The global Climate TRACE coalition harnesses remote sensing and software intelligence to monitor human-caused greenhouse gas emissions in near real time around the world and share this dataset openly for all to build with.

I connected with Mapbox alumni Mikel Maron and Jonni Walker to learn more about Climate TRACE and how it applies data science, data visualization, and location intelligence to climate action.

What is different about Climate TRACE in the field of climate action?

Climate TRACE is the most detailed facility-level global inventory of greenhouse gas emissions ever produced. By using satellite imagery, machine learning and ground truth technologies ClimateTrace provides a comprehensive and verifiable picture of global greenhouse gas (GHG) emissions. This radical data transparency, combined with the recency, coverage, and granularity of the data, helps organizations, governments, and individuals take action towards a net zero economy. 

Also, Climate TRACE is a neutral data provider, co-founded by a global coalition of non-profits, and we share this dataset as free and open to all to use. By exposing facility-level GHG data, we aim to spark curiosity and inspire others to use it to make meaningful climate action faster and easier.

The Climate TRACE map is very striking, why did you choose a map as the primary interface for the dataset?

We built the map experience to encourage users to delve into the nuanced details, download the data, and build on top of it. The map visualizes the Climate TRACE data in a compelling way that exposes both its granularity and its versatility.

The map is also the primary way that several core audiences will interact with Climate TRACE data. For journalists and others who want to use the data for storytelling purposes, the map presents an intuitive and easily accessible way to understand and communicate the emissions data. For policy-makers and the general public, the map view is a friendly interface to explore the information, from the global level down to the local. And even for data scientists, developers, and others who ultimately will work with the more comprehensive data (via CSV data downloads), the map is an initial entry point to see what is in the data and get inspiration about ways to apply the data.

There is an incredible amount of data displayed on the map. Can you tell us more about how you built this in the backend?

First, we had to get the data ready to be brought together on the map. Climate TRACE is built by a coalition of data scientists and researchers from the private sector, academia and nonprofit sectors. The data covers emissions from every geography and many different sectors. It started with annual national-level inventories, but the most recent release added source-level emissions insights from individual assets and facilities. As a result, the sources, methodologies and formats vary enormously, so a big part of this project was developing and aligning data to a unified schema. We import data from the sectors as CSV, and you can also access these directly from our downloads. 

This is imported into a data warehouse, after applying automated quality checks to validate the structure of all of that data. For the production database, the data is restructured to make serving up tiles and the most common queries from the front end very fast. We use PostgreSQL/PostGIS, along with Rust to implement the API.

Right now, there are 80,188 sources of emissions on one map. Rendering that much data on the front end was never an issue with vector tiles and Mapbox GL JS. Almost all of the map interface is driven by the vector tiles, including the emission source detail cards at the bottom of the map. At load, the site gathers details on every source from the tiles using Mapbox GL JS queryRenderedFeatures, and dynamically builds the card listing from the stored results. This works very fast and avoids additional duplicative API calls.

We relied heavily on Mapbox Studio to design the map layers, and quickly iterate between the data engineers, map design teams, and front end developers. Since we made our own vector tiles, we had to package up samples as mbtiles and upload as a small tileset to Studio for styling. Once the design was ready there, we copied the style properties from Studio and shared with the front end team to implement in code.

Speaking of map design, the visualization of the Climate TRACE data is complex cartographically, with so much different data to display. How did you keep the visual design user-friendly?

We face two major cartographic tasks with Climate TRACE. First, working with our data requires accurately representing a large number of different carbon emissions sources, across many geographies, with several orders of magnitude in emissions values across sources. Second, we had to thread the needle of making an effective map and attractive UI, while not making emissions look too pretty, as the data visualized on the map represents harmful emissions we want to reduce.

The visual interaction of tens of thousands of points, polygons and lines on a global map can become cluttered, making it difficult to clearly understand the sources and distribution of emissions. We went through many iterations - experimenting with clustered points, dot grids, hot spots, and line segments connecting the things that move. We ultimately decided that using scaled points at the global scale illustrated the spread and depth of data, with natural clustering. 

The points at the global level resolve to more detailed geometry at higher zooms. For land use emissions, we plotted centroids of grid squares as dot grids at medium zoom, and illustrated as polygons at high zooms. For ship tracks, we also used a grid, and collected all emissions attributed to ships passing through that grid to a centroid of that square and scaled the opacity by total emissions. This provided an impression of major shipping routes and their relative contribution to emissions.

The fine-tuned application of color intentionally strikes a balance between serious, but not apocalyptic. The palette spans 8 different colors for different sectors, each distinctive and complementary, without being overwhelming. This palette was placed over a basemap that provides needed geographic context of terrain and boundaries, while cleanly faded into the background. The map projection also adjusts based on view, so at the global level we use an Equal Earth projection that then flattens out into the more familiar Mercator at closer zoom levels, something that is made incredibly using using Mapbox adaptive projections.

Speaking of the base map, we knew from the start that this visualization would need a minimalist base map design in order to not distract from the data layers added to it. We used Mapbox Studio extensively to customize zoom-level dependent styling for base map features in order to optimize the balance of visual simplicity with location context that is helpful for orientation and understanding the data. The base map palette is based on a white, low-poly-esque feel that would be compatible with virtually any color overlayed on top of it in the data layers. To add subtle geographic features at a global level, we used a terrain hill shade layer. Then as a user zooms in on the map, additional layers begin to appear, like place labels and shading for urban areas. This helps to ‘personalize’ the data for the viewer at the local level, helping them understand what emission sources are near to them such as a power station or farm.