Next Article in Journal
Using Naturalness for Assessing the Impact of Forestry and Protection on the Quality of Ecosystems in Life Cycle Assessment
Previous Article in Journal
Financial Literacy and Alcohol Drinking Behavior: Evidence from Japan
 
 
Search for Articles:
Title / Keyword
Author / Affiliation / Email
Journal
Article Type
 
 
Advanced Search
 
Section
Special Issue
Volume
Issue
Number
Page
 
Journals
Sustainability
Volume 13
Issue 16
10.3390/su13168857
sustainability-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

Uncertainty Analysis of Multisource Land Cover Products in China

by
Longhao Wang
1 and
Jiaxin Jin
1,2,*
1
College of Hydrology and Water Resources, Hohai University, Nanjing 210024, China
2
National Earth System Science Data Center, National Science & Technology Infrastructure of China, Beijing 100101, China
*
Author to whom correspondence should be addressed.
Sustainability 2021, 13(16), 8857; https://doi.org/10.3390/su13168857
Submission received: 7 July 2021 / Revised: 4 August 2021 / Accepted: 6 August 2021 / Published: 8 August 2021
Browse Figures
Versions Notes

Abstract

:
Satellite-based land cover products play a crucial role in sustainability. There are several types of land cover products, such as qualitative products with discrete classes, semiquantitative products with several classes at a predetermined ratio, and quantitative products with land cover fractions. The proportions of land cover types in the grids with coarse resolution should be considered when used at the regional scale (e.g., modeling and remote sensing inversion). However, uncertainty, which varies with spatial distribution and resolution, needs to be studied further. This study used MCD12, ESA CCI, and MEaSURES VCF land cover data as indicators of qualitative, semiquantitative, and quantitative products, respectively, to explore the uncertainty of multisource land cover data. The methods of maximum area aggregation, deviation analysis, and least squares regression were used to investigate spatiotemporal changes in forests and nontree vegetation at diverse pixel resolutions across China. The results showed that the average difference in forest coverage for the three products was 8%, and the average deviation was 11.2%. For forest cover, the VCF and ESA CCI exhibited high consistency. For nontree vegetation, the ESA CCI and MODIS exhibited the lowest differences. The overall uncertainty in the temporal and spatial changes of the three products was relatively small, but there were significant differences in local areas (e.g., southeastern hills). Notably, as the spatial resolution decreased, the three products’ uncertainty decreased, and the resolution of 0.1° was the inflection point of consistency.

1. Introduction

Multisource land cover products are the basis for studying the Earth’s surface and atmosphere [1] and have had a profound impact on the survival and development of humanity. Reasonably classified and accurate land cover products have been essential for studying Earth surface processes and information [2]. Land cover products play a crucial role in studying terrestrial surface processes, ecosystem assessments, environmental modeling, geographic national condition monitoring, remote sensing inversion, etc. [3,4]. In short, land cover products are the basis for sustainability research [5].
In recent decades, the rapid development of remote sensing technology and algorithm improvements have resulted in many land cover products [6,7]. Twenty-three categories of global land cover products and 41 categories of regional land cover products were summarized by Grekousis [8]. Among them, the most widely used global land cover products include IGBP-DISCover from the US Geological Survey, UMD from the University of Maryland, CCI-LC from the European Space Agency, and FROM-GLC from Tsinghua University in China, etc. Due to the diverse types of data, different data sources, different classification methods, and different systems, the quality of these products is not the same, and the differences are large [9]. High-resolution products have high accuracy at regional and global scales [10] and have an extensive imaging range, low cost, and wide application [11,12,13]. When research has been conducted by scholars, products of a certain resolution have been selected, and the uncertainty brought about by spatial resolution has needed to be considered [14]. The types of land cover products can be divided into three categories: Qualitative products, semiquantitative products, and quantitative products. Qualitative products are represented by MCD12 products with discrete classes [15]; semiquantitative products are represented by ESA CCI products with several classes at a predetermined ratio [16]; and quantitative products are represented by VCF products with land cover fractions. The percentage of land cover types for each pixel of VCF products has been recorded, and the area of land cover types can be calculated quantitatively, which is a quantitative product. Land cover products act as an essential foundation for scholars to carry out remote sensing inversion and model simulation. Land cover products have essential applications when subgrid schemes, parameter assignment, and parameterization scheme determination are considered—the uncertainty of different products in these processes needed to be explored. For example, in the leaf area index (LAI) inversion, land cover data are needed. Due to the significant differences in multisource land cover products’ spatial distribution and accuracy [17], different land cover data had greater accuracy in influencing LAI inversion. Researchers have pointed out that the main land cover products for China’s land area cannot meet the needs of land surface process simulation, and there is great uncertainty [18]. With different spatial resolutions of land cover data at regional and global scales, gross primary productivity (GPP) model simulation results have considerable differences [19]. Land cover is the most significant uncertainty factor in estimating carbon storage and release in terrestrial ecosystems [20]. Due to the limited spatial resolution of remote sensing images, the existence of subpixels needs to be considered. That is, several different land cover types may be included in a single pixel [21], which would bring greater uncertainty to the research results [22,23]. This uncertainty varies with spatial resolution [24]. Therefore, how the uncertainty of multisource land cover products changes with resolution urgently needs to be quantitatively explored. It is of great significance for scholars to carry out remote sensing inversion and numerical simulation.
Some scholars have realized these issues, and the uncertainty of land cover products at the regional scale has begun to be explored. Shao et al. studied the consistency of GlobeLand30 and FROMGLC in the Beijing-Tianjin-Hebei region of China, and their differences in land cover types and landscape pattern indexes were compared and analyzed [25]. A single resolution was selected in their study, only two sets of products were selected, and the area differences of the land cover types in the two types of products were directly compared, without focusing on spatial differences between the products. Bayer et al. explored five sets of land use models and found that different regions were very inconsistent and had high uncertainty [26]. A method to resolve these differences was proposed using high-resolution land cover datasets [27]. Bai et al. studied the consistency of five global land cover datasets in China. The five datasets were compared in pairs. The results showed that the forest classification differed significantly, and the consistency was the lowest [28]. These studies mainly focused on comparisons between qualitative products, they rarely involved quantitative products and semiquantitative products, and they did not involve the impact of resolution on the uncertainty of land cover products. The uniqueness of this study was that it considered three completely different types of land cover products, including qualitative products, semiquantitative products, and quantitative products. It explored the impact of resolution on the uncertainty of land cover products.
Land cover products have a wide range of sources and different classification methods. There may be differences when the same area is described by different products, and there is high uncertainty in the land cover types in areas with large differences. Before these land cover products with different resolutions are used by researchers, the uncertainty of various products must be evaluated. That is, the difference in the area and the uncertainty of the change must be determined with different spatial resolutions [29]. Therefore, this study used three types of land cover data, and the methods of maximum area aggregation, deviation analysis, and least squares regression were used to analyze the difference in the areas and spatiotemporal changes of forest and nontree vegetation of MCD12, ESA CCI, and VCF land cover classification products. China was chosen as the study area due to its large variations in land cover, the junction of vegetation cover types, and a large number of mixed pixels, making it an ideal study area. When different products and resolutions are selected by scholars for research, our research helps reduce the uncertainty of multisource land cover products.

