Nowcasting Using Geostationary Weather Satellites 

A Data Science Research Project through Summer STEM Institute 2020

Mentor: Ethan Weber, MIT Computer Science and Artificial Intelligence Library (CSAIL)

Function: My goal is to design a model that could accurately predict both the rise in sea levels geographically and the rescission of the ice caps. I want to use weather data that I have captured from my model and make it open-source, so anyone can add relevant data to make it much more accurate.  I want to end up creating a program that, based off of user location, will contextualize sea level rise and show people that this is an important topic and one that we need to address. I also want to be able to identify local precipitation events which is helpful to small communities without broad internet access.

Inspiration: Sea levels are rising around the world due to climate change. We are being forced to answer two very difficult questions: to what levels will they rise and by when? My idea is to use my project and final solution to help others easily understand how relevant climate change and sea levels are to their everyday life.

Methods: Trial and error; I installed a satellite disk on my roof to receive data/images of the earth that pertain to weather, i.e., cloud patterns and formations, etc. I’ve created a neural network that will process this information and determine locations of clouds and make weather predictions for me.

I have collected almost 8 months of data, and am working on compiling all of my images into a website which allows visitors to easily and quickly understand what the weather will be doing over the next few hours.

Outcome: TBD

Failures and challenges:

• I built an antenna by attaching a 2.4GHZ dish to an old tripod with a 3D printed mount I designed. I set it up on my roof and it was too top-heavy and constantly fell over. I needed to weight the base and a few sandbags did the trick.

• The wildfires burning all over California (summer of 2020) and the resulting smoke created an issue with data collection. The images pulled down from the satellite were occluded by smoke. The smoke cleared and I resumed data collection.

• I had many issues with my algorithm, but have been working through them. I now have a successful algorithm, with a 99% success rate. (99% of the images are in good enough shape to be processed and categorized)