Statistics Seminars: Emulating the output of large climate simulators
3 November 2011 17:00 in CM221
Climate simulators contain large numbers of parameters whose meaning is obscure, as they stand in for missing or uncertain processes: physical, chemical, of biological. Usual practice (eg for the IPCC archive) is to run the simulator only in its standard parameterisation, which presupposes that the contribution to total climate uncertainty of simulator parametric uncertainty is small enough (relative to other sources of uncertainty such as future scenario uncertainty) to be negligible. This is testable, indirectly, by constructing an emulator of the simulator from a training set of runs. This emulator also provides an opportunity to check the code itself ("code verification"), and to screen the parameters to identify the important ones. I illustrate with a multivariate emulator of mid-Holocene summer temperatures anomalies over North America. This is joint work with Tamsin Edwards (Bristol) and Mat Collins (Exeter), as part of the NERC Palaeo-QUMP project.
This is an RSS NE group meeting.
Contact email@example.com for more information