Zone II • Versions EN1
Abstract: Fractional vegetation cover (FVC) is a quantitative indicator to measure surface vegetation cover, which provides basic data for studying ecological environment, water and soil conservation, and climate change. In the land-surface process model, FVC is not only the key parameter for calculating precipitation interception in vegetation canopy, but also controls the phenological period of Leaf Area Index (LAI). A quantitative inversion of FVC is conducive to ecological environment treatment and eco-system construction. Based on existing literature, this study collected remote sensing models used to estimate FVC over different underlying surfaces in China from 1980 to 2016. The study covers a spatial range of 24.49° – 51.42°N, 80.23° – 128.95°E, including northwest regions (Inner Mongolia, Xinjiang, Qinghai, Tibet and Gansu), north regions (Beijing, Hebei, Shandong and Henan), as well as southwest and south regions (Yunnan, Guangxi and Jiangxi). The dataset covers the major underlying surfaces in typical Chinese terrestrial ecosystems, including forests, shrubs, grasslands, wetlands, deserts, farmlands, unban lands and karst regions. It provides model statistics for quantitative research of ecology, water and soil conservation, hydraulic engineering, and vegetation. It can also be used to support ecosystem service evaluation, regional eco-environmental protection, and conservation redline delineation.
Keywords: China; underlying surface; vegetation coverage; models
|Title||Remote sensing models for fractional vegetation cover estimation over different underlying surfaces in China|
|Data authors||Liu Erhua, Zhou Guangsheng, Zhou Li|
|Data corresponding author||Zhou Guangsheng|
|Time range||1980 – 2016|
|Geographical scope||24.49° – 51.42°N, 80.23° – 128.95°E|
|Data volume||42 KB|
|Data service system||<http://www.sciencedb.cn/dataSet/handle/674>|
|Sources of funding||The CAMS Basic Research Fund (Grant No. 2018Z008); China Special Fund for Meteorological Research in the Public Interests (Grant No. GYHY201506001-3).|
|Dataset composition||The dataset includes 115 FVC models covering the different underlying surfaces of China’s typical terrestrial ecosystems during 1980 – 2016, supplemented with respective model construction specifications and model profile information including geographical location, underlying surface type, study period, data sources, and so forth.|
Vegetation refers to the general term of plant communities on the ground surface, such as forests, shrubs, grasslands and agricultural crops,1 and it is a major type of land cover.2 Vegetation plays an important role in energy exchange, biogeochemical and hydrological processes for global change.3,4 As an important biophysical parameter involved in the surface processes, it is the prerequisite for Numerical Weather Prediction, regional and global climate modelling, and global change monitoring.5 A change in vegetation is subject to the influence of both natural environmental changes and human activities, among which climate change is the main factor. In turn, vegetation can also regulate local climate.6 It is the natural element connecting soil, atmosphere and water7 and an important parameter for ecological environment assessment. Fractional Vegetation Cover (FVC) is an index for describing vegetation coverage conditions, referring to the ratio of the vertical projection area of vegetation (including leaves, stalks, and branches) on the ground to the total vegetation area. It reflects the vegetation growth status and trend. As a comprehensive quantitative indicator8,9 to measure surface vegetation cover, FVC plays a crucial role in the study of regional ecosystems, including ecological environment, water and soil conservation and climate change.10,11 Quantitative inversion of FVC is of instructive values for ecological environment detection and management, as well as ecological construction. Jia (2013) argued that FVC studies help advance earth system modelling and global change research.12 Chen (2014)found that the changes of meteorological elements simulated in the climate model are closely related to the changes of vegetation parameters, especially to FVC and LAI.13
Previous studies used diversified ecosystem models for FVC estimation, which can be attributed to the following factors: first, China is a country with geographical complexity, with spatially heterogeneous climates that vary by geographical location, climate environment and human activities. Second, the use of varied data sources resulted in different approaches to FVC estimation, ranging from visual interpretation to instrument estimation and remote sensing interpretation.14 Especially, FVC can be monitored on a large scale in a more timely and accurate manner15 to reflect the vegetation coverage at different spatial scales and changing trends,12 with the application of remote sensing technologies in FVC estimation.16 Third, the FVC estimation methods are varied. Among them, two types of methods are most common: the first type consists of empirical model methods (e.g., regression model method, vegetation index method), mixed pixel decomposition methods (e.g., pixel dichotomy model method), physical model methods (e.g., spectral gradient difference method, model inversion method), and FCD model cartography.2,8 The second type mainly refers to machine learning methods involving spatial data mining technologies, including neural network algorithm, support vector machine and decision tree algorithm.8,17 Among them, empirical model and dimidiate pixel model are the most widely used. Fourth, the FVC models may also result from the different vegetation types on the underlying surfaces they are used to estimate.
