CS 6404
Data assimilation: inverse modeling with differential equations

Fall 2015

Adrian Sandu

In solving the foward problem one runs a model (based on the physical laws
that govern the system) to predict (observations of) reality. In solving
the inverse problem one uses observations of reality to infer the
properties of the model. Inverse problems are tremendously important
in many fields, from biology to nuclear engineering to numerical weather
prediction.

This class introduces different computational methods for solving data
assimilation inverse problems, including ensemble Kalman filters,
variational methods, and nonlinear and non-Gaussian techniques. We will
build a strong background in inverse problems by covering topics such as
statistical estimation theory, ill-conditioning and regularization
techniques, differential equations and adjoint modeling, and numerical
optimization.