Mary Huang
Biome

Earthshot Labs
2022-2023
UI/UX Design Lead

Data is essential to the conservation of the world's forests. Biome is a modern app for tree inventories. The tedious task of measuring trees is often done by scientists and field technicians to collect essential data for managing forest projects and estimating biomass and carbon sequestration.

Biome uses machine learning and AR to greatly innovate the user experience. Our initial field test showed that using the app is 5-10x faster than conventional methods.

Earthshot Labs was awarded an NSF SBIR grant for this work.
I worked with a small R&D team of 3 engineers to research user needs, design and build the first working alpha, scope future features, and run a field test in Panama.

Domain Context

In order for tree data to be useful in scientific models, trees must be measured in plots of defined size. Scientists may use a range of methodologies, but a common practice is to define a plot by a center location and radius.

While there has been growing investment in remote sensing in the last decades, there has been little improvement in scientific field tools. Measurements are still collected with measuring tape, pencil, and notebook. 

In the field, scientists are juggling these supplies while hiking down a rainforest ravine and avoiding fire ants. And then it starts raining.

The work is tedious and prone to inaccuracy.

Organize and view plot data

A common technique for tree inventories is to have 'permanent monitoring plots', where the same plots are measured in subsequent years. I included this in the design as a primary expected flow.

The information heirarchy of Projects > Studies > Sites > Plots > Trees was proposed by members of our science team, and aligned with data formatting from scientific literature.

Unity prototype

Record forest sample plots using AR
  • Adjustable plot radius.
  • Save GPS location for center of plot.
  • Mark relative locations of trees in plot. Accurately see and record trees on the boundary line.
  • Works on hilly terrain. Plot boundary is visualized perpendicular to world orientation.

While field testing, we were delighted to find that the AR circle is perfectly accurate when standing on a hillside. (Scientists usually do a bunch of trigonometry to straighten the ellipse to compensate for the slant of the hill.)

The challenge of the feature is maintaining a tracked AR session for the entire duration of measuring the plot, which can be 30 min. We also needed to allow the user to optionally add DBH, height, and species for each tree within that AR session.

I worked on this prototype in Unity with our engineers. In Panama, I made sure to test in a variety of natural terrain. The AR tracking worked even in tall grass!

Measure tree diameter with computer vision

  • Biome uses computer vision to segment tree trunks from the background.
  • Works on odd-shaped trunks, trees with vines, and thorny trees.
  • Measure trunks as small as 5cm diameter.
  • Record multiple measurements from different angles for odd-shaped trees.
  • Record measurements for multi-trunk trees.

Alpha test build

Measure tree height with AR
  • As accurate as a clinometer, but faster and easier.
  • Good visibility in forest environments.

This feature worked surprisingly well even for large trees. While there is a margin of error, it proved to be within the acceptable range compared to the traditional method using a clinometer.

We explored several UX flows for this feature, and the simplest version proved to be the most reliable.

Alpha test build - small tree

Alpha test build - large tree