracy of clustering. ESA-CCI-LC had the lowest accuracy results due to it having the coarsest resolution among all the datasets. The coarser grid increased the probability of mixed pixels and may have introduced positional inaccuracies. The results of this study emphasized the importance of selecting datasets that are appropriate for a particular study area. The high accuracy of HL-LUC may be related to its customization for the HeLong Region, which provided insights for other studies that should consider regional specificity when selecting datasets. By comparing different LUC products, this study revealed the limitations of the existing dataset and provided directions for improving the existing dataset, especially in terms of improving resolution and classification accuracy.

4.2. Comparative Analysis of Terraces and Orchard Land Use Type in HL-LUC

Terraces and orchards are not only significant features of the Loess Plateau region but are also indispensable components of ecological conservation and land management in the area. In the past few decades, to effectively alleviate the severe soil erosion problem, the Loess Plateau region implemented a series of long-term slope-to-terrace projects. These projects not only altered the physical morphology of the surface but also profoundly influenced the LUC patterns. Therefore, conducting an in-depth analysis of the spatiotemporal transitions of terraces and orchards not only enables a deeper understanding of the dynamic changes in these land use types over different periods but also holds significant importance for evaluating and optimizing the effectiveness of ecological restoration projects. To demonstrate the originality and novelty of the land use products in this study, we further explored the differences between HL-LUC, FROM-GLC, GLC-FCS30, and ESA-CCI-LC. This study conducted a comparison of these four datasets in the local area (Figure 12). Overall, the FROM-GLC, GLC-FCS30, and ESA-CCI-LC products failed to identify terraces and orchards. FROM-GLC exhibited a high accuracy, similar to that of HL-LUC, in providing the spatial distribution of land cover and could roughly outline the contours of terraces and orchards. However, FROM-GLC had difficulty distinguishing terraces and orchards from cropland, whereas HL-LUC could clearly differentiate between them (Figure 12-(2)). GLC-FCS30 could identify fragmented patches but still failed to recognize continuous, complete terraces and orchards. Additionally, compared with GLC-FCS30, HL-LUC showed more detail and coherence in depicting cropland, thereby improving the readability of the classification results (Figure 12-(1)). Thanks to its higher resolution, HL-LC exhibited a significantly higher accuracy than ESA-CCI-LC in identifying bare lands, terraces, and orchards, providing more detailed and accurate classification results. Therefore, only HL-LUC identified the terraces and orchards in the Helong Region and provided long-term spatial distribution data. HL-LUC’s ability to identify specific land use types, especially terraces, and orchards, demonstrated its innovation and uniqueness in land use datasets. HL-LUC was able to provide long-term spatial distribution data of terraces and orchards in the HeLong Region, which was valuable for monitoring land use change trends and formulating related policies.

5. Conclusions

This study designed a land use classification system for the Helong Region that includes terraces and orchards, and it employed the RF method to map LUC data for eight periods from 1986 to 2020. By examining the changes in LUC area, the rate of area changes in spatial grids, and the LUC transition matrix, this study analyzed the spatiotemporal patterns of LUC changes from 1986 to 2020, including those of terraces and orchards. The conclusions were as follows:
(1)
Based on the GEE platform, we obtained 1060 Landsat images with less than 10% cloud coverage and used the RF algorithm to acquire LUCC classification data for eight periods from 1986 to 2020. The results indicated an average overall accuracy of 87.54% and an average Kappa coefficient of 76.94%. In the study area, HL-LUC demonstrated a higher classification accuracy than FROM-GLC, GLC-FCS30, and ESA-CCI-LC, and only HL-LUC identified the terraces and orchards in the Helong Region, providing long-term spatial distribution data. Therefore, HL-LUC products could effectively identify terraces, orchards, and other land use types with a high accuracy.
(2)
Through a comprehensive analysis of the time series of land use types, the land use transition matrix, and the spatial distribution of area change rates, this study delved into the spatiotemporal evolution patterns of the LUC types in the Helong Region from 1986 to 2020. The results revealed a notable increase in the areas of forest and grassland. The growth of forest areas primarily stemmed from grassland and cropland conversion, while the expansion of grasslands was mainly attributed to cropland conversion. This was largely due to the implementation of the GGP and the NFCP. Additionally, the rapid increase in impervious surfaces (2038.19 km2) was primarily attributed to conversions from cropland, reflecting the implementation of the Western Development Strategy and the Rise of Central China policy.
(3)
Through a spatiotemporal analysis of land use transfer, we observed that the total area of terraces and orchards in the study area increased by 1334.14 km2 from 1986 to 2020. This growth was primarily due to the conversion of forest lands, grasslands, and croplands into terraces and orchards. From the late 1970s to the late 1990s, the Loess Plateau implemented a policy of constructing terraces on gentle slopes. Subsequently, the GGP was implemented, and the second phase, which started in 2014, encouraged farmers to develop economic forestry, such as planting fruit trees. These policies significantly promoted the growth of terraces and orchards during 1985–1990 and 2015–2020. The expansion of terraces and orchards not only improved the ecological environment but also enhanced economic efficiency.
The above results provide scientific and data support for ecological conservation, land resource management, and future sustainable utilization policies in the Loess Plateau region under changes in land use types such as terraces and orchards.

