First Openly Available High-Res Crop Map of China Developed By UMD GLAD Lab
Researchers with the Global Land Analysis and Discovery (GLAD) lab at the University of Maryland have published the first openly available 10-meter resolution maps of maize and soybean crops across China. The developed maps were shown to be consistent with sample-based area estimates across the country and have greater than 90% overall accuracy. Using the European Space Agency’s Sentinel-2 satellite platform, the team at GLAD developed an operational workflow allowing for the creation of publicly available, high resolution, annual crop maps for China. The Director of the GLAD lab Dr. Matt Hansen and GLAD affiliated member Dr. Xiao-Peng Song also serve as the Crop Mapping Co-Leads for the NASA Harvest Consortium.
The maps were created using a workflow that was originally developed for crops within the United States, before being further tested in Argentina, Pakistan, and at the continental scale for all of South America and its most recent application in China. Using a stratified random sampling method, the GLAD team collected 17,858 ground data points to train the machine learning models. They used a probability sample of 450 ground points to validate the resulting maps. From their analysis, the GLAD team estimated approximately 330,609 sq km of maize and 78,107 sq km of soybean within China.
Traditionally, countries have acquired national-scale crop production information using statistical ground surveys. However these methods are labor and time intensive. Crop mapping methods using remote sensing can dramatically reduce these resource requirements and provide a secondary output to compare with survey results. Additionally, the high temporal and spatial resolutions of the Sentinel-2 sensor (combined with its global coverage and free access) allow for numerous data collections to build robust models.
Difference between mapped crop area and planted crop area reported by official statistics at provincial and prefectural level. (a) maize at provincial level; (b) soybean at provincial level; (c) maize at prefectural level; (d) soybean at prefectural level.
With the modeling workflow showing previous success at national and continental scales, the expansion of the methodology to China can improve market transparency and knowledge on maize and soybean production within the country.
Read the full article to learn more about the data and methods of the GLAD research team, their results, and how they plan to continue their work.