Skillful Coupled Atmosphere-Ocean Forecasts on Interannual to Decadal Timescales Using A Linear Inverse Model

Abstract:

Improvements to forecasts on interannual to decadal timescales face two major challenges: (1) consistently initializing the coupled system so that variability is not dominated by initial imbalances, and (2) having a large sample of different initial conditions on which to test forecast skill. The second challenge requires consideration of time periods not only outside the recent period of intensive ocean observation, but also before the instrumental era, which increases the importance of the first challenge. Forecasting atmospheric and oceanic conditions prior to the 1850s also isolates internally generated sources of variability by removing the majority of anthropogenic forcing, yet the sparse observational record cannot capture low-frequency variability, further emphasizing the importance of both challenges and paleoclimate proxy data.

This research addresses these two challenges by using a multivariate linear inverse model (LIM) and recent data assimilation (DA) results that extend the observational record with annually-resolved atmospheric and oceanic variables via a low-cost forecast that taps into ocean memory. The reconstructions provide data throughout the last millennium to initialize, validate, and calibrate the LIM. This work tests the forecast skill of LIMs trained on GCM simulations and on paleo-data assimilated reconstructions. Forecasts are initialized and verified on the reconstructions over 1000-2000. Both the DA and GCM-analog LIMs are found to have skill on interannual to decadal timescales that surpasses damped persistence. In addition, all LIMs have positive spatial skill for forecasted variables (maximized over the tropical Pacific and Southern Indian Ocean) as well as high out-of-sample skill scores when validated against instrumental temperature data, with correlations above 0.8 for one-year forecasts.