Laterite taking geospatial interpolation efforts further
Laterite brought a case study on geospatial interpolation to the the AI4SG seminar at Dagstuhl in February 2022. Rik Linssen (Lead Economist at Laterite) and Jose Rubio Valverde (Associate Economist at Laterite) teamed up with Ruby Sedgwick and Jose Pablo Foch from Imperial College London and Raghu Rajan from the University of Freiburg, to work on a proof-of-concept to predict school dropout rates in Rwanda.
“We have since moved on from this proof-of-concept phase within Laterite’s Geolab. Our Geolab is working to expand its capabilities in applying machine learning techniques to geospatial interpolation as well as selecting proper comparison groups for our analyses. And we’re still doing that together with Ruby”, says Rik Linssen.
Jose Rubio Valverde, who is leading the Geolab adds: “Laterite’s Geolab is combining ground-truth survey data collected during Laterite’s projects, with a range of different data-sources such as census data, administrative data, and geospatial data (e.g., information on land use or wealth proxies stemming from remote sensing). We then overlay a highly granular hexagonal grid across Tanzania, Rwanda, Ethiopia, and Kenia that combines these layers of data.”
Ruby on the problem to which ML might offer a solution: “Laterite’s survey data is usually spatially scattered (i.e., they only field surveys in a limited number of villages in the research projects they run), but there are similarities in the outcomes that this survey data covers. Thus, they have highly detailed information from survey data in some villages and only limited information from their geospatial data for other (neighboring areas). Together with Laterite, I’m now exploring how we can use ML-based interpolation techniques to estimate or interpolate proxies for wealth and poverty for these non-surveyed areas.”
“We are currently in the early stages of developing and testing these models in Rwanda but hope to expand to other countries Laterite works in and broaden this approach to interpolating different outcomes over the next year”, Rik Linssen concludes with a look towards the future.