Author Contributions

Y.C. was responsible for proposing the original idea and providing technical guidance; J.L. was responsible for data compilation, processing, computation, and writing; Y.G. and M.W. were responsible for data curation; Y.Z. was responsible for reviewing and editing. All authors have read and agreed to the published version of the manuscript.

Funding

This study was supported by the Natural Science Foundation of China (Grant No. U2243227).

Data Availability Statement

The raw data supporting the conclusions of this article will be made available by the authors on request.

Conflicts of Interest

The authors declare no conflicts of interest regarding the publication of this paper.

References

  1. Liu, H.; Gong, P.; Wang, J.; Clinton, N.; Bai, Y.; Liang, S. Annual Dynamics of Global Land Cover and Its Long-Term Changes from 1982 to 2015. Earth Syst. Sci. Data 2020, 12, 1217–1243. [Google Scholar] [CrossRef]
  2. Li, J.; Chen, Y.; Zhu, Y.; Liu, J. Study of Flood Simulation in Small and Medium-Sized Basins Based on the Liuxihe Model. Sustainability 2023, 15, 11225. [Google Scholar] [CrossRef]
  3. Zhu, Y.; Chen, Y.; Zhao, Y.; Zhou, F.; Xu, S. Application and Research of Liuxihe Model in the Simulation of Inflow Flood at Zaoshi Reservoir. Sustainability 2023, 15, 9857. [Google Scholar] [CrossRef]
  4. Zhao, Y.; Chen, Y.; Zhu, Y.; Xu, S. Evaluating the Feasibility of the Liuxihe Model for Forecasting Inflow Flood to the Fengshuba Reservoir. Water 2023, 15, 1048. [Google Scholar] [CrossRef]
  5. Lambin, E.F.; Meyfroidt, P. Global Land Use Change, Economic Globalization, and the Looming Land Scarcity. Proc. Natl. Acad. Sci. USA 2011, 108, 3465–3472. [Google Scholar] [CrossRef]
  6. Jones, K.R.; Venter, O.; Fuller, R.A.; Allan, J.R.; Maxwell, S.L.; Negret, P.J.; Watson, J.E.M. One-Third of Global Protected Land Is under Intense Human Pressure. Science 2018, 360, 788–791. [Google Scholar] [CrossRef]
  7. Marques, A.; Martins, I.S.; Kastner, T.; Plutzar, C.; Theurl, M.C.; Eisenmenger, N.; Huijbregts, M.A.J.; Wood, R.; Stadler, K.; Bruckner, M.; et al. Increasing Impacts of Land Use on Biodiversity and Carbon Sequestration Driven by Population and Economic Growth. Nat. Ecol. Evol. 2019, 3, 628–637. [Google Scholar] [CrossRef] [PubMed]
  8. Han, X.; Tsunekawa, A.; Tsubo, M.; Li, S. Effects of Land-Cover Type and Topography on Soil Organic Carbon Storage on Northern Loess Plateau, China. Acta Agric. Scand. Sect. B—Soil Plant Sci. 2010, 60, 326–334. [Google Scholar] [CrossRef]
  9. Douangsavanh, L.; Polthanee, A.; Katawatin, R. Food Security of Shifting Cultivation Systems: Case Studies from Luang Prabang and Oudomxay Provinces, Lao PDR. J. Mt. Sci. 2006, 3, 48–57. [Google Scholar] [CrossRef]
  10. Tang, G.; Ge, S.; Li, F.; Zhou, J. Review of Digital Elevation Model (DEM) Based Research on China Loess Plateau. J. Mt. Sci. 2005, 2, 265–270. [Google Scholar] [CrossRef]
  11. Xu, Y.; Yang, B.; Tang, Q.; Liu, G.; Liu, P. Analysis of Comprehensive Benefits of Transforming Slope Farmland to Terraces on the Loess Plateau: A Case Study of the Yangou Watershed in Northern Shaanxi Province, China. J. Mt. Sci. 2011, 8, 448–457. [Google Scholar] [CrossRef]
  12. Gong, P.; Wang, J.; Yu, L.; Zhao, Y.; Zhao, Y.; Liang, L.; Niu, Z.; Huang, X.; Fu, H.; Liu, S.; et al. Finer Resolution Observation and Monitoring of Global Land Cover: First Mapping Results with Landsat TM and ETM+ Data. Int. J. Remote Sens. 2013, 34, 2607–2654. [Google Scholar] [CrossRef]
  13. Chen, J.; Chen, J.; Liao, A.; Cao, X.; Chen, L.; Chen, X.; He, C.; Han, G.; Peng, S.; Lu, M.; et al. Global Land Cover Mapping at 30 m Resolution: A POK-Based Operational Approach. ISPRS J. Photogramm. Remote Sens. 2015, 103, 7–27. [Google Scholar] [CrossRef]
  14. Zhang, X.; Liu, L.; Chen, X.; Gao, Y.; Xie, S.; Mi, J. GLC_FCS30: Global Land-Cover Product with Fine Classification System at 30 m Using Time-Series Landsat Imagery. Earth Syst. Sci. Data 2021, 13, 2753–2776. [Google Scholar] [CrossRef]
