Research Brief — R04: Environmental Footprint of Cannabis Cultivation

Report ID: R04 (Wave 1, macro foundation — final curriculum slot; executed sixth) · Brief version: v1.0 — FROZEN September 6, 2026 on Danny’s confirmation. Both research runs execute this version and must cite it; the critique stage records it. Date drafted: September 6, 2026 (v0.9; v0.95 same day: six calibrations from ChatGPT’s editorial review — comparison rule no longer demands identical systems/populations, eleventh provenance field made conditional, evidence type separated from evidence quality with the measured-outranks-modeled rule removed, mitigation ranked within impact categories with stated baselines, single energy-update cutoff fixed, search-stopping language bounded to the documented search; v1.0 same day: three consistency fixes from ChatGPT’s read of the live v0.95 — energy-cutoff exceptions made explicit as a shared §B rule, “unmeasured” no longer auto-graded as weak evidence, “asserted” redefined by absence of traceable data rather than by source type — plus the section rename to “public claims to test”) · Drafted by: Claude (editor) · Protocol: v1.4

Methodological spine (one sentence): name the impact category → fix the system boundary and functional unit → name the production population and system → separate measured from modeled from asserted → compare only on compatible boundary + unit + category + method → attribute effects to practices and systems, never to labels or legal status.

Assigned to: ChatGPT deep research + Gemini deep research (independently, same frozen brief; both runs cite the frozen version)

Assignment

Produce an evidence review of the environmental footprint of cannabis cultivation across all impact categories the record actually covers: energy and greenhouse gases, water (quantity, source, and timing), land use and habitat, chemical inputs and their fate (pesticides, fertilizers, nutrient losses), air emissions (biogenic volatiles, CO₂ enrichment, odor), and waste and material streams (growing media, plastics, plant waste, packaging at the farm gate). The goal is a boundary-disciplined map of what has been measured, what has been modeled, and what is merely asserted — and a ranked account of which mitigation levers have documented effect.

This report inherits, does not repeat, R05’s energy/carbon axis. R05 (signed off 2026-08-28) established the indoor-vs-outdoor carbon spread (R05-C001–C010), the boundary-incompatibility finding (R05-C033), functional-unit sensitivity (R05-C032), and the water-figure lore problem (R05-C014). R04’s job on energy and carbon is to (a) update — work published or materially revised from January 1, 2025 onward that is not already in the R05 corpus (the energy-update cutoff; its exceptions are stated once, as the shared rule at the head of §B); (b) contextualize — how cannabis compares to other crops and to other consumer products on compatible boundaries and units; and (c) trace the provenance of the most-circulated energy claims. Everything else — water, land, chemicals, air, waste — is R04’s primary ground.

Tag every load-bearing claim by evidence type (the R03 discipline, R03-C001), keeping the types distinct:

1. Measured impacts — metered energy, gauged diversions, assayed residues, surveyed land conversion.
2. Modeled impacts — LCAs, scenario models, extrapolations from license counts or canopy. Every model carries its boundary, functional unit, impact-assessment method, and input assumptions, or it is not a finding. The common hybrid — measured inputs → modeled outcome (e.g., metered electricity × grid emission factor) — is tagged as such, naming which part was measured and which was modeled.
3. Asserted impacts — claims with no traceable supporting data or analysis, wherever they appear. “Asserted” describes the claim, not the outlet: a journalistic, advocacy, or industry report that contains inspectable measurements has those measurements evaluated on their merits (as measured or modeled evidence), with the source’s institutional interests recorded separately as context. Assertions generate questions, not evidence.

Evidence type is not evidence quality. A measurement can be unrepresentative and a model can be transparent and well-bounded; quality is graded separately (see Methodology).

