From Remote to Field Intelligence: Scaling Tropical Agroforestry with Smallholder Farmers
Session Description
This session started as a PUR Project proposal and became a joint one once Kobo, whose KoboToolbox underpins much of PUR's field data collection, reached out to co-lead. PUR shared 16 years of experience delivering agroforestry projects with smallholder farmers, using an iceberg image to make the case that reporting and dashboards are only the visible tip, while the real work, farmer engagement, on-the-ground adaptation, training, sits below the waterline and rarely reaches clients directly. A 2022 story anchored the session: PUR discovered they were losing more trees than expected, and only found out why by going back into the field and asking farmers directly, some had given saplings away, some hadn't planted at all. They also traced a specific cause: for their first 12 years, a six-month gap between delivery and first monitoring meant roughly 30% of trees were never planted, and closing that gap has driven their best-ever survival rates since 2023.
The session closed with three facilitated small-group discussions, rotated so each attendee contributed to all three: what indicators are needed to unlock investment at scale, what the operational and social challenges are in scaling agroforestry with smallholders, and what solutions exist for combining field data, technology and remote sensing at scale.
Speakers
Bianca Ximenes, Chief Data and AI Officer, PUR Project
Caroline Favart, Senior Data Product Manager, PUR Project
Tino Kreutzer, Chief Operating & Innovation Officer, Kobo
Watch the Session Recording
Key Takeaways
Remote sensing alone consistently misses what's actually happening on smallholder land. Field data revealed that trees were being given away, left unplanted, or affected by farmers' own autonomous decisions, none of which shows up from satellite imagery.
Land in these projects belongs to the people living on it, not to the project developer. Farmers are free to plant, not plant, reassign, or change crops entirely, which means social and human context (including basic safeguards like checking school attendance to guard against child labour) has to be built into the data model from the start.
Shortening the gap between input delivery and first field monitoring had a direct, measurable effect. PUR's six-month monitoring gap in their early years was linked to a 30% tree loss rate, and closing that gap materially improved survival outcomes.
Field intelligence and remote intelligence aren't substitutes, they're most effective as a continuous back-and-forth: remote data screens for eligibility and flags anomalies at scale (deforestation, land-use change, weather events), while field data confirms and explains what's actually driving those signals.
Smallholder tropical agroforestry has structural characteristics, small and fragmented plots, dispersed fields, heterogeneous pre-existing vegetation, that make it a particularly hard case for scalable remote-sensing-led MRV, and the sector still lacks shared, public evidence on where the accuracy-versus-scale trade-off is acceptable.
PUR's operating model increasingly screens land for eligibility before engagement, meaning some communities are only approached if a non-carbon client or funding structure fits, since carbon methodologies wouldn't pass audit on that land otherwise. This has direct implications for which communities get access to which kinds of project funding.