2. Materials and Methods

2.1. Land Cover Data

The multisource land cover products used in this study were representatively qualitative, semiquantitative, and quantitative land cover products. The source, resolution, and land cover classification systems, and classification methods of the three products were different. The parameters are shown in Table 1.
The MODIS land cover products (MCD12) classification system has 17 categories. The overall evaluation accuracy was 78.3%, and the spatial resolution was 1 km. It is a commonly used land cover data product of the IGBP system [30]. The product was obtained based on the artificial neural network and decision tree classification method, which can be downloaded at https://modis.gsfc.nasa.gov/data/ (accessed on 22 October 2014). It was a qualitative product with discrete classes.
ESA CCI land cover data divided the land surface into 37 land cover categories according to the United Nations Land Cover Classification System (UN-LCCS), with an overall accuracy of 71.5%. The data came from http://maps.elie.ucl.ac.be/CCI/viewer/download/ (accessed on 22 December 2019) with a spatial resolution of 300 m. The data had high user accuracy for paddy fields, evergreen broad-leaved forests, cities, bare land, water bodies, and permanent glaciers [31]. In addition, the proportion of plant functional types (PFTs) in qualitative land cover types was considered in this study. Therefore, the initial data of 37 types of ESA CCI were summarized into 0.5 × 0.5° resolution and then converted into PFT data. The PFTs included forests, shrubs, farmland, and grassland [32]. Combined with the actual situation in the study area, Classes 2 and 1 were calculated through the PFT cross-walking table, as shown in Table 2. The ESA CCI PFT-based land cover products were converted from qualitative pixel types to land cover percentages and from qualitative products to quantitative products, which are semiquantitative products.
MEaSURES vegetation continuous field (VCF) products were divided into three layers: Tree cover, nontree vegetation, and bare ground. VCF products represent a new revolutionary method to characterize plant land cover [33]. Traditional land cover maps assigned each pixel to a single land cover category. VCF products used fractional vegetation coverage (FVC), which was a primary means of measuring global forest cover change [11,34,35]. FVC was defined as the ratio of the vertical projection area of green vegetation on the ground to the total area. The spatial resolution was 5 km, and the data came from https://lpdaac.usgs.gov/productsearch/ (accessed on 19 February 2020). Since the percentage of vegetation coverage was directly recorded by the VCF product, it was a qualitative product.

2.2. Method

Three aspects were included in the uncertainty evaluation of land cover products: Area uncertainty, spatial location uncertainty, and spatiotemporal change uncertainty. Area uncertainty was used to evaluate the quantitative characteristics of land cover products, spatial uncertainty was used to evaluate the spatial distribution of land cover products, and spatiotemporal uncertainty was used to evaluate changes in land cover types over time. The methods used in this study mainly included the following:
(a) Data preprocessing: Before the uncertainty analysis of land cover products, data preprocessing was carried out, including research area tailoring, upscaling transformation, and classification system merging. First, three kinds of land cover products were obtained based on the ArcGIS software, and land cover products with the same boundary were obtained. The maximum area aggregation method was used to unify products with different spatial resolutions to the same resolution for research. It aligned the raster of the output dataset (coarse raster) with the input data raster (fine raster) and assigned the type with the most occurrences in the input raster to the output raster. The three multisource land cover products have different land cover classification systems, they needed to be merged into a new unified and generalized classification system: Forest, nontree vegetation, and bare land.
(b) Area uncertainty analysis: According to the forest and nontree vegetation in the study area from 2001 to 2012, the change in the area of the three products in the past 12 years was calculated. In order to explain the uncertainties and differences among the different land cover products, this study used relative deviation to express the difference of forest and nontree vegetation. The coefficient of deviation formula was used to calculate the coefficient of deviation of the three products [36]. The deviation coefficient D was an effective index to measure the classification accuracy of land cover products.
D x k = ( x k k ¯ − 1 ) × 100
In formula (1), x is the land cover classification product; k is the land cover type of the product; k ¯ is the mean value of the area of the three products of a certain type k ; and D x k is the deviation coefficient of the area of the type k in product x [37]. The mean value (MEAN) and root mean square error (RMSE) were used to calculate the variation in the resolution consistency in the entire study area [38] using Formula (2).
R M S E = 1 m ∑ i = 1 m ( y i − y ^ i ) 2
(c) Spatial uncertainty analysis: The area uncertainty analysis can quantitatively give the difference between different products, but the result was statistical information, and the spatial characteristics were removed. Therefore, this study adopted the method of spatial superposition with three products. The difference was made pixel-by-pixel, and the difference in different products was judged pixel-by-pixel.
(d) Temporal and spatial uncertainty analysis: The changing trend of forest and nontree vegetation from 2001 to 2012 was evaluated, and the entire area was analyzed pixel-by-pixel. The 12-year trend of forest and nontree vegetation coverage of each pixel was analyzed by the least squares method. For example, using a moving window of 18 × 18 pixels (about 0.5°) across the entire area, we calculated the proportion of forest in the window year by year. Then, a linear regression on the forest proportion was performed, and the regression slope was assigned to the window pixels. The temporal and spatial distribution of forest and nontree vegetation changes for different products in the study area were obtained.