  15. Xu, Y.; An, X.; Yang, B.; Liu, P. Synthetic analyses of effects of famland terracing on the Loess Plateau:A case study of Yangou Watershed. Sci. Soil Water Conserv. 2010, 8, 1–5+12. [Google Scholar] [CrossRef]
  16. Hill, R.D.; Peart, M.R. Land Use, Runoff, Erosion and Their Control: A Review for Southern China. Hydrol. Process. 1998, 12, 2029–2042. [Google Scholar] [CrossRef]
  17. Gao, P.; Deng, J.; Chai, X.; Mu, X.; Zhao, G.; Shao, H.; Sun, W. Dynamic Sediment Discharge in the Hekou–Longmen Region of Yellow River and Soil and Water Conservation Implications. Sci. Total Environ. 2017, 578, 56–66. [Google Scholar] [CrossRef]
  18. Kumar, L.; Mutanga, O. Google Earth Engine Applications Since Inception: Usage, Trends, and Potential. Remote Sens. 2018, 10, 1509. [Google Scholar] [CrossRef]
  19. Perilla, G.A.; Mas, J.-F. Google Earth Engine—GEE: A Powerful Tool Linking the Potential of Massive Data and the Efficiency of Cloud Processing. Investig. Geográficas 2020, n.101, e59929. [Google Scholar] [CrossRef]
  20. Goward, S.N.; Williams, D.L.; Arvidson, T.; Rocchio, L.E.P.; Irons, J.R.; Russell, C.A.; Johnston, S.S. Landsat’s Enduring Legacy: Pioneering Global Land Observations from Space. Photogramm. Eng. Remote Sens. 2022, 88, 357–358. [Google Scholar] [CrossRef]
  21. Masek, J.G.; Wulder, M.A.; Markham, B.; McCorkel, J.; Crawford, C.J.; Storey, J.; Jenstrom, D.T. Landsat 9: Empowering Open Science and Applications through Continuity. Remote Sens. Environ. 2020, 248, 111968. [Google Scholar] [CrossRef]
  22. Gorelick, N.; Hancher, M.; Dixon, M.; Ilyushchenko, S.; Thau, D.; Moore, R. Google Earth Engine: Planetary-Scale Geospatial Analysis for Everyone. Remote Sens. Environ. 2017, 202, 18–27. [Google Scholar] [CrossRef]
  23. Rabus, B.; Eineder, M.; Roth, A.; Bamler, R. The Shuttle Radar Topography Mission—A New Class of Digital Elevation Models Acquired by Spaceborne Radar. ISPRS J. Photogramm. Remote Sens. 2003, 57, 241–262. [Google Scholar] [CrossRef]
  24. Cao, B.; Yu, L.; Naipal, V.; Ciais, P.; Li, W.; Zhao, Y.; Wei, W.; Chen, D.; Liu, Z.; Gong, P. A 30 m Terrace Mapping in China Using Landsat 8 Imagery and Digital Elevation Model Based on the Google Earth Engine. Earth Syst. Sci. Data 2021, 13, 2437–2456. [Google Scholar] [CrossRef]
  25. Clausi, D.A.; Jernigan, M.E. A Fast Method to Determine Co-Occurrence Texture Features. IEEE Trans. Geosci. Remote Sens. 1998, 36, 298–300. [Google Scholar] [CrossRef]
  26. Zhang, Y.; Yang, H.; Xin, Z.; Lu, L. Extraction of Small Watershed Terraces in the Hilly Areas of LoessPlateau Through UAV Images with Object-oriented Approach. J. Soil Water Conserv. 2023, 37, 139–146. [Google Scholar] [CrossRef]
  27. Teluguntla, P.; Thenkabail, P.S.; Oliphant, A.; Xiong, J.; Gumma, M.K.; Congalton, R.G.; Yadav, K.; Huete, A. A 30-m Landsat-Derived Cropland Extent Product of Australia and China Using Random Forest Machine Learning Algorithm on Google Earth Engine Cloud Computing Platform. ISPRS J. Photogramm. Remote Sens. 2018, 144, 325–340. [Google Scholar] [CrossRef]
  28. KIM, K.E. Adaptive Majority Filtering for Contextual Classification of Remote Sensing Data. Int. J. Remote Sens. 1996, 17, 1083–1087. [Google Scholar] [CrossRef]
  29. Ma, H.; Gao, X.; Gu, X. Random Forest Classification of Landsat 8 Imagery for the Complex Terrain Area based on the Combination of Spectral, Topographic and Texture Information. J. Geo-Inf. Sci. 2019, 21, 359–371. [Google Scholar] [CrossRef]
  30. Flood, N. Seasonal Composite Landsat TM/ETM+ Images Using the Medoid (a Multi-Dimensional Median). Remote Sens. 2013, 5, 6481–6500. [Google Scholar] [CrossRef]
  31. Azzari, G.; Lobell, D.B. Landsat-Based Classification in the Cloud: An Opportunity for a Paradigm Shift in Land Cover Monitoring. Remote Sens. Environ. 2017, 202, 64–74. [Google Scholar] [CrossRef]
