Clark University’s Graduate School of Geography is developing a scalable, fast, and cost-effective land cover mapping platform that seamlessly fuses human and machine intelligence. This combined human-machine approach aims to overcome many current obstacles to mapping complex agricultural landscapes, and to approach the limits of detail and accuracy that can be achieved with satellite-based mapping.
With the ultimate goal of mapping the distribution and characteristics of cropland across sub-Saharan Africa with unprecedented accuracy, Omidyar Network is supporting Clark University to create a next-generation, open source mapping platform. The platform harnesses the deluge of data collected by commercial satellite fleets and applies human judgement to guide machine learning to better interpret satellite data.
Clark University believes that the job of mapping Africa’s croplands should be done by Africans and that local knowledge leads to better maps. Clark is partnering with a Kenyan-based geospatial company that employs local workers, including residents of informal settlements, to train and test the algorithm. Clark is also working with a Ghana-based land documentation company, to collect field data on the ground to provide an extra level of map validation. By the end of Clark’s grant, the University will demonstrate the capacity to accurately map Ghana’s cropland with this combined approach within 7 days.