Zone II • Versions EN1
Abstract: Land use/cover data reflect the cultivation and coverage state of land surface. They are the bases for geographic and ecological research. To produce land use/cover data of the Loess Plateau, this study employs Landsat 8 OLI satellite images with a cloud coverage of less than 10%. The Loess Plateau region is divided into sub-regions by using Esri ArcGIS according to the availability of qualified satellite imageries. Object-oriented classification is used to extract land use/cover classes assisted by eCognition 8.7 software. After multi-scale segmentation, a collection of parameters are established as classification traits, including metrics on objects' adjacent relationship, spectrum, texture and shape. Based on these parameters, a rule set is constructed for conducting an automatic classification and a manual check and modification. Classification accuracy is assessed by using 1,028 field samples. Results show that the data set has an overall accuracy of 85.4% with a corresponding Kappa coefficient of 0.807. The classification results for each sub-region are exported as vector data in shapefile format. Then a vector-grid conversion is conducted through Arc/info AML procedure. Stitching operation on all obtained grid data is accomplished by using ArcGIS. The land use/cover grid data of the Loess Plateau, with a resolution of 250m, are generated as results of a subsequent clipping and resampling procedure.
Keywords: Loess Plateau; land use; land cover; Landsat 8; object-oriented classification; grid data
|English title||Grid data on land use and land cover of the Loess Plateau region (2015)|
|Corresponding author||Liu Yu (email@example.com)|
|Data authors||Zhu Yuan, Liu Yu, Zhao Liang|
|Geographical scope||The dataset covers the whole Loess Plateau in China (33°43′N – 41°16′N, 100°54′E – 114°33′E).|
|Spatial resolution||250 m||Data volume||16.1 MB|
|Data format||Geotiff||Time range||2015|
|Data service system||<http://www.sciencedb.cn/dataSet/handle/404>|
|Sources of funding||Major Program of the National Natural Science Foundation of China – "Spatial Pattern of Ecological Carrying Capacity and Its Adjustment on the Loess Plateau" (Grant No. 41390464)|
|Dataset composition||This dataset includes land use/cover data of the Loess Plateau in 2015 and corresponding shapefiles about the ground check points. The dataset contains the following files:|
(1) "ld2015_v1_250m.zip", which includes land use/cover data of the Loess Plateau in 2015, in ESRI grid format.
(2) "验证点[Verification Points].zip", made up of data on ground check points in shapefile format which are sampled for land cover/use classification validation. In the .dbf table, the field "class_name" records class codes for land use/cover types.
(3) "code_I.xlsx", which includes class codes and Chinese class names for the land use/cover data.
Land use/cover change (LUCC) is one of the direct causes of regional- to global-scale ecosystem changes. It has significant effects on water cycling, environment quality, biodiversity, net primary production of terrestrial ecosystems and the ecosystems' adaptability to climate changes. LUCC hence stands at the center of global change studies.1 Land use/cover (LUC) constitutes the fundamental data for researches on large-scale carbon cycle, water cycle, biodiversity assessment and ecosystem service assessment.2–4 High-quality LUC data are thus critical for generating credible scientific results in these fields. In 2014, a major program of the National Natural Science Foundation of China – "Spatial Pattern of Ecological Carrying Capacity and Its Adjustment on the Loess Plateau" (Grant No. 41390464) – was launched to explore the interaction between vegetation and hydrology at regional scale on the Loess Plateau, China. One of the fundamental tasks of this program is to produce LUC data covering the whole Loess Plateau. It is expected to provide land cover parameters for modelling water resource states and changes and for assessing the vegetation carrying capacity of regional hydrologic systems.
Previous studies by the Chinese Academy of Sciences have produced LUC datasets covering the whole terrestrial territory of China. These data are recorded at a temporal interval of 5 years from 1980s to 2010, and has a resolution of 30 m.5 Recently, an LUC dataset named "Globeland30" was released by the Geographical Information Center of China.6 To date, there have been six LUC datasets for open access globally, which are DISCover (USGS), Global MODIS LUC with a resolution of 1 km (Boston University), GLC2000 (European Commission Joint Research Center), GlobCover2 (European Commission Joint Research Center), and Globeland30-2010 (Geographical Information Center of China). However, most of these datasets are produced from the spectral information of satellite images at pixel level and follow a traditional supervised classification method, which have clear disadvantages in removing "pepper and salt" effects.7 Spatial relationship among objects on land surface and relevant geographical knowledge are often ignored. Besides, the classification is all too often inconsistent because of too much manual intervention, which leads to a rather low data accuracy.8 In our product, we depend on rule sets to conduct object-oriented classification based on the spectral characteristics of objects and their spatial arrangement. Then manual revision is adopted. By this approach, images are segmented into objects of relatively homogeneous spectral characteristics and textures for further classification. Both spectral information and the spatial relationship among objects are included for promising a higher accuracy.
