At the end of April, the world's first global data set was published, which shows the boundaries of agricultural fields. It is the result of an 18-month campaign by geospatial data specialists from industry and academia. Initiative, which was led by the non-profit organization Taylor Geospatial and the Microsoft AI for Good Lab, created an open and publicly available data set, which can be used to ensure food security, carbon accounting, precision farming and water quality analysis.
For reference: Taylor Geospatial was founded in 2025 year to catalyze the development and commercialization of geospatial AI. The company's focus is on the application of AI, machine learning and computer vision to satellite imagery to create global datasets and publish training data, models and results.
Microsoft AI for Good Lab is a Microsoft research lab (founded in 2018 r.), which uses AI, machine learning models and data to solve global problems. The laboratory works on sustainable development projects, health care’I, protection of human rights, humanitarian aid and liquidation of the consequences of natural disasters.
Taylor Geospatial Vice President of Strategic Innovation Programs Jennifer Marcus said, that the Fields of the World project has identified certain difficulties in the analysis of satellite data around the world with the help of computers and that this process needs to be improved. The initiative started with the joint efforts of geospatial experts from the private sector and academia to develop an educational database. One of the key limitations was the lack of verified data outside the US and EU, since only these regions actively publish government information, suitable for training models. Companies have taken it upon themselves to create a more global training dataset. Participants selected the best model architecture and published the results. As Marcus explained, a special indicator was added to the data set, which honestly shows, in which region the model works well, and where its results are less reliable.
Food and Agriculture Organization of the United Nations (FAO) and NASA Harvest is a group, which promotes the implementation of satellite imagery for food security and agricultural production — is already using this dataset. Marcus said, that the project has moved on to gathering feedback from practitioners: their corrections will help automatically improve model accuracy in the future.
Other global datasets can be created in a similar way. Taylor Geospatial's upcoming project called Features of the World will study the infrastructure, which can be mapped on a global scale.
Source: https://spacenews.com