As FVC can substantially vary by the types of underlying surfaces, it would be of great value to sort out existing FVC models as per their geographical areas and underlying surface types. Due to surface complexity and heterogeneity, it is often difficult to establish a FVC model applicable to an entire region. As each ecological zone has its own vulnerability and environmental carrying capacity, researchers need to choose an optimal FVC model for accurate ecosystem security estimation. Despite a large number of valuable FVC models, there has been no systematic studies dedicated to building a comprehensive dataset synthesizing previous research achievements. Given this, we collected and collated the FVC models for the different underlying surfaces of typical Chinese ecosystems, which is conducive to ecological environment assessment. The dataset provides data support for quantitative research on such fields as regional ecology, water and soil conservation, plants, etc., thereby advancing natural environment research in general. The FVC model database presented by this paper can be used for the construction of ecological security in different administrative and natural regions across the country.
2.1 Data sources
As this is a literature-based study aiming to collect data on the remote sensing FVC estimation models for China’s typical ecological systems from 1980 to 2016, CNKI (China National Knowledge Internet) was used as the main source of data extraction. The keywords used for data retrieval included “fractional vegetation cover [植被覆盖度]”, “grassland fractional vegetation cover [草地植被覆盖度]”, “regression model method [回归模型法]” and “dimidiate pixel model [像元二分模型]”. Certain criteria were set up for data screening: first, the remote sensing images should reflect the precise location and spatial scale of their ground sampling sites. Second, the sampling sites should be representative of the region and reflect its overall vegetation and soil status, with a maximal coverage of different species and types. Third, in describing each FVC model, the literature should have the underlying surface type specified. Data eligibility was then determined in accordance with the above criteria, and the eligible FVC models were sorted out and categorized as per their surface type, including forestlands, shrubs, grasslands, wetlands, desertification grasslands, farmlands, towns and rocky desertification areas.
2.2 Data processing
After being collected, the FVC models were sorted out, supplemented by profile information including model expression, vegetation type, and geographic location. If the model was of a regional scope, the longitudes and latitudes were averaged, and the averaged values were taken and recorded as the sampling site. If the model was of multi years from a single site, the FVC models for a specific time each year. In case of multiple FVC models for a single site, each model was correlated with its vegetation type before a statistical analysis was performed. Any initial missing information, for example, on geographic location or ecosystem type, was supplemented later through a further data retrieval. Specific steps are shown in Figure 1.
This dataset contains 115 FVC models built from dimidiate pixel model and regression model, mainly involving surface types such as forestlands, grasslands, desert grasslands, and farmlands, as well as shrubs, wetlands, urban areas and rocky desertification areas.
The dataset consists of an Excel file with two worksheets, which store FVC models and supplementary information (e.g., references), respectively. The data items are arranged in the following order as per our dataset: sequence number, surface type, partition, distribution, latitude, longitude, time period, FVC model, vegetation index, temporal resolution, spatial resolution, data source, modelling method, validation data, R2 , model precision and references (Table 1). The FVC models vary by the type of underlying surface. The example in Table 1 shows a FVC model built for the underlying surface of forestland, by the dimidiate pixel model method based on FVC and forestland vegetation index. 0.42 and 0.83 represent the minimum and maximum vegetation indices of the study area, respectively. When compared with the FVC model of the rocky desertification area, forestland has larger maximum and minimum vegetation indices. On the other hand, two regions with the same surface type (e.g., forestland) may have different FVC values, as FVC is also affected by regional specificities and surface subtypes.
