Introduction

Identifying Clouds in Panoramic SETI Data with Machine Learning

Motivation

• To increase the observatory’s duty cycle, we may observe on nights that have a potential for occasional clouds.
  • Example: first half of the night is clear, second half has occasional cloud cover.
  • Since panoseti telescopes have large solid angles, if clouds are present over Lick observatory we will likely see them in our data.
  • Clouds are a source of interference, so we need a way to identify them whenever they exist in our data.
    • When the observatory is scaled up, it will be impractical to manually search for clouds in our data.
  • → we need some kind of automatic system to identify when clouds have affected our data.

Path to a Solution

• Other groups have found success using additional equipment for cloud detection, such as an all-sky camera or LIDAR sensors. (See https://ui.adsabs.harvard.edu/link_gateway/2020AJ....159..178M/doi:10.3847/1538-3881/ab744f.)
• However, panoseti instrumentation has more than enough sensitivity and time resolution to observe even the faintest and fastest moving clouds in the sky, suggesting a panoseti-based cloud-detection system is feasible.
  • Using panoseti telescopes for cloud detection has the following benefits.
    1. No additional equipment & engineering is required.
    2. No changes to DAQ software are needed.
    3. Data from different telescopes looks consistent, even at different sites, an important feature for ML systems.
    4. It’s the most accurate way to determine if and when clouds affect our data.
• Plots of detector-array currents can suggest the presence of clouds. However, as each current value corresponds to 64 pixels and is sampled at ~1 Hz, this metric, by itself, is too coarse to precisely identify which pixels are influenced by clouds.
• Image classification is a standard ML application, so there was a good chance it would work for this project.

Project Objective

<aside> ☁️ Develop an automatic system capable of using Panoseti data to classify every second of an observation as cloudy or clear.

</aside>

• Example

Software Architecture & Database Schema

Software Architecture

Database Schema

Labeling & Features

Jupyter Notebook Labeling Interface