State estimates for the tropical Pacific: a reanalysis for evaluating the model, observations, and mass, heat, and salt fluxes

The First Report of TPOS 2020 (Tropical Pacific Observing System 2020, tpos2020.org) recommended several process studies, including Pacific Upwelling and Mixing Physics (PUMP), and Air-sea Interaction at the eastern edge of the Warm Pool (WPEE). We are proposing a modeling and assimilation study in support of these process studies, at the large-scale end of an expected hierarchy of models. The state estimates will form a reanalysis for siting smaller-scale studies and the adjoint model can be used to probe sensitivities that can inform sampling. Data withholding experiments will also be performed to assess the values of different components of the observing systems, as well as their usefulness for testing and improving models. The overall goal of these studies is to improve the ocean models and initializations for better predictability of the coupled Pacific climate system in support of better environmental prediction, to help refine the TPOS 2020 implementation, and to assess an assimilation system as a key component of the observing system.
We will produce a series of overlapping Four-dimensional variational (4D-Var) state estimates for the tropical Pacific covering 2010-2019 at a resolution of 1/3 degree or better. Each estimate is a free forward model run that has had initial conditions, forcing, and other controls adjusted so that it is consistent with observations and can provide diagnoses of mass, heat, and salt fluxes. They will be used to support embedded process studies and to assess regions of good and bad model skill to provide guidance for model improvement and designing observing systems. The model domain will include the entire tropical Pacific, and so will contain both the PUMP upwelling region and the warm pool region. We will work collaboratively with the NOAA labs and other investigators to provide a dynamically-consistent, property-conserving, large-scale context and boundary conditions for the hierarchy of process models as mentioned in the Call.
The state estimates will enforce the model dynamics over assimilation windows of up to 4 months or longer, to be determined experimentally as part of the research. Each fit tests the model as a hypothesis for the dynamical explanation of the observations, and the controls estimated as part of the assimilation process will be examined to identify model errors. Cross-validation will come from comparisons to withheld observations and from forecasts beyond the time range of each estimate. In this way, the assimilation serves as a process experiment by directly testing the compatibility of the proposed dynamics, as quantified in the model, with the observations. The state estimates can be used as part of a diverse set of methods and models to provide the outer context and large-scale budgets for the inner models.