Zone II • Versions EN1
Abstract: Based on massive ortho-rectified Landsat images, the 30 m Global Mosaic Map (One-Map) of 2018 with consistent color is made using several key techniques like semi-automatic data selection, automatic mosaic, color balancing, etc. The data is projected as WGS84 latitude/longitude projection and saved in GeoTiff format. For users’ convenience, the data is also re-tiled and has inner pyramids pre-built. The valid area of One-Map covers all land from 60°S to 80°N except Greenland, and the candidate scene of each Path/Row is the one with the lowest cloud cover in growing period or the composed of all scenes in 2018 using the median value. Of high geometry precision and great color consistency, this dataset can serve as a base map for applications in Remote Sensing, Virtual Reality, etc.
Keywords: global map; 2018; mosaic; color balance; massive data; Landsat
|Title||30-meters global mosaic map of 2018|
|Data corresponding author||He Guojin (email@example.com)|
|Data authors||Yin Ranyu, He Guojin, Wang Guizhou, Long Tengfei|
|Geographical scope||All land except Greenland (60°S - 80°N)|
|Data volume||1.04 TB|
|Data service system||<http://www.sciencedb.cn/dataSet/handle/865>|
|Sources of funding||Strategic Priority Research Program of the Chinese Academy of Sciences (XDA19090300)|
|Dataset composition||The dataset consists of 354 data files in total. Each file is an image block named after the latitude and longitude in the upper left corner.|
At present, Landsat series satellites can provide global data with a long time-series resolution of 30 meters, and have been widely used in agriculture, forestry and land surveys, ecological environment monitoring, energy and resource management, urban management planning, disaster monitoring and recovery, etc. 1 For large-scale remote sensing data, the data pre-processing is technically difficult, as it requires a lot of computing resources. Especially it would take a lot of time and energy for researchers in the fields of non-remote sensing image processing, The current strong 30-meter resolution global remote sensing image has a wide range of uses in implementing the 2030 Agenda for Sustainable Development, tackling urgent issues concerning climate change, sustainable management of forests and water resources, and virtual reality. The production of publicly shared high-quality large-scale image basemaps, in this sense, is critical in that it frees scientific researchers from the tedious labor of data processing, allowing more time to invest into the main line of scientific research.
This article presents the technical process of making a map of the global land area based on Landsat images, through which to produce a dataset of global map with a resolution of 30 meters in 2018. First, the best image covering the global land area is selected, which is then stretched and reprojected, color-balanced, and mosaicked to generate a global map of 2018 covering all land areas except Greenland (60 ° S–80 ° N). For users’ convenience, it has been divided into blocks and a pyramid has been generated.
2.1 Data collection
The One-Map is mainly based on Landsat 8 satellite OLI sensor data. According to the operation plan of the satellite, it can cover all land areas of 60 ° S–80 ° N 2. Greenland was excluded when making the One-Map. The overall data coverage is shown in Figure 1. Each polygon in the figure represents a Path / Row (PR), also called a Scene, which usually collects 24 images per year, and generating the One-Map involves 9,595 PRs. However, as the mapping process requires images to be as cloudless as possible, only a small proportion of images can be used due to the influence of clouds, snow and ice.
For each PR, the growing season images with minimal clouds are selected, which ensures color consistency in subsequent large-scale mapping. Still, many PRs do not have acceptable images, in which circumstances the annual composite images by Google Earth Engine (GEE) are adopted for widely concerned areas. In general, the images covering China and its surrounding areas come mainly from the data received by the China Remote Sensing Satellite Ground Station, while the rest come from the US Geological Survey website (USGS) and GEE.
2.2 Data processing
The technical route for data processing is shown in Figure 2. It mainly consists of three steps: color preprocessing, re-projection, and mosaicking with color balancing. For PRs that do not meet quality requirements, GEE annual median composite images are used instead.