Methodology requirements

Boundary-and-unit discipline (the methodological spine — extends R05-C032/C033 and R08-C027). No two environmental figures may be compared, ranked, or averaged unless they share compatible (i) system boundary (cradle-to-farm-gate vs. operational-only vs. cradle-to-consumer), (ii) functional unit (per kg dried flower, per g cannabinoid, per m² canopy-year, per plant, per $ revenue), (iii) impact category, and (iv) characterization or measurement method. Production system and population are not matching requirements — they are the subjects of comparison. Indoor vs. outdoor, licensed vs. unlicensed, cannabis vs. tomato are legitimate comparisons when boundary, unit, category, and method line up; the comparison must then name the system/population difference as what is being compared. Where sources differ on (i)–(iv), report them side by side with the difference named — never harmonized by the researcher’s own arithmetic unless the arithmetic is shown.

Impact-category integrity (the R08-C027 lesson). The corpus’s one post-signoff audit finding to date was a percentage assigned to the wrong impact category (metal depletion read as global-warming potential) inside a single paper that had been registered twice as two sources. For R04 this is the primary failure mode to pre-empt: every quantitative LCA claim names the impact category and the characterization method (e.g., GWP100 under which IPCC report; ReCiPe / TRACI / CML midpoint), and every source is registered once (PROTOCOL rule 11). Corroboration claims (“multiple LCAs find…”) must pass the independence check (rule 12): the same dataset, the same facility, or a review citing the original is one source, not two.

Population and system discipline (R01-C033, R05-C017, R03-C023). Environmental effects attach to practices and systems (sealed warehouse, supplemental-light greenhouse, light-dep hoop, full-sun container, full-sun in-ground; trespass public-land grow, unpermitted private-land grow, licensed farm) — never to “cannabis,” “legal,” “illegal,” “organic,” or “sun-grown” as labels. The best-documented environmental harms in the corpus belong to one population (P3, trespass grows: rodenticides, carbofuran, diversions — R03-C023) whose share of production is unknown; those harms must not characterize unlicensed production generally, and licensed production gets no halo (R05-C023). Every claim names its population and system.

Evidence type and evidence quality, graded separately. Each load-bearing claim carries one of the three type tags above (or the measured-inputs → modeled-outcome hybrid). Quality is then assessed on four separate axes regardless of type: relevance (does it answer the question asked, on the boundary and unit asked?), representativeness (facility, region, season, population — does it stand for more than itself?), transparency (can the number be reconstructed from stated inputs and methods?), and reported uncertainty (range, sensitivity, or none). There is no blanket rule that measured outranks modeled: a single-facility meter reading and a transparent multi-facility model each carry their own limits. What does disqualify is opacity — a precise figure whose origin cannot be reconstructed is estimate-lore whether it is called a measurement or a model (the R03 provenance rule). For modeled claims, the provenance record must make the model reconstructable (see 11-field record below).

Comparators must be commensurable. “Cannabis uses more water than wine grapes,” “a joint equals X miles of driving,” “cannabis consumes 1% of US electricity” are comparisons across different boundaries, units, and years. For every comparator claim: state both sides’ boundary, unit, year, and source; if they cannot be aligned, say the comparison is not defensible rather than repeating it with a caveat.

Scale contextualization is required, not optional. Absolute impact (total sector emissions, total acreage, total diversions) and intensity (per kg, per m²) answer different questions and are routinely swapped in public discourse. Report both where both exist, and always situate cannabis against the relevant sector baseline (total agricultural land, total state electricity, total watershed withdrawals) so the reader can tell a large intensity from a large share.

Practice–effect separation. Split the practice (“media reuse,” “LED conversion,” “condensate recovery,” “drip vs. hand-watering”) from its claimed environmental effect and grade each separately. Adoption alone does not establish environmental benefit: how widely a practice is used is evidence about the practice, not about its effect. Where the effect has been measured or modeled, that evidence receives its own quality assessment on the four axes above; where no measured or modeled effect is located, the effect claim stays unverified — which is a status, not a grade of weakness.

Jurisdiction- and date-stamping. Grid carbon intensity, pesticide approvals, water law, waste-destruction mandates, and energy-reporting rules are all jurisdictional and moving. Every regulatory and empirical claim carries jurisdiction and as-of date. Prefer primary sources: peer-reviewed LCAs and field studies; state agencies (CDFA/DCC, CA Water Boards, CA DPR, Colorado MED, Massachusetts CCC energy reporting