COnvective Cores

Background

A convective core is the circular motion of air within a convective cloud. Convective cores are created when warm, humid air rises through cooler air, creating updrafts which carry the warm air upwards. The "core" is the most intense area of vertical updrafts. Once the warm air reaches a height where it is no longer buoyant, the air begins to fall back down. This creates a cycle which in the study of atmospheric science is called a convective core.

The convective cores data set we use is meant to represent a convective core as a three dimensional set of velocities over a period of time. This project thus works with a four dimensional array which is manipulated to study various features and attributes which convective cores present. The visual to the right depicts how convective cores move in space over time. Darker areas indicate larger velocities or where the center of the core would be located. Our stream uses a variety of approaches to manipulate the data including water-shedding and transecting, as well as a couple of clustering algorithms . Each of these approaches allows for specific insights into the data set, each having advantages and disadvantages.

Results

Transecting Method

The transecting method was based on a simulation done in a paper by the Journal of Advances in Modeling Earth Systems. This method attempts to model a research aircraft passing through a convective core collecting data such as size, maximum velocity, and average velocity information about the convective cores it penetrates along the way. In order to mimic such a situation the 2021 cohorts Sam Kinloch and Diego Pantelis iterated through each vertical level and each time step and collected the necessary data for analysis. They then split up the data into percentiles so that we may see where the extreme data points lie.

For each of the graphs the vertical axis represents height in kilometers. The colors are labeled in the legend as 50th, 90th, or 99th percentile. From left to right the graphs depict maximum velocity, average velocity and size.

The transecting method is useful to mimic a one-dimensional collection of data. This may be useful for certain approaches, however convective cores are three-dimensional phenomenons.

Water-shedding Method

The water-shedding method attempts to model the data as an "image", allowing for a three-dimensional analysis of the data in two dimensions. Each pixel of the "image" has a value (represented as the color of the pixel) assigned to it, creating a three-dimensional representation of each horizontal slice for each height. One way to think of it is as a topographic map (right), which depicts the altitude and shape of a three-dimensional landscape in two dimensions. The water-shedding method uses the same approach except that instead of the peaks representing higher altitudes, they represent higher velocities.

Using this water-shedding algorithm on the data, 2021 cohorts Ethan Pak and Yasmine Torres were able to identify the maximum and average velocities as well as the size of convective cores. Their final products differed from the transecting method's products, however produced a similar shape indicating that this method analyzes the data in a similar but different way than the transecting method. These graphs are set up the same way as the transecting method's graphs except for the x-axis of the size graph. Notice that the x-axis of the size graph is in kilometers squared and not just kilometers.

Morphology

Peer mentor Ryan Moore used a 2D morphology algorithm which allowed him to segment the data at each vertical level into clusters which were then labeled as individual cores. This method was then applied in 3D to create a three-dimensional visual of all the cores identified. Each of the colors represent an individual core.

Core Statistics based on Morphology


This graph shows the distribution of core volume at each time step in the simulation. There's a clear peak in the maximum value around the middle of the simulation, but the median and quartile values stay relatively constant throughout.


This graph shows the distribution of each core's maximum velocity at each time step. There is not a clear trend.

Clustering Methods

Peer mentor Ritvik took a different approach towards separating and labeling different cores. He used a K-means clustering algorithm which uses a feature vector in order to isolate the data into 7 groups. These 7 groups were each assigned a unique color and each represent a single cluster, with the white triangle representing the centroid of the data. Future work informed by Ryan's work above will involve increasing the number of clusters searched for.