Climatologist Derrick Muheki, a doctoral student at Vrije Universiteit Brussel in Belgium, had to travel by plane, boat, and motorcycle along dirt roads to reach a remote site in the Democratic Republic of the Congo, where the National Institute of Agronomic Research stores thousands of pages of old climate records at a facility with no access to electricity. The researcher, who took enough batteries with him to keep his digital camera charged, spent two months digitizing pages filled with numerical climate data. Back in Brussels, he extracted the information using a machine-learning tool called MeteoSaver, described in a paper shared on the EGUsphere scientific repository. Initial tests showed the tool read the documents with 75% accuracy, which improved to over 90% with refinement of the neural network used to recognize the handwritten text. The work is an example of how artificial intelligence can help recover climate records that are currently beyond the reach of large-scale analyses (Nature, September 16).
Republish