Green Gauge

Inspiration

We were inspired to make this project because living in an urban environment like Boston we understand how important trees are to the environment which can improve things like air quality, help reduce carbon, and mitigate rising temperatures. So we wanted to create Green Gauge to help inform the community and policy makers of this problem to help increase the environmental quality of Boston.

What it does

Green Gauge works by allowing users to type in an address anywhere in the world. It then takes a satellite image of that location and is run by our pre-trained AI model and the results of that analysis get sent to an LLM to provide actionable insights to policy makers and the community. It shows things like tree coverage percentage, number of trees, weather of that location, air quality.

How we built it

We built our project using React + Vite as our main front-end tech stack. We used a few APIs, such as OpenAI vision and AirVisualAPI and MapBox, to request the satellite images and generate relevant information based on the satellite image, such as recent environmental conditions in the area.

Challenges we ran into

The biggest challenge we encountered was figuring out how to analyze the satellite images using a pre-trained model. We could not find a model that would score high enough to provide accurate results about the image. Moreover, there was not enough time to train a CNN model from scratch, which limited us to only using a vision model that detected trees rather than multiple land use classes, such as roads, rooftops, etc.

Accomplishments that we're proud of

We are proud of the accurate readings outputted by our program using our pre-trained CNN model and the recommendations our program outputted after using OpenAI Vision.

What we learned

We learned how to use Mapbox-GL's API as well as converting the images generated by Mapbox to a useful format that DeepForest and OpenAI Vision could use.

What's next for Green Gauge

We are looking to incorporate better data and higher resolution satellite photos in order to glean more impactful insights. We also hope to develop our own model to evaluate greenery and experiment with transformers, deep learning, and SVMs. Finally, we want to use MongoDB in order to store snapshots of notable regions so that users can see how the green space has been changing over time.