Assimilation of surface AWS using 3DVAR and LAPS and their effects on short-term high-resolution weather forecasts

Abstract

The progress in data assimilation techniques that incorporate weather observations into high-resolution numerical weather prediction models is challenging because of handling surface data in terrain misrepresentation, balance approximations, instrument errors and sensor representativeness. In the framework of operational numerical weather prediction, two data assimilation systems are compared using conventional observations from surface Automatic Weather Stations (AWS), a three-dimensional variational analysis (3DVAR) and the Local Analysis and Prediction System (LAPS). The goal is to study the ability of these two systems to assimilate data from AWS and to assess which performs better for near-surface wind and temperature fields to initialize a short-range 1-km resolution forecast with the Weather Research and Forecasting (WRF) model. Results show that the 3DVAR assimilation patterns are unrealistic given the inhomogeneous nature of the near-surface fields in complex terrains. In contrast, LAPS analyses show a heterogeneous assimilation pattern, more consistent with the complexity of the terrain and the observations. During the model spin-up period, simulations initialized using both data assimilation methods approach rapidly the control simulation, initialized without assimilation. However, the 1km resolution forecasts initialized with LAPS exhibit a significant improvement, particularly for the wind field module.