In our last post, we highlighted the first successful Innovation Bridge Initiative. In this post, we’d like to explain what we’re looking to do in Phase 2, and invite you to join us! We’ll be doing an in-person kick-off in St. Louis, and we’ll continue to expand online collaborations throughout the initiative.
The long-term vision is to build a framework of components that form a digital commons to enable many different organizations to innovate by using the Fields of The World (FTW) components as a foundation for further unique insights and analysis.
We’ll first review the components that were started in Phase 1, and then share where we’re looking to go in Phase 2 and beyond.
Foundational Components
The first phase really focused on establishing the core components, creating a number of new open projects that work in concert and are each useful as independent tools.
- The Fields of The World (FTW) benchmark dataset aggregates and harmonizes a number of open datasets into 1.6 million parcel boundaries and over 70,000 samples covering diverse agricultural landscapes across 4 continents and 24 countries. The goal is to enhance the development of machine learning models for instance segmentation of agricultural field boundaries.
- The FTW Baselines Codebase provides tools for data pre-processing, model training, and evaluation, making the process of creating and using models from FTW much more accessible. It works seamlessly with the FTW dataset, and any data that is formatted to match the FTW datasets can also be used.
- FTW Pre-trained models are the initial release of models trained on the whole dataset (plus one just on the CC-BY licensed subset), enabling anyone to see results. The ftw-tools command-line (part of the baseline codebase) enables anyone to easily run the model and generate common GIS formats. More details on all FTW components can be found in this post.
- Fiboa is a common language for all agriculture data, to help enable more interoperability between everyone creating field boundaries. More data in fiboa enables easier incorporation into the FTW benchmark dataset. Over 100 million field boundaries are now in the fiboa format. See this post for more.
Phase 2 Goals
If the theme of phase one was ‘foundations’, the theme of the next initiative is ‘users’. Using our Innovation Bridge lingo, the ‘foundations’ are the bridge from academia to industry and commercial uses. In Phase 2, we want to put the model results in front of real users and get the right feedback loops to make all components of the ecosystem better. This may lead to improved models, more benchmark data, enhanced tools to make it easier to create or use the data, and/or better core schemas and extensions. Those improvements will be based on real user needs, so we’ll be establishing a strong ‘stakeholder’ group that will review outputs and give feedback.
Our goal is to engage with a group of stakeholders with a strong focus on sustainable agricultural practices and food security. In Phase 1 we had a lot of interest from organizations who are focused on the EU Regulation on Deforestation-free Products (EUDR), as well as organizations interested in measuring things like cover crops and tillage. One clear need for FTW to support EUDR compliance is to include field boundaries for tree crops like palm, rubber, and coffee, which are not represented in the FTW benchmark dataset to date.
Our goal in the Phase 2 initiative is to continue to bridge the academic research teams and industry. We are adding a focus on producing data in collaboration with users in sustainable agriculture and food security. We are looking to add two roles to our community: the first is the role of sponsor; the second is the role of stakeholder. With sponsors, our interest is to engage directly with people from organizations with an interest in influencing the direction of FTW because they can see a future benefit from it improving. We want our sponsors to shape our priorities and give us feedback on early progress by engaging weekly with our community.
In the stakeholder category, we’re looking for people interested in staying aware of our progress and providing some feedback in the middle and end of our active effort. For this group, we’ll host progress sessions every other month where we can incorporate general feedback.
The key to a successful innovation bridge is to do everything in concert with the commercial ecosystem. The goal is not to ‘compete’ with any proprietary field boundary datasets or models. Our goal is to get more people collaborating on the ‘commons’, to enable easier innovation on top of that commons. So we are actively looking for companies to be users, sponsors, stakeholders and direct contributors. Together we can build better foundations, and save everyone time on the elements that are shared challenges, to reach the higher order and higher value analyses, as a rising tide that lifts all boats.
Phase 2 Topics
We have not fully locked in exactly what we’ll work on in Phase 2, as we want our stakeholders to shape the direction, and we’re also running everything like the best open source projects, where collaborators may choose to expand the core in different ways. But the set of topics will likely include the following:
- Production of FTW boundary data – Since all the foundations are now laid, a key goal will be to use the models to generate high-quality field boundary data.
- Model improvements – Produce outputs from the baseline models and then enhance the models and the underlying data to achieve better results, in tight feedback loops with users.
- Scale & accessibility – Make it easy to run models and get good results on larger areas, and build the tooling (QGIS plugins? Web interface?) to make it easy for anyone to get results.
- Feedback loops – Explore how to go from users finding issues with the results to directly contributing data updates that are incorporated into the models.
- Increased field diversity – Most of the FTW baseline data is on traditional annual crops, and the models understandably do best there. To meet more needs, the plan is to incorporate pastures and permanent crops into the baseline data and test output.
- LLM querying of data – The LLM demo from Phase 1 generated a lot of excitement, so Phase 2 will explore further enhancements so it is more powerful and robust.
- Fiboa core evolutions & extensions – The fiboa schema language has proven useful, but needs to evolve to support its use. A goal is to flesh out more ‘extensions’, especially ones focused on sustainability and detections of insights from space.
- Grow the collaborative community – The projects have all attracted new collaborators who found us online. We want to be sure that continues, making sure the components have momentum beyond just the TGE initiatives.
Get Involved!
We’re really excited about this next phase of work, and we’d love to have you join us. Since the beginning, we’ve made a concerted effort to make the core component projects truly open and welcoming to outside contributors. We’re proud of the fact that some of our best contributors joined us directly through the open projects, so just jumping into the GitHub repositories is one of the best ways to get involved immediately. You can start with github.com/fieldsoftheworld for the data, models and related tools, or github.com/fiboa/ for schemas, data converters or related tools.
If you’d like to get updates on ways to get involved in this next phase and/or are excited to join us in St. Louis then please fill out this Google form. We’ll be emailing everyone who submits there with details on how the communications will work during this phase, so even if you’re just interested in following along, then please submit there. Though I can’t promise we won’t try to get you to contribute! With such ambitious goals, we know the only way we’ll get there is by all working together.