Efficient optimization algorithms for nonlinear least-squares and inverse problems

I will present an on-going project with Simon Tett, Mike Mineter and Kuniko Yamazaki (School of GeoSciences, Edinburgh University) that investigates automatically tuning relevant parameters of a standard climate model to match observations. The resulting inverse/least-squares problems are nonconvex, expensive to evaluate and noisy which makes them highly suitable for derivative-free optimisation algorithms. We successfully employ such methods and attempt to interpret the results in a meaningful way for climate science.