That idea has been sitting with us for a reason: this year, for the first time, we actually asked our agroforests something instead of just assuming we knew the answer - and it talked back.

Team updates and fire season are still in here too, further down. But we wanted to lead with the results of a pilot biodiversity study we just got back, including a bird we don’t have enough data on, and a clear-cut answer to a question we’d been too scared to ask: is our agroforestry model actually better for biodiversity than the conventional farming it replaces, or have we just been assuming that because it sounds right?

We Finally Asked the Land What It Thinks of Us

About six months ago, we started working with researchers Pooja Choksi, of Ficus Research Consulting, and Sarika Khanwilkar to help us answer a question we’d been thinking about for a while - is our agroforestry actually doing anything for biodiversity, or are we just planting trees and hoping very hard?

The ecologists spent January crawling around our plots with sound recorders and camera traps so that we could stop assuming and start measuring.

We now have an answer. It’s better than we hoped, wrapped in exactly as many caveats as good science requires.

The setup, and the guessing game we lost

Eight agroforestry sites (treatment), four monoculture cropland sites (control), and four naturally occurring open natural ecosystem sites (benchmark), all wired with acoustic recorders and camera traps across two districts in Maharashtra.

Two of our urban/peri-urban biodiversity hubs came along for the ride as supplementary sites too. Where our regular agroforestry plots are planted with an eye toward what farmers can actually harvest and sell - fruit trees, timber, livelihood species - biodiversity hubs are planted densely and deliberately for habitat alone, no economic use case required.

On the final results call, the team played us two audio clips - one from a treatment site, one from a control site - and asked us to guess which was which. We confidently, collectively, guessed wrong.

First spectrogram below is from the control site. Second one is from the treatment.

Turns out telling a thriving agroforestry plot from a quiet monoculture field by ear is genuinely hard for an untrained human, most likely because these agroforests are young and of small sizes. Which is, conveniently, the entire reason you hire ecoacousticians instead of just guessing.

What the data actually said

Monoculture cropland contributed zero unique bird species. Every single species detected in our control sites was also found in either the treatment or benchmark sites - meaning the cropland wasn’t adding anything to the local bird community that wasn’t already there, somewhere else, better. Our agroforestry and natural sites, meanwhile, each had their own species that showed up nowhere else.

Acoustics out-birded the humans. We detected 75 avian species acoustically versus 65 through old-fashioned point counts - a person, standing still, listening, the way birding has worked for a couple hundred years. Acoustic monitoring was especially good at catching migratory species: 27 detected acoustically versus 21 by point count, and this was the result when we used only a portion of the data for species detection. And control sites didn’t harbour any migratory species that the treatment and benchmark sites did not already host. Even though acoustics detected more species, point counts and observations are invaluable to our understanding of the relationship between sites and birds, moving beyond simply the presence or absence of a species. Worth flagging too: a lot of the migratory species we found are water birds, so it’s likely a mosaic of different land uses and land covers across the landscape, not agroforestry alone, that’s driving this result.

We found a bird that we don’t have enough data on. The Rock Bush-Quail - a species that the State of India’s Birds defines as “data deficient” because we currently lack sufficient data to track long-term and annual trends in its population - was detected in our treatment sites, and only acoustically. No one at Farmers for Forests has ever seen it on these plots (though villagers here probably have - it’s distributed across parts of Maharashtra). Sharing the soundscape below.

Camera traps turned up their own surprises, too. Across the sites where they were deployed, every single one picked up at least one Schedule I mammal - the highest legal protection category India gives a species. One of the more delightful finds was the Indian crested porcupine, showing up at one of our agroforestry sites.

The fine print

This is a small pilot. Sites weren’t formally statistically matched, and treatment sites got double the sampling effort of control and benchmark. So we’re not declaring “agroforestry saves biodiversity, science says so” and calling it a day.

What we can say is this: these are encouraging signs that show the promise of agroforestry, which - coupled with the open natural ecosystems that have historically dominated this landscape - could provide real conservation value here. Worth building on with more seasons, tighter site-matching, and, per the research team’s recommendation, a locally-trained AI classifier so we’re not simply relying on global classifiers that can often miss or misclassify birds found in this landscape.

More seasons of this are already being planned. We’ll keep you posted, presumably with even better footage next time.

Meet the Humans Currently Keeping This Ship Afloat

If you read Hornbill Dispatch #11, you’ll remember we said the next chapter was about scaling without losing the plot. Turns out scaling mostly means “hiring people so that no single person has to simultaneously run carbon accounting, forest fire response, and a biodiversity pilot.” Groundbreaking management insight, we know.

Introducing them properly would take another 1,000 words. So please just enjoy this yearbook of people you'll be hearing about for years to come (next Dispatch, we will introduce you to many more new team members).

300 Fires, Zero Chill

While the recorders were quietly picking up birdsong, our field teams were busy with something considerably less relaxing: fire season.

This was also the first year we got to run forest protection and forest fire prevention at real scale, thanks to a grant from Salesforce that we’re genuinely thrilled about. What made it more interesting than a typical grant is that Salesforce wanted this treated less like straightforward implementation and more like a live research question - testing whether structured fire prevention actually moves the needle, evaluated against a control, the same instinct that got our research pilot with Project Dhvani off the ground in the first place. So alongside the actual on-the-ground patrolling, we’re tracking reductions in burnt area against historical baselines, which is what let us put real numbers on this season instead of just a lot of tired anecdotes from our field team.

Some context first: Gadchiroli’s forests go through a fire season every year, roughly February through May, before the monsoon arrives and puts a natural stop to it. Nationally, well over 90% of India’s forest fires are set by people, and Gadchiroli consistently ranks among the worst-hit districts in the country. Most of these fires start small and deliberate - forest floor set alight to make it easier to collect mahua flowers off the ground and the new flush of tendu leaves, or to clear visibility and encourage fresh grass growth. Left alone, and without anyone around to catch them early, these same small ground fires are exactly the kind that spread into the larger, uncontrolled fires that take out saplings, undergrowth, and years of regeneration in a single dry season.

Across our forest fire prevention sites in Gadchiroli, our patrol and field teams logged and responded to 300 individual fire incidents (as of June 17th) this season. Median time from detection to the fire being controlled: about 53 minutes. Most of these were small ground fires in mixed forest - the kind that, left alone, don’t stay small for long.

The overwhelming majority were intentional fires - most tied to how mahua flowers and tendu leaves get harvested here, where a quick burn makes picking easier and faster in a short seasonal window. This isn’t carelessness. For many families in these forests, tendu and mahua collection is one of the only sources of cash income all year, and burning is simply the fastest way to make a physically hard, poorly-paid harvest a little less hard.

So alongside patrolling, we tried something more direct this year: paying people to collect without burning. Through door-to-door surveys across 300 households, we found nearly everyone was willing to gather mahua the fire-free way if it didn't cost them time or money to do it. So we hired local daily wagers to manually clear the leaf litter under mahua trees by hand, and to trim tendu plants the old-fashioned way instead of torching them for a flush of new leaves. Across the season, that added up to roughly 1,800 person-days of paid work at ₹312 a day - real income, not just a fire-prevention line item. And it turns out manual clearing isn't just fire-free, it's better: no ash residue, better flower quality, and it let our teams pull out invasive species while they were at it. Our approach isn't about stopping the harvest; it's about making it fire-free.

We don't have a tidy fix for the deeper tension between that immediate income need and a forest's longer-term health, and we're not going to pretend patrolling and buy-back programs alone resolve it. But we can tell you what the sustained patrolling, awareness work, fire-free harvesting, and community agreements added up to this season:

• Shankarpur: a 22% reduction in area burnt compared to the 2020–2025 average, which works out to roughly 182 hectares saved from burning (this number is still being finalized as data collection wraps up)

• Gewardha: a 43% reduction, roughly 424 hectares saved from burning

Put together, that’s over 600 hectares still standing because someone was patrolling at the right time, with the right water, tree branches, or blower in hand. Some of this work is winding down for the season; some of it - awareness sessions, gram sabha agreements, monitoring protocols - continues year-round, because fire season doesn’t actually end so much as go quiet for a few months.

This is the view our drones caught during fire season:

The Thing We’re Not Sleeping Great About

We’d be lying if we said we weren’t watching El Niño forecasts with a low hum of dread. A stronger El Niño season generally means a drier, hotter window for us - which is bad news twice over. It stresses the young trees we’ve spent years getting into the ground, and it makes the fire seasons we just described considerably worse.

We don’t have a tidy resolution to offer here. We’re adjusting irrigation schedules, thinking harder about drought-tolerant species mixes, and bracing our fire teams for what could be a longer, meaner season next time around. We’re also quietly bracing our budget: we already cover the cost of replanting whatever doesn’t survive its first year, and we’re expecting mortality to run higher than usual this cycle, which means replantation costs that don’t look like a normal year’s either. If you’ve read this far and are the praying type, or the “please let it rain on schedule” type, we’ll take it.

Here’s the practical version of that same worry: most of our funding comes tied to specific projects - a grant for this many hectares, a partnership for that particular intervention - which is great, until reality doesn’t stick to the plan. Unexpected mortality from a rough El Niño year is exactly the kind of thing project budgets aren’t built to absorb. It’s the unrestricted, no-strings donations that actually let us respond to a year like this one instead of watching saplings die because the money for them was earmarked for something else. This is one of those years where that kind of giving will matter more than usual.

