Development of a classification algorithm for ice crystal habit using machine learning
The radiative effects of ice crystals in the visible, infrared and microwave light spectrums strongly depend on the size and habit (or shape) of the ice crystal. This dependence has important implications for remote sensing estimates of precipitation rates and cloud radiative properties.
Thus, accurately identifying these properties is essential for accurate estimates of precipitation and cloud-climate interactions. Furthermore, the habit that ice crystals acquire during their lifetime depends on the environment that they grow in with respect to the ambient temperature and humidity. As such, the origin and history of an ice crystal can be assessed through its habit. With the recent increase of in situ measurements of ice crystal habit and size, there is a need to automatically determine the habit of ice crystals for a better assessment of cloud microphysical pathways, and the impact of habit on cloud radiative properties and precipitation estimates. Therefore, the aim of this project is to develop an automatic method to classify ice crystals based on their habit.
The proposed project will use a mixture of artificial and field collected datasets to train a neural network to automatically classify ice crystal habit. Once the algorithm is developed, the prospective candidate will participate in a field campaign to acquire a dataset that the algorithm is applied to.
The prospective candidate should have experience with machine learning, neural networks, handling large datasets and an interest in developing training datasets by classifying the habit of ice crystals manually. A candidate interested in participating in a field campaign is a bonus but not necessary. The project will start in Fall 2019, with the potential field campaign occurring in Fall 2020.
Beck et al. (2018): Impact of surface and near-surface processes on ice crystal concentrations measured at mountain-top research stations