Environmental Data Scientist, Machine Learning
Grassroots Carbon · Austin and San Antonio, TX
Grassroots Carbon pays ranchers for the carbon their land stores when they change how they graze it, and sells that carbon as credits to companies such as Microsoft and Boeing. Measuring that carbon is the center of the company's science. It rests on three kinds of record: soil samples taken across each ranch; what the ranchers themselves log about their grazing, through PastureMap, the company's own software (which pastures the herd was on and for how long); and public data that describes the land from above and around it, satellite imagery of how it's growing, daily weather, soil surveys, and elevation. On the data science team, my work is the modeling. We estimate how much carbon a ranch's soil is gaining and, just as important, how much it would have gained anyway under the old grazing: only the difference counts as a credit. A typical model follows the grass through its seasons. From the years before the change, it learns how much the land grows from the sunlight, weather, and cover it gets; it then runs those same rules forward to show what the old grazing would have produced. Others answer the questions that come before and after it: how much grass is standing on a ranch right now, estimated from satellite imagery and the field measurements that calibrate it; how deep a ranch's soil can be credited, worked out from where its own cores stopped and what the terrain and soil maps say about the ground beneath; and how to divide a ranch into zones that sample alike, so that fewer cores carry less uncertainty into the credit. Around the models sit the shared data systems the team maintains (one grid, one set of sources, so any ranch in the country can be run the same way) and the checks that test our estimates against research sites where grazing never changed.