  32. Louwagie, G.; Gay, S.H.; Sammeth, F.; Ratinger, T. The Potential of European Union Policies to Address Soil Degradation in Agriculture. Land Degrad. Dev. 2011, 22, 5–17. [Google Scholar] [CrossRef]
  33. Li, X.H.; Yang, J.; Zhao, C.Y.; Wang, B. Runoff and sediment from orchard terraces in southeastern China. Land Degrad. Dev. 2014, 25, 184–192. [Google Scholar] [CrossRef]
  34. Hua, G. Study on Coordination and Quantification of Ecological Protection and High Quality Development in the Yellow River Basin. IOP Conf. Ser. Earth Environ. Sci. 2021, 647, 012168. [Google Scholar] [CrossRef]
  35. Sheykhmousa, M.; Mahdianpari, M.; Ghanbari, H.; Mohammadimanesh, F.; Ghamisi, P.; Homayouni, S. Support Vector Machine Versus Random Forest for Remote Sensing Image Classification: A Meta-Analysis and Systematic Review. IEEE J. Sel. Top. Appl. Earth Obs. Remote Sens. 2020, 13, 6308–6325. [Google Scholar] [CrossRef]
  36. Tassi, A.; Vizzari, M. Object-Oriented LULC Classification in Google Earth Engine Combining SNIC, GLCM, and Machine Learning Algorithms. Remote Sens. 2020, 12, 3776. [Google Scholar] [CrossRef]
  37. Rong, W.; Li, Z.; Zhang, W.; Sun, L. An Improved Canny Edge Detection Algorithm. In Proceedings of the 2014 IEEE International Conference on Mechatronics and Automation, Tianjin, China, 3–6 August 2014; pp. 577–582. [Google Scholar]
  38. Gonenc, A.; Ozerdem, M.S.; Acar, E. Comparison of NDVI and RVI Vegetation Indices Using Satellite Images. In Proceedings of the 2019 8th International Conference on Agro-Geoinformatics (Agro-Geoinformatics), Istanbul, Turkey, 16–19 July 2019; pp. 1–4. [Google Scholar]
  39. Huang, S.; Tang, L.; Hupy, J.P.; Wang, Y.; Shao, G. A Commentary Review on the Use of Normalized Difference Vegetation Index (NDVI) in the Era of Popular Remote Sensing. J. For. Res. 2021, 32, 1–6. [Google Scholar] [CrossRef]
  40. Galvão, L.S.; dos Santos, J.R.; Roberts, D.A.; Breunig, F.M.; Toomey, M.; de Moura, Y.M. On Intra-Annual EVI Variability in the Dry Season of Tropical Forest: A Case Study with MODIS and Hyperspectral Data. Remote Sens. Environ. 2011, 115, 2350–2359. [Google Scholar] [CrossRef]
  41. Gao, B. NDWI—A Normalized Difference Water Index for Remote Sensing of Vegetation Liquid Water from Space. Remote Sens. Environ. 1996, 58, 257–266. [Google Scholar] [CrossRef]
  42. Zhang, Y.; Odeh, I.O.A.; Han, C. Bi-Temporal Characterization of Land Surface Temperature in Relation to Impervious Surface Area, NDVI and NDBI, Using a Sub-Pixel Image Analysis. Int. J. Appl. Earth Obs. Geoinf. 2009, 11, 256–264. [Google Scholar] [CrossRef]
  43. Ma, X.; Li, C.; Tong, X.; Liu, S. A New Fusion Approach for Extracting Urban Built-up Areas from Multisource Remotely Sensed Data. Remote Sens. 2019, 11, 2516. [Google Scholar] [CrossRef]
  44. Xiu, L.; Yan, C.; Li, X.; Qian, D.; Feng, K. Monitoring the Response of Vegetation Dynamics to Ecological Engineering in the Mu Us Sandy Land of China from 1982 to 2014. Environ. Monit. Assess. 2018, 190, 543. [Google Scholar] [CrossRef] [PubMed]
  45. Huang, L.; Zhou, W.; Li, H.; Zhou, F.; Yang, H. Effect of land use/cover change on grassland NPP in grassland ccosystem of Ordos City. Bull. Soil Water Conserv. 2018, 38, 46–52+59+2. [Google Scholar] [CrossRef]
  46. Guo, S. Research on the Coupled Coordination of Ecosystem Health and Urbanization in the Yellow River Basin. Ph.D. Thesis, Jilin University, Changchun, China, 2023. [Google Scholar]
  47. Gao, Y.; Zhang, Z.; Wei, S.; Wang, Z. The Impact of Urbanization on Food Security in China: An Empirical Analysis Based on Provincial and Regional Panel Data. Resour. Sci. 2019, 41, 1462–1474. [Google Scholar]
  48. Chen, X.; Li, R.; Zhao, J. China’s food security policy: Evolutionary trajectory, internal logic and strategic orientation. Economist 2020, 117–128. [Google Scholar] [CrossRef]
  49. Wang, J.; Liu, J.; Song, Z.; Huang, L.; Fang, Y.; Li, Z. Strategies of ecosystem protection and territory spatial utilization for high-quality development in the Yellow River Basin. J. Nat. Resour. 2022, 37, 2930–2945. [Google Scholar] [CrossRef]
