CMCC Seminar
CMCC-Bo meeting room (2nd floor) and via Zoom
April 22, h. 12:00 CEST


Speaker:
Matt Newman
Atmosphere-Ocean Processes and Predictability Division, NOAA


Abstract:
Seasonal to interannual forecasts made by coupled general circulation models (CGCMs) undergo strong climate drift and initialization shock, driving the model state away from its long-term attractor. Here we explore initializing directly on a model’s own attractor, using an analog approach in which model states close to the observed initial state are drawn from a “library” obtained from prior uninitialized CGCM simulations. The subsequent evolution of those “model-analogs” yields an ensemble forecast, without additional model integration. This technique is applied to a few dozen CGCMs from the CMIP6 database. By selecting from these long control and/or historical (externally forced) runs those model states whose monthly SST and SSH anomalies best resemble the observations at initialization time, hindcasts are then made for leads of 1-36 months since the late 1800’s. Deterministic and probabilistic skill measures of these model-analog hindcasts are comparable to, and in some regions better than, traditionally assimilation-initialized CGCM hindcasts, for both the individual models and the multi-model ensemble; where the model-analog skill is higher, it suggests that CGCM skill is degraded by initialization shock, which may allow for future diagnosis of impactful model errors.
A hybrid method that integrates deep learning with model-analog forecasting is also introduced. This hybrid framework employs a convolutional neural network to estimate state-dependent weights to identify the analog states, yielding about a 10% skill improvement over the model-analog approach alone. It also reveals state-dependent initial error sensitivity linked to various seasonally varying physical processes, including the Pacific Meridional Modes, equatorial recharge oscillator, and stochastic wind forcing.
This study suggests that with little additional effort, sufficiently realistic and long CGCM simulations may offer skillful seasonal to interannual forecasts of global SST anomalies, even without sophisticated data assimilation or additional ensemble forecast integrations. The model-analog method could provide a baseline for forecast skill when developing future models and forecast systems and can provide a practical metric of global climate models and their ability to reproduce nature’s attractor.

The seminar will be held in the CMCC-Bo meeting room (2nd floor) and via Zoom.