CAZMEX 2017: Improving Monsoon Precipitation Forecasts in Northwest Mexico and Southwest United States

North American monsoon precipitation occurs from early July until mid-September and impacts the water supply, severe weather, droughts, and wildfires in the arid regions of southwest U.S. and northwest Mexico (Adams and Comrie 1997). In this project, we seek to improve the forecasts of convective precipitation over the Sierra Madre Occidental (SMO) and its propagation west later in the day and evening into lower terrain and towards the Gulf of California (GoC) as mesoscale convective systems (MCSs). Moker et al. (2018) defines a strongly (weakly) forced day as the presence (absence) of inverted trough (IV) over northwest Mexico as an IV is an indicator of dynamic forcing. We focus on the days where there is weak dynamic forcing as they have the most potential to benefit from the data assimilation (DA) of precipitable water vapor (PWV) observations from Global Positioning System (GPS) meteorological stations. The PWV observations used in this project are obtained from 18 installed GPS PWV sites that collected data from June to September 2017 as part of the Consortium for Arizona-Mexico Arid Environments (CAZMEX) field campaign in northwest Mexico. A total of 39 weakly forced days 9 strongly forced days are observed during the campaign.

In a similar configuration to Moker et al. 2018, we use the Advanced Research Weather Research and Forecasting (WRF-ARW) model to simulate convection explicitly on a 2.5 km grid with initial conditions and lateral boundary forcing provided by the Global Forecasting System (WRF-GFS) and the North American Mesoscale (WRF-NAM) models.

We ran the model on all weakly forced days during the field campaign for 24 hours beginning at 1200 UTC. To verify the rainfall, out of GPM IMERG Early, GPM IMERG Final, CMORPH, and PERSIANN satellite-derived rainfall products, the GPM IMERG Final has the lowest RMSE when compared to rain gauge data from 25 meteorological sites. Using the GPM IMERG Final as “truth”, WRF-NAM performed better than the WRF-GFS in terms of 24-hour total of precipitation and hourly precipitation.

Using the community ensemble-based DA software Data Assimilation Research Testbed (DART), we assimilate GPS PWV data for27 July (strongly forced day) and 9 August (weakly forced day). The assimilation of PWV observations in a high-resolution convective-permitting model may have some positive impact on numerical weather prediction forecasts (Serra et al., 2016). At 00 UTC, we perturb 20 ensemble members in the largest domain then run a 6-hour spin-up for those perturbations to reach the convective-permitting domain. Then, we run six hourly DA cycles to initialize the model. We also run the ensemble members without assimilation during these 6 hours. The 24-hour deterministic forecast is then initialized from the mean of the ensemble members. Although the PWV RMSE is reduced, we find that the DA performs more poorly than without DA in terms of rainfall. To address this, we are currently updating our DA algorithm and model configuration.