Terrain

The on-ramp to environmental data science I wish I'd had. Free datasets, guided R and Python projects, and faculty to reach out to, built for students who don't know where to start.

Comment

Inspiration

We tried to get into environmental data science as a first-year and hit a wall. There was no single place that told us which datasets to use, which professors to email, or what a good first project even looked like. We pieced it together slowly and inefficiently. Terrain is the thing we wish had existed when we started, built so the next student doesn't have to go through the same cold start, with a real focus on making it reachable for first-gen, low-income, and underrepresented students who don't already have a way in.

What it does

Terrain is a complete on-ramp to environmental data science, taking a student from zero experience to real, portfolio-ready work. It brings together four things beginners usually have to piece together on their own:

  • A 100+ hour skills curriculum covering the relevant tools and techniques for environmental data science, structured so students build competency in a deliberate order rather than guessing what to learn next
  • Guided projects with pre-made Google Colab notebooks that reinforce each skill on real datasets (CalEnviroScreen, EPA AirNow, NOAA, NASA Earthdata, USGS, and UCLA's own emissions data), so students apply what they learn immediately, right in the browser with no setup, and finish with something they built
  • A faculty directory with ready-to-use cold-email templates, so reaching out to a professor stops being intimidating and becomes a fill-in-the-blank first step
  • A live carbon tracker that takes each UC campus's verified annual emissions (from around 2015 on), fits several trend models, and validates them with leave-one-out cross-validation. Models that go unstable or negative are dropped, and the survivors are blended into a weighted ensemble (weighted by inverse validation error) to project emissions to 2030 and 2045 with a 95% prediction band that widens further out. A built-in what-if simulator starts from that baseline and subtracts the coupled savings of different intervention scenarios, so students can see what it would actually take to hit the 2045 goal

Everything runs on free tools and assumes no prior experience.

How we built it

The site is plain HTML, CSS, and JavaScript with no framework and no build step, which keeps it fast, free to host, and accessible on any machine. The carbon tracker turns each campus's verified annual emissions into a forecast: it fits multiple trend models on the short series, scores them with leave-one-out cross-validation, drops any that go unstable or negative, and blends the survivors into an ensemble weighted by inverse validation error. It projects to 2030 and 2045 with a 95% prediction band that widens with distance, and a what-if simulator layers coupled intervention savings on top of that baseline. The curriculum and project libraries are data-driven, so the team can add lessons, Colab notebooks, and datasets without rebuilding pages. Beyond the code, Terrain runs as an organization: an internship program of 20+ students across marketing, design, and technical teams builds and grows the platform. It's hosted on Netlify and open on GitHub.

Challenges we ran into

The hardest part was scaling the project into an organization while keeping the product simple. Coordinating 20+ students across three teams meant building content and workflows that stay consistent no matter who contributes, which is why the curriculum and notebooks are standardized and the site stays dependency-free. On the technical side, making the carbon tracker genuinely useful rather than decorative meant working from verified public reports and building an ensemble honest enough to stand behind, with cross-validation and sanity bounds instead of a single model drawing a nice-looking line to zero.

Accomplishments that we're proud of

Turning a personal frustration into a working platform and a real team. Terrain offers a 100+ hour curriculum, hands-on Colab projects, faculty outreach templates, and a live carbon tracker with cross-validated ensemble forecasting, all free and accessible on any laptop. It's officially partnered with UCLA DataX, and it's run by an internship program of 20+ students who are actively growing it. Most of all, we built something structured to outlast any one person and to genuinely lower the barrier for students who don't start with an advantage.

What we learned

How to turn a vague "someone should build this" into both a product and an organization. On the technical side, structuring content so it scales, building forecasting models that are honest rather than flashy, and keeping everything accessible with no paid tools. On the human side, running teams across marketing, design, and engineering, and learning that a lot of the hard work is editorial: deciding what a real beginner needs first and cutting everything else.

What's next for Terrain

Terrain is already partnered with UCLA DataX, so the focus now is reach. That means hosting a Terrain event on campus through DataX to bring in more students, continuing to grow the internship program, and expanding the curriculum and project library. Longer term, the goal is to take Terrain beyond UCLA and toward lower-income high schools and colleges, so students anywhere can use it as a free, no-barrier way into environmental data science.

Share this project:

Updates