3. Results

3.1. Area Uncertainty

A comparison of the forest and nontree vegetation areas of the three land cover products for 2001–2012 was shown in Figure 1, and the area percentage and deviation coefficient of the land cover types are shown in Table 3. In general, the three products were highly consistent in the study area, and nontree vegetation was the main land cover type. Nontree vegetation accounted for approximately 40% of the total area in the study area, and forests accounted for approximately 13–18%.
The uncertainty in the land coverage for the three products was explored. The overall deviation of the range was 2–17%. The uncertainty between the quantitative VCF product and the semiquantitative ESA CCI product was small for the forest area. The difference in the area was 8%, and the deviation range was 4.56 and 12.22%, respectively. When studying the area of nontree vegetation, the uncertainty between the qualitative MODIS product and the semiquantitative ESA CCI product was small, and the consistency was strong. The area difference was 8%, and the deviation ranges were 2.87 and 9.52%, respectively. Semiquantitative products performed better when the area coverage was monitored since semiquantitative products are the conversion of qualitative products to quantitative products, and as a result have both qualitative and quantitative properties. The time series changes in forest cover were monitored. The results showed that between 2001 and 2012, the forest coverage area of MODIS products continued to increase, with an average annual increase of 1.3%. VCF products first increased and then decreased, and they increased overall, with an average annual increase of 0.6%. The increase in ESA CCI products was not obvious, and the total amount remained unchanged. The time series change in nontree vegetation cover was monitored. The results showed that between 2001 and 2012, the nontree vegetation coverage area of MODIS products continued to decrease, with an average annual decrease of 0.3%. VCF products first decreased and then increased, and the total amount remained unchanged. The increase or decrease in ESA CCI products was insignificant, and the total amount remained unchanged. The three products had large uncertainties in the area and time changes, and the opposite conclusions were obtained. The quantitative VCF product was more sensitive to land cover changes, while the semiquantitative ESA CCI product was not sensitive to land cover changes.

3.2. Spatial Uncertainty

The spatial uncertainty of the forest cover of the three products was studied. The percentage of the forest cover of the three products was calculated first, and then the two products were differentiated pixel-by-pixel. The products were divided into three groups with three resolutions, as shown in Figure 2a. The spatial distribution of the forest cover of the three products was relatively consistent but differed significantly in different regions. The North China Plain, Northeast Plain, and Xinjiang Basin in China have good consistency, with a difference in the area of less than 10%. The Greater Khingan Mountains, Southeast Hills, Hengduan Mountains, and Yunnan-Guizhou Plateau had poor consistency and more significant spatial uncertainty, with a difference in the area of more than 20%. The reason for this difference was the influence of mixed pixels in hilly areas. The impact of resolution on the uncertainty between products was explored. The percentage of various pixels varied with the resolution, as shown in Figure 2b. In general, the uncertainty between products decreased as the resolution decreased. Pixels with high uncertainty between products were counted (the difference between products was >40%). When the resolution was reduced from 0.1 to 0.3°, such pixels were reduced by 3% on average, and the uncertainty was reduced. The low-uncertainty pixels were counted (the difference between products was >10%). This type of pixel increased by 5.8% as the resolution decreased. CCI-VCF had the fewest high-uncertainty pixels, accounting for 0.1% of the total pixels, while the low-uncertainty pixels were the most numerous, accounting for 81% of the total pixels. The semiquantitative ESA CCI product had the highest consistency with the qualitative VCF product.
The spatial uncertainty of the nontree vegetation cover of the three products was studied. The percentage of the nontree vegetation cover of the three products was calculated first, and then the two products were differentiated pixel-by-pixel. The products were divided into three groups with three resolutions, as shown in Figure 3a. More significant differences appear in different regions, with great uncertainty. For example, the nontree vegetation of VCF products in the Qinghai-Tibet Plateau was more than 40% different from that of the other two products. The reason is that under the VCF classification system, the percentage of nontree vegetation in the Qinghai-Tibet Plateau was considered to be large, while the other two products showed very little nontree vegetation there. The nontree vegetation coverage of ESA CCI products and MODIS products in the North China Plain was relatively consistent, with an area difference of less than 10%, while the difference between VCF products and these two products in the North China Plain was 10–20%. The percentage of various pixels varied with the resolution, as shown in Figure 3b. In general, the uncertainty between products decreased as the resolution decreased. As the resolution decreased from 0.1 to 0.3°, pixels with high uncertainty were reduced by 2.6% on average, and pixels with low uncertainty increased by 10%. Pixels with a difference greater than 0.2 decreased by 7.8% as the resolution decreased. According to the histogram of the pixel difference, the two products of CCI-MODIS had the fewest high-uncertainty pixels (13%), and the consistency was strong, while the two products of VCF-MODIS had the highest uncertainty pixels (18%), and the consistency was the worst.
In general, the consistency of nontree vegetation cover was worse than that of forest cover since the classification system of nontree vegetation was quite different. The spatial consistency of the three products in the study area was strong, but there were significant differences in local areas. The consistency of plains and basins was better than that of mountains and hills; the consistency of northern China was better than that of southern regions; and the consistency of the inland areas was better than that of the coastal areas. In general, the uncertainty between products decreased as the resolution decreased. When the forest cover was studied, the semiquantitative ESA CCI product and qualitative VCF product had the least uncertainty. When the nontree vegetation cover was studied, the qualitative MODIS product and semiquantitative ESA CCI product had less uncertainty. This conclusion was the same as that obtained when discussing the area uncertainty.
Since the new unified classification system had only three classes, more classes were worthy of investigating. We investigated the spatial uncertainty of the grass cover and crop cover. The products were divided into three groups with three resolutions, as shown in Figure 4. The spatial distribution of the grass and crop cover of the three products was relatively consistent but differed significantly in different regions. The North China Plain and Xinjiang Basin in China have good consistency, with a difference in the area of less than 10%. The Greater Khingan Mountains, Hengduan Mountains, and Yunnan-Guizhou Plateau had poor consistency and greater spatial uncertainty, with a difference in the area of more than 20%. The reason for this difference was the influence of mixed pixels in hilly areas. In general, the uncertainty between products decreased as the resolution decreased.

