Nick's MiSpace Hackathon Ice Predictor and Route Creator

Inspiration

I really wanted this project to be unique and fun to build. While the challenge was fairly narrow, I tried to build out the frontend in a way that nobody else would. This resulted in fairly complex features that took a long time to build but are extremely fun to play with.

What it does

If you navigate to the website, you will be able to hover over any ice coverage and understand the ice cover prediction. There's also toggles in the top right corner to switch between days by either selecting days or sliding the slider. You can also click on the major hubs to learn about the shipping and ice breakers located in that area. Finally, if you toggle on the route creator, you have the opportunity to create a route for the ships to go on. It will spit out the gps data which you can then plug into large shipping navigation websites!

How we built it

I developed the machine learning pipeline in Python using Google Colab, loading ice and weather data in the NetCDF format provided. After aligning the coordinate systems, I trained a Linear Regression model on 17 days of historical data that predicts tomorrow's ice from today's ice and temperature, achieving 2.7% daily error. For the frontend, I built a React app with Leaflet that renders the forecasts as canvas overlays for performance. I added interactive features like day selection, hover ice readings, route planning with water detection, and GPX export so users can load routes into marine GPS systems.

Challenges we ran into

I originally started the ML model with all of the weather points. Unfortunately, I couldn't get it to work as the error was consistently over 5%. Thus, I decided to tone back and allow the model to input just the two most important datapoints. Additionally, formatting the data to have the correct longitude and latitude was a complete pain but I was able to pull through in the end. Finally, getting the data to overlay at the correct coordinates was a many hour process and something I don't wish to do again.

Accomplishments that we're proud of

While I wish I turned in the project at the start, I still feel proud that I finished the project in its entirety. Additionally, I made large strides compared to my original mvp with a fairly good prediction model and extremely advanced UI.

What we learned

I learned all about linear regression and machine learning. This was one of my first cracks at ML and it took many youtube videos, online blog posts, and chatgpt consultations to understand how I'd go about building it.

What's next for Nick's MiSpace Hackathon Ice Predictor and Route Creator

I'll look to increase the accuracy of the prediction model by adding a larger quantity of the current data as well as expanding to more datapoints. Additionally, I'd like to update the frontend to take in more inputs such as the gps of the ships and other weather data (in real time) to be able to update the ships in real time to help them choose routes in the most efficient manor.