Online Model Parameter Estimation With Ensemble Data Assimilation in the Real Global Atmosphere

When and where will it rain, and how heavy? – this is a central question that meteorology tries to address. Numerical weather prediction (NWP) is a major approach using data assimilation with mathematical models to predict the weather. The NWP models discretize the earth atmosphere into 3-dimensional grids, and compute the evolution of the atmospheric states based on physical processes (e.g., fluid dynamics, radiation, and water phase changes). The NWP models cannot include subgrid-scale and complex processes explicitly, and these complex phenomena are represented by simplified equations so-called “parameterization”. Parameterization usually contains tunable model parameters, and manual tuning of these parameters is a tedious but important task. This study explores an objective and autonomous approach to optimizing these parameters using data assimilation. We chose a parameter of the subgrid-scale parameterization of raindrop initiation processes of a global NWP model. We successfully mitigated the overproduced precipitation of the NWP model by estimating the model parameter with satellite-observed precipitation data.

Figure: Global precipitation forecasts (mm 6h-1) at 0000 UTC on 16 June 2014 by an NWP model. (left) control experiment with the default model setting, (middle) test experiment with the model parameter estimation, and (right) satellite observation, respectively. Overproduced precipitation over ocean in the control experiment is successfully mitigated by the model parameter estimation.