3.3. Uncertainty of Temporal and Spatial Changes

The uncertainty of the spatial and temporal changes in the forest cover of the three products was studied, the pixel-by-pixel forest coverage of different resolutions from 2001 to 2012 was subjected to least squares regression, and the results are shown in Figure 5a. The forest increased slightly (<0.02/yr). The increase in forest mainly occurred in the northeast plain and inland China, and the loss mainly occurred on the southeast coast. The spatiotemporal changes in the forest were consistent, but the spatiotemporal changes in the forest derived from different products in different regions were quite different, and there was great uncertainty. In the ESA CCI products, there was an increase in forests (<0.02/yr) in China’s Xinjiang Basin and Qinghai-Tibet Plateau. The other two products did not show this change. In the VCF product, there was a larger area of forest increase in the North China Plain (<0.04/yr), while the other two products did not show this change. When the spatiotemporal changes in the forests were considered, the qualitative VCF product was better, and the result was between qualitative products and semiquantitative products, which can qualitatively reflect the spatiotemporal changes in the forest. The impact of the resolution on product uncertainty was explored, and the percentage of various pixel changes with the resolution was counted. As the resolution decreased, the products tended to be consistent. The average percentage of pixels with forest increases of less than 0.02/yr for the three products was 50%. At a resolution of 0.1°, the percentages of pixels with forest increases of less than 0.02/yr for the ESA CCI and MODIS products were 38 and 73%, respectively. When the resolution was reduced to 0.3°, their pixel ratios were 46 and 62%. As the resolution decreased, they both approached 50%. As the resolution decreased, the uncertainty between products also decreased. The various pixels of VCF products did not change much with changes in the spatial resolution and were not sensitive to changes in the resolution. When the scale increased and decreased, there was no significant uncertainty.
The uncertainty of the spatial and temporal changes in the forest cover of the three products was studied, and the results are shown in Figure 6a. The spatial and temporal variations in nontree vegetation derived from different products in different regions were quite different, the consistency of each product was low, and the uncertainty was large. The loss of nontree vegetation in coastal areas was severe (>0.02/yr), and the increase mainly occurred in western China, such as the Northeast Plain, Xinjiang Basin, and Qinghai-Tibet Plateau (<0.1/yr). The opposite conclusion may be drawn for different products in the same area. In the ESA CCI products, there was a decrease in nontree vegetation on the North China Plain (>0.02/yr), while in the MODIS products, there was an increase (0–0.02/yr). VCF qualitative products were highly sensitive to the temporal and spatial changes in nontree vegetation, and MODIS qualitative products were the least sensitive. The impact of resolution on product uncertainty was explored, and the percentage of various pixel changes with the resolution was counted. As the resolution decreased, the products tended to be consistent. The various pixels of the quantitative VCF product did not change much with the change in the spatial resolution and were not sensitive to the change in the resolution. When the scale increased and decreased, there was no significant uncertainty.
The uncertainty of the spatial and temporal changes in the grass and crop cover were investigated. The pixel-by-pixel grass and crop coverage of different resolutions from 2001 to 2012 were subjected to least squares regression, as shown in Figure 7. The impact of the resolution on product uncertainty was explored. As the resolution decreased, the products tended to be consistent. The crop and grass increased slightly (<0.02/yr). The increase in grass mainly occurred in the Qinghai-Tibet Plateau. The spatiotemporal changes in the crop derived from different products in different regions were quite different, and there was great uncertainty. In the MODIS products, there was an increase in crop (<0.02/yr) in North China Plain. The other product ESA CCI did not show this change.
Overall, the spatial and temporal changes in the three products in the research areas had strong consistency. However, the differences in local areas were considerable, the uncertainty was high, and the conclusions were the opposite. It is worth noting that the quantitative VCF product was not sensitive to changes in resolution and did not produce large uncertainties when the scale increased and decreased.

3.4. Resolution Uncertainty

The uncertainty among multisource land cover products decreased as the resolution became coarser. Although the uncertainty decreased, the coarser resolution inevitably led to a decrease in the classification accuracy of land cover products. Precision and consistency were two irreconcilable contradictions. The average and root mean square error of the percentage of land cover types were calculated, and the results are shown in Figure 8.
As shown in the figure, the overall consistency of quantitative VCF and ESA CCI products was the best, indicating that the uncertainty of VCF and ESA CCI was the smallest. As the resolution decreased, the difference in each product decreased rapidly at first and then slowly, and the consistency increased rapidly at first and then slowly. In the research area, the resolution was reduced, and the consistency was enhanced. The resolution of 0.1° was the inflection point of uncertainty and accuracy. Before 0.1°, the uncertainty of multisource land cover products decreased significantly as the resolution decreased, and after 0.1°, the uncertainty decreased slowly as the resolution decreased.

4. Discussion

In this research, the forest cover and nontree vegetation cover of three sets of land cover products in China from 2001 to 2012 were discussed to explore the uncertainty between the products. As shown in the conceptual illustration of the uncertainty of land cover products (Figure 9), the three selected products each had different characteristics: MODIS was a qualitative product with discrete classes in a pixel. This model was easy to express, but its shortcomings were quite obvious. When there were mixed pixels or the resolution was low, the accuracy would be very low. VCF was a quantitative product with land cover fractions. This qualitative feature had obvious advantages. It was quite sensitive to changes in land cover types but not sensitive to changes in spatial resolution. The disadvantage was that its data had only three layers: Tree cover, nontree vegetation, and bare ground. The ESA CCI land cover product was a semiquantitative product, which could be regarded as a transformation from qualitative to quantitative and had the properties of both qualitative and quantitative products (Figure 9b). The characteristics of these three representative land cover products are visually shown in the conceptual illustration (Figure 9). When different land cover types were studied, semiquantitative products were consistent with qualitative products or quantitative products. When the forest coverage in the study area was monitored, the semiquantitative ESA CCI product and the quantitative VCF product had good consistency (Figure 9c). When the nontree vegetation coverage in the study area was monitored, the semiquantitative ESA CCI product and the qualitative VCF product had good consistency (Figure 9a). The consistency increased when the resolution became coarser, but the coarser resolution inevitably decreased the classification accuracy. The decrease in uncertainty inevitably sacrificed the accuracy, including the accuracy and consistency of land cover products, which were two irreconcilable contradictions (Figure 9d). The overall consistency in the study area was counted with the change in resolution. As the resolution decreased (and the accuracy decreased), the uncertainty of multisource land cover products also decreased (and the consistency increased). This change was not linear, and there was an inflection point, which was approximately 0.1° (Figure 9e).
In exploring spatial consistency, we found that areas with more significant uncertainty were plateaus or hilly areas. The reason for this pattern was that there was a problem of mixed pixels in low-resolution remote sensing data, which led to considerable differences among the dataset. In the research, the pixel corresponded to a large area on the ground (>5 km). There were multiple vegetation types on the hills. The influence of mixed pixels was also difficult to eliminate using semiquantitative ESA CCI products and qualitative VCF products. The impact was further magnified in the complex mountainous areas. In this study, the areas with more significant spatial uncertainty included the Greater Khingan Mountains, the southeast hills, the Hengduan Mountains, and the Yunnan-Guizhou Plateau. The areas with less uncertainty included the North China Plain, the Northeast Plain, and the Xinjiang Basin. In addition, the pixel difference was quantified with the change in spatial resolution, which would help scholars select products with suitable resolution in combination with the uncertainty of different regions when performing regional remote sensing inversion and model simulation. The developing trend of land cover products was quantitative. Quantitative land cover monitoring was provided by VCF products, which solved the problem that land cover products could not capture subpixel and intrapixel changes. VCF products were also determined by the scientific community as earth system data records (ESDRs), but the current VCF data version only has three data layers, and further detailed land cover type monitoring cannot be provided.
The reliability of our results was confirmed by similar studies. Different land cover product classification systems, sources, and precisions were different, and the differences between products were also significant, especially the differences in nontree vegetation, and the classification systems of different products were different, such as in the Qinghai-Tibet Plateau. MODIS products had little vegetation coverage, and some products had more grass coverage [39]. In this research, the semiquantitative ESA CCI product based on PFTs was more reliable. First, compared to other products, the ESA CCI had good spatial resolution [40]. Due to the limitation of spatial resolution, the image contained mixed pixels of different land cover types. This may have led to uncertain results, but the PFT-based method quantified each land cover type, giving it quantitative characteristics [41]. Due to its qualitative and quantitative characteristics, it had good consistency with qualitative and quantitative products.
In order to enrich our study, more products were compared. Since only one semiquantitative product and one quantitative product were available at present, we selected an additional qualitative ChinaLC product for comparison. The ChinaLC products used a large-scale classification approach to produce a decadal 5 km resolution land cover dataset from 1981 to 2010. A total of 19 classes of training and validation samples were obtained. The ChinaLC dataset has an average overall accuracy of approximately 75% [42]. However, the years of these products were not continuous, and it was difficult to conduct the trend analysis. We used the ChinaLC products to compare with the other three products. The percentage of the forest cover of the four products was calculated first, and then the two products were differentiated pixel-by-pixel, the resolution was 0.2°, as shown in Figure 10. The spatial distribution of the forest cover of the four products was relatively consistent but differed significantly in different regions. The semiquantitative ESA CCI product had the highest consistency with the qualitative VCF product.
This study used three types of land cover products in China from 2001 to 2012 and explored their uncertainties. The development of land cover products was as follows: Qualitative to quantitative and rough to fine. Quantitative land cover products could be used to quantitatively describe the climate, carbon, and sustainability [43]. However, only three data layers were included in the current quantitative VCF product, version 1.0, and the time frame reached only to 2015. Due to the lack of long-term records and the lack of intermediate years, the semiquantitative ESA CCI product is currently a better choice. VCF products had fewer data layers, the “shrubs,” “grassland,” and “farmland” in ESA CCI products were all classified as “nontree vegetation.” In fact, more detailed and specific category consistency is worthy of further study. We investigated the uncertainty of the grassland and farmland of MODIS and ESA CCI and found that the conclusions reached were consistent with forests and nontree vegetation. In the current research, we noticed that the quantitative VCF product was more sensitive in characterizing land cover changes, leading to differences in landscape consistency. These differences will be quantified using the landscape index to obtain the uncertainty of different land cover products in our future research. In addition, the consistency of land cover products on a global scale will be studied to enrich our research system.

