Radar Data Assimilation for Wintertime Atmospheric River Related Precipitation Events

Landfalling atmospheric rivers (ARs) during winter along the US West Coast can have significant societal and economic impacts. Improving the predictability of these events and their associated cascading impacts, such as flash flooding, debris flows, and wind gusts, is crucial for providing accurate and timely warnings to the region. Accurate heavy precipitation forecast in the short range (0–18 h) is critical for supporting water resource management decision-making. However, the precipitation forecast skill in the short range over states along the western US is still lower than that over the eastern US, which may be due to complex terrain and observation gaps. Radar data assimilation (DA) is essential to improve short-range precipitation forecasts given its high spatial and temporal resolution.

This study aims to investigate the impact of radar data assimilation on short-range quantitative precipitation forecasts through two high-impact AR cases (one in January and the other in October 2021) over California, using a new radar DA modeling framework developed at the Center for Western Weather and Water Extremes (CW3E). The new DA framework has been built upon the Gridpoint Statistical Integration (GSI) to assimilate radar data, including both the radial winds and reflectivity. Direct assimilation of radar reflectivity is achieved through the augmenting control variables in GSI. The modified GSI system is applied to generate analysis for the CW3E’ regional model called West-WRF, which is based on the Weather Research and Forecast (WRF) and developed for the US West weather predictions. The radial winds are from the Next Generation Weather Radar (NEXRAD) level II data, while the reflectivity is 3D gridded national reflectivity mosaics from the Multi-Radar Multi-Sensor (MRMS).

Model outputs are validated using the MRMS Pass 2 precipitation product, California-Nevada River Forecast Center quality-controlled rain gauge observations, and Stage-IV precipitation. It's worth noting that significant inconsistency in precipitation amount and location was observed between Stage-IV and MRMS products, highlighting the need for independent, reliable precipitation observations from multiple sources to validate hourly forecasts. Five intermittent DA experiments with a 6-h interval are conducted for each case, which spans the heavy precipitation period. In eight out of ten pairs of experiments, the radar DA experiments generally improve the precipitation forecast in the first hour compared to the experiments without radar DA. In two pairs of experiments, the precipitation forecasts in radar DA experiments are degraded due to further underestimation, and tuning key components (e.g., observation errors, ensemble sizes, localization scales) in the data assimilation process does not fully resolve the issue. Analysis of the simulated and observed reflectivity revealed discrepancy, that composite reflectivity in observations is smaller than that in the background even for regions where observed precipitation is larger than simulated precipitation, which may have contributed to the precipitation degradation through the reduction of hydrometeors in DAs. To test the hypothesis, two more sets of radar DA experiments are performed that only either negative or positive reflectivity innovations are assimilated, and results reveal that the degradations are mainly a result of assimilation of negative innovations. Another set of experiments in which reflectivity is adjusted so that median reflectivity in the observation is substituted by its counterpart in the background, further corroborates the hypothesis. More AR cases will be studied in the future to test the general applicability of this hypothesis and optimize the DA system.