2.1 Geographic coverage of the data
The Loess Plateau is located in middle of the Yellow River basin, with a latitude and longitude range of 33°43′N – 41°16′N and 100°54′E – 114°33′E, respectively. It neighbors Mt. Qin to the south and Mt. Yinshan to the north. It reaches Taihang Mountain in the east and is bordered by Wuqiao Mountain and Riyue Mountain in the west. It covers 287 counties of Shannxi, Shanxi, Inner Mongolia, Ningxia, Qinghai, Gansu and Henan provinces. The Loess Plateau has a total acreage of 620,000 km2, accounting for 6.69% of China's terrestrial territory. Pushed by ecosystem remediation and restoration policies, as well as rapid economic development since the 1990s, the Loess Plateau stepped into an era featured by large-scale and rapid revegetation, which reshaped regional land use/cover significantly.9 As a result, the Loess Plateau became a hotspot for LUCC studies in China.
2.2 Sources and features of the imageries
This land use/cover dataset is derived from imageries obtained by Landsat 8 OLI sensor. It has a temporal range of 2014 – 2015, and a resolution of 30 m. In total, 117 imageries were collected during winter (Dec. 2014 – Feb. 2015), summer (Jun. 2015 – Sept. 2015), and autumn (Oct. 2015). The spatial coverage of the imageries is shown in Figure 1. All the imageries are collected by National Aeronautics and Space Administration (NASA) of the US and downloaded from the USGS website (http://earthexplorer.usgs.gov).
2.3 Data processing
Object-oriented methodology is used for land use/cover classification, where eCognition8.7 is taken as the platform. We adopt an approach combining ruleset-based automatic classification and manual revision. Specific steps for the land use/cover data production are given below.
(1) Subregion division. Before land use/cover classification, ArcGIS is employed to divide the whole Loess Plateau region according to the availability of qualified imageries, wherein cloud coverage is a critical metric for qualified image selection.
(2) Multi-resolution image segmentation. Before classification, imageries are segmented into polygons (objects) with relatively homogeneous spectral and textural characteristics. A threshold of scale factor is obtained through test, which is then used in image segmentation. Results show that 70 is the most suitable scale factor in the classification of this data set.
(3) Classification traits establishment. Credible land use/cover classification relies on effective traits to distinguish and classify objects. In this study, indicators of spectral characteristics, texture, shape and adjacent relationship are selected to establish the ruleset. NDVI, mean value and standard variation of pixel DN within objects, GLCM homogeneity, length of shared border, and shape index of objects are included as classification metrics.
(4) Land use/cover classification. A preliminary list of land use/cover categories are extracted after multi-resolution image segmentation according to the spatial pattern, spectral and textural characteristics of objects using a rule set based on classification metrics. Then manual revision is conducted.
(5) Accuracy evaluation. Classification results of this study are validated using the confusion matrix and the Kappa coefficient. 1,028 sampling points, collected through ground survey in 2016, are used in data accuracy validation. Validation is performed by combining the high quality imageries of 2015 and real land use/cover categories. Patches with an area greater than 3 pixels × 3 pixels (8100 m2) and without changes in land use/cover types during 2015 – 2016 are selected. In the map, points are located in the center of patches.
(6) Esri grid data generation. Land use/cover data of each sub-region are exported from eCognition 8.7 in shapefile format. Then all the shapefiles are converted into Esri grid format with a spatial resolution of 30 m by using Arc/info.
(7) Clipping and mosaicking. Esri grid data of all the sub-regions are mosaicked and clipped using the "mosaic" and "clip" modules in the ArcGIS Data Management to produce a Geotiff file. Data are then resampled by the Majority method to produce a 250 m Geotiff dataset.
3.1 Description of the classification system
When establishing the data classification system, we referred to the National Ecosystem Survey and Assessment of China (2000 – 2010),10 and adopted the seven land use/cover types, i.e., forest land, shrub land, grassland, arable land, built-up area, waterbodies and other land (Table 1).
|1||Forest||Land covered by natural or artificial woodland with a canopy density greater than 30% and a height greater than 2 m|
|2||Shrub||Land covered by primitive or secondary shrubs or bushes with a coverage greater than 20% and a total vegetation coverage no less than 30%|
|3||Grassland||Land whose vegetation is dominated by herb species, with a total vegetation coverage greater than 20%|
|4||Arable land||Land used for grain or other food cultivation, including irrigated and non-irrigated farmland|
|5||Built-up area||Land used for residential habitats, commercial and manufacturing industries, traffic infrastructures, etc.|
|6||Waterbodies||Land covered by natural waterbodies or man-made ponds and reservoirs|
|7||Other||Land with a maximum seasonal vegetation coverage less than 20%, including barren areas, deserts and unused land|
3.2 Data samples
Traditional land use/cover mapping through satellite imageries mainly depends on the spectral information of land objects. It doesn't take full use of other information of the imageries. For example, relationship among adjacent pixels or patches (land objects), shape of the patches, and geographical information of the objects are often ignored, which limit the accuracy of classification.8 This study follows the object-oriented classification approach. This approach makes use of not only the spectral information of the imageries but also the spatial relationship and arrangement of land objects. This methodology guarantees the reliability of the resulted land use/cover dataset (Figure 2).