5. Conclusions

This study used three multisource land cover products from 2001 to 2012 in China to explore their area, spatiotemporal changes, and resolution uncertainty, revealing the essential differences between qualitative, semiquantitative, and quantitative products. The uncertainty of land cover products in different regions and the uncertainty of resolution have been explored, laying the foundation for the uncertainty evaluation of remote sensing inversion, model simulation, and subpixel estimation. The results showed that the overall consistency of the three land cover products was relatively strong, the area uncertainty was relatively small, and they were relatively consistent in China. Considering forest coverage, VCF and ESA CCI products had the slightest uncertainty, and considering nontree vegetation, ESA CCI and MODIS products had the slightest uncertainty. However, the spatial distributions of the three products were locally inconsistent, and different regions may show opposite conclusions regarding spatiotemporal changes. The regions with significant spatial differences were plateaus or hills. The products with the best overall consistency were VCF and ESA CCI, and the worst were VCF and MODIS. The semiquantitative ESA CCI product had both qualitative and quantitative characteristics and was a better choice. As the resolution increased, the consistency of each product was enhanced, but the accuracy was inevitably reduced. The contradiction between accuracy and resolution was balanced through an inflection point of resolution. When scholars conduct remote sensing inversion, model simulation, and subpixel program research, the uncertainty system of multisource land cover products can assess the reliability of different studies. Therefore, in future research, we will establish a global-scale, multiproduct consistency evaluation system.

Author Contributions

Conceptualization, J.J.; Data curation, L.W.; Formal analysis, L.W.; Methodology, J.J.; Resources, J.J.; Visualization, L.W. All authors have read and agreed to the published version of the manuscript.

Funding

This work was funded by the Fundamental Research Funds for the Central Universities (grant number B200202016).

Institutional Review Board Statement

Not applicable.

Informed Consent Statement

Not applicable.

Data Availability Statement

The study did not report any data.

Acknowledgments

We would like to acknowledge Yingying Ji, Hongbo Wang, Liyang Sun, Yanan Bai and Han Gao from Hohai University for the constructive suggestions. Data of the MCD12 product used are available at https://modis.gsfc.nasa.gov/data (accessed on 22 October 2014), data of the ESA CCI product are available at http://maps.elie.ucl.ac.be/CCI/viewer/download (accessed on 22 December 2019), and data of MEaSURES VCF are available at https://lpdaac.usgs.gov/productsearch/ (accessed on 19 February 2020).

Conflicts of Interest

The authors declare no conflict of interest.