Kimmerer writes about a pond near her house, and how she came to understand that everybody lives downstream of everybody else - that what she does at her end of the water eventually reaches a lake she’ll never see. We think about our work the same way lately. None of this - the birds, the fires, the saplings - happens in isolation. It’s all one system, and right now that system could use some help.

If this letter was worth your time, here’s one way to do something with that. If you’re in the US, donations are tax-exempt through our 501(c)(3); if you’re in India, there’s a tax-exempt route too. Details for both are on our support us page.

-Arti Dhar, Co-Founder

Last October, I got the opportunity to experience something pretty cool: attending Reid Hoffman's Masters of Scale Summit in San Francisco, listening to and meeting some of the foremost thinkers and doers of our times. And learning new terms like "zero-shot learning," and "AlphaFold moment.” I was there as one of 40 early-stage founders selected to attend, oscillating between taking notes and wondering if the person next to me could tell I’m furiously Google-ing every other word coming out of Fei-Fei Li’s mouth.

On the second day, Reid Hoffman interviewed Aza Raskin, Co-Founder of Earth Species Project - an organization using AI to decode how animals communicate. And Aza told us about a 1994 University of Hawaii study that made my brain short-circuit.

“Do something you’ve never done before. Together.”

That was the instruction researchers gave to dolphins.

Researchers had taught dolphins two gestures. The first gesture meant: “do something you’ve never done before” - to essentially “innovate.” The dolphins understood this. They could remember everything they'd done previously, understand the concept of negation, and invent new behaviors on command.

Pretty cool. But then the researchers taught them a second gesture: “do something together.”

And then they combined them: “Do something you’ve never done before. Together.”

Aza explained: “The dolphins would go down, exchange sonic information, come up and do the same thing that they had never done before at the same time.”

If dolphins are this sophisticated - coordinating innovation through sonic communication - then what else are we missing? How many other species have complex communication systems we simply can’t perceive or decode?

Turns out that’s the question driving the Earth Species Project (ESP). And Aza explained that answering it requires building fundamentally new AI tools. They’re developing large-scale foundation models - NatureLM - designed to work across the entire tree of life, bringing AI’s power to understanding non-human communication.

Wait. Foundation models? For animal communication? That’s... exactly what we’d been building. Just for trees instead of dolphins.

The Foundation Model Parallel

For the past four years, we’d been building tree detection AI. Training models on drone imagery. Creating ground truth datasets. Figuring out how to make models work across different species, different terrains, different lighting conditions.

I’d always thought of this as forestry work - restoration monitoring, carbon accounting, agroforestry optimization. But sitting there listening to Aza, I realized: we weren’t just solving forestry problems. We were solving the exact same AI problems that ESP was tackling for whale clicks and crow calls.

The technical challenges were identical. Here’s what Aza described about ESP’s approach - and how familiar each one sounded:

Problem 1: The 70% We’ve Been Missing

Aza told us about their work with crows at the University of León. Crows have remarkable social structures - collective child rearing like a commune, their own dialects, and they even take in outside adults and teach them the local dialect before those newcomers start participating in childcare.

“We work with the university to put backpacks on crows,” Aza explained. “And we get to record what they’re saying all the time.”

(Yes. Crow backpacks.)

“One of the things we’ve been finding is that around 70% of the calls that they [crows] make are actually quiet, intimate calls. When we brought these to the scientists we were working with, they said: ‘We haven’t seen these calls before.’”

Think about that. Scientists have been studying crows for decades. But they’d only been listening to the loud stuff - the alarm calls, the territorial announcements, the shouts. They were missing 70% of crow communication happening in quiet, close-range conversations.

“We’ve been trying to understand the language of another species by only listening to their shouts,” Aza said. “It’s like trying to understand humans by only listening to Twitter.”

We’d had the exact same problem with forests. For years, the restoration monitoring world has focused on what’s easy to detect from far away - satellite imagery showing large-scale deforestation, broad forest health metrics, regional carbon stock estimates. The “shouts.”

But what about the quiet signals? The subtle stress indicators in individual saplings? The microclimate variations that determine which trees thrive?

Satellites can’t see that granular detail. They miss significantly large parts of the story happening at the scale where restoration actually happened for us - on small farms, with individual trees, in the hands of farmers making daily decisions.

ESP put backpacks on crows to hear the quiet calls. We use cameras on drones for the same reason - to get close enough to see what satellite monitoring misses.

Problem 2: Cross-Species Transfer Learning

“The same tools we are building for crows, these methods can scale,” Aza explained. “The way AI works - learning about one species teaches you about all species. In human language, if I learn German and French and Japanese, I can learn very quickly Esperanto and Aramaic, because each language has a pattern. And the same thing is true of the animal kingdom.”

This is where the foundation model approach becomes powerful. ESP isn’t building a different AI for every animal. They’re building NatureLM - a model trained on diverse animal sounds that can transfer learning across species. The same architecture that detects beluga whale communication patterns can be adapted for crow calls, for elephant rumbles, for bat echolocation.

“We’re building these tools that are crossing the entire tree of life,” Aza continued. “And we’re heading into AlphaFold moments where we will be able to show that many species have names, or many species have compositional language, or many species have abstract thinking going from abstract to specific.”

(For the non-AI nerds: AlphaFold was the AI breakthrough that solved protein folding - a 50-year-old problem in biology. It was a watershed moment. Aza was predicting similar breakthroughs in understanding animal communication.)

We’d built the same approach for trees. We started with organized agroforestry plots in the tropical and subtropical zones of western Maharashtra - rows of sweet lime and mango saplings, predictable spacing, open canopy. Relatively simple detection scenarios. But the real test of a foundation model isn’t how well it works on training data that looks like itself. The test is: can it generalize to ecological contexts it’s never seen?

Because the world needs forest monitoring in places that don’t look like neat agroforestry rows. Dense old-growth sal forests in eastern Maharashtra where individual tree crowns blur into continuous canopy. Bamboo plantations in the Philippines where culms grow in tight clumps. Mixed species agroforestry in Tanzania where farmers are integrating indigenous trees with cash crops in patterns we’ve never encountered.

We can’t build a different detection model from scratch for every terrain type, every tree species, every altitude, every biome, every forest structure. The computational cost would be prohibitive - tens of thousands of new annotations per context, months of retraining, siloed models that can’t learn from each other’s improvements.

Instead, we’re piloting our Maharashtra-trained models in new regions - bamboo in the Philippines, mixed agroforestry in Kenya, dense forest in Gadchiroli - and using ground validation from local partners to understand where the model works well and where it needs fine-tuning. A few hundred validated examples from partners who know their landscapes intimately, rather than rebuilding everything from zero.

The foundation is transferable. Not because all forests look the same, but because the core computer vision task - detecting photosynthetically active biomass in overhead imagery, distinguishing individual plants from background, estimating structure from projections - follows learnable patterns that transcend specific contexts.

Just like ESP’s NatureLM can go from crow calls to whale songs, our detection models can go from agroforestry to old-growth forest. Different ecosystems, different structures, different challenges. Same foundational architecture, adapted through transfer learning rather than rebuilt from zero.

Problem 3: The Cocktail Party Problem (Or: Why Are Orcas Shouting at Each Other?)

“Here’s an interesting puzzle,” Aza said. “Orcas can communicate up to 150 kHz. We can hear up to 20 kHz. But in water, high frequencies don’t go very far. So why are they communicating at such a high pitch?”

The theory: because they’re often communicating in close calls. Intimate conversations, while moving around each other, and often touching. High frequency means short range, which means private. They’re not broadcasting - they’re texting.

But this creates the “cocktail party problem” - when you record whales in the wild, you don’t get clean, isolated vocalizations. You get chaos. Multiple animals calling at once, at different frequencies, overlapping. ESP had to build AI that could separate individual animal voices from this acoustic mess.

We had the exact same problem - just visual instead of acoustic.

When you fly a drone over an agroforestry plot, you don’t get neat rows of evenly-spaced trees. You get chaos. Trees overlapping with each other. Shadows creating false signals. Farmers intercropping saplings with vegetables - brinjal with their characteristic beautiful purple flowers emerging from fields of young papaya trees (see video below from a F4F farmer plot), sweet lime between rows of soybean, custard apple surrounded by pigeon pea. Varying canopy densities. Different heights creating occlusions where larger trees hide smaller ones.

And then there’s the camouflage problem: young saplings get visually lost in the spectral signature of whatever’s growing around them. Noxious weeds like Parthenium hysterophorus grow aggressively in the monsoon, completely obscuring saplings. Cover crops that farmers intentionally plant for soil health look remarkably similar to young trees in RGB imagery. Ground vegetation responding to the same irrigation that’s feeding the trees creates a uniform green background that makes crown detection nearly impossible without careful analysis of leaf texture, growth patterns, and spectral signatures.

Our model learned to separate individual tree crowns from dense, overlapping canopies. To distinguish between a tree and a tree’s shadow. To detect small saplings next to large mature trees without the large tree’s canopy drowning out the small one’s signature.

