2.2 Use of an Artificial Neural Network to Forecast Thunderstorm Location

Deterministic Numerical Weather Prediction (NWP) models integrate the conservation equations of atmospheric mass, heat, motion, and water. With respect to gridpoint NWP models, the difference terms in the equations are approximated as taylor expansions and are integrated forward in time (Pielke 2002). Processes resolved on the grid-scale are referred to as model dynamics. Sub-grid scale processes in NWP models must be accounted for—otherwise the quality of the numerical predictions will rapidly degrade with time. These sub-grid scale processes—which by definition cannot be explicitly determined by the model—are parameterized in terms of the grid-scale. These parameterizations are referred to as model physics (Kalnay 2003). The parameterization important to this paper is convection those processes related to shower and thunderstorm activity. The purpose of convective parameterization (CP) is to reduce atmospheric instability to prevent the model from generating excessive grid-scale precipitation. Precipitation is simply a by-product of the CP process. Thus, such convection is not explicitly predicted. However, if model grid-spacing is decreased to around 2-km (the mesoscale/microscale boundary), convection can be explicitly predicted thus rendering convective parameterization mute. However, such an increase in model resolution will require enormous computing resources (CyRDAS 2004) -unrealistic for operational applications at present. Further, it is unclear whether increasing the horizontal resolution will improve forecast accuracy. Mass et. al (2002) suggest that increasing horizontal resolution of NWP models to 4-km may not provide additional accuracy. According to Fabry (2006), the exact location of convective cells that develop during the daytime is generally determined by the location of updrafts on the mesoγ (2-20 km) or smaller scales. According to Orlanski (1975), individual deep convective cells occur on the micro-α scale (200m-2km). However, based on 2-D deterministic numerical atmospheric simulations, Zeng and Pielke (1995) found that vertical velocities—induced by surface heterogeneity on flat terrain—are generally unpredictable on length scales less than 5-km.