Report out from the Earth Index offsite
One of the core values of Earth Genome is to work extremely closely with users, to understand their context and data needs at critical junctures in conservation work. We’re sharing this report out as part of our continuing commitment to this approach, and are eager to hear back reflections on the direction of Earth Index.
Setting up the offsite
Last month was the perfect moment to pull back to plan the future of Earth Index. We completed a big run of development that started at the beginning of 2024, and have just started bringing in users from the Earth Index waitlist. We have secured a strong pot of resources from grants and paid work to drive forward core Earth Index development. 2025 is coming up soon. We need a solid plan.
Ben, Hutch, Cam, Brad, Kwin, Ryan and Mikel got together in DC. It’s good to get together in person … it always is these days! We are a remote team, and in some ways a new team. The front end developers that put a motor on Earth Index development this year have rolled off the team as planned. Other Earth Genome developers have been extremely busy with other projects but have been eager to get involved, and are finally seeing their time open up for Earth Index. We are all excited and aligned in our gut about what we’re building, but missing how we would organize ourselves in a fresh configuration and lacking the explicit priorities of what we need to build in the next year. That’s really the fundamental problem humans face again and again — set our direction and figure out how to cooperate to get there.
We have established a pattern for organizing offsites. We’ve borrowed from what we’ve learned from great facilitators like Aspiration Tech. In the couple weeks before, everyone involved reflected on a few questions to drive what is going to make our time together as valuable as it can be. We talked as a group the week before, and key topics and questions emerged. We set up a blank time schedule for the 1.5 days, with work sessions of about an hour length and plenty of breaks in between to stay fresh. Then Ben filled up the schedule, but not too much, since we iteratively adapt the schedule as we go depending on where the work is directing us. We stayed all together at a big nearby AirBnB, and we gathered to work at our usual workspace of Eaton Workshop DC, a beautiful comfortable place that goes out of its way to make space for social impact organizations. We have norms to help us maximize the time. Mind the clock, and draw work sessions to a close on time with concrete outcomes (or open work defined to pursue later). Give space for everyone to contribute with the “Rule of N” (in a group of N people, talk 1/Nth of the time). Close your devices. Assume good intent. Bias ideas for action.
Reflection
The starting point was purposefully enjoyable — share proud moments from the last year. We found billions of chickens with Earth Index, validating usefulness with partners. That led to starting work with Climate TRACE on cattle CAFOs. We have great partnerships with philanthropies and paying customers. The infrastructure to assemble and index imagery is incredibly responsive, building a cloud free global mosaic and taking very short time to respond to user AOI requests. We can now search the entire Amazon and are on the way to global coverage. Amazon Mining Watch relaunched with Mozilla support, and sharing Earth Index at MozFest delighted the audience and converted AI skeptics. Demos regularly delight, with literal responses like “THAT WAS SO COOL”, and winning over new folks ready to sign up to build with us. The team came together fast in 2024, ramped up and was very effective in a short time.
Then to a harder reflection — challenges. And the top challenge is exactly the inverse. We are resource constrained to hit the ultimate vision. The last 6 months have been incredibly productive — can we keep up the pace? As a team we’re stretched across building the product, doing business development to fuel the work, and finding time for the underlying science to solve tough challenges in the technology. We need to keep momentum up, but it is challenging to focus with other work priorities across Earth Genome. On honest reflection, the product today is impressive, but has clear technical barriers to scaling and requires a user to remain intensively engaged throughout a search experience. . We had not settled on what priority features will allow usage to freely scale, and have a strong product fit across to unmet environmental monitoring needs. We have good success stories with mining and CAFOs, but there are so many other needs to uncover. People have limited attention span, and meanwhile the entire field of Earth Observation and AI is moving fast. This is the moment for something like Earth Index, and it won’t be open for long.
We took a little time to look into our crystal ball, considering possible good and bad fates for Earth Index. We sticky noted the heck out of this and came up with two broad possible futures. The future isn’t binary, but these two possibilities articulated the extremes of what we projected.
First possible future: Earth Index becomes the go to source for searching the Earth. We have the technical features nailed that drives user adoption. We’re part of the right conversations and partnerships to place Earth Index into users’ hands. We’re bringing in consistent resources to make it happen.
Second possible future: A competitor comes in and out executes the idea. We face growing pains in building the team and can’t deliver. EI ends up being used for bad purposes contrary to our mission. The technology isn’t good enough to meet the need, or it’s too specific and overfit to a niche purpose.
Beyond the next year of development, we need to define in detail what “stable” looks like for the Earth Index technology and business model, identify other needs with Earth Observation, AI and user experience beyond the Earth Index approach, and keep orienting this work with the vision, mission and theory of change of Earth Genome as a whole
Back to the plan for 2025, this reflection lays out the three fundamental questions of what’s needed for the year ahead. What resources as in people and money do we have available? What exactly do we target to build? How do we organize ourselves to get there?
Resources
We have a mix of direct grants to support Earth Index development, and work with directed output where part of the work supports core development. This is a huge challenge but a very useful one — deliver on specific requirements, but also drive the fundamental product forward. The advantage is that the Earth Index is forced to be useful to real needs right now. The problem is finding the commonality across needs and balancing our time and attention.