Different modality - sound waves vs. pixels. Identical challenge - separating individual signals from noisy, overlapping data.

Problem 4: Open-Source or It Doesn't Scale

ESP needs to decode communication across thousands of species - whales, crows, elephants, dolphins, birds, bats, primates. Each with different vocal structures, different contexts, different behavioral repertoires. They can’t build proprietary models for every species and lock them behind paywalls. The problem is too big.

We need to monitor forests across every biome on Earth - tropical, temperate, boreal, arid. Across every terrain type, every tree species, every stage of restoration. We can’t build proprietary detection systems for every context and charge licensing fees. The problem is too big.

The only way solutions to these problems scale is if everyone can build on everyone else’s work. If a researcher in Tanzania can take ESP’s crow model and adapt it for local bird species. If an implementer in the Philippines can take our Maharashtra model and fine-tune it for bamboo. If improvements flow back into the commons, making the foundation stronger for everyone.

That’s why ESP is making everything open-source - benchmark datasets like BEANS (Benchmark for Animal Sounds) that allow researchers to standardize evaluation methods, foundation models like AVES (Animal Vocalization Encoder based on Self-Supervision) that work across species, and the acoustic datasets they’ve collected.

And that’s why we had independently arrived at the exact same conclusion early last year. Not because it’s idealistic. Because it’s the only approach that works at the scale these problems demand.

This is where we stand…

…currently with Project Anāhitā:

Our white paper is published and freely available - the complete technical methodology behind AI-based carbon stock calculation in agroforestry plantations, including our validation results and honest discussions of limitations.

Our ground truth dataset - 7,033 manually measured trees across Maharashtra - is open-sourced on GitHub. GPS coordinates, species identification across 23 species, field-verified DBH measurements, age estimates. Hundreds of person-hours in 45°C heat, freely available to anyone who wants to build on it.

Our complete codebase - the tree detection models, the Gaussian Process implementations, the data processing pipelines - is coming. We’re documenting it properly, which is taking longer than we’d like. But it’s coming, with documentation that we genuinely hope is better than the usual “it works on my machine” approach.

And today, we're releasing TreeLens - free for anyone to use. More on that below.

Launching TreeLens

For the past four instalments of this series, we’ve been telling you about our technical journey:

• Part 1: How muddy QR codes taught us that manual tree monitoring doesn’t scale

• Part 2: How we taught AI to see trees using DeepForest and Detectron2

• Part 3: How Gaussian Process Regression taught our AI to be honest about uncertainty

• Part 4: How ground truth data from 7,033 manually measured trees became the foundation for everything

Today, we’re releasing the platform that makes all of that accessible - to anyone with a drone and a plantation to monitor.

TreeLens is simple to use. Here’s how it works: you fly a drone over your plantation, create an orthomosaic from the images (free tools like WebODM or a paid one like Pix4D can do this), and upload it to TreeLens at farmersforforests.com/tree-lens. The platform accepts .tiff, .png, and .jpg files up to 100 MB, and works best with orthomosaics at around 2.5 cm resolution - the kind any standard agricultural drone can produce. If you also have a Digital Surface Model (DSM) file for your plot, upload that too: it’s what gives us accurate tree height, and from height, biomass and carbon estimates.

What you get back: individual tree detections and counts, species identification across 9 species, height and crown size for each detected tree, and carbon sequestration estimates.

Right now, our models are optimised for managed agroforestry plantations - the organised plots we’ve been training on for four years. Natural forest detection is under active development.

A note on access: TreeLens is free to use - anyone can create an account at farmersforforests.com/tree-lens. Currently, each account supports up to 3 file uploads (max 100 MB each), which is enough to run a meaningful analysis on your plots. You can test it immediately using our pre-loaded drone images, or upload your own orthomosaic. For any questions or concerns, please reach out at tech@farmersforforests.com.

Detect. Debt. Deploy.

At an event a few weeks ago, I met someone from Pula - an organisation making crop insurance work for smallholder farmers across Africa and Asia.

Their insight was deceptively simple. The reason smallholder farmers couldn’t access insurance wasn’t that they were uninsurable. It was that nobody had the data to price their risk. Traditional insurers need to know what’s on the ground - what was planted, how it’s growing, what a normal season looks like - before they can underwrite anything. Pula’s solution was to generate that data and connect it to insurers who could now, for the first time, see the risk clearly enough to price it.

Nature-based projects face a version of the same problem. The trees are real. The carbon is real. The farmer’s commitment is real. But from the perspective of a lender, an insurer, or a government scheme trying to direct subsidies efficiently - none of it is legible. You can’t collateralise what you can’t verify. You can’t underwrite what you can’t measure. This is why, despite enormous enthusiasm for nature-based climate solutions, a lot of financing that reaches farmers still comes from philanthropy - not because commercial capital doesn’t care about trees, but because the verification layer has been missing.

TreeLens is an attempt to build that missing layer. Consider a smallholder agroforestry farmer in Vidarbha with 400 trees on her land, three years into a five-year project. Today, she is largely invisible to formal financial systems. But with verified, timestamped drone data showing exactly which trees are alive, how tall they are, how much carbon they’ve sequestered - she suddenly has something that looks, to a bank or insurer, like an asset with a track record. Maybe better terms on a crop loan. Maybe parametric insurance tied to tree survival rates, not just rainfall. Maybe a government scheme that can verify outcomes rather than just inputs. The same detection layer that produces a carbon estimate can underpin a microinsurance product, a green loan. The asset is the same. What changes is who can now see it.

We think of this as Detect, Debt, Deploy. Detection creates the verified data layer. That layer makes nature-based projects legible to capital - reducing perceived risk enough that commercial and blended finance can flow in alongside philanthropy. That capital then deploys at the scale philanthropy alone will never reach.

None of this is solved yet. The financial instruments are nascent, the integration between monitoring technology and financial services is still being built, and the last-mile challenges of getting capital to smallholder farmers are every bit as hard as the technical ones. But the detection layer - the piece that has historically been missing - now exists.

What We’re Really Asking

When I was listening to Aza Raskin talk about dolphins, what struck me most was the instruction the researchers gave: Do something you’ve never done before. Together.

The dolphins understood something that a lot of organisations working on climate and restoration still struggle with: that the most interesting things - the breakthroughs, the new behaviours, the solutions nobody has tried yet - only emerge when you share information and act in coordination. The dolphins went underwater, exchanged sonic information, and surfaced doing the same new thing at the same time. Coordinated novelty. That’s what open-source is, for us.

We spent four years building this, funded by patient philanthropists who believed in a possibility rather than a product. We didn’t build it to keep it. We built it because the problem - smallholder farmers excluded from the very climate solutions their land stewardship makes possible - is too large and too urgent for any one organisation to solve.

Project Anāhitā is now live in pieces: the white paper is published, the dataset is on GitHub, the codebase is coming, and TreeLens is open to partners who want to try it. We’re not done. We’re probably not even halfway. But we’ve learned enough to know that the next phase of this work won’t be built by us alone.

If you’re an implementer working with smallholder farmers and want to try TreeLens on your plots - reach out. If you’re a researcher who sees something interesting in our methodology or our dataset - build on it, and tell us what you find. If you’re working on the financing side - the insurance products, the green loans, the blended finance vehicles - and you’re trying to figure out what verified tree data could unlock for you, we’d genuinely love that conversation.

And if you've made it to the end of a five-part series about drone orthomosaics and Gaussian Process Regression - honestly, we're a little worried about you. But also, thank you.

- Arti Dhar, Co-Founder

“Farmers for Forests is a bogus company.”

My colleague’s voice crackled over the phone line. This was in June 2025, I was spending a few months in Oakland, California, working with my co-founder Arti. My colleague in Pune was telling me over a phone call about how one of our government partners was furious with F4F about a road side plantation project that had gone awry.

So this is what had happened: the stretch of road where we had planted trees and installed recycled-plastic benches had been overrun by miscreants. They trampled the saplings while gleefully unscrewing the benches and carrying them away to sell. They littered the area with packets of snacks and empty beer bottles. They smashed the barbed-wire fencing installed by the government, leaving gaps wide enough for cows, goats, and raggedy (but adorable!) stray dogs to wander in and feast on or crush the baby saplings we had planted.

I was sitting in the quiet and orderly Oakland neighbourhood, a far cry from the government offices I usually frequent for meetings. The noise, the humidity, the overpowering smells. People jostling to get their requests heard by a government officer. It all felt far away and unreal. My brain, instead of focusing on the gravity of the situation, was strangely fascinated by the government officer’s use of the word “bogus”. It’s one of those English words that has cutely made its way into regular Marathi lingo - like “time-pass” or “scene-scenery” or “tension”.

That morning, before the phone call with my Pune colleague, Arti and I had just completed a successful meeting where we’d presented our tech platform to a funder who was impressed with the progress we’d made. But, here we were, the truth staring us in the face, (more bitter than the morning beverage Arti makes and manages to pass off as coffee):

Having amazing carbon measurement technology means nothing if the trees we’ve planted don’t survive.

