Stochastic parameterization identification using ensemble Kalman filtering combined with maximum likelihood methods

My colleague Manuel Pulido from Reading University (formerly the university of Buenos Aires) has led an effort to apply and test maximum likelihood methods to estimate model error, and specifically parameters of stochastic subgrid parameterization, with an EnKF. The estimation methods are based on either the expectation–maximization (EM) or the Newton–Raphson (NR) on top of the EnKF. The techniques are successfully evaluted on low-order but quite sophisticated models with stochastic parameterizations, representationg coarse and small-scale dynamics. The proposed methods, the EnKF-EM and the EnKF-NR, are promising efficient statistical learning methods for developing stochastic parameterizations in high-dimensional geophysical models.

The published article is freely accessible here .