2.2.1 Color preprocessing
While the Landsat 8 satellite OLI sensor data have a digital number (DN) value ranging between 0–65535, the range required for image mapping is 0–255, which means the original pixel value needs to be mapped to 0–255. In our study, linear stretching is used to convert the DN value range. Because of the varied DN quantization parameters of the scene images, it is impossible for us to stretch the images with a consistent maximum / minimum value. In order to ensure a consistent stretching, the original data is processed into the top of atmosphere (TOA) reflectance, and the minimum value (0) and the maximum value (0.6) are used to linearly stretch all the data. The Red-Green-Blue channel uses short wave infrared 1 (SWIR1), near infrared (NIR) and red (RED) bands of the OLI sensor.
2.2.2 Re-projection and mosaicking with color balancing
Landsat 8 raw data results from the UTM projection, which is not suitable for global mapping. So when making the global mosaic map, we adopted the WGS 84 longitude and latitude projection with a resolution of 0.00025 degrees, and cubic convolution for resampling. The projection is converted by using a spatial data conversion module at the PCI GXL high-performance processing platform.
Still, there will be slight color deviations between adjacent images due to varied sensing periods and different atmospheric conditions, so color balancing is required when mosaicking. For color balancing, the binding algorithm is used. The algorithm first uses “block constraining” to adjust the overall mean and standard deviation of each image, based on adjacent image overlaps; then it automatically calculates dodge points between image pairs and makes local adjustments. As high-quality mosaic lines are critical to the natural transition between mosaic images, the least squared difference is used to determine the mosaic lines. This method calculates the squared difference of the pixel values of all bands between adjacent image overlaps and determines the mosaic line by minimizing the squared difference.
Finally, the mosaic output is divided into blocks. The next section gives more details on the block and naming rules.
For the convenience of users, the One-Map is saved in GeoTiff format, divided into 10 ° × 10 ° (40000 × 40000) blocks. It contains three bands, corresponding to SWIR1, NIR, and RED of the OLI sensor, with a resolution of 0.00025°. To facilitate image browsing, a pyramid has been built into the TIF file. The naming format of the files is shown in Figure 3.
The data used to make the One-Map consist of L1TP class products (66.95%), L1GT class products (3.23%), and GEE median composite images (29.82%). The distribution of each type is shown in Figure 4, respectively. According to the USGS report, up to February 28, 2019, 92.2% of the L1TP products, among all archived satellite data, have RMSEs less than or equal to 12 meters 4. GEE median composite images will have some spatial details missing, and their positioning accuracy is close to that of L1TP products. The One-Map shows a good color consistency after color balancing. Figure 5 is an overall view of the map.
This article is the first in its kind to launch a global map of 2018 with a resolution of 30 meters. The map is of high geometric accuracy and good color consistency. It can be used as a base map for remote sensing applications, virtual reality, etc. It is also useful for agricultural, forestry and mining surveys, ecological environment monitoring, energy and resource management, urban management planning, disaster monitoring and recovery, global change research, implementation of the Belt and Road Initiative, and actions of the “2030 Agenda for Sustainable Development”.
Thanks go to our colleagues for their assistance in exploring the data selection method: Gong Chengjuan, Guo Yantao, He Haipeng, He Chen Linqiu, Huang Liting, Leng Wanchun, Li Mozhen, Pu Dongchuan, Sun Jiayue, and Zheng Qin.
Jiang G, Han B, Gao Y et al. Review of 40-year earth observation with Landsat series and prospects of LDCM. Journal of Remote Sensing 17 (2013): 1033-1048. DOI: 10.11834/jrs.20132296.
Wulder MA, White JC, Loveland TR et al. The global Landsat archive: Status, consolidation, and direction. Remote Sensing of Environment 185 (2016): 271-283.
Gorelick N, Hancher M, Dixon M et al. Google Earth Engine: Planetary-scale geospatial analysis for everyone. Remote Sensing of Environment 202 (2017): 18-27.
1. Yin R, He G, Wang G et al. 30-meter Global Mosaic Map of 2018. Science Data Bank, 2019. (August 29, 2019). DOI: 10.11922/sciencedb.865.
How to cite this article
Yin R, He G, Wang G et al. 30-meter global mosaic map of 2018. China Scientific Data 4(2019). DOI: 10.11922/csdata.2019.0048.zh.