Technology might not be able to save us, after all

When we last left you off, our technology could precisely identify every tree in our program and estimate how much carbon it was sequestering. Our Gaussian Process Regression model could predict trunk diameter from crown size and height, then tell you exactly how confident it was in that prediction. Quite a neat piece of tech, no? And honestly, it’s easy to get absorbed in the “technology-making” side of things - polishing graphics, playing with fonts, figuring out the most efficient lines of code, designing a slick UI. It’s rewarding work: effort in, progress out. Predictable, linear.

But none of it matters if the cows eat the trees.

Field work is anything but linear. There are thousands of farmers and landowners to engage with - each with their own personalities, priorities, and wish lists. There are crafty sapling vendors, delayed material deliveries, unpredictable weather, tricky government officials, and field team members who come with their own quirks and limitations. As our co-founder and COO Aditya often says: “It’s not that easy, yaar, to just go and do a project successfully.”

This is the uncomfortable truth that sits at the intersection of technology and implementation: measurement precision doesn’t automatically translate to impact. You can have 92% accuracy in tree detection and uncertainty quantification that would make a statistician weep with joy, and automated carbon accounting that processes forests in hours instead of months. But if your field operations are “bogus,” you’re just precisely measuring failure.

Maybe that’s why so much capital is flowing into the technology space, funding things like remote sensing, AI-driven risk models and remote measurement of capital. It feels like there’s far less for tackling the messy question of how to get innovation and efficiency at the last mile. Even when good ideas exist, like water saving irrigation methods, drone-assisted reforestation, biodegradable sapling bags, the challenge is: how do you make them cheap and convenient enough for widespread use? How do you ensure these products reach farmers and then how do you ensure that farmers actually use them?

Don’t get us wrong - measurement is critically important. As the saying goes: what gets measured gets managed. But measurement alone doesn’t equal success. As we learned the hard way: getting things done on the ground demands creativity, resilience, collaboration - oh and tons of patience and a stomach for sticky sweet tea. Our field team knows this intimately. When people started dozing off during our agroforestry and biodiversity training, they improvised, setting up skits, street plays, serving tea and snacks. All small, human touches that connected people emotionally to the issues at hand.

Of course, technology helps. Our FieldHQ system lets us centrally manage operations: assigning farmers to field staff, tracking sapling inventory, monitoring compost and irrigation systems, and storing all our field data for funder reports and dashboards. But here’s the lesson we learned the hard way: a good tech stack doesn’t automatically guarantee smooth operations. Training teams on tedious data entry, managing field team expectations, streamlining logistics, building relationships with vendors, this is the behind-the-scenes, unglamorous work that matters just as much.

And yet, it’s precisely this kind of work that gets less attention, and even lesser funding. Why would anyone invest in “boring” internal software or field-team training, when drones and AI-powered platforms look so much cooler? These supposedly dull investments are just as critical - and sometimes require far more creativity than precise carbon-monitoring tech.

Which brings us to what might seem like a contradiction: if we just spent several paragraphs arguing that implementation matters more than measurement, why are we about to release a measurement dataset?

Because the 7,033 trees in our ground truth dataset represent both.

When we last told you about teaching our AI to quantify uncertainty through Gaussian Process Regression, we glossed over what that actually required on the ground. It wasn’t just elegant mathematics and clever algorithms. It was our field team trudging through Maharashtra fields in 45°C heat with measuring tapes. It was wrapping those tapes around 7,033 individual tree trunks. Recording GPS coordinates. Identifying species. Estimating ages. All while managing the same messy reality we’ve been describing - delayed equipment, uncooperative weather, negotiations with farmers.

Each of these 7,000+ data points required someone to do the unglamorous work: locate the specific tree, physically measure it, log the data correctly, and verify it later. This wasn’t remote sensing from a comfortable office. This was getting mud on your shoes and sweat on your shirt, tree after tree after tree.

So when we say our AI is “honest about uncertainty,” what we really mean is: our AI learned honesty from field teams who did the hard, boring, unglamorous work of measuring reality. The algorithm provides the precision. The humans who wrapped 7,000+ measuring tapes provided the honesty - documenting what they actually found, not what we hoped to find.

Today, we’re open-sourcing that entire ground truth dataset. 7,033 trees, manually measured across Maharashtra:

• GPS coordinates (latitude/longitude) for every tree

• Species identification (23 different species including Mango, Custard Apple, Papaya, Jamun, Guava, Tamarind, and more)

• Actual DBH measurements (field-verified with measuring tapes)

• Tree age estimates

This dataset represents hundreds of person-hours in the field. It’s the foundation that taught our Gaussian Process model to be honest about uncertainty. And we believe it should be a public good.

But here’s the uncomfortable question that keeps us up at night: does doing the right thing make for a bad business?

Of value, values, and valuation

A founder friend once told me: “The stronger your values, the lower your valuation.” It resonated! Because let’s be honest: no one wants to hear about the difficulties of implementation or the hard truth - “hey scaling this might take 5 years and not 2”.

Investors want grand visions: global platforms, exponential scale, big returns. It’s not about doing one 5,000-hectare project well - it’s about imagining 500,000 hectares across continents. And to be fair, ambition is good. Ambition is needed, if we have to create impact at scale. But ambition shouldn’t blind us to the truth: creating anything of real value takes time - and is a bit boring.

Back in 2019, a chance encounter led to the founding of Farmers for Forests. Arti and I, friends from college who hadn’t really met properly since 2010, randomly found ourselves working out of adjacent rooms at a Regus co-working space. She was at Social Finance, working on an education impact bond; I was at J-PAL, exploring Public Distribution System reform with the Maharashtra government. Nine years since we’d last really met, we found ourselves reunited over endless cups of chai and conversation in the bylanes behind Mumbai’s Secretariat building on HT Parekh Marg, and from that reunion, Farmers for Forests was born.

I sincerely believe that HT Parekh Marg might have some of the best street food in Mumbai. In addition to there being plenty of cats and dogs to hang out with, food carts line the road, serving adrak-elaichi chai, paani puri, sandwiches - you name it. The MLA canteen offers all sorts of hot meals from thalis to dosas. Delicious, affordable, and reliable. Vendors work on a simple cost-plus-margin model, and customers keep coming back because they see value in a hot cup of tea and nice spicy Bombay Sandwich after a stressful government meeting.

The value or price for most of the things we consume like chai or clothes is usually easy to define with a cost + margin model. Valuing things like companies or nature or carbon on the other hand is infinitely complex - things people make entire careers out of. Let’s take the example of Regus for example, the co-working space we were spending our working hours in Mumbai from. In 2019, as Arti and I chatted away, market observers were puzzled. Regus’s parent company, IWG, had a market cap of $3.7 billion. Its competitor WeWork meanwhile was valued at over $47 billion, despite having half the revenue and being massively unprofitable.

We now know why: Adam Neumann’s storytelling. WeWork was not merely a real estate company, it was a tech company - a platform that was going to revolutionise the way people worked and lived. Neumann sold an audacious vision of the future. WeWork was a company that was going to see hyper growth and dominate global markets. Exciting - yes, but often hard to implement quickly or at scale.

[How] Should we financially value nature?

The Marathi language has this beautiful habit of using endearments for nature and its creatures. Cats are maternal aunts, squirrels and sparrows are older sisters, the crow is an older brother and the moon is an uncle. Deep indications that our ancestors saw themselves as inextricably linked with nature - viewing nature almost as a headstrong but proficient relative, one to be looked upon with fond indulgence and relied on for everyday survival.

Nature is sacred and so putting an economic value on it can rightly seem blasphemous. I was personally horrified at the thought at first, but the practical reality is - we buy water, pay for safaris in natural parks, purchase flowers, fruits and vegetables. Maybe it’s not the pricing of nature that’s the problem, but the grossly inaccurate pricing of it.

Insects pollinate more than 75 per cent of global crops and provide ecosystem services valued at $577 billion annually, according to the Intergovernmental Science-Policy Platform on Biodiversity and Ecosystem Services. Unfortunately, our global economic models have been shamefully bereft of accurately accounting for these indispensable services.

The result? In just the last 27 years, insect populations have dropped by 75%! That’s a nightmare-inducing statistic really. Why isn’t this the only thing all of us are talking or thinking about?

Carbon credits and biodiversity credits are at least a start, an attempt to value nature in ways our current economic models can understand. Flawed as those models may be, they remain our best short-term bet. A carbon credit system lets major emitters offset some of their damage through verified sequestration, channelizing much needed finance to players like us on the ground. Ideally, companies should first decarbonize their supply chains as much as possible, and only then purchase offsets for what remains too difficult to eliminate. The danger is in treating offsets as a free pass - after all, we can’t just offset our way out of all the 50 billion tonnes of emissions we put out every year.

Still, carbon standards are grounded in solid, peer-reviewed science, with methodologies that try to quantify these impacts. Combined with open-source technology, realtime, AI-powered technology this could become a game-changer - directing more finance to nature-based projects, while ensuring that finance actually reaches implementers on the ground.

And the open-source part is crucial. Without it, capital risks pooling in the hands of those who already control both technology and money. With it, communities and implementers can be fairly compensated, encouraged to keep innovating, and supported in ensuring projects actually endure. Because carbon is only sequestered if trees survive for 100 years or more, and if farm fires and stubble burning really stops.