Examining our available budget, and monthly burn rate for the team with majority focus on Earth Index, we have a solid development runway through to late 2025. There are a couple variables to consider further. One is team growth. We could choose to staff up to take advantage of the moment, and move even faster. This could allow us to do more research on the machine learning algorithms, investigate more data sources and ways to process them, and work with new models. Another variable is more funding, through opportunities in our current pipeline or yet to be explored options. We could make bets on expected growth, and expand our plans as we land new support.
Requirements
So we then spent time looking closely at what our project pipeline requires, and how much time to work on core development they give us. We also all did pre-work before the offsite, to give Earth Index a thorough run and gather our feedback.
A few key themes emerged.
- Monitoring. There’s high demand for timely ongoing detections. That takes more automation for processing. We also need to define a reasonable target for baseline availability and more frequent updates as a more bespoke offering. And the approach to modeling and detecting change needs more fundamental investigation.
- Comprehensiveness. The common need is to find every instance of a search target. Frequently that means working with the Earth Index backend outside of the UI, iteratively using initial search results to seed a random forest model to better score results across the entire AOI. Can we move this to something that runs our infrastructure in an async process? What UX patterns are needed to validate such a large scale of data?
- New data. Bespoke projects are going to require higher resolution imagery, sources with different band definitions, and very different sources like radar. The models need to adapt to these. Tile sizes need to vary. New visualization techniques needed for human review steps. This is all very much in the R&D realm, but needs best possible support in the core infrastructure.
- Customization. Since we expect to have a combination of baseline product and bespoke services, can we enable user permissions within the core to easily support custom embeddings for some users, make the UI flexible to variations like tile size, use Earth Index to directly deliver data results even if prepared in a different environment, and provide visualizations for custom work that avoids forking code or doing one offs.
From our individual pre-work, similar and complementary needs emerged. For newcomers, it’s difficult to know how to start a search, or know which kinds of searches would do well; the ability to seed searches or explore demos would be useful. It’s challenging to judge results, and unclear how much labeling and validation is needed. You can get spoiled after a first easy result, setting an expectation that the whole process is easy (it does take some labor). It’s unclear when a search is “done”, and when it’s time to bring the results into another system for further processing. The lack of global embeddings is a core frustration. There’s lots of UI rough edges to smooth out.
From all of that, what do we prioritize? Where are we heading in the bigger picture, and what is the core set that is needed right now? Looking at both our general user baseline needs, and premium development needs, what signal are we getting on where to go?
After laying it all out, chewing, discussing, sticky-noting, we came up with 2 immediate Product Epics to focus on, and 3 Research Epics to investigate which will inform product development after the first two.
We are still in motion on detailed technical planning for the two Product Epics, but from rough discussion, mid 2025Q1 seems a likely timeline for these to converge on a delivery milestone.
Product Epic 1: Global coverage for the last 2 years. The number one blocker for users having a good experience with Earth Index is lack of a relevant AOI. Essentially we need the globe. Infrastructure investigations make this a promising near term goal. There are several front end implications as well, since a specific search should still have a limited AOI.
Product Epic 2: Easy Earth. We want baseline users to work as much as possible with little oversight. This encompasses a number of small/medium tasks, organized into Documentation / Content, One Click Sign Up, and small and large UI/UX tweaks. Content builds on the existing docs, to potentially include videos, walk-throughs, case studies, and tooltips in the app. The One Click Sign Up would mean granting a user access to the system, and triggering email intro and usage tracking, with minimum admin effort. The small UI/UX tweaks encompass smoothing rough edges from our backlog and based on a fresh UX audit, and cleaning up or removing change detection for the time being. Larger tweaks would be something like seeding search with a data upload, and light refactoring of the front end code so draw done some accumulated tech debt.
Research epics will cover Exhaustive Change Detection, Exhaustive Search, and incorporating Multimodel and New Data Sources.
‘Rganizing
The team then spent some time assessing our pain points, our ideal working state, and the systematic needs to get there. We need clear, intentional systems to organize the work. We need clear documentation so it’s easy to jump in and out of development streams. There’s a lot of different systems (Asana, GitHub, Google Docs, Slack) in use across Earth Genome, and how they are used needs explicit definition.
Focusing on engineering work, we want to have one place where work can be tracked, making it easy to refer to issues and plans, and keep it lean so it’s not a burden. GitHub Projects can be used to organize issue tickets across repositories, in a single view, that can be further grouped into Epics and assigned timing for planning of milestones / releases. It’s useful to have this on screen during meetings to have a common reference for backlog grooming and planning. Daily stand ups and the weekly meeting are the right place. Slack works just fine. Google Drive works fine for docs, but needs better organization, or at least a top level Earth Index doc that explains where everything is at. Above all, for any of these to work, there needs to be a champion of the process to center the team on using these tools and help everyone hold themselves accountable.
We’re writing up a document that details the development process and identifies who is responsible for what, and detailing out the Product Epics into our GitHub project roadmap.
More soon
We’ll continue reporting out and sharing as we build. Stay tuned.