Uncertainty Quantification for Microwave Limb Sounder Retrievals

Uncertainty Quantification for the Microwave Limb Sounder (MLS) is just getting started. Our first task is to build an emulator for the MLS forward full-physics model. An emulator is a fast statistical surrogate for the more computationally-intensive full-physics code. We need the emulator because the Monte-Carlo experiments we anticipate running (e.g., simulation-based uncertainty quantification) require calling the forward model code many times. To be useful, emulators must be fast, accurate, and report their own uncertainties. We built a Gaussian Process emulator for the MLS forward model. Some early results are shown below.


Comparison of an emulated Band 2 radiance prediction to the full forward model prediction (black points) across pressure levels for five channels. Gray bands show emulator uncertainty. Comparison of emulated Band 2 radiance predictions to full forward model predictions (black points) at three pressure levels across all channels. Gray bands show emulator uncertainty. Emulator error across pressure levels and channels for a Band 2 radiance.

Selected Publications

No papers match yet.