The Bayesian computer model calibration method has proven to be effective ina wide range of applications. In this framework, input parameters are tuned bycomparing model outputs to observations. However, this methodology becomescomputationally expensive for large spatial model outputs. To overcome thischallenge, we employ a truncated basis representations of the model outputs. Wethen aim to match the model outputs coefficients with the coefficients fromobservations in the basis representations; we also optimize the truncationlevel. In a second step, we enhance the calibration with the addition of theINLA-SPDE technique. We incorporate the nonstationary behavior and thederivative information of the spatial field into the calibration by insertingtwo INLA-SPDE parameters into the calibration. Several synthetic examples and aclimate model illustration highlight the benefits of our approach for modeloutputs distributed over the plane or the sphere.
展开▼
