Investigating the Performance of Convection-Allowing Hindcast Simulations during the North American Monsoon with and without GPS-PWV Data Assimilation

Abstract

The North American monsoon (NAM) is responsible for summer severe weather in the Southwest U.S. and northwest Mexico and its associated rainfall contributes the highest percentage of yearly precipitation to this region. Short-term convection-allowing model forecasts have shown difficultly in replicating the diurnal cycle of NAM convective precipitation. Generally, convection initiating over the Sierra Madre Occidental (SMO) mountain range in the early afternoon may later organize into mesoscale convective systems (MCSs) that propagate west towards the lower elevations and Gulf of California (GOC). MCSs account for the greatest proportion of all NAM-related precipitation. In my dissertation, I investigate the performance of daily short-term WRF hindcasts on moisture and precipitation with and without the assimilation of precipitable water vapor (PWV) measurements from Global Positioning System (GPS) ground receivers in the NAM GPS Transect Experiment 2013. In Chapter 2.1, I investigate the overall performance of daily hindcasts during the 2013 monsoon season without data assimilation. The 2.5-km convection-allowing hindcasts consistently display a moist bias in their initial conditions compared to GPS-PWV observations; this leads to diurnal convection beginning 3-6 hours earlier than observations. Because the precipitation forecast skill varies with the proximity of an inverted trough (IV), I compare the days when an IV is present (“strongly forced”) to days when an IV is not present (“weakly forced”). I find that strongly forced days display higher precipitation forecast skill than weakly forced days especially in the slopes of the northern SMO west of the crest that is associated with MCSs. In a case study spanning 8-9 July 2013, when nearly identical MCSs evolved over northern Sonora on consecutive days, the MCS is poorly simulated on the first day (weakly forced) when the IV is east of the SMO while a salient MCS is simulated on the second day (strongly forced) when that IV is over the SMO. I find a greater ensemble-based sensitivity to the initial specification of PWV for the weakly forced day when compared to the strongly forced day. Therefore, GPS-PWV data assimilation has the potential to benefit weakly forced days the most. In Chapter 2.2, for the weakly forced day (8 July 2013), I explore the impact of ensemble data assimilation of GPS-PWV observations to the model fields and the hindcast simulation of an MCS that occurred 9-15 hours after forecast initialization. I find that GPS-PWV DA improves location and intensity of the MCS. For all experiments, the GPS-PWV DA reduces the PWV root-mean-square-error to within the GPS-PWV observation error of 1-2 mm at initialization and reduces the initial wet bias. Although there is a short “memory” of these adjustments in that the PWV RMSE across the sites rises quickly and approaches the RMSE of the non-DA experiment after 2 hours, this is due to advective effects near the GOC in that the adjustments move quickly away or toward the sites. Assimilating GPS-PWV observations lowers the moisture (water vapor mixing ratio) error in the lower atmosphere to where is it within the instrument error aboard the radiosonde. From my sensitivity analyses, I conclude that increasing the covariance localization cutoff radius improves the MCS when adjusting all state variables but degrades it when adjusting only thermodynamic variables. Also, I recommend assimilating a mean hourly observation (“superobbing”) rather than individual 5-min observations as it allows for more stable adjustments. Lastly, I note that having a 12-h spin-up improves the MCS simulation because the initial conditions have a chance to make their way to the convection-allowing grid before GPS-PWV DA adjusts the MCS towards the observation. In a region of complex terrain that suffers from unreliable observations and poor convective forecasts during the NAM, I have shown utility from GPS-PWV observations in a) the diagnosis of wet model bias, b) the improvement of the initial conditions via convection-allowing ensemble data assimilation, and c) the improvement of MCS simulation. The results of this dissertation point to a need for more observations in the vertical and a deeper understanding of sensitivities of atmospheric variables to one another, such as what can be gained with with a network of ground-based lidars that continuously monitor the boundary layer.