Back to that “bogus” accusation

In the end, we went back and replanted those trees and worked hard on mending our broken relationship with the government officer. This time, we’re adding sturdy tree guards and planted significantly taller saplings, so they wouldn’t be trampled or eaten, and slowly the little patch has begun to look green again. The government officer who once dismissed us as “bogus” now greets us without anger - a small but meaningful reminder that persistence on the ground matters as much as all the technology we are developing.

Our sophisticated AI can now measure carbon with unprecedented precision. Our Gaussian Process model can quantify uncertainty in ways that make forest monitoring more honest than it’s ever been. And we’re open-sourcing it because we believe democratizing measurement technology can unlock better financing for restoration.

But we’ve also learned that none of this matters if we don’t also get the unglamorous parts right: the tree guards, the community engagement, the vendor relationships, the boring internal systems that track whether saplings actually get planted, within budget and survive in the long-run.

It’s these everyday acts of resilience - combined with the right financial and technological innovations - that will decide whether efforts like ours can scale.

- Krutika Ravishankar

In part #5 of this 5-part series, we’ll release TreeLens - a platform where anyone can upload drone imagery of their plantation and get back tree detection, carbon sequestration estimates, and species identification. No coding required, no technical expertise needed. We’ll also explore how blended finance and open-source technology can unlock the trust and resources needed to truly sustain nature-based solutions at scale, and introduce our principle of “Debt, Detect, and Deploy” - because measuring carbon honestly is only step one. Making that measurement accessible to everyone is step two. Getting it financed sustainably is step three.

We were feeling pretty smug about ourselves. Our AI could spot trees, measure their heights, and even identify a few species. We'd gone from muddy QR codes to sophisticated computer vision in just two years. Time to pop the champagne, right?

Wrong.

We still weren’t sure how to measure the carbon in our trees. Here’s why.

The Hidden Measurement Problem

Picture this: You're trying to estimate how much money is in someone's wallet by looking at them from across the street. You can see their height, you can guess their age. You might even know what kind of job they have(?). But the actual cash? That's hidden in their back pocket.

This was exactly our problem with trees.

(Yes, we're comparing trees to people's wallets. And yes, we couldn’t come up with a smarter analogy. Stay with us though!)

The amount of carbon in a tree is a function of the size of the tree trunk and the density of the wood. And so, to measure the carbon accurately we would need to know the diameter of the tree trunk and also know what species of tree it is, so that we can factor in the wood density.

Our drones flying overhead could see everything except the most important measurement for calculating carbon storage: the trunk diameter. It's literally hidden beneath the canopy, invisible to our aerial eyes. And without trunk diameter (or DBH - Diameter at Breast Height, as forestry folks call it), calculating how much carbon a tree stores without knowing the DBH is like trying to bake a cake without knowing how much flour to use in the recipe (lame analogy #2).

The run-of-the-mill solution would be to re-send our field team to these 55,000 trees we had annotated (to train the algorithm to detect trees) - but this time with measuring tapes to wrap around every single trunk. And that too in the horrid Maharashtra heat.

But instead, we asked ourselves what if we could teach our models to make predictions from visible data? Really, really sophisticated predictions.

Enter the Mathematician

In early 2024, we realized we needed someone who spoke fluent "uncertainty." Someone who could teach our models not just to predict trunk diameter from crown size and height, but to also say "I'm 85% confident in this prediction" or "This tree is weird, I'm only 60% sure."

Enter Dr. Abhishek Gupta from the Indian Institute of Technology (IIT) Goa, our mathematician who speaks fluent "uncertainty."

The breakthrough came in a serendipitous brainstorming session between Abhishek and Makarand Datar, our VP of Wildlife & Tech. As they discussed where uncertainty quantification would be most useful in our pipeline, old academic memories started surfacing. Makarand's master's thesis had involved something about Gaussian Processes and vehicle dynamics - dusty knowledge from years past that suddenly seemed relevant to trees.

The conversation evolved, ideas bounced back and forth, and eventually they landed on applying Gaussian Process Regression to our DBH prediction challenge.

The concept, as Abhishek eventually explained, was beautifully simple. Traditional AI is like a confident student who always gives you definitive answers, even when it’s guessing. Gaussian Process Regression is like an honest student who admits uncertainty - providing both a prediction and a confidence level based on how similar the current problem is to ones it has encountered before.

The Honest AI

Here's how we taught our AI to be honest about uncertainty:

Step 1: Collect Some Ground Truth Data: We did send our field team with measuring tapes. But instead of 55,000 trees (which was our original training data), they only visited 7,000 trees. For each tree, we recorded three things: i) Tree Crown size (how wide the leafy part is ii) Tree Height (from drone data) iii) DBH (from wrapping measuring tape around the trunk)

Step 2: The Log Transformation Trick: Here's where it gets mathematically interesting. Trees don't grow in nice, predictable patterns. Some skinny trees are tall, some fat trees are short. The relationship between crown size and trunk diameter is... messy. Abhishek showed us something cool: instead of trying to predict DBH directly, we predict the logarithm of DBH.

"Why?" we asked.

"Because trees follow something called a log-normal distribution," he explained. It's like how most people have average incomes, but a few people are super rich. Most trees have average trunks, but some are really thick. The logarithm transforms this skewed data into a nice, normal bell curve that the math can handle.

Which means, instead of dealing with a skewed distribution, we transform it into a normal distribution that is better suited for the model to learn from.

Step 3: Teaching Uncertainty: This is where Gaussian Process Regression gets magical. Traditional machine learning gives you a single prediction: "This tree's DBH is 15.3 cm."

Gaussian Process Regression gives you a prediction plus uncertainty: "This tree's DBH is 15.3 cm, plus or minus 2.1 cm, and I'm 90% confident it's in that range."

But here's the really clever part - the AI learns when to be uncertain. If you show it a tree that looks similar to ones it's seen before, it'll be confident. If you show it something weird, it'll admit it's guessing.

The Aleatoric vs Epistemic Uncertainty Problem

Abhishek introduced us to two types of uncertainty that sound like Harry Potter spells but are actually crucial for forest monitoring:

Epistemic Uncertainty: Which means "I don't know because I haven't seen enough examples."

Solution: Collect more training data.

Aleatoric Uncertainty: Which means "I don't know because the world is genuinely random."

Solution: Accept that some things are inherently unpredictable.

Our tree data had both types of uncertainty. Some trees were just inherently weird - same crown size, same height, but completely different trunk thickness. Maybe one grew in rocky soil, another in rich earth. Maybe one got more water.

Who knows? The world is random that way.

The Gaussian Process basically learned to say: "For trees with this crown size and height, I've seen trunk diameters ranging from 12-18 cm. Based on the patterns, I think this one is probably 15 cm, but there's natural variation I can't account for."

The Coverage Plot (Or: How We Knew Our AI Wasn't Lying)

Here's how we tested whether our AI was actually being honest about uncertainty:

We took our validation data (trees the AI had never seen during training) and asked it to make predictions with confidence intervals. For example:

• I'm 60% confident this tree's DBH is between 12-16 cm

• I'm 80% confident this tree's DBH is between 11-17 cm

Then we checked: when the AI said it was 60% confident, was it right 60% of the time? When it claimed 80% confidence, was it right 80% of the time?

We plotted these results on what's called a "coverage plot." A perfectly calibrated AI would show a straight diagonal line - 60% confidence = 60% accuracy, 80% confidence = 80% accuracy.

Our AI showed a slope of 1.08 - meaning it was slightly overconfident, but remarkably well-calibrated.

To put it simply, our AI wasn't just making good predictions - it was being honest about when it was guessing.

From Trees to Carbon: The Final Calculation

With honest DBH predictions, we could finally calculate carbon storage for every single tree:

The formula:

1. Biomass = exp(-1.996 + 2.32 × ln(DBH)) (from Verra’s VMD0001 methodology document)

2. Total Tree Biomass = Above-ground Biomass × 1.27 [accounting for roots]

3. Carbon Storage = Total Biomass × 0.5 × (44/12) [molecular weight conversion]

But here's the beautiful part - because our DBH predictions came with uncertainty estimates, our carbon calculations also came with uncertainty. We could say: "This forest has stored 45.3 tons of carbon, plus or minus 3.2 tons, with 90% confidence."

For the first time, we had automated carbon accounting with built-in honesty.

The Validation Moment

The real test came when we compared our AI predictions to manual field measurements across two case studies (two distinct intervention plots):

Case Study 1: AI predicted 15,480 kg of biomass. Manual measurements (with extrapolation): 18,142 kg. Difference: 15% overestimate.

Case Study 2: AI predicted 5,680 kg of biomass. Manual measurements: 5,693 kg. Difference: 0.2% overestimate.

Our AI was consistently within the uncertainty bounds it claimed. When it said "plus or minus 20%," the real answer was indeed within that range.

And more importantly, we could process entire forests in hours instead of months, and provide uncertainty estimates that manual sampling simply can't match.

..but wait, what about Tree Species?

By December 2024, we had achieved something quite remarkable: an AI system that could count trees, measure their dimensions, estimate carbon storage and even tell us how confident it was in its own calculations. Fully automated. Completely transparent.

But once we could measure carbon, a different question emerged: what if we could also teach our models to recognize different tree species? Imagine an algorithm that not only says “here’s a tree,” but also tells you this one’s mango, that one’s custard apple, and over there, that’s a bael tree.

