Introduction

FRED is a conversational decision-support system for forest biomass procurement in California. Ask it a question in plain English, get back a defensible recommendation grounded in spatial data.

#What FRED does

Forest biomass procurement is a high-stakes spatial decision. Where do you site a facility? How much biomass can you actually source within an economic haul radius? What does it cost per ton once you account for terrain, transport, and harvest system? How does the answer change if you also want to prioritize wildfire risk reduction? How does it evolve over a 10-year operating horizon as nearby supply depletes?

Each of these questions has historically required a specialist with a seat license to FRCS, access to GIS data, and a lot of patience. FRED collapses that workflow into a conversation.

You ask FRED a question. It reasons about what data it needs, pulls the right spatial layers, runs the necessary calculations, and returns an answer grounded in the same models published in peer-reviewed research. The answer comes back as a recommendation card with a map, a supply curve, and a written justification you can act on.

#Who it's for

FRED is designed for three audiences with overlapping but distinct needs:

  • Biomass operators evaluating where to site a new facility or whether an existing site has enough sustainable supply. FRED returns the answer in minutes instead of weeks.
  • Researchers studying biomass economics, fuels treatment, or fire risk reduction. FRED gives reproducible numbers against published prediction layers.
  • Policy analysts quantifying the tradeoff between procurement cost and wildfire risk reduction. FRED's Pareto-weighted framework was designed for exactly this question.

#A typical question

FRED is most useful for questions that combine location, scale, and a preference. For example:

NOTE
"Where should I site a 20MW biomass facility near Redding if I want to prioritize fire risk reduction?"

FRED will geocode Redding, find candidate cluster supply within a defensible radius, evaluate the cost-vs-fire tradeoff across that supply, and return a specific recommended location with the full Pareto curve. A typical answer takes about 3 to 5 seconds.

You can keep going from there. Follow-ups like "what if I push the radius out to 60km?" or "how does the cost change over 10 years?" work without restating the original context. FRED remembers the conversation.

#What makes FRED different

Three things, mainly.

It thinks in spatial layers, not text. Most AI assistants retrieve text passages from documents and reason over them. FRED retrieves machine-learning predictions about cost, transport, and fire risk, then composes them at query time. The architecture, Compositional Spatial RAG, is purpose-built for this kind of question.

Every answer is traceable. When FRED tells you a facility should cost $87.40 per BDT, you can see exactly which clusters contributed, what the surrogate predicted for each, what the haul distances were, and how the fire weighting shifted the result. There is no opaque embedding step.

The underlying research is published. FRED is built on three peer-reviewed contributions: a learned harvest cost surrogate validated against FRCS, an empirical transport circuity model derived from 445,000 truck-profile routes, and a fire-aware Pareto framework built over USDA's burn probability data. See the Methodology section for the details.

#Where to go next

If you want to get started right away:

If you want to understand the system first: