The land as a living document
UnderStory treats a working forest, ranch, or watershed as a living document — an integrated data structure that domain experts author over time, that machines and sensors continually update, and that can be rendered into grants, reports, partner updates, and interactive views on demand.
The science of a given piece of land is fine-grained and place-specific. The paperwork that surrounds it asks for that science in summary form, repeatedly, for different audiences. UnderStory treats the translation layer between the two as a software problem: author once, render many.
Customized annotation tools
The authoring surface for UnderStory is a set of customized map annotation and forest planning tools — built around the way forestry consultants, arborists, and land managers actually think about place. Inside these tools, a domain expert can:
- Draw geographic regions and associate them with treatment plans, prescriptions, and phasing.
- Log timed observations about forest ecology — stand composition, understory, pest pressure, burn history, wildlife, hydrology.
- Configure which public APIs to sample for each region, and at what cadence.
- Attach notes, photos, sensor feeds, and reference documents.
These tools are not generic GIS. They are shaped to the specific workflows of forestry, arboriculture, and rangeland management, so a treatment plan author can work at their real level of detail rather than fighting a spreadsheet.
The UnderStory Digital Twin
Everything authored inside the annotation tools lands in an underlying data structure — the UnderStory Digital Twin. Alongside the human-authored layer, the same data structure holds:
- Public data layers — USGS elevation and slope, SSURGO soils, NLCD land cover, CAL FIRE hazard zones, Open-Meteo climate, USGS watersheds.
- Sensor streams — soil moisture, weather stations, camera traps, drone imagery, and other on-the-ground signals as they become available.
- Metric pipeline outputs — carbon, forest health, fire risk, and invasive species models computed on top of the integrated structure.
The Digital Twin is the single place where all of this is reconciled, versioned, and kept live. Nothing in UnderStory is a one-shot document — the data structure is the source of truth, and every deliverable is a view into it.
Views into the living document
Because the data structure is the source of truth, any number of views can be rendered from it on demand:
- Grant proposals — CAL FIRE Forest Health narratives, budgets, timelines, and supporting maps, generated from the tagged regions and their shared context.
- Partner updates — short, land-specialized briefs for landowners, NGOs, and funders, reusing the same integrated data in a lighter template.
- Annual reports — carbon and agriculture accountability templates that roll up pipeline outputs against the baselines established at project start.
- Interactive dashboards and maps — the current treatment planner, with land intelligence cards, region comparisons, and drawing tools.
- Interactive question answering (coming soon) — natural-language queries against the integrated data, so a reviewer, collaborator, or landowner can ask "how much of this parcel is in high fire hazard severity zones?" and get a grounded answer with citations back to the underlying regions and sources.
Each view is specialized to the specific land and treatment areas it describes. When the underlying data changes — a new observation, a revised treatment plan, an updated public data layer — every view changes with it.
Software pipelines + LLM generation
UnderStory's modeling, tracking, and aggregation are traceable, deterministic computations — every key calculation and core metric is consistent, transparent, and reproducible. The software engine itself is AI-generated and continually updated, then verified and tested by human experts on real projects.
On top of this engine, UnderStory supports report generation pipelines that use multi-step LLM generation. Each step is supplied with detailed, map-derived context about land, statistics, and treatment plans drawn directly from the Digital Twin's data structures.
How it is built
Several design choices shape everything else:
Custom map annotation and treatment planning tools. Authoring is a first-class step in the pipeline, not an afterthought. Forestry consultants, arborists, and land managers work inside purpose-built interfaces — drawing regions, tagging them with treatment prescriptions, logging timed observations, and attaching notes, photos, and references. These tools are shaped to the way domain experts actually think about place, so the resulting data is richer and more trustworthy than anything a generic GIS session would produce.
KML as the portable data structure. UnderStory reads and writes KML files that carry far more than geometry — treatment tags, per-region metadata, observation histories, and references to public data and sensor samples all ride along inside the file. Because every page is a single self-contained HTML file that can load and save KML directly in the browser, collaborators can create, update, and share intermediate results without standing up a server, account, or database. A consultant can open a file on a laptop, draw regions, save, and email the result to a partner who opens it on their phone — no infrastructure in between.
Filter, aggregate, compute on the regions. Once the KML is loaded, UnderStory can slice it: filter regions by treatment tag or property, aggregate acreage and biomass across categories, compute per-region statistics against the public data layers and sensor samples, and roll results back up for dashboards, reports, and grant narratives. The same file is simultaneously a map, a database, and a handoff format.
Narrow LLM prompts over wide ones. Rather than handing a model the whole project and asking for a proposal, each narrative section gets its own prompt, its own context window of relevant data from the twin, and its own schema. The model does the language work; the pipelines do the numerical work. This keeps the outputs auditable and the errors localized.
Self-contained HTML as the deliverable format. Interactive pages ship as single files — Leaflet maps, canvas charts, embedded GeoJSON, all inlined. The files are small enough to email, survive archival, and read identically on a grant reviewer's laptop and a landowner's phone. Combined with KML load/save in the browser, this gives UnderStory a genuinely serverless toolchain for creating, updating, and sharing treatment plans across teams.
Why we are sharing this
The tools are built against real land — working ranches, vineyards, a BLM-adjacent forest in the Sierra foothills — and real grant cycles. The vision is larger than the current scope, and we are actively looking to collaborate with climate researchers, software engineers, and founders working on adjacent problems.
If any of the above is interesting — whether you want code access, or are working on something we should know about — write to info@deepforest.org.