Using Initialized Hindcasts to Assess Simulations of 1970–2009 Equatorial Pacific SST, Zonal Wind Stress, and Surface Flux Trends

Abstract

Abstract Initialized decadal hindcasts are used to assess simulations of 1970–2009 equatorial Pacific SST, zonal wind stress, and surface flux trends. Initialized hindcasts are useful to assess how well the models simulate observed trends, as well as how simulations of observed trends (due primarily to natural variability) differ from ensemble-mean forecasted trends (due to the response to an increase in external forcing). All models forecast a statistically significant warming trend in both the warm-pool and cold-tongue regions. However, while the warm-pool warming trend is within the observed estimates, the cold-tongue warming trend is an order of magnitude larger than an ENSO residual estimated using SST instrumental reconstructions. Multimodel ensemble means formed using forecasts 6–10 years from initialization with 40 ensemble members do not produce an unambiguous zonal SST gradient response to an increase in external forcing. Systematic biases are identified in forecasts of surface fluxes. For example, in the warm-pool region all year-1 forecasts produce SST trends similar to observations but ocean mixed layer and net surface heat flux trends with an opposite sign to air–sea datasets. In addition, year-1 forecasts produce positive shortwave feedbacks on decadal time scales, whereas 6–10-yr forecasts produce negative or statistically insignificant shortwave flux feedbacks on decadal time scales, suggesting sensitivity to circulations forced by the initialized ocean state. In the cold-tongue region initialized ensembles forecast positive net radiative flux trends even though shortwave flux trends are negative (i.e., for increasing cloudiness). This is inconsistent with air–sea datasets, which uniformly show that the net surface radiative flux feedback is a damping of the underlying SSTs.