Using Multiple Fields of Observational Data to Identify and Track Deep Convective Clouds

Automated feature detection and tracking algorithms have become increasingly popular with the growing availability of sub-hourly satellite data and high-resolution model data. Most tracking algorithms are field-specific; however, some can be used on user-specified datasets. tobac (Tracking and Object-based Analysis of Clouds) is an open-source tracking algorithm developed in recent years with a primary goal of identifying and tracking deep convective clouds (DCCs) within satellite and model data. Isolated DCCs are easily identifiable in satellite data and infrared brightness temperature data is often used to identify their height, which in turn can be used to estimate parameters such as updraft intensity. This process becomes challenging in cases where DCCs are obscured by upper-tropospheric clouds, as is often the case with clustered convection. This makes tracking individual convective cores challenging or, in some cases, even impossible. Data from the Geostationary Lightning Mapper (GLM) can be helpful in these cases, as the highest lightning flash densities are typically found within DCCs and only somewhat attenuated by high clouds. However, only a subset of DCCs ever produce lightning, making GLM unsuitable for convective cloud studies that are not limited to lightning-producing clouds. We introduce a multivariate approach that improves tobac’s existing feature detection technique. Our approach combines features from multiple fields, and features from different fields that occupy a similar location at the same time are considered to belong to the same physical feature. The end result is a list of features that can be tracked continuously, even if they are not continuously identifiable in all fields throughout their lifecycles. A major advantage of this process is that the feature linking step that creates the cell tracks requires minimal modification. The algorithm can also be used on a wide range of fields, permitting that their times overlap. In this presentation, we will show how multivariate feature detection allows for more robust tracking of DCCs in varying environments throughout their lifecycle, compare these results against single-variable tracking on a diverse set of variables, and demonstrate the novel applications of this new multivariate tracking algorithm.