Ground sampling is conducted to collect samples for evaluating the accuracy of the land use/cover dataset. Randomness and representativeness are two principles for the sampling route design, which ensure a full coverage of the land use/cover types as well as enough samples for each type. The route goes through the core areas of the Loess Plateau. It starts at Taiyuan, passes through Xinzhou, Yulin, Erdos, Yan'an, Yinchuan, Qingyang, Tongchuan, Linfen and Jinzhong, and ends in Taiyuan. Besides, we collected land use/cover information of 78 vegetation survey plots.11 Totally, 1,028 samples were collected. These samples cover all land use/cover types across the Loess Plateau (Figure 3).
The selected samples consist of 277 forest sites, 70 shrub sites, 118 grassland sites, 379 arable land sites, 154 built-up sites, 20 waterbody samples and 10 barren and desert samples. The result of accuracy assessment is given by Table 2.
|Land use/cover||Forestland||Shrub Land||Grass land||Arable land||Built-up area||Waterbodies||Other land||Sum||Producer accuracy|
(Total accuracy = 85.4％; Kappa coefficient = 0.807)
In Table 2, user accuracy of each type is defined as the ratio of samples correctly classified to total samples used (indicated by columns); producer accuracy indicates the proportion of objects of each type that match correct samples among total samples located in this type (indicated by rows). Total accuracy refers to the ratio of samples of all types correctly identified to the number of samples used. Kappa coefficient delineates the consistency of classification, wherein 1.0 indicates the highest consistency. As shown in Table 2, this land use/cover dataset has a good quality with a total accuracy of 85.4% and a corresponding Kappa coefficient of 0.807.
This dataset is valuable for assessing ecosystem functions and services over the Loess Plateau, particularly for researches on carbon sequestration, landscape pattern evolution and its impacts on hydrology and water resource. Since this dataset focuses on the status of land surface cover with a classification system consistent with the National Ecosystem Survey and Assessment of China (2000 – 2010), it can be used as basic data for similar assessments on the Loess Plateau and its sub-regions. It should be noted that this dataset is suitable for studies involving a spatial scale of over 10000 km2 because of its 250-m spatial resolution.
This work is sponsored by the National Natural Science Foundation of China (Grant No. 41390464). We give our appreciation to Miss Dou Yuehan from Wageningen University for polishing this paper.
Li X. A review of the international researches on land use/land cover change. Acta Geographica Sinica 51 (1996): 553 – 558.
Liu JY, Shao QQ, Yan XD et al. Geobiophysical effects of land use change on climate change. Chinese Journal of Nature 36 (2014): 356 – 363.
Fu BJ & Zhang LW. Land-use change and ecosystem services: Concepts, methods and progress. Progress in Geography 33 (2014): 441 – 446.
Verburg P, Neumann K & Nol L. Challenges in using land use and land cover data for global change studies. Global Change Biology 17 (2011): 974 – 989.
Liu JY, Kuang WH, Zhang ZX et al. Spatiotemporal characteristics, patterns and causes of land use changes in China since the late 1980s. Acta Geographica Sinica 69 (2014): 3 – 14.
Du GM, Liu M, Meng FH et al. Fine classification method study of large-scale land use/cover based on geoscience knowledge. Journal of Geo-information Science 19 (2017): 91 – 100.
Gong P, Wang J, Yu L et al. Finer resolution observation and monitoring of global land cover: First mapping results with Landsat TM and ETM+ data. International Journal of Remote Sensing 34 (2013): 2607 – 2654.
Aguirre-Gutiérrez J, Seijmonsbergen A & Duivenvoorden J. Optimizing land cover classification accuracy for change detection, a combined pixel-based and object-based approach in a mountainous area in Mexico. Applied Geography 34 (2011): 29 – 37.
Lü Y, Fu B, Feng X et al. A policy-driven large scale ecological restoration: Quantifying ecosystem service changes in the Loess Plateau of China. PLoS ONE 7 (2012): e31782.
Zhang L, Li X, Yuan Q et al. Object-based approach to national land cover mapping using HJ satellite imagery. Journal of Applied Remote Sensing 8 (2014): 464 – 471.
1. Zhu Y, Liu Y & Zhao L. Grid data on land use and land cover of the Loess Plateau region (2015). Science Data Bank. DOI: 10.11922/sciencedb.404
How to cite this article
Zhu Y, Liu Y & Zhao L. Grid data on land use and land cover of the Loess Plateau region (2015). China Scientific Data 2 (2017). DOI: 10.11922/csdata.170.2017.0137