  50. Qin, Y.; Yang, D.; Gao, B.; Wang, T.; Chen, J.; Chen, Y.; Wang, Y.; Zheng, G. Impacts of Climate Warming on the Frozen Ground and Eco-Hydrology in the Yellow River Source Region, China. Sci. Total Environ. 2017, 605–606, 830–841. [Google Scholar] [CrossRef]
  51. Song, X.; Yang, G.; Yan, C.; Duan, H.; Liu, G.; Zhu, Y. Driving Forces behind Land Use and Cover Change in the Qinghai-Tibetan Plateau: A Case Study of the Source Region of the Yellow River, Qinghai Province, China. Environ. Earth Sci. 2009, 59, 793–801. [Google Scholar] [CrossRef]
  52. Li, R. Review and Enlightenments of Soil and Water Conservation on Loess Plateau in Past 70 Years. Bull. Soil Water Conserv. 2019, 39, 298–301. [Google Scholar] [CrossRef]
  53. Pi, H.; Zhang, M.; Xia, J. Ecological Compensation Standards of the Grain for Green Project: Based on the survey data of farmers in Jingyuan County of Ningxia. Issues For. Econ. 2018, 38, 39–44+104. [Google Scholar] [CrossRef]
  54. Li, J.; Peng, S.; Li, Z. Detecting and Attributing Vegetation Changes on China’s Loess Plateau. Agric. For. Meteorol. 2017, 247, 260–270. [Google Scholar] [CrossRef]
  55. Shi, S.; Zhang, M.; Ru, K. Research on Agroforestry-Grassland Composite Model for Slope Terraces in Loess Hills Gully Area. Hanxi Waterre Sources 2000, 31–32. [Google Scholar]
  56. Li, W.; MacBean, N.; Ciais, P.; Defourny, P.; Lamarche, C.; Bontemps, S.; Houghton, R.A.; Peng, S. Gross and Net Land Cover Changes in the Main Plant Functional Types Derived from the Annual ESA CCI Land Cover Maps (1992–2015). Earth Syst. Sci. Data 2018, 10, 219–234. [Google Scholar] [CrossRef]
Remotesensing 16 03738 g001
Figure 1. Location of Helong Region.
Figure 1. Location of Helong Region.
Remotesensing 16 03738 g001
Remotesensing 16 03738 g002
Figure 2. Workflow of this study.
Figure 2. Workflow of this study.
Remotesensing 16 03738 g002
Remotesensing 16 03738 g003
Figure 3. Number of Landsat scenes used in the GEE image synthesis.
Figure 3. Number of Landsat scenes used in the GEE image synthesis.
Remotesensing 16 03738 g003
Remotesensing 16 03738 g004
Figure 4. Anomaly screening and repair of Landsat data (a) filling missing data (b) repairing Landsat 7 image gaps.
Figure 4. Anomaly screening and repair of Landsat data (a) filling missing data (b) repairing Landsat 7 image gaps.
Remotesensing 16 03738 g004
Remotesensing 16 03738 g005
Figure 5. Distribution of the training sample polygons at different times and in different categories.
Figure 5. Distribution of the training sample polygons at different times and in different categories.
Remotesensing 16 03738 g005
Remotesensing 16 03738 g006
Figure 6. Validation of the spatial distribution of the sample set.
Figure 6. Validation of the spatial distribution of the sample set.
Remotesensing 16 03738 g006
Remotesensing 16 03738 g007
Figure 7. Temporal distribution of area changes for various LUC types in the Helong Region (the proportion of change on the right axis is relative to the area change ratio with respect to the base year (1986)).
Figure 7. Temporal distribution of area changes for various LUC types in the Helong Region (the proportion of change on the right axis is relative to the area change ratio with respect to the base year (1986)).
Remotesensing 16 03738 g007
Remotesensing 16 03738 g008
Figure 8. Spatial distribution of LUC change rates in the Helong Region. (Through linear regression, we calculated the area ratio change rates for each category within each grid (0.1°) from 1986 to 2020, and the spatial distributions of the area ratio changes that were found to be significant (p < 0.05) are displayed. In the figure, gray grids represent results with insignificant changes or changes below 0.1% per year (−0.1 to 0.1)).
