Article ID: 2018-056
The feasibility of regional reanalysis assimilating only conventional observations was investigated as an alternative to dynamical downscaling to estimate the past three-dimensional high-resolution atmospheric fields with long-term homogeneity over about 60 years. The two types of widely applied dynamical downscaling approaches have problems. One with a serial long-term time-integration often fails to reproduce synoptic-scale systems and precipitation patterns. The other with frequent reinitializations underestimates precipitation due to insufficient spin-up. To address these problems maintaining long-term homogeneity, we proposed the regional reanalysis assimilating only the conventional observations. We examined it paying special attention to summer precipitation, through one-month experiment before conducting a long-term reanalysis.
The system is designed to assimilate surface pressure and radiosonde upper-air observations, using the Japan Meteorological Agency's nonhydrostatic model (NHM) and the local ensemble transform Kalman filter (LETKF). It covers Japan and its surrounding area with a 5-km grid spacing and East Asia with a 25-km grid spacing, applying one-way double nesting in the Japanese 55-year reanalysis (JRA-55).
The regional reanalysis overcomes the problems with both types of dynamical downscaling approaches. It reproduces actual synoptic-scale systems and precipitation patterns better. It also realistically describes spatial variability and precipitation intensity. The 5-km grid spacing regional reanalysis reproduces frequency of heavy precipitation and describes anomalous local fields affected by topography such as circulations and solar radiation better than the coarser reanalyses.
We optimized the NHM-LETKF for long-term reanalysis by sensitivity experiments. The lateral boundary perturbations derived from an empirical orthogonal function analysis of JRA-55 brings stable analysis, saving computational costs. The ensemble size of at least 30 is needed because further reduction significantly degrades the analysis. The deterministic run from non-perturbed analysis is adopted as first guess in LETKF, instead of the ensemble mean of perturbed runs, enabling reasonable simulation of spatial variability in the atmosphere and precipitation intensity.