Climate & Computing

Some organisms store information about the climate in their skeletons. By examining these structures, we can get a glimpse into the climatic past. In this project, we modeled the relationship between Oxygen-18 isotopes in organism skeletons and localized temperature. Both our organism data and our 5x5 gridded temperature data came from large NOAA databases.

That is, given the annual Oxygen-18 composition of a coral or tree's skeleton, can we predict the temperature of that year for the 5x5 lat/long grid that organism lived in? We used three models: A simple linear regression, a fully-connected neural network, and a recurrent neural network. The RNN's results were pretty gorgeous, performing better than the models used by paleo-oceanographers today.