Figure 8. Spatial distribution of LUC change rates in the Helong Region. (Through linear regression, we calculated the area ratio change rates for each category within each grid (0.1°) from 1986 to 2020, and the spatial distributions of the area ratio changes that were found to be significant (p < 0.05) are displayed. In the figure, gray grids represent results with insignificant changes or changes below 0.1% per year (−0.1 to 0.1)).
Remotesensing 16 03738 g008
Remotesensing 16 03738 g009
Figure 9. Spatial and temporal distributions of forests, grasslands, and croplands transformed into terraces and orchards.
Figure 9. Spatial and temporal distributions of forests, grasslands, and croplands transformed into terraces and orchards.
Remotesensing 16 03738 g009
Remotesensing 16 03738 g010
Figure 10. Heat map of the transitions of LUC types in two adjacent periods.
Figure 10. Heat map of the transitions of LUC types in two adjacent periods.
Remotesensing 16 03738 g010
Remotesensing 16 03738 g011
Figure 11. Comparison of the accuracy of HL-LUC, FROM-GLC, CLC-FCS30, and ESA CCI-LC.
Figure 11. Comparison of the accuracy of HL-LUC, FROM-GLC, CLC-FCS30, and ESA CCI-LC.
Remotesensing 16 03738 g011
Remotesensing 16 03738 g012
Figure 12. Comparison of HL-LUC-2015 with the three other datasets.
Figure 12. Comparison of HL-LUC-2015 with the three other datasets.
Remotesensing 16 03738 g012
Table 1. Landsat datasets used in this study.
Table 1. Landsat datasets used in this study.
PlatformSensorDatasetSpatial Resolution (m)Acquisition Year
Landsat 5TMLandsat Collection1 Tier1 surface reflectance301986, 1990, 1995, 2000, 2005
Landsat 7ETM+Landsat Collection1 Tier1 surface reflectance302010
Landsat 8OLILandsat Collection1 Tier1 surface reflectance 302015, 2020
Table 2. LUC classification system of this study.
Table 2. LUC classification system of this study.
CodeCategoryMeaning and Interpretation of Symbols for Various Land Use and cover TypesRemote Sensing Image
1ForestForest with a canopy density exceeding 30%. Forest in the study area is primarily distributed in the southern part of the Yanhe River, in Shiwangchuan, on the Lvliang Mountain, and near the eastern boundary. There are also scattered distributions in the moist and well-watered Liangmao gullies. In false-color composite images, colors appear as deep red, bright red, and brownish-red, with hues or directions consistent with the terrain orientation. Moreover, there is a slight difference in hue between the shady and sunny slopes of mountains. In particular, the shrub forest at the edges of the forested areas in this study are classified as forest, and they have a lighter hue than the forested areas.Remotesensing 16 03738 i001
2GrasslandVarious types of grassland with vegetation coverage above 20%. Mainly distributed in plains, on steeper slopes, and within valleys. Showing as brown-red, dark red, light red, etc.Remotesensing 16 03738 i002
3Impervious SurfaceThe land surface formed by human construction activities, including various types of residential areas such as towns, industrial and mining areas, and transportation roads. It is primarily characterized by a grayish-white color, with clear boundaries, regular shapes, and rough textures.Remotesensing 16 03738 i003
4WaterIncluding rivers, lakes, reservoirs, and ponds, located in open channels and ravines. Mainly dark blue and black, as well as blue and light blue, with a uniform and smooth texture.Remotesensing 16 03738 i004
5Bare landAt least 60% of the area is low-vegetation land with less than 10% vegetation coverage, such as bare rocks and sandy land. Bare rocks are primarily located on both sides of the Yellow River Basin. They appear as black or brown in color, with irregular shapes. Sandy land is mainly located in the northwest of the study area, appearing as gray or brown in color. Remotesensing 16 03738 i005
6CroplandLand used for cultivating crops, including paddy fields, vegetable plots, pasturelands, and greenhouse land. It is distributed on slopes and flat terrain, appearing in shades of magenta, light green, and dark red.Remotesensing 16 03738 i006
