Proposal Title: Linear Inverse Modeling for Coupled Analysis and Forecasting in the Unified Forecast System
Principal Investigator: Gregory J. Hakim (U Washington)
Co-Investigators:
Matthew Newman (NOAA/OAR/ESRL and CIRES/CU)
Stephen G. Penny (NOAA/OAR/ESRL and CIRES/CU)
Chris Snyder (NCAR)
ABSTRACT:
Progress on coupled data assimilation (CDA) is limited by the computational burden to conduct experiments. Here we propose a framework for rapidly prototyping novel techniques for this problem using emulators of the forecast model for coupled DA in the Unified Forecast System (UFS). Specifically, we propose to use established methods for linear inverse models (LIMs) in novel CDA experiments with three objectives: (1) constructing coupled atmosphere-ocean LIMs for week 3-4 forecasts; (2) developing a prototype framework for current and future development of CDA in the UFS, and (3) improving upon current week 3-4 forecasts with research results developed in objectives (1) and (2). The computational economy of this approach facilitates a range of experiments with LIMs calibrated on different sources applied to large samples of independent forecasts. We propose to develop LIMs from three sources, two from global forecast models and one from observational products. Four sets of CDA experiments will use these LIMs as the forecast models connecting the assimilation of observational data projected onto the LIM basis. By exploring a range of LIM configurations (variables and time averaging), we will identify the CDA approach that has the highest forecast skill during the target period. Furthermore, the LIM basis will allow us to diagnose the source of forecast skill, so we will be able to identify the strongly and weakly coupled modes of variability that have the highest skill.
The result will be a flexible tool that will likely have more predictive skill at S2S timescales than the parent model, be computationally efficient, allow rapid prototyping of new CDA approaches in the UFS, and identify new directions for cost-efficient future investment in CDA with the parent model. The approaches developed here will likely also be applicable to forecasting for seasonal and longer timescales. Moreover, the results may also provide new scientific insight into predictable coupled modes of variability on S2S timescales.
