A1658
Title: Generative modelling and stochastic parametrizations for a rotating shallow water system
Authors: Oana Lang - Imperial College London (United Kingdom) [presenting]
Dan Crisan - Imperial College London (United Kingdom)
Alexander Lobbe - Imperial College London (United Kingdom)
Abstract: The stochastic parametrization of small-scale processes is essential in the estimation of uncertainty when trying to represent systematic model errors which arise from small-scale fluctuations (e.g. weather and climate predictions). In a stochastic partial differential model, the noise can be calibrated in a way that is consistent with such subgrid-scale parametrizations using a principal component analysis (PCA). The focus is on the explanation of how the PCA technique can actually be replaced by a generative diffusion model technique - this allows avoiding the imposition of additional constraints on the increments. This methodology is applied to a stochastic rotating shallow water model, using the elevation variable of the model as input data.