7Terraces and OrchardsA mixture of artificial vegetation grown for agricultural purposes with a slope greater than 3°, mainly in the loess table district, including terraces and orchards. They exhibit dark red, light green, grayish-white, and blue-yellow colors. Remotesensing 16 03738 i007
Table 3. Results of testing different classification methods.
Table 3. Results of testing different classification methods.
Classification Method2010
Overall Accuracy (%)Kappa Coefficient (%)
Classification and Regression Tree (CART)86.0780.86
Random Forest90.1686.37
Naive Bayes48.5137.72
Minimum Distance43.1532.40
k-Nearest Neighbor (KNN)86.4881.36
Table 4. Auxiliary features and factors involved in classification.
Table 4. Auxiliary features and factors involved in classification.
Feature AttributeFeature NameFeature Description
Spectral features6 Landsat spectral bandsBlue, green, red, near-infrared, shortwave infrared 1, and shortwave infrared 2
Texture features (Blue Nir)ASMAngular Second Moment (ASM) expresses the texture fineness and the uniformity of gray level distribution.
ContrastContrast expresses the texture depth and image sharpness
CorrCorrelation expresses the consistency of texture
IdmInverse Difference Moment (IDM) expresses the homogeneity of texture
EntEntropy expresses the non-uniformity or complexity of texture.
DissDissimilarity expresses the degree of difference in texture within an image
CannyCanny edge detection helps identify boundaries of different land cover types
Vegetation index featuresNDVI N D V I = ( N I R − R e d ) / ( N I R + R e d )
NDMI N D M I = ( N I R − S W I R ) / ( N I R + S W I R )
NDBI N D B I = ( S W I R − N I R ) / ( S W I R + N I R )
EVI E V I = 2.5 ( N I R − R e d ) / ( N I R + 6 × R e d − 7.5 × B l u e + 1 )
MNDWI ( G r e e n − S W I R ) / ( G r e e n + S W I R )
BSI B S I =   [ ( S W I R + R e d ) − ( N I R + B l u e ) ]   [ ( S W I R + R e d ) + ( N I R + B l u e ) ]
Terrain factor featureselevationElevation refers to the height of a location above a reference point.
slopeSlope refers to the steepness or incline of the terrain
aspectAspect refers to the orientation or direction that a slope faces on the terrain surface
TopDiversityTopDiversity refers to the variability in terrain elevation within a specific area, indicating the degree of variation in elevation across the landscape
Table 5. Classification accuracy and F1 scores for the four periods based on independent stratified random sampling validation points.
Table 5. Classification accuracy and F1 scores for the four periods based on independent stratified random sampling validation points.
YearOA/%Kappa Coefficient/%ForestGrasslandImpervious SurfaceWaterBare LandCroplandTerraces and Orchards
200586.0876.140.910.910.80.910.910.680.69
201085.1172.890.870.920.690.910.910.610.68
201589.3279.820.950.950.840.80.830.670.65
202089.6478.890.890.940.750.770.940.750.76
Table 6. LUC transition matrix from 1986 to 2020 (km2).
Table 6. LUC transition matrix from 1986 to 2020 (km2).
1986–2020ForestGrasslandImpervious SurfaceWaterBare LandCroplandTerraces and Orchards
Forest9415.241321.232.671.760.05135.9813.68
Grassland2631.8450,413.96626.4757.51388.462698.812263.94
Impervious surface0.7170.9274.308.371.3247.972.71
Water1.4951.2556.72341.572.6592.3310.85
Bare land0.466750.35275.9723.901613.481641.38262.50
Cropland550.8013,469.17857.0375.87247.546951.091001.67
Terraces and Orchards30.032167.5611.491.543.73275.911116.77
Table 7. LUC datasets for comparison.
Table 7. LUC datasets for comparison.
DatasetTime RangeSpatial ResolutionData SourceData Provider
HL-LUC1986-2020 (every 4 or 5 years)30 mLandsatThis study
FROM-GLC2010, 2015, 201730 mLandsat[12]
GLC-FCS301985–202230 mLandsat[14]
ESA-CCI-LCYearly from 1992 to 2020 300 mPROBA-V, Sentinel-3, and other multi-source data[56]
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

