Relating CMIP5 Model Biases to Seasonal Forecast Skill in the Tropical Pacific

We examine links between tropical Pacific mean state biases and El Niño/Southern Oscillation forecast skill, using model‐analog hindcasts of sea surface temperature (SST; 1961–2015) and precipitation (1979–2015) at leads of 0–12 months, generated by 28 different models from the fifth phase of the Coupled Model Intercomparison Project (CMIP5). Model‐analog forecast skill has been demonstrated to match or even exceed traditional assimilation‐initialized forecast skill in a given model. Models with the most realistic mean states and interannual variability for SST, precipitation, and 10‐m zonal winds in the equatorial Pacific also generate the most skillful precipitation forecasts in the central equatorial Pacific and the best SST forecasts at 6‐month or longer leads. These results show direct links between model climatological biases and seasonal forecast errors, demonstrating that model‐analog hindcast skill—that is, how well a model can capture the observed evolution of tropical Pacific anomalies—is an informative El Niño/Southern Oscillation metric for climate simulations. Research institutions around the world have developed computer models to conduct simulations of Earth's climate to support the Intergovernmental Panel on Climate Change Assessment. Therefore, it is necessary to assess aspects of these models using historical observations. Here, we develop a new model metric that involves making forecasts over several seasons, using existing long simulations of Earth's preindustrial climate conducted by many research institutions and provided to the community for use in climate studies. Within each simulation, we find the best matches, or “model‐analogs,” to historical observations of tropical Indo‐Pacific Ocean surface temperatures and sea level. The evolutions of these analogs over the next several months within the simulation are then used as forecasts. Long‐term accuracy of the forecasts indicates how well a model reproduces the time evolution of the observations. Precipitation forecast accuracy (regardless of lead) and sea surface temperature forecast skill (at 6‐month or longer leads) emerge as key identifiers of the models that best simulate tropical Indo‐Pacific climate. Skill of monthly ENSO hindcasts (1961–2015) of 28 CMIP5 models, determined from model‐analogs, is related to tropical Pacific model biasesModels with poorer precipitation hindcast skill generally have larger cold and dry biases in the equatorial cold tongue regionModel‐analog ENSO hindcast skill may be a useful metric for CMIP6 models, given its correspondence with CGCM mean and variance errors