Impacts of Assimilating GPS-PWV in Convective-permitting Model on Forecasting Monsoon Precipitation over Arizona Complex Terrain

Forecasting monsoon precipitation over Arizona is challenging due to its complex terrain since the model grid structure may misrepresent topographic details and the sparse observation network is insufficient for initialization of the model at the scale of the topography, particularly the spatial distribution of moisture. Our study aims to evaluate the monsoon precipitation forecast skill over Arizona by conducting an Observing System Experiment (OSE), or “data denial study” using the Data Assimilation Research Testbed (DART) to assimilate Global Positioning System precipitable water vapor (GPS-PWV) into the advanced version of the Weather Research and Forecast (WRF) model. The High-Resolution Rapid Refresh (HRRR) model is used as the initial and boundary conditions. The hourly GPS-PWV data were collected from 30 sites across Arizona characterized by a very nonuniform distribution with clusters of observations separated by large spatial gaps.

The precipitation event of interest occurred on 16 August 2021 with convective initiation developing over the Mogollon Rim in the afternoon and precipitation occurring over Flagstaff, Sedona, and Prescott as the increasingly well-organized mesoscale convective system propagated southwestward to the Arizona-California border. The amount of total precipitation recorded by NOAA’s MRMS (Multi Radar – Multi Sensor) system was 25 - 60 mm within the 12- hour period of 00 UTC 16 August to 12 UTC 16 August. In this study we initiated the forecast at 06 UTC 15 August with 40 ensemble members and assimilated the hourly GPS-PWV data over the 6h period from 1200-1800 UTC, after which we ran a deterministic forecast using the mean ensemble data assimilation analysis at 18 UTC as the initial condition for this “free forecast”.

We discovered that this forecast and assimilation system was sensitive to the specification of the initial state of the atmosphere, the radius of influence in the Ensemble Kalman Filter data assimilation system, and the model physics. Therefore, we tested the simulation using a variety of horizontal and vertical localizations and microphysics schemes to find a configuration resulting in the least-bias PWV. We used this optimal configuration to forecast 24 other precipitation events occurring in the 2021 monsoon season.

Our results show: 1) GPS-PWV data assimilation reduced forecast PWV errors across the model domain. 2) GPS-PWV data assimilation increased instability (due to moistening) of the pre-convective atmosphere over the Mogollon Rim and southeastern Arizona by as much as 1000 J/kg. 3) GPS-PWV data assimilation maintained these more favorable atmospheric conditions for convection and improved precipitation forecasts for at least 6 hours into the free forecast period but became too moist afterward. 4) The results revealed a surprising dry bias of 3-4 mm PWV in the HRRR model (used for initial conditions) compared to the actual GPS-PWV values, and this bias was maintained in the WRF model control run without GPS-PWV data assimilation for at least 18h.