The possibilities are powerful. Species detection doesn’t just sharpen carbon estimates - it creates living inventories of trees, mapped by location. Suddenly, it might be possible to get remote data to feed into decisions like setting up a custard apple pulp processing unit in a region with abundant supply, or identifying landscapes where certain species are in decline.

Take the bael tree (Aegle marmelos), revered in Hindu tradition - its leaves offered to Lord Shiva and its fruit turned into delicious candies and jams. Despite its cultural and ecological importance, the bael population in India has shrunk by 25% in just a few decades, landing it on the IUCN’s Near Threatened list.

AI-driven species detection could become a tool not only for planning and economics, but also for conservation, helping us protect what’s sacred before it disappears.

The Platform Decision

As we began explaining our tech to folks, other on-ground implementers started asking: "Can we use this too?"

We faced a choice. We could keep this as our competitive advantage. Build a business around proprietary forest AI.

Or we could do something different.

We remembered why we started this journey - standing ankle-deep in monsoon mud, watching QR codes wash away, frustrated by how expensive collecting and analyzing high quality and granular data in this sector was. That problem wasn't unique to us. Every nature organization in the world faces similar challenges.

We were also acutely aware of how we'd been able to develop this technology in the first place. It was primarily funded by patient philanthropic capital - funders who believed in us and our vision from year 1 - when the entire F4F team was just the co-founders, one field team member, and a two-person tech team (one of whom was volunteering his time with us). These funders invested not in a guaranteed product, but in a possibility.

We truly think this should be a public good. Given the history, hoarding it as proprietary IP felt wrong. So we have decided to open-source everything.

Announcing: Project Anāhitā by Farmers for Forests

Named after Anahita, the ancient Persian goddess who was considered a guardian deity, symbolizing life-giving forces, Project Anāhitā is our attempt to democratize forest monitoring. Because, apparently, we needed a suitably grand name for what amounts to years of work that we're now cheerfully handing out to anyone with a conservation and restoration dream.

We're releasing four resources that represent everything we've learned about teaching machines to count trees:

1. The White Paper (Available Today)

Our complete technical methodology, published as "AI Based Automated Calculation of Carbon Stock in Agroforestry Plantations." Everything from our AI training process to validation results to honest discussions of limitations. The full scientific process behind teaching machines to count trees and estimate carbon storage with uncertainty quantification.

2. Ground Truth Dataset (Coming in 2 weeks)

The crown jewel of our data collection efforts: annotated ground truth data with GPS coordinates, real and estimated DBH measurements, species identification, and age estimates for select trees. This dataset represents hundreds of person-hours of our field team measuring trees in 45°C heat. We're sharing it because we believe good training data shouldn't be a competitive moat.

3. TreeLens Platform (Coming in 2 weeks)

A web interface where any organization can upload drone footage and get automated analysis. No setup required, no technical expertise needed. Just upload your orthomosaic and get back geo-tags, tree counts, carbon estimates, and some species identification.

4. Complete Codebase (Coming soon)

All our tree detection models, Gaussian Process implementations, and data processing pipelines. It will be available on GitHub with documentation that we hope is better than the usual "it works on my machine" approach.

Why We're Doing This

Ecosystem restoration often fails globally because monitoring is too expensive and accountability is too low. Organizations plant trees, report 'success', and hope for the best because proper monitoring would cost more than the tree planting itself.

We're not trying to build the next unicorn. We're just tired of watching good restoration projects fail because nobody can afford to monitor them properly.

By making professional-grade forest monitoring accessible, we hope to:

• Increase survival rates through early intervention when something goes wrong

• Improve funding flows through transparent, granular impact data

• Scale restoration efforts by reducing monitoring costs from prohibitive to affordable

• Build trust in nature-based climate solutions by moving beyond "plant and pray"

The Invitation

Starting with today's research paper, we invite you to dive into our methodology. Poke holes in our approach. Tell us if we got anything wrong.

But we would also love to know if we got something really right!

In the coming weeks, as we release the dataset, platform, and code, we'll walk you through exactly how to use each resource. Whether you're a researcher wanting to build on our models, an implementer needing to monitor your own trees, or just someone curious about how AI can count trees from the sky.

Because the future of forest monitoring isn't going to be built by one organization in central India. It's going to be built by everyone who cares about getting this right.

And maybe, just maybe, we can finally move beyond the "plant and pray" model that's been failing farmers and communities for decades.

In Part #4 of 5 coming in couple of weeks, we'll walk you through our ground truth dataset and share the stories behind some of those measurements. We will also launch TreeLens - our attempt to make drone-based forest monitoring as easy as uploading a photo to Instagram - minus the beauty filters!

When we last left off our riveting tale, our seemingly ingenious QR code tree tagging method had failed quite miserably and we left with “mud on our face, a big disgrace.. QR code debacle put us back into our place!” (Bonus points if you sang that line out in your head while you read it - ha!)

Chucklesome GIFs aside, while our QR code method had failed, we’d chanced upon DeepForest and were now actively using it in combination with drone imagery to count trees.

The first few climbs up our technical mountain

The DeepForest model has been primarily trained on high resolution RGB imagery at the 10cm / pixel resolution from the National Ecological Observatory Network (NEON) tree crown dataset.

The base DeepForest model took one look at our Maharashtra drone images and basically shrugged. We created orthomosaics by stitching together numerous drone images for the same plot, correcting for angles and perspective - ensuring that the final image is to scale and akin to viewing it perpendicularly from above. The model gave us precision and recall values of less than 0.5 for most images.

Precision measures how many of the objects the model predicted as trees were actually trees, while recall measures how many of the real trees the model was able to find. So a number of less than 0.5 means the model was missing nearly half the trees.

The visual characteristics of tree crowns and surrounding landscapes can vary significantly across geographic regions. As a result, the default DeepForest model, which was trained on a specific dataset, might not always perform as well when applied to new environments with different vegetation types, lighting, and background features.

The solution was transfer learning - essentially giving DeepForest a crash course in Indian agroforestry landscapes. We needed to fine-tune this pre-trained model on new data to better suit this new context. DeepForest is particularly useful in this regard, as it offers a ready-to-use structure that can be further trained using local imagery to improve accuracy and fit it to our use case.

What followed was the most tedious phase of our technical journey. We had drone imagery from ~80 land parcels that were representative of the kinds of landscapes where the model will ultimately be deployed. In the case of all 80 images, the drone was flown at an altitude of approximately 250-300 feet and the resulting overlapping images were stitched together to create georeferenced orthomosaics with a pixel resolution of about 3-4 cm. These orthomosaics were then manually annotated using the labelImg tool to mark individual tree crowns. In total, we annotated approximately 55,000 trees, and this dataset was used to retrain the DeepForest model.

The images below demonstrate the improvement in the model post training on our local data from Maharashtra.

Our team then experimented with different image sets, probability thresholds, and training epochs to finally achieve precision and recall scores close to a solid 0.9.

But why just stop at tree counting?

Once all the trees in an image are detected, the height and crown size of each tree can also be extracted. If that is possible - then our team wondered, can we also accurately estimate how much carbon each tree is removing from the atmosphere?

The height of the trees comes from using the Digital Elevation Model (DEM) feature of the drone imagery we captured. We had been using the Mavic M series of DJI made drones. This drone uses image overlap to generate elevation maps using structure from motion technique method.

This DEM thus gives the height of each pixel in the image. Once a bounding box containing each tree is obtained using a tree detection algorithm, the crown size of the tree can be easily calculated based on the size of the bounding box in pixels and the ground sampling distance Mapping this bounding box on to DEM, tree height can also be extracted.

We were now able to automatically identify trees in drone imagery, geotag each one, measure their height, and calculate crown size - all without setting foot in the field! What once took thousands of hours of manual tagging could now be done in just a couple of hours with drone flights and data processing.

Our field team was thrilled: no more mind-numbing tree counting. And our data team was just as happy: finally, clean, comprehensive data on every single tree, without having to sift through and fix thousands of messy rows.

Beyond Mountains there are Mountains

Just as the Haitian proverb predicts, no sooner had our team solved one problem that quite a few more cropped up.

Now that we were detecting trees, could we also maybe detect other things like drip irrigation pipes? We spend nearly 50% of our per plot budget on drip irrigation - snaking networks of pipes that reduced water usage and labor for farmers. But we had no efficient way to verify what was actually installed versus what was gathering dust in warehouses.

If we could identify not only the number of trees planted and surviving on farms, but also the meters of drip irrigation installed, we’d gain a much clearer view of our operations and our inventory:

• how many saplings and metres of drip pipe we bought,

• how many were deployed on farmer lands and,

• how many lay in inventory in stock at our local warehouses?

In essence, we’d be moving from just counting trees to monitoring over 60% of our cost items with unprecedented precision.

Enter Detectron-2

Detecting irrigation pipes with bounding boxes is like trying to gift-wrap a garden hose - the box either misses half the pipe or includes half the farmland. Bounding boxes are effective for detecting objects like trees, which have relatively consistent and compact shapes. Drip lines snake across fields in irregular patterns, too long and skinny for standard object detection. We needed pixel-level precision, not approximate boxes.

