Linking GLM Flash Rate Observations and HWRF Modeled Atmospheric Parameters

Abstract:

Currently, lightning observation operators in Data Assimilation (DA) systems are based on regressions between lightning flash rate and a combination of storm-related properties, such as maximum vertical updraft, vertical graupel flux, and vertically integrated ice. These regressions, however, only statistically agree with lightning flash rate, effectively creating biases and prompting a need for additional optimizations of empirical parameters in order to improve the fit to lightning observations in individual DA cycle. Additionally, the available regressions are developed for global applications or convective storms over land, and their relevance to tropical cyclones is of limited value. In this study, an Artificial Intelligence Deep Neural Network (DNN) model is developed to help create a link between GLM lightning features and modeled tropical cyclone properties as an attempt to improve tropical cyclone lighting DA. Using the HWRF model dynamical and cloud variables as input, a DNN model is trained to produce accurate representations of GLM-derived lightning flash rate. When testing for the best input features, total column values and profiles of ice hydrometeors, specific humidity, super-cooled liquid water and temperature are considered. Results suggest that classification model can easily achieve the overall accuracy of 70% in predicting the occurrence of GLM positive lightning flash rate.