Improving Daily and Sub-seasonal Precipitation Forecasts in Arid Regions Through Convective-Permitting Modeling and Data Assimilation

Abstract

Precipitation in arid and semi-arid regions is important for both human and natural systems, but the extreme weather events associated with the precipitation can pose threats to the growing population in the regions, especially due to the changing of climate. Thus, there is a need for reliable weather forecast systems that are able to provide early warnings to reduce fatalities and socio-economic loss. This dissertation demonstrates two forecasting techniques that are able to improve the current operational forecast systems to predict extreme precipitation events in arid and semi-arid environments. The first technique is utilizing a data assimilation system in convective-permitting models. The data assimilation implements the ensemble Kalman filter scheme that integrates precipitable water vapor (PWV) data derived from Global Positioning System (GPS) sensors into a numerical weather prediction (NWP) model and updates the modeled state variables sequentially. The updated modeled PWV is expected to maintain the errors between the model and the observations low at the model initial condition. This potentially generates more accurate weather forecasts since the initial model biases is minimized. We implemented this technique during the North American monsoon (NAM) GPS Hydrometeorological Network field campaign in summer 2017 over northwest Mexico. The GPS-PWV data from the field campaign was assimilated into a 30-member ensemble convective-permitting (2.5 km) model. The results show that assimilating GPS-PWV improves the initial condition of the modeled PWV, most unstable convective available potential energy (MUCAPE), and 2-meter dewpoint temperature (Td2). This also leads to an improvement in capturing nocturnal convection of mesoscale convective systems (MCSs; after 0300 UTC) and to an increase by 0.1 mm h-1 in subsequent precipitation during the 0300-0600 UTC period relative to no assimilation of the GPS-PWV (NODA) over the area with relatively more observation sites. The results demonstrate that this technique could be implemented over areas where traditional observations for data assimilation, such as radar and radiosonde, are not available. The second technique is applying convective-permitting (CP) modeling to a sub-seasonal weather forecast model. In this technique the model’s horizontal grids are downscaled from 80 km to 4 km that is a convective-permitting scale, in which precipitation can be represented and resolved explicitly without any convective scheme. The sub-seasonal time frame indicates that the forecasts go beyond two weeks but less than two months. We applied this technique over the Arabian Peninsula (AP) during winter months (October to April) of 1999 to 2018. We simulated 18 extreme precipitation events that directly impacted Jeddah. Each simulation is initiated at one-, two-, and three-week lead times. The model lateral boundary condition is taken from the European Centre for Medium-range Weather Forecasts sub-seasonal to seasonal (ECMWF S2S) reforecasts. The results show that the CP model outperforms the low-resolution ECMWF S2S reforecasts in predicting the extreme precipitation events at all lead times, especially the events associated with the extratropical synoptic regime. The CP model high forecast skills in precipitation are likely attributed to the better representation of dynamics and thermodynamics in CP model. The promising results demonstrate the feasibility of predicting extreme weather events at sub-seasonal time scales over the AP.