|Data item||Data unit||Data type||Example|
|Partition||-||Character||Coniferous forest in the cold temperate zone of Daxing’an mountains|
|Distribution||-||Character||North of Hulunbuir, Inner Mongolia Autonomous Region|
|Time period||-||Character||September 6, 2000|
|FVC model||-||Character||FVC =（NDVI-0.42/(0.83-0.42)|
|Vegetation index||-||Character||Normalized difference vegetation index (NDVI)|
|Data source||-||Character||Landsat 5 TM image data|
|Modelling method||Character||Dimidiate pixel model|
|Validation data||-||Character||August 2013 survey data|
This dataset drew sources from the published literature. Data quality control was performed throughout the process from source database selection, keyword retrieval, literature screening, to data extraction and sorting.
Model quality control : To ensure the quality of literature sources, authoritative databases such as CNKI were used for model retrieval, in tandem with the dimidiate pixel model and regression model methods. Meanwhile, a unified procedure and specification was used to obtain VIsoil and VIveg , two key input parameters for the pixel dichotomy model method, as their accuracy would impact subsquent FVC estimation.18,19 The formula is expressed as follows:
where FVC represents fractional vegetation cover and VI represents vegetation indices. VI can be NDVI or an alternative vegetation index. VIsoil is the VI value in the area with bare soil, while VIveg is the VI value in the area with vegetation cover. For most bare surfaces, the theoretical value of VIsoil should be 0, but the measured value was generally between −0.1 and 0.2 due to atmospheric effect, surface moisture, etc. The value of VIveg also vary depending on the type of vegetation cover, as well as the methods used for estimation.20 At present, there are four methods to obtain the parameters: (1) Visual interpretation or endmember extraction. (2) Analyze the correlation between measured data and vegetation indices of their corresponding pixel, based on sample plot surveys. (3) Obtain the vegetation index values for full vegetation pixel and bare soil pixel, respectively, based on cumulative probability values by attributing the pixel to a land use type as per remote sensing data and land classification maps. (4) Obtain the maximum and minimum values of a vegetation index within the confidence interval of vegetation index grayscale distribution. Due to a lack of large-scale measured data, a common approach to determining VIsoil and VIveg is to obtain VI data through statistical analysis of land use map, soil map and topographic map.21 Figure 2 shows the steps for building the dimidiate pixel model.
Firstly, obtain the remote sensing image maps of VI and use remote sensing image processing software to summarize into a table the VI distribution by pixel. Secondly, perform statistical processing on the table. Count the total number of pixels and summarize the number of pixels for VI in each scene. Cumulative probability distribution was obtained as the accumulated number of pixels for VI in each scene divided by the total number of pixels. Lastly, determine the maximum and minimum cumulative probability and the confidence interval for VIveg and VIsoil , respectively, so that any outliers could be detected.22 All values should fall onto the intervals though exact values are determined by image size, image definition, etc.23 Studies have shown that the dimidiate pixel model method has a high accuracy in estimating FVC. It is relatively easier, more applicable to the monitoring data of regional scale.24 It can also minimize the impact of atmosphere, soil and vegetation type on FVC estimation.25
Model review : after the FVC model database was initially built, the same staff reviewed and checked each model in reference to its source of origin, which was then cross-checked by multiple reviewers in a random fashion. After all, the dataset was forwarded to experts for final review to ensure a high data accuracy and reliability.
Model quality assessment : deviations could have occurred in the FVC estimation process due to the complexity of data sources, vegetation types and study regions. For remedies, we identified and analyzed the following possible types:
(1) Some of the FVC remote sensing estimation models, which were derived from somehow simpler but more computationally efficient dimidiate pixel models, and occasionally, linear or non-linear models, were only applicable to certain vegetation types of certain areas. This is because these are empirical or semi-empirical remote sensing estimation models built upon a particular data source, vegetation cover, and study area.13
(2) The FVC estimated by the dimidiate pixel model method could be inaccurate both theoretically and physically. Theoretically, the raw data for model input could be inaccurate due to subjective selections of ground sampling points; systemic errors could occur during remote sensing, as a result of sensor inclination, atmospheric instability, etc.; the dimidiate pixel model method has its own deficiencies which could also cause errors during the inversion process. Physically, as FVC parameters are themselves directional, FVC could vary by angle, resulting in possible deviations.