References

  1. Giri, C.; Pengra, B.; Long, J.; Loveland, T. Next generation of global land cover characterization, mapping, and monitoring. Int. J. Appl. Earth Obs. Geoinf. 2013, 25, 30–37. [Google Scholar] [CrossRef]
  2. Sexton, J.O.; Song, X.; Feng, M.; Noojipady, P.; Anand, A.; Huang, C.; Kim, D.; Collins, K.M.; Channan, S.; DiMiceli, C.; et al. Global, 30-m resolution continuous fields of tree cover: Landsat-based rescaling of MODIS vegetation continuous fields with lidar-based estimates of error. Int. J. Digit. Earth 2013, 6, 427–448. [Google Scholar] [CrossRef] [Green Version]
  3. Yongke, Y.; Pengfeng, X.; Xuezhi, F.; Haixing, L.; Xiao, C.; Weiding, F. Comparison and assessment of large-scale land cover datasets in China and adjacent regions. J. Remote. Sens. 2014, 18, 453–475. [Google Scholar]
  4. Erb, K.H.; Luyssaert, S.; Meyfroidt, P.; Pongratz, J.; Don, A.; Kloster, S.; Kuemmerle, T.; Fetzel, T.; Fuchs, R.; Herold, M.; et al. Land management: Data availability and process understanding for global change studies. Glob. Chang. Biol. 2017, 23, 512–533. [Google Scholar] [CrossRef] [PubMed] [Green Version]
  5. Lambin, E.; Ehrlich, D. Land-cover changes in sub-Saharan Africa (1982–1991): Application of a change index based on remotely sensed surface temperature and vegetation indices at a continental scale. Remote Sens. Environ. 1997, 61, 181–200. [Google Scholar] [CrossRef]
  6. Giglio, L.; Loboda, T.; Roy, D.P.; Quayle, B.; Justice, C.O. An active-fire based burned area mapping algorithm for the MODIS sensor. Remote Sens. Environ. 2008, 113, 408–420. [Google Scholar] [CrossRef]
  7. Ganguly, S.; Friedl, M.; Tan, B.; Zhang, X.; Verma, M. Land surface phenology from MODIS: Characterization of the Collection 5 global land cover dynamics product. Remote Sens. Environ. 2010, 114, 1805–1816. [Google Scholar] [CrossRef] [Green Version]
  8. Grekousis, G.; Mountrakis, G.; Kavouras, M. An overview of 21 global and 43 regional land-cover mapping products. Int. J. Remote Sens. 2015, 36, 5309–5335. [Google Scholar] [CrossRef]
  9. Hua, T.; Zhao, W.; Liu, Y.; Wang, S.; Yang, S. Spatial Consistency Assessments for Global Land-Cover Datasets: A Comparison among GLC2000, CCI LC, MCD12, GLOBCOVER and GLCNMO. Remote Sens. 2018, 10, 1846. [Google Scholar] [CrossRef] [Green Version]
  10. Zhang, W.; Tang, P.; Zhao, L. Fast and accurate land cover classification on medium resolution remote sensing images using segmentation models. Int. J. Remote Sens. 2021, 42, 3277–3301. [Google Scholar] [CrossRef]
  11. Friedl, M.A.; McIver, D.K.; Hodges, J.C.F.; Zhang, X.Y.; Muchoney, D.; Strahler, A.H.; Woodcock, C.E.; Gopal, S.; Schneider, A.; Cooper, A.; et al. Global land cover mapping from MODIS: Algorithms and early results. Remote Sens. Environ. 2002, 83, 287–302. [Google Scholar] [CrossRef]
  12. Price, J.C. Comparing MODIS and ETM+ data for regional and global land classification. Remote Sens. Environ. 2003, 86, 491–499. [Google Scholar] [CrossRef]
  13. Ling, F.; Li, W.; Du, Y.; Li, X. Land Cover Change Mapping at the Subpixel Scale With Different Spatial-Resolution Remotely Sensed Imagery. IEEE Geosci. Remote Sens. Lett. 2011, 8, 182–186. [Google Scholar] [CrossRef]
  14. Urbanski, S.P.; Hao, W.M.; Nordgren, B. The wildland fire emission inventory: Western United States emission estimates and an evaluation of uncertainty. Atmos. Chem. Phys. 2011, 11, 12973–13000. [Google Scholar] [CrossRef] [Green Version]
  15. Lu, L.; Huang, Y.; Di, L.; Hang, D. Large-scale subpixel mapping of landcover from MODIS imagery using the improved spatial attraction model. J. Appl. Remote Sens. 2018, 12, 046017. [Google Scholar] [CrossRef]
  16. Argles, A.P.K.; Moore, J.R.; Huntingford, C.; Wiltshire, A.J.; Harper, A.B.; Jones, C.D.; Cox, P.M. Robust Ecosystem Demography (RED version 1.0): A parsimonious approach to modelling vegetation dynamics in Earth system models. Geosci. Model Dev. 2020, 13, 4067–4089. [Google Scholar] [CrossRef]
  17. McCallum, I.; Obersteiner, M.; Nilsson, S.; Shvidenko, A. A spatial comparison of four satellite derived 1 km global land cover datasets. Int. J. Appl. Earth Obs. Geoinf. 2006, 8, 246–255. [Google Scholar] [CrossRef]
  18. Ran, Y.; Li, X.; Lu, L. Evaluation of four remote sensing based land cover products over China. Int. J. Remote Sens. 2010, 31, 391–401. [Google Scholar] [CrossRef]
  19. Schaefer, K.; Schwalm, C.R.; Williams, C.; Arain, M.A.; Barr, A.; Chen, J.M.; Davis, K.J.; Dimitrov, D.; Hilton, T.W.; Hollinger, D.Y.; et al. A model-data comparison of gross primary productivity: Results from the North American Carbon Program site synthesis. J. Geophys. Res. Biogeosci. 2012, 117. [Google Scholar] [CrossRef] [Green Version]
  20. Levy, P.; Friend, A.; White, A.; Cannell, M. The influence of land use change on global-scale fluxes of carbon from terrestrial ecosystems. Clim. Chang. 2004, 67, 185–209. [Google Scholar] [CrossRef]
  21. Gao, Y.; Liu, L.; Zhang, X.; Chen, X.; Mi, J.; Xie, S. Consistency Analysis and Accuracy Assessment of Three Global 30-m Land-Cover Products over the European Union using the LUCAS Dataset. Remote Sens. 2020, 12, 3479. [Google Scholar] [CrossRef]
  22. Zhu, Q.; Jin, J.; Wang, P.; Ji, Y.; Xiao, Y.; Guo, F.; Deng, C.; Qu, L. Contrasting Trends of Forest Coverage between the Inland and Coastal Urban Groups of China over the Past Decades. Sustainability 2019, 11, 4451. [Google Scholar] [CrossRef] [Green Version]
  23. Ji, Y.; Jin, J.; Zhu, Q.; Zhou, S.; Wang, Y.; Wang, P.; Xiao, Y.; Guo, F.; Lin, X.; Xu, J. Unbalanced forest displacement across the coastal urban groups of eastern China in recent decades. Sci. Total Environ. 2020, 705, 135900. [Google Scholar] [CrossRef] [PubMed]
  24. Mertens, K.; Verbeke, L.; Westra, T.; De Wulf, R. Sub-pixel mapping and sub-pixel sharpening using neural network predicted wavelet coefficients. Remote Sens. Environ. 2004, 91, 225–236. [Google Scholar] [CrossRef]
  25. Mingchao, S.; Hongli, S.; Ming, S.; Hongtao, H.; Yimeng, S. Multisource land-cover data category accuracy evaluation and the difference of landscape patterns: A case study of Beijing-Tianjin-Hebei region. Chin. J. Ecol. 2020, 39, 3881–3889. [Google Scholar]
  26. Bayer, A.D.; Lindeskog, M.; Pugh, T.A.M.; Anthoni, P.M.; Fuchs, R.; Arneth, A. Uncertainties in the land-use flux resulting from land-use change reconstructions and gross land transitions. Earth Syst. Dyn. 2017, 8, 91–111. [Google Scholar] [CrossRef] [Green Version]
  27. Hartley, A.J.; MacBean, N.; Georgievski, G.; Bontemps, S. Uncertainty in plant functional type distributions and its impact on land surface models. Remote Sens. Environ. 2017, 203, 71–89. [Google Scholar] [CrossRef]