Turns out, what we needed was a more precise approach called object segmentation, where each pixel of the object is identified and labeled. This method allows for more accurate mapping of the pipe’s exact shape and position on the ground, especially when they wind through varied terrain or intersect with other visual elements in the imagery.

Our search for semantic segmentation tools led us to Detectron2 - an open-source computer vision library that seemed perfectly suited for this pixel-level precision work. If it could handle the complex segmentation tasks it was designed for, maybe it could tackle our drip pipe challenge.

Detectron-2 is an open-source computer vision library developed by Facebook AI Research (FAIR) that helps computers detect and understand objects in images or videos. It can do things like find people, animals, or cars in a photo, draw boxes around them, and even outline their exact shapes. It's widely used for tasks like object detection, segmentation, and keypoint detection.

Detectron2 was going to be viable for object segmentation (or at least it provided the necessary machinery for the purpose). But when we annotated our full images with long snakelike annotation objects skirting our drip lines, the model performed very poorly. The annotations were still too lanky and each image provided only a few annotations based on the number of drip lines.

Here, some creative problem solving came to the rescue. What if we break the image into small square shaped tiles and annotate the part of the drip that was seen in that tile? This annotation would be much stubbier and when even during prediction, the model would need to predict only on one small tile at a time: much easier to predict, easier to annotate and more annotated tiles possible. This was a real breakthrough and we were able to effectively detect most of our drip lines on the images that were not used in model training.

However there was another challenge, prediction on a single image was taking 2 hours. First, the image was broken into tiles, then prediction on each tile and then recombining the results to get plot level detection! Something had to be done that allowed parallelizing this on a GPU. Here came another soldier from the world of open-source awesomeness: SAHI: A model agnostic tiling/slicing code. Now, we were able to predict in 2-3 minutes what was taking us 2 hours before. We could detect drip in a series of sections.

Detectron2 + SAHI used for drip detection had revealed something new: We didnt have to train even the tree detection model on large images. Only small sections of images could be included in the training and inference on sections of the images could be made and results recombined.

Training directly on full orthomosaic images isn't practical - they're extremely large (often 1 GB or more, with millions of pixels) and contain 300 to 800 trees each, all of which would need to be manually annotated. That would take a huge amount of time. But because the trees and backgrounds in each orthomosaic are quite similar and repetitive, annotating just a small portion is usually enough to train an effective model. This greatly reduces the annotation effort and computational load.

It was making a lot of sense then to move from using DeepForest to using a comprehensive, unified model Detectron-2 for all of small tree, big tree and drip irrigation counting and that’s what our team did.

Carbon: the final frontier

So we were in a pretty good place now - we were successfully, remotely and automatically measuring drip pipes, counting trees, their height and calculating their crown size. But something was still missing.

What if we could use the data we have to actually measure the carbon in each and every tree? And what if we could also understand how healthy each of these trees were, so that we could provide this precision advisory to our farmers to further improve their tree survival rates and fruit yields.

Just as we’d summited a few mountains, another one, and a fairly massive one at that, had emerged.

In Part 3 of 5 next week, we’ll share how a mathematician came to our rescue, how we taught AI to be honest about what it doesn't know, why it’s important to look beyond what’s visible to the eye, and also…bael candy.

Our team stood ankle-deep in mud on a farmer's plot in Maharashtra in August 2021, watching weeks of painstaking work literally blow away in the monsoon wind. The laminated tag we'd carefully affixed to this particular tree was now dangling by a thread - the crisp black squares already blurred by moisture.

This was supposed to be our solution to the lack of accountability and measurability in our sector. Each QR code represented hours of work - recording GPS coordinates, documenting growth patterns, creating what we proudly called "individual tree profiles." We were going to bring transparency by giving every single tree a digital identity. Or so we thought.

The inadequacy of our "simple" and “transformative” scanning process soon dawned on us: an enthusiastic volunteer kept messaging us for months, begging to work with us in the field. We were at capacity, but his persistence eventually wore us down. "Fine," we said, "come help with tree monitoring."

His first task: scanning QR codes and updating tree data. Day one in the village of Dhawalgaon went OK. Day two, he trudged along. Day three? He disappeared. Just vanished. No goodbye, no explanation - he'd realized "individual tree documentation" actually meant tedium and drudgery, and wisely decided he had urgent business elsewhere. We never heard from him again. His disappearing act should have been our first warning sign.

But standing there in the rain, watching our revolution wash away, the truth was obvious: we'd solved transparency but murdered scalability.

The 10,000-Tree Question

It started innocently enough. At Farmers for Forests, we were working with smallholder farmers across Maharashtra to plant mixed agroforestry systems - not the monoculture plantations that dominate industrial forestry - but diverse ecosystems that could generate high quality carbon credits while providing farmers with harvest income.

We'd also made a promise to farmers: annual payments linked directly to tree survival rates on their land from year one. It wasn't a huge amount - maybe enough to cover a month or two of household expenses. But it represented something bigger. Agroforestry takes three to five years before trees start yielding fruit or timber income. Our payments were meant as a bridge - recognition of the ecosystem services farmers were providing while waiting for their trees to mature.

But our commitment to fairness had created a technical challenge. We had to know exactly how many trees survived on each farmer's plot. Not estimates. Not statistical samples. Actual counts.

But there was another layer to our challenge. We couldn't just trust field reports that claimed "90% survival rate on Plot A." We needed verification that our field teams had actually visited every tree, observed its condition, and documented its status. The QR codes became our guardrail - a "trust but verify" system that ensured accountability. When field teams submitted survival data, each entry had to include a scanned QR code as proof of visit.

No scan? No count.

Every tree gets approached. Every tree gets photographed and GPS-tagged. When trees died, we'd remove their codes. When new ones were planted, we'd add them to the system. We could calculate survival rates down to the individual tree level. Perfect transparency, perfect accountability.

The first 100 trees were exhilarating. Farmers gathered around as we explained how our field team’s phones could now tell them exactly which trees were being counted for their payments. "This is the first time someone has properly documented our trees," farmers would tell us. For the first time, the connection between land stewardship and payment appeared tangible.

The next 1,000 trees were satisfying. We were building something unprecedented - a database where every individual tree had its own digital profile.

By tree #10,000, the cracks were showing. The physical labor was immense. Our field team hated the work. Every quarterly visit meant checking thousands of QR codes. The QR codes needed constant maintenance. Sometimes QR codes kept getting mixed. When we would find the culprit, the reason often was “just for fun.”

The data entry was consuming entire days. We were spending more time documenting forests than helping farmers build them.

At tree #50,000ish - that muddy day in the monsoon - we finally asked the question we'd been avoiding: Why are we doing this?

We were like punctilious accountants who insisted on counting every penny by hand.

But we couldn't abandon the tree-level payments. The transparency we'd created was valuable. We just needed a better way to achieve it.

The Smallholder Invisibility Problem

Our QR code experiment had revealed something important to us about nature-based solutions (NbS) and the carbon credit sector: it's not built for small farmers. While we were hand-tagging individual trees, the big carbon projects at that time were using manual monitoring and satellite data to monitor vast forest concessions. Satellites can track deforestation across vast parcels of land, but they can't see the 50 young trees on a farmer's two-acre plot. They can monitor established forests and agro-forests, but they can't accurately detect small saplings in their critical first years of growth.

This creates a cruel irony: the communities most vulnerable to climate change - smallholder farmers - were often getting effectively excluded from climate solutions. Their plots are too small for satellite monitoring and too scattered for rigorous yet efficient manual checks.

From 2020 itself, our team was researching alternatives. We talked to remote sensing experts who explained why more affordable satellite data can't reliably detect small trees. We spoke with carbon project developers who described the economics that favor large-scale plantations over diverse smallholder agroforestry. And the statistical sampling approach of calculating carbon sequestration in carbon credit projects introduced uncertainty that buyers penalized with paltry payments.

The DeepForest Moment

By mid 2022, we knew one thing: we had to automate tree detection.

Our first attempt used satellite data. We'd secured 3-meter resolution PlanetScope imagery and spent weeks analyzing the images, searching for individual trees. The resolution wasn't there. Even at 3 meters per pixel, the young agroforestry trees we worked with simply didn't show up as distinct objects. Higher resolution satellite data exists, but is prohibitively expensive. Just monitoring itself would have quadrupled our cost. Also, satellite imagery comes in standard tiles covering very large areas - but our farmer plots are small (1-2 acres), scattered, and irregularly shaped.

That’s when our team read about drone based photogrammetry and thought of experimenting with it. We captured our first orthomosaic - higher-resolution aerial images where individual trees were clearly visible.

Finally, we could see what we needed to count. But seeing and automatically counting were different problems entirely.

When Pravin Mulay (who is today our VP of Research & Tech) joined us in early 2022, one of his first tasks was figuring out how to automate tree counting from these drone orthomosaics.

Instead of building detection algorithms from scratch, he discovered something called DeepForest - a pre-trained model from the University of Florida that could identify individual tree crowns in aerial photographs, achieving 70-75% accuracy across different forest types.

We got excited. Here was exactly what we needed: a computer vision system designed for the problem we were trying to solve. The team worked together to deploy and adapt the DeepForest code. What if we could give every tree a digital identity without physically tagging it? What if computer vision could replace QR codes? What if drone photogrammetry could scale our documentation approach to hundreds of thousands of farmers?