Li, J.; Chen, Y.; Gu, Y.; Wang, M.; Zhao, Y. Remote Sensing Mapping and Analysis of Spatiotemporal Patterns of Land Use and Cover Change in the Helong Region of the Loess Plateau Region (1986–2020). Remote Sens. 2024, 16, 3738. https://doi.org/10.3390/rs16193738

AMA Style

Li J, Chen Y, Gu Y, Wang M, Zhao Y. Remote Sensing Mapping and Analysis of Spatiotemporal Patterns of Land Use and Cover Change in the Helong Region of the Loess Plateau Region (1986–2020). Remote Sensing. 2024; 16(19):3738. https://doi.org/10.3390/rs16193738

Chicago/Turabian Style

Li, Jingyu, Yangbo Chen, Yu Gu, Meiying Wang, and Yanjun Zhao. 2024. "Remote Sensing Mapping and Analysis of Spatiotemporal Patterns of Land Use and Cover Change in the Helong Region of the Loess Plateau Region (1986–2020)" Remote Sensing 16, no. 19: 3738. https://doi.org/10.3390/rs16193738

APA Style

Li, J., Chen, Y., Gu, Y., Wang, M., & Zhao, Y. (2024). Remote Sensing Mapping and Analysis of Spatiotemporal Patterns of Land Use and Cover Change in the Helong Region of the Loess Plateau Region (1986–2020). Remote Sensing, 16(19), 3738. https://doi.org/10.3390/rs16193738

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