  28. Bai, Y.; Feng, M.; Jiang, H.; Wang, J.; Zhu, Y.; Liu, Y. Assessing Consistency of Five Global Land Cover Data Sets in China. Remote Sens. 2014, 6, 8739–8759. [Google Scholar] [CrossRef] [Green Version]
  29. Qionghuan, L.; Yili, Z.; Linshan, L.; Lanhui, L.; Wei, Q. Accuracy evaluation of the seven land cover data in Qiangtang Plateau. Geogr. Res. 2017, 36, 2061–2074. [Google Scholar]
  30. Friedl, M.A.; Sulla-Menashe, D.; Tan, B.; Schneider, A.; Ramankutty, N.; Sibley, A.; Huang, X. MODIS Collection 5 global land cover: Algorithm refinements and characterization of new datasets. Remote Sens. Environ. 2010, 114, 168–182. [Google Scholar] [CrossRef]
  31. 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] [Green Version]
  32. Li, W.; Ciais, P.; MacBean, N.; Peng, S.; Defourny, P.; Bontemps, S. Major forest changes and land cover transitions based on plant functional types derived from the ESA CCI Land Cover product. Int. J. Appl. Earth Obs. Geoinf. 2016, 47, 30–39. [Google Scholar] [CrossRef]
  33. DeFries, R.; Hansen, M.; Steininger, M.; Dubayah, R.; Sohlberg, R.; Townshend, J. Subpixel forest cover in central Africa from multisensor, multitemporal data. Remote Sens. Environ. 1997, 60, 228–246. [Google Scholar] [CrossRef]
  34. Hansen, M.C.; Potapov, P.V.; Moore, R.; Hancher, M.; Turubanova, S.A.; Tyukavina, A.; Thau, D.; Stehman, S.V.; Goetz, S.J.; Loveland, T.R.; et al. High-Resolution Global Maps of 21st-Century Forest Cover Change. Science 2013, 342, 850–853. [Google Scholar] [CrossRef] [PubMed] [Green Version]
  35. Kim, D.; Sexton, J.O.; Noojipady, P.; Huang, C.; Anand, A.; Channan, S.; Feng, M.; Townshend, J.R. Global, Landsat-based forest-cover change from 1990 to 2000. Remote Sens. Environ. 2014, 155, 178–193. [Google Scholar] [CrossRef] [Green Version]
  36. Cai, Z.; Sawamoto, T.; Li, C.; Kang, G.; Boonjawat, J.; Mosier, A.; Wassmann, R.; Tsuruta, H. Field validation of the DNDC model for greenhouse gas emissions in East Asian cropping systems. Glob. Biogeochem. Cycles 2003, 17. [Google Scholar] [CrossRef] [Green Version]
  37. Hu, Y.; Zhang, Q.; Dai, S.; Huang, M.; Yan, H. Agreement analysis of multi-sensor satellite remote sensing derived land cover products in the Europe Continent. Geogr. Res. 2015, 34, 1839–1852. [Google Scholar]
  38. Masiliunas, D.; Tsendbazar, N.; Herold, M.; Lesiv, M.; Buchhorn, M.; Verbesselt, J. Global land characterisation using land cover fractions at 100 m resolution. Remote Sens. Environ. 2021, 259, 112409. [Google Scholar] [CrossRef]
  39. Yu, W.; Li, J.; Liu, Q.; Zeng, Y.; Zhao, J.; Xu, B.; Yin, G. Global Land Cover Heterogeneity Characteristics at Moderate Resolution for Mixed Pixel Modeling and Inversion. Remote Sens. 2018, 10, 856. [Google Scholar] [CrossRef] [Green Version]
  40. An, R.; Zhang, L.; Wang, Z.; Quaye-Ballard, J.A.; You, J.; Shen, X.; Gao, W.; Huang, L.; Zhao, Y.; Ke, Z. Validation of the ESA CCI soil moisture product in China. Int. J. Appl. Earth Obs. Geoinf. 2016, 48, 28–36. [Google Scholar] [CrossRef]
  41. Bontemps, S.; Herold, M.; Kooistra, L.; Groenestijn, A.V.; Hartley, A.; Arino, O.; Moreau, I.; Defourny, P. Revisiting land cover observation to address the needs of the climate modeling community. Biogeosciences 2012, 9, 2145–2157. [Google Scholar] [CrossRef] [Green Version]
  42. Li, H.; Xiao, P.; Feng, X.; Yang, Y.; Wang, L.; Zhang, W.; Wang, X.; Feng, W.; Chang, X. Using Land Long-Term Data Records to Map Land Cover Changes in China Over 1981–2010. IEEE J. Sel. Top. Appl. Earth Obs. Remote. Sens. 2017, 10, 1372–1389. [Google Scholar] [CrossRef]
  43. Swetnam, R.; Fisher, B.; Mbilinyi, B.; Munishi, P.; Willcock, S.; Ricketts, T.; Mwakalila, S.; Balmford, A.; Burgess, N.; Marshall, A.; et al. Mapping socio-economic scenarios of land cover change: A GIS method to enable ecosystem service modelling. J. Environ. Manag. 2011, 92, 563–574. [Google Scholar] [CrossRef] [PubMed]
Sustainability 13 08857 g001 550
Figure 1. Interannual variabilities of (a) forests and (b) nontree vegetation for MODIS, CCI, and VCF products.
Figure 1. Interannual variabilities of (a) forests and (b) nontree vegetation for MODIS, CCI, and VCF products.
Sustainability 13 08857 g001
Sustainability 13 08857 g002 550
Figure 2. Spatial difference in forest coverage among the three products. Panel (a) shows the resolution and spatial difference between products, “CCI−MODIS” represents the difference between the forest percentage of ESA CCI and MODIS products, the horizontal is the product consistency comparison, and the vertical is the consistency comparison of different resolutions. The darker the color, the greater the product difference. Panel (b) shows the PFT statistics and the percentage of various pixels with the change in resolution.
Figure 2. Spatial difference in forest coverage among the three products. Panel (a) shows the resolution and spatial difference between products, “CCI−MODIS” represents the difference between the forest percentage of ESA CCI and MODIS products, the horizontal is the product consistency comparison, and the vertical is the consistency comparison of different resolutions. The darker the color, the greater the product difference. Panel (b) shows the PFT statistics and the percentage of various pixels with the change in resolution.
Sustainability 13 08857 g002
Sustainability 13 08857 g003 550
Figure 3. The spatial difference in nontree vegetation coverage among the three products. Panel (a) shows the resolution and spatial difference between products, “CCI−MODIS” represents the difference between the forest percentage of ESA CCI and MODIS products, the horizontal is the product consistency comparison, and the vertical is the consistency comparison of different resolutions. The darker the color, the greater the product difference. Panel (b) shows the PFT statistics and the percentage of various pixels with the change in resolution.
Figure 3. The spatial difference in nontree vegetation coverage among the three products. Panel (a) shows the resolution and spatial difference between products, “CCI−MODIS” represents the difference between the forest percentage of ESA CCI and MODIS products, the horizontal is the product consistency comparison, and the vertical is the consistency comparison of different resolutions. The darker the color, the greater the product difference. Panel (b) shows the PFT statistics and the percentage of various pixels with the change in resolution.
Sustainability 13 08857 g003
Sustainability 13 08857 g004 550
Figure 4. Spatial difference in crop and grass coverage of the ESA CCI and MODIS products. The darker the color, the greater the product difference.
Figure 4. Spatial difference in crop and grass coverage of the ESA CCI and MODIS products. The darker the color, the greater the product difference.
Sustainability 13 08857 g004
Sustainability 13 08857 g005 550
Figure 5. The spatiotemporal difference in forest coverage changes among the three products. Panel (a) shows the resolution and spatiotemporal difference between products, the horizontal is the product consistency comparison, and the vertical is the consistency comparison of different resolutions. Green indicates forest increase, red indicates forest loss, and the color shading indicates the degree. Panel (b) shows the PFT statistics and the percentage of various pixels with the change in resolution.
Figure 5. The spatiotemporal difference in forest coverage changes among the three products. Panel (a) shows the resolution and spatiotemporal difference between products, the horizontal is the product consistency comparison, and the vertical is the consistency comparison of different resolutions. Green indicates forest increase, red indicates forest loss, and the color shading indicates the degree. Panel (b) shows the PFT statistics and the percentage of various pixels with the change in resolution.
Sustainability 13 08857 g005
Sustainability 13 08857 g006 550
Figure 6. The spatiotemporal difference in nontree vegetation coverage changes among the three products. Panel (a) shows the resolution and spatiotemporal difference between products, the horizontal is the product consistency comparison, and the vertical is the consistency comparison of different resolutions. Blue indicates forest increase, red indicates forest loss, and the color shading indicates the degree. Panel (b) shows the PFT statistics and the percentage of various pixels with the change in resolution.
Figure 6. The spatiotemporal difference in nontree vegetation coverage changes among the three products. Panel (a) shows the resolution and spatiotemporal difference between products, the horizontal is the product consistency comparison, and the vertical is the consistency comparison of different resolutions. Blue indicates forest increase, red indicates forest loss, and the color shading indicates the degree. Panel (b) shows the PFT statistics and the percentage of various pixels with the change in resolution.
Sustainability 13 08857 g006
Sustainability 13 08857 g007 550
Figure 7. The spatiotemporal difference in grass and crop coverage changes of the products. Green indicates increase, blue indicates loss, and the color shading indicates the degree.
Figure 7. The spatiotemporal difference in grass and crop coverage changes of the products. Green indicates increase, blue indicates loss, and the color shading indicates the degree.
Sustainability 13 08857 g007
Sustainability 13 08857 g008 550
Figure 8. Overall variability of the uncertainty with an increase in the spatial resolution for forests and nontree vegetation of the three land cover products ((a,c) represent the mean value of the difference; (b,d) represent the root mean square error of the difference).
Figure 8. Overall variability of the uncertainty with an increase in the spatial resolution for forests and nontree vegetation of the three land cover products ((a,c) represent the mean value of the difference; (b,d) represent the root mean square error of the difference).
Sustainability 13 08857 g008
Sustainability 13 08857 g009 550
Figure 9. Conceptual illustration of (a–c) three types of land cover product characteristics and (d,e) consistency variations with resolution.
Figure 9. Conceptual illustration of (a–c) three types of land cover product characteristics and (d,e) consistency variations with resolution.
Sustainability 13 08857 g009
Sustainability 13 08857 g010 550
Figure 10. Spatial difference in forest coverage changes among the four products “ChinaLC−MODIS” represents the difference between the forest percentage of ChinaLC and MODIS products. The darker the color, the greater the product difference.
Figure 10. Spatial difference in forest coverage changes among the four products “ChinaLC−MODIS” represents the difference between the forest percentage of ChinaLC and MODIS products. The darker the color, the greater the product difference.
Sustainability 13 08857 g010
Table
Table 1. Land cover products used in this work.
Table 1. Land cover products used in this work.
ProductOrganizationResolutionLand Cover CategoryClassificationType
MCD12Boston University1 KM17Decision treeQualitative
ESA CCIEuropean Space Agency300 M37Unsupervised classificationSemi-quantitative
MEaSURES VCFUniversity of Maryland5 KM3Linear Model TreesQuantitative
Table
Table 2. PFT-based integrated land cover category for the ESA CCI land cover data.
Table 2. PFT-based integrated land cover category for the ESA CCI land cover data.
Class 1Class 2Class 1Class 2
ForestEvergreen broad-leaved forestNontree vegetationDeciduous broad-leaved shrub
Deciduous broad-leaved forestEvergreen needle-leaved shrub
Evergreen needle-leaved forestDeciduous needle-leaved shrub
Deciduous needle-leaved forestGrassland
Nontree vegetationEvergreen broad-leaved shrubCrop
Table
Table 3. Sample selection categories and data statistics.
Table 3. Sample selection categories and data statistics.
Land Cover TypeESA CCIMCD12VCF
Area (%)Coefficient of DeviationArea (%)Coefficient of DeviationArea (%)Coefficient of Deviation
Forest13.24−0.122217.620.167714.40−0.0456
Nontree vegetation48.760.095245.810.028739.01−0.1239
Publisher’s Note: MDPI stays neutral with regard to jurisdictional claims in published maps and institutional affiliations.

Share and Cite

MDPI and ACS Style

Wang, L.; Jin, J. Uncertainty Analysis of Multisource Land Cover Products in China. Sustainability 2021, 13, 8857. https://doi.org/10.3390/su13168857

AMA Style

Wang L, Jin J. Uncertainty Analysis of Multisource Land Cover Products in China. Sustainability. 2021; 13(16):8857. https://doi.org/10.3390/su13168857

Chicago/Turabian Style

Wang, Longhao, and Jiaxin Jin. 2021. "Uncertainty Analysis of Multisource Land Cover Products in China" Sustainability 13, no. 16: 8857. https://doi.org/10.3390/su13168857

APA Style

Wang, L., & Jin, J. (2021). Uncertainty Analysis of Multisource Land Cover Products in China. Sustainability, 13(16), 8857. https://doi.org/10.3390/su13168857

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