The more we experimented with DeepForest, the more excited we became. Computer vision could detect trees regardless of plot size or shape. Machine learning could potentially identify not just trees, but species, health, growth rates - all the data we'd been manually collecting, but extracted automatically from aerial images.

This wasn't just a better way to solve our inventory problem. This was a way to democratize forest monitoring itself.

Our QR Legacy

Our QR experiment had failed as a scalable solution, but it succeeded as a research project. Those 100,000 manually documented trees taught us what farmers actually needed from tree monitoring technology.

When our field teams scanned QR codes and logged each tree's condition into the Kobo Toolbox, the value that emerged from the data were the patterns we started observing when we aggregated individual tree data over time and space.

Tree #47 thrived in particular soil conditions while tree #52 did not, but only when we tracked growth rates across seasons did the correlation with microclimate become clear.

We realized that comprehensively documenting what human eyes could already observe - survival, visible growth, obvious stress - was so valuable. And there was so much more potential in what we couldn't see: early disease detection, nutrient deficiencies before they became visible, and stress patterns that would only show up in multispectral analysis. Computer vision could potentially detect problems weeks before farmers would notice them, identifying subtle changes in leaf color or growth patterns that predict future issues.

The Technical Mountain

By the end of 2022, we knew what we wanted to do next. Basically, instead of manually tagging millions of trees, we now needed to teach machines to see trees the way humans do.

But the challenges were daunting. Automating tree identification meant solving computer vision problems across dozens of tree species, varying light conditions, different growth stages, and diverse agricultural settings. We'd need to detect individual trees in dense canopies, distinguish between crops and trees in agroforestry systems, and track growth over time.

Then there was the "hidden measurement problem." Drones flying overhead could see tree crowns and estimate height, but they couldn't see trunk diameter - a very important measurement for calculating carbon storage. The trees' most valuable data was literally hidden beneath their canopies.

Finally, there was the inherent randomness. Two trees of the same species, same age, same growing conditions could have completely different carbon storage. Any automated system would need to quantify not just measurements, but uncertainty - how confident could we be in our estimates?

From QR codes to bounding boxes, we had a massive technical mountain in front of us. But we also had something valuable: nearly 1 million trees worth of ground truth data (that we ourselves had planted), and a clear understanding of what success would look like from the farmer's perspective.

As we scanned that #50,000ish QR code on that muddy day, we had already begun to imagine a different way forward. Instead of physically tagging trees, we'd digitally detect them. Instead of manual measurements, we'd have automated analysis. Instead of excluding smallholder farmers from carbon markets, we'd leverage and build upon inclusive open-source technologies. We will teach machines to count what matters.

In part 2 of 5, next week we will share how we climbed the technical mountain - from computer vision breakthroughs to solving the ‘hidden measurement’ problem with AI that quantifies its own uncertainty.

Dear friends,

It’s been a while. Too long, actually. We’ve been busy on the ground - wading through devil’s thorn, dodging errant mosquitoes, sitting with farmers and sipping excruciatingly sweet chai - and somewhere along the way, we stopped writing. Not because there wasn’t anything to say. But because every time we sat down to write, another fire needed putting out - sometimes literally, often emotionally, and almost always urgently.

So we’re moving to Substack - partly to make it easier for you to find us. But mostly to remind ourselves that even in the chaos, we have to keep telling the story.

The story of climate change isn’t one of constant headlines anyways. It’s a kind of “slow violence” - a steady unraveling. Death by a thousand cuts. Some days, it’s animals collapsing from heat stress. Other days, it’s yet another infrastructure project swallowing one of India’s few remaining old-growth forests. And sometimes, it’s just driving through town after town with no trees, no shade, and wondering: how did we let this become normal?

Why doesn’t the sheer scale of loss - of lives, trees, animals, futures - provoke more outrage, more heartbreak, more action? Why does something so devastating to us feel, at times, so abstract to others?

Maybe that’s part of our work we need to take more seriously. Not just fundraising to roll out our programs on the ground, but also finding better ways to communicate what’s at stake. To tell stories that certainly cut through our fatigue, and hopefully will cut through somebody else’s as well.

Since our last letter in August 2022 (yes, it was that long ago!), a lot has changed. We’ve more than quadrupled in size. We now work across five States in India (as opposed to just the one in 2022). Our tech is getting really good (that needs its own separate Substack post). And we’re making some bold internal shifts - because we want to grow without burning out or turning on each other. Turns out, scaling an organisation is not unlike growing a forest: it takes intention, spacing, sunlight, and patience.

Also - hi again! I came back to F4F full-time earlier this year after nearly two years in a part-time advisor role. This felt like the moment. Over that time, I had the privilege of proudly watching my co-founders and the early team build something truly transformative, and surprisingly sturdy.

Now, the model is holding up. The field teams are moving fast. The farmers are with us. And we’re at an inflection point where it’s not just us saying the model works - we’re collaborating with researchers and peer organisations to test and validate that.

The next chapter is about scaling - without compromising quality, losing our grounding in the field, or missing the chance to shape what a truly resilient, farmer-first, regenerative, and climate change-resilient land restoration model looks like for the long haul. One that doesn’t just remain a case study, but lasts on the land.

We’re here. Still occasionally chaotic. Still mostly hopeful.

- Arti Dhar (Co-Founder, Farmers for Forests)

From Joining Us, To Building with Us

Aditya Avinashe joined F4F as Chief Operations Officer in 2023 , and it didn’t take long for us to realise we’d lucked out. He brought calm to chaos, built systems with soul, and poured himself fully into the work. So after 1.5 years of being absolutely indispensable, it only made sense to make it official: he’s now a F4F co-founder! Oh, and somewhere along the way, he also became an Acumen Fellow. No big deal!

New Builders for the Next Chapter

Meet the new folks helping steer this slightly wild ship - our senior management team who’ve joined us over the past 1 year. We’re growing fast, and so is our leadership bench (with more to come!).

Star Farmer Event

In January, we hosted our very first F4F Farmer Event - a gathering of over 100 farmers, some already part of our program, others curious. Our chief guest was Mr. Satyajit Bhatkal, CEO of Paani Foundation (F4F’s partner organisation). We came together to talk about trees, water and soil, yes, but most importantly to celebrate our “star farmers”- the ones who didn’t just plant trees, but cared for them like family, and became organic evangelists of F4F in their communities. It was a joyful and sentimental experience.

We organised this event not just to felicitate, but to say clearly: land stewardship deserves public respect and celebration - so that the next generation sees it as work worth aspiring to.

Medals, Mics, Mild Panic…

Arti pitched our hearts out at Fast Forward’s Demo Day in San Francisco (with fancy slides and only minor existential dread).

Krutika became a Mulago Fellow (Henry Arnhold Fellow) and got F4F into the Mulago portfolio while she was at it.

Aditya Avinashe pitched F4F at the India Fundraising Conference 2025, organized by India Leaders for Social Sector (ILSS) - and won!

By the Numbers (Because Feelings Need Data)

Since we’ve started, we’ve worked on transitioning nearly 5000 acres of land into agroforestry and forestry.

We’ve planted nearly 2 million trees - and yes, they’re doing well (turns out, obsessing over aftercare works).

Through our various programs, we’re working with nearly 25,000 farmers.

We’re active in five states (and counting).

We have 25,000 acres already in our pipeline to transition to agroforestry.

Meanwhile, in Elephant Country…

Most people know us for planting trees. Fewer know that some of our most important work is making sure those trees don’t get trampled - literally. In Gadchiroli in eastern Maharashtra (where we have been working for the last few years on forest protection and forest restoration), elephants have returned after centuries - wandering in from the neighbouring state of Chhattisgarh in search of food and forest. It’s a shift in the landscape - awe inspiring, yes, but also risky to communities whose homes and crops might lie in the elephants’ paths.

So we started The Elephant Project. It's quiet, long-term work - talking to people, using thermal drones to track elephant movements at night (in partnership with the Forest Department), setting up early warning systems, and reminding people that elephants don’t necessarily have to be enemies.

The good folks at The Grasslands Trust, along with the support of Center for Wildlife Studies and Pernod Ricard India Foundation, made a mini documentary called ‘Crossing Paths: The Elephants of Eastern Maharashtra’, that profiles the important work happening on the ground. We really encourage you to watch it!

From Sapling to…Slightly Bigger Sapling

We recently closed hiring for over 30 new full-time team members - which means we’ve officially crossed the 100-person mark (yes, we’re also in shock). As we grow, we’re learning how to make the slow, slightly awkward transition from scrappy startup to something more… grown-up-ish.

Last month, we brought the whole team together for an off-site. We danced. We balanced cups on our heads. We played cricket with our own rules. And we finally put faces to names we mostly knew through Whatsapp threads.

Regional team loyalties ran deep, but so did pride in our collective work - no matter where it’s rooted. We also spent time reflecting on which of our organisational values we truly live by, which ones need work, and what we hope to build together this year.

If words aren’t your thing - we made a little video (with music, of course).

If you made it this far - thank you. We promise not to ghost you again (or at least not for too long).

Until next time,

Team F4F

Subscribe now

Subscribe now