We use AI to map land use at regional and national scales.
Our workflow starts with the aggregation and alignment of multiple satellite imagery sources, targeted to the region of interest (ROI). The imagery is annotated using Geosurvey, a scalable web platform for visual annotation, in which experts rapidly outline and classify land use across a random subset of the ROI. The resulting data is used to train multiple machine learning models to infer relationships between land cover use and satellite imagery. Top performing models are then extrapolated across the ROI and also validated by ground teams, to produce estimated cropland maps at 10 meter resolution.
Cropland maps are essential to the development of agricultural intensification and surveillance. Example activities that are informed by cropland maps include soil sampling expeditions, plant health surveillance, agro-advisory extension programs, placement of agri-input distributors and retailers, and budget resource allocation for ministries of agriculture.
Building footprints at national scale provide essential intelligence for a broad swath of industries. Utility companies rely on building footprints for electrical grid planning. Urgent vaccine delivery requires building footprints for logistics planning. Agricultural businesses use building footprints, in conjunction with cropland and forestry maps, to plan expansion strategies.
Time series of remote sensing layers enables the visualization and prediction of spatial-temporal trends in urbanization and agricultural intensification, providing key market intelligence for governments and emerging industry players.
Nigeria has the largest workforce and economy in Sub-Saharan Africa (SSA), possessing over 25% of Africa's population. With the recent plunge in the price of oil, Nigeria is urgently seeking strategies to diversify its economy. Strong attention is being placed toward Nigeria’s agricultural sector, but many basic indicators necessary for strategic investment, such as the area of croplands under production, remain missing. By combining artificial intelligence with agronomic expertise and ground-based validation campaigns, QED has produced up-to-date maps of croplands across all 36 states of the federation, along with derived statistics that can be used to guide fertilizer policies and resource allocation. Above is an interactive slider showing a sample of our cropland maps, which exhibit an accuracy and precision exceeding 85%.
QED is at the forefront of R&D in automated field boundary detection, particularly for smallholder farming systems in Africa and Asia with small plots and high spatial variation. Field boundaries are critical for generating maps of crop types, fallow land, and estimated yields. They also support land registry and titling efforts, which are still missing in many countries. Above is an interactive example showcasing our automated field boundary detection.
Building footprints are critical for planning interventions in the energy sector and health sectors, such as for optimizing the efficient layout of power grids in rural areas in need of electrification, and designing sampling frames and randomized control trials for medical interventions. Above is an interactive example from Malawi where you can see auto-detected building footprints highlighted in teal, and surrounding croplands highlighted in magenta.