(3) When using the FVC estimation models in the dataset, users should pay special attention to their sources of origin, and they are encouraged to use existing data to further verify the accuracy of the models. For example, this dataset contains multiple FVC models for a single study area and underlying surface, which might indicate the impact of data sources on the model’s quantitative forms.
(4) Statistical analysis errors. Systemic errors could be introduced during data analysis due to lack of consideration on the ecosystem, vegetation type and data source peculiarities of the sampling site, when researchers could have used different methods of data elimination, correction and interpolation. During data collection, we found researchers used varied modelling methods, data sources, and data spatio-temporal resolutions for analysis. In view of this, we partitioned the models as per their underlying surface types to facilitate more nuanced model selection for the users’ convenience. It should be noted that the vegetation index values for pure soil and vegetation pixels obtained by the dimidiate pixel model can vary by study area and vegetation type. This dataset mainly contains models used for annual, monthly and daily scales.
FVC provides a scientific basis for research on ecological construction and sustainable development. By synthesizing existing literature, this study maps the spatial distribution of FVC models for China’s typical ecological zones as per the type of underlying surface, providing data support for the evaluation of ecological pattern, function and quality at both national and regional levels. This is the first FVC model database at ecosystem scale generated from a synthesis of existing research findings, with a geographical focus on the Chinese territory. It includes 115 FVC models which can provide data support for evaluating the benefits of regional ecosystem vegetation, assessing the carrying capacity of ecological environment, carrying out the studies on global changes, or optimizing biogeochemical circulation models. Attention should be paid to the following points when this dataset is used:
(1) There has been no uniform specifications, techniques or methods for FVC model processing. More often, studies used different FVC model estimation methods most suitable to regional conditions, and as a result, each model differed in its construction method and extremum setting. When selecting a model for studying FVC of a particular region, users are suggested to consider their study area and surface vegetation type in the first place. Users’ data processing method should also be in conformity with their chosen model.
(2) FVC is known for its spatio-temporal differentiation.21 This study synthesizes multi-year FVC models across different observation areas, which covers varied data sources and underlying surfaces. When selecting a model for regional FVA estimation, users should pay special attention to their pertaining data sources and underlying vegetation types.
(3) When applied to different spatio-temporal scales, remote sensing inversion products are usually incommensurable, and hence cannot be used interchangably.14 Users are recommended to select FVC models that meet the spatio-temporal resolution requirements of their study.
Thanks go to our colleagues for their contribution to data collection, collation and analysis!
Zhang X, Liao C, Li J et al. Fractional vegetation cover estimation in arid and semi-arid environments using HJ-1 satellite hyperspectral data. International Journal of Applied Earth Observation and Geoinformation 21 (2013): 506 – 512.
Xing Z, Feng Y, Yang G et al. Method of Estimating Vegetation Coverage Based on Remote Sensing. Remote Sensing Technology and Application 24 (2009): 849 – 855.
Zhou G, Zhang X, Gao S et al. Experiment and Modelling on the Responses of Chinese Terrestrial Ecosystems. Acta Botanica Sinica 39 (1997): 879 – 888.
Wang H, Li W, Du G et al. Research on the change of grassland vegetation coverage using 3S technology in Gannan. Acta Prataculturae Sinica 21 (2012): 26 – 37.
Jimenez-Munoz J, Sobrino J, Plaza A et al. Comparison Between Fractional Vegetation Cover Retrievals from Vegetation Indices and Spectral Mixture Analysis: Case Study of PROBA/CHRIS Data Over an Agricultural Area. Sensors (Basel) 9 (2009): 768 – 793.
Lu X, Zhao H, Yang S et al. Vegetation coverage estimated by the environmental satellite CCD image. Geomatics ＆ Spatial Information Technology 38 (2015): 44 – 47.
Zhang Y, Li X, Chen Y et al. Overview of field and multi-scale remote sensing measurement approaches to grassland vegetation coverage. Advance in Earth Sciences 18 (2003): 85 – 94.
