Recently, the Forecast Systems Laboratory (FSL) developed an improved version of the Local Analysis and Prediction System (LAPS). As with previous versions, this improved version uses virtually all sources of operational meteorological data, including satellite imagery and three-dimensional radar data, to produce a complete atmospheric analysis in a matter of minutes on a "low-end" (e.g., Linux PCs, small, single-processor Unix workstations) computing platform. In addition, a complete analysis of clouds and precipitation contribute to a diagnosis of water in all phases (vapor, cloud liquid, cloud ice, precipitating ice, rain, and snow) over the entire volume. The LAPS upgrade also includes application of variational constraints which allow the analyzed cloud to impact the mass, moisture, and momentum fields so that the initial fields will sustain the clouds through the first few time steps of model integration. Details of the analysis technique are discussed in companion papers, but the resulting analyses have been shown to be ideally suited for initializing cloud-resolving numerical forecast models.
At FSL, a capability to use these analyses to initialize both the state variables and the microphysical species has been developed and implemented in our real-time system. By initializing the model with these improved LAPS analyses, we show a dramatic improvement in the short range explicit forecasts of clouds and precipitation, without the need for any extended model "nudging" period, which also saves computing resources. Qualitatively, the user of the model output sees clouds and precipitation present in the initial condition that closely match what was actually occurring, and those fields maintain themselves in a realistic manner rather than dissipating due to a lack of dynamic balance. Quantitative verification demonstrates dramatic improvement compared to other initialization methods for the first 0-6 hours of the forecast period and significant improvement throughout 6-12 hour period, after which the skill scores of the various techniques converge due to the domination of the lateral boundary condition errors within our limited domain. These results are significant and have positive implications for the future of short-range mesoscale numerical weather prediction.