EGU22-10892
https://doi.org/10.5194/egusphere-egu22-10892
EGU General Assembly 2022
© Author(s) 2022. This work is distributed under
the Creative Commons Attribution 4.0 License.

Mining Large Climate Model Datasets to Make Multi-Year Initialized ENSO Forecasts with Actionable Skill

Matthew Newman1, Hui Ding1, Jiale Lou1, Sam Lillo1, Michael Alexander2, Andrew Hoell2, and Andrew Wittenberg3
Matthew Newman et al.
  • 1University of Colorado, CIRES, NOAA/PSL, Boulder, United States of America
  • 2NOAA/PSL, Boulder, United States of America
  • 3NOAA/GFDL, Princeton, United States of America

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 CGCMs either used operationally by NCEP or as part of the CMIP6 dataset. By selecting from these long control 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 during 1958-2019. Deterministic and probabilistic skill measures of these model-analog hindcasts are comparable to, and in some regions better than, traditionally assimilation-initialized CGCM hindcasts after 1982, for both the individual models and the multi-model ensemble.

On average, ENSO skill of AC>0.5 exists for forecast leads of 18 months for forecasts initialized in summer. More important, we find that not only were some notable ENSO events predictable two years (or more) ahead of time, but that we can both identify forecast “hits” and avoid “false alarms” -- at the time of forecast -- by using a simple forecast signal-to-noise metric (SNR; root-mean-squared ensemble mean divided by ensemble spread), determined from the large (O(100) member) model-analog ensemble. That is, our analog ensemble approach can be used to make actionable ocean predictions, where forecasts of opportunity can be identified well in advance.

Since these long-lead hindcasts do not require full-field initialization, they have also been extended back prior to 1900. We find that while there has been considerable multi-decadal variation in seasonal ENSO skill, there has been no long-term trend for leads up to about 6-9 months. However, while multi-year ENSO skill appears to have also occurred in the past for a few large ENSO events, in the past thirty years it has occurred with considerably greater frequency, raising the possibility that it is a more recent phenomenon.

How to cite: Newman, M., Ding, H., Lou, J., Lillo, S., Alexander, M., Hoell, A., and Wittenberg, A.: Mining Large Climate Model Datasets to Make Multi-Year Initialized ENSO Forecasts with Actionable Skill, EGU General Assembly 2022, Vienna, Austria, 23–27 May 2022, EGU22-10892, https://doi.org/10.5194/egusphere-egu22-10892, 2022.

Displays

Display file