Jiapaer G, Chen X, Bao A et al. A comparison of methods for estimating fractional vegetation cover in arid regions. Agricultural and Forest Meteorology 151 (2011): 1698 – 1710.
Jing X, Yao W, Wang J et al. A study on the relationship between dynamic change of vegetation coverage and precipitation in Beijing’s mountainous areas during the last 20 years. Mathematical and Computer Modelling 54 (2011): 1079 – 1085.
Godínez-Alvarez H, Herrick J, Mattocks M et al. Comparison of three vegetation monitoring methods: Their relative utility for ecological assessment and monitoring. Ecological Indicators 9 (2009): 1001 – 1008.
Zhai Y, Zhou Q, Zhang C et al. Study on the changes of the vegetation coverage in Baoji city in recent 15 years. Journal of Baoji University of Arts and Sciences (Natural Science) 36 (2016): 31 – 36.
Jia K, Yao Y, Wei X et al. A review on fractional vegetation cover estimation using remote sensing. Advances in Earth Science 28 (2013): 774 – 782.
Chen H, Jia G, Feng J et al. Remote sensing estimates of key land surface vegetation variables used in climate model: A review. Advances in Earth Science 29 (2014): 56 – 67.
Qin W, Zhu Q, Zhang X et al. Review of vegetation covering and its measuring and calculating method. Journal of Northwest A＆F University(Natural Science Edition) 34 (2006): 163 – 170.
Wang A & Zhao Z. Research advances on the method of estimating vegetation coverage based on remote sensing. Journal of Green Science and Technology (2015): 10 – 15.
Li Y, Jia W, Wei X et al. Fractional vegetation cover estimation in northern China and its change analysis. Remote Sensing for Land and Resources 27 (2015): 112 – 117.
Liang S, Zhang J, Chen L et al. Production and Application of Remote Sensing Products for Global Change. Beijing: Science Press, 2017.
Wang C, Zhang D & Ren W. Comparison of vegetation coverage extracting based on MODIS data. Journal of Atmospheric and Environmental Optics 5 (2010): 457 – 463.
Wang M & Wang W. Research on extraction method of vegetation coverage. Journal of Yellow River Conservancy Technical Institute 25 (2013): 23 – 27.
Li Z, Sun R & Zhang J. Temporal-spatial analysis of vegetation coverage dynamics in Beijing-Tianjin-Hebei metropolitan regions. Acta Ecologica Sinica 37 (2017): 7418 – 7426.
Liu Y, Huang B, Cheng T et al. Vegetation coverage in upper Huaihe River basin based on binary pixel model of remote sensing. Bulletin of Soil and Water Conservation 32 (2012): 93 – 98.
Lou J, Sun Z, Zhou M et al. A remote sensing monitoring of the dynamic change on vegetation coverage in Kailu country, Tongliao city based on NDVI. Journal of Inner Mongolia Agricultural University (Natural Science Edition) 36 (2015): 48 – 57.
Liu Y, Ren Y, Chen T et al. Comparisons of estimating methods of vegetation fraction based on “BJ-1” Microsatellite imagery. Remote Sensing Technology and Application 22 (2007): 581 – 586.
Zhou Z, Zeng Y, Zhang L et al. Remote sensing monitoring and analysis of traction vegetation cover in the water source area of the middle route of projects to divert water from the south to the north. Remote Sensing for Land and Resources 24 (2012): 70 – 76.
Liu X, Liu R, Liu A et al. Study on information extraction and the dynamic monitoring of grassland coverage in three river source area. Acta Agrestia Sinica 18 (2010): 154 – 159.
1. Liu E, Zhou G & Zhou L. Remote sensing models for fractional vegetation cover estimation over different underlying surfaces in China. Science Data Bank, 2018. (2018-11-06). DOI: 10.11922/sciencedb.674.
How to cite this article
Liu E, Zhou G & Zhou L. Remote sensing models for fractional vegetation cover estimation over different underlying surfaces in China. China Scientific Data 4(2019). DOI: 10.11922/csdata.2019.0069.zh.