Technical achievement and summary of the Geo-AI Challenge cropland extent mapping 2023

Cropland extent serves as a critical determinant for advancing various Sustainable Development Goals (SDGs), and satellite-derived data has been widely used to generate cropland extent maps. To further address the global mission of high-resolution cropland extent mapping, the Food and Agriculture Organization (FAO) and the United Nations Office on Drugs and Crime (UNODC) have proposed "cropland extent mapping" as a focal theme for the 2023 Geo-AI Challenge. Organized by the International Telecommunication Union (ITU) in collaboration with the Zindi platform, tasks of this challenge are divided into two components: (1) annual cropland extent mapping in Sudan and Iran, and (2) temporal cropland mapping in Afghanistan, with pre-provided training samples across all three test regions. Throughout the challenge duration, a total of 74 participating teams submitted their solutions, with the top five teams selected based on classification accuracy of cropland extent maps, innovative methodology, and effectiveness of oral presentation. Several key scientific questions were addressed during the challenge, including optimal classification feature selection, comparative analysis of diverse Machine Learning (ML) models, and fine-tuning of ML algorithms. Importantly, all datasets, scripts, and technical reports resulting from the challenge are openly accessible to the public domain, thereby fostering collaborative advancements within the agricultural remote sensing community.