Tuning of length-scale and observation-error for radar data assimilation using four dimensional variational (4D-Var) method

Abstract

The effects of tuning of length-scale and observation-error on heavy rainfall forecasts are investigated. Length scale and observation error are tuned based on observation minus background (O - B) covariances and theoretically expected cost function values, respectively. Tuned length scale and observation error are applied to radar data assimilation using the Four Dimensional Variational (4D-Var) method. Length-scale tuning leads to improved Quantitative Precipitation Forecast (QPF) skill for heavy precipitation, better analyses, and reduced errors of wind, temperature, humidity, and hydrometeor forecasts. The effects of observation-error tuning are not as significant as those of length-scale tuning, and they are limited to improvements in QPF skill. This is because tuned observation errors are close to pre-assumed values. Proper tuning of length-scale and observation-error is essential for radar data assimilation using the 4D-Var method.