Impact Analysis of Variable Resolution of MPAS on Intrinsic Predictability Using Bred Vectors

Abstract

Variable resolution configuration is a defining feature of the NCAR MPAS (Model for Prediction Across Scales) model, which allows us to smoothly vary the horizontal resolution for taking a closer look at an area of interest. In this study, we aimed to analyze the impact of variable resolution on intrinsic predictability using bred vectors. Thus, the breeding cycles of the MPAS model with and without variable resolution configuration were implemented and tested with two different rescaling intervals of 6 h and 1 day. Rescaling within our breeding cycles were centered by the nature run, thus we could deal with the intrinsic predictability limited only by the initial error growth. We confirmed reasonable estimates of fast-growing errors by bred vectors at two different scales of convective and synoptic systems. We then found that the variable resolution configuration gave consistent improvement of intrinsic predictability not only over the high-resolution area but also outside. A quantitative analysis showed that an improvement with the variable resolution could be found in general for most vertical levels for both rescaling interval experiments. Additionally, we present the computational cost and experience of performing the variable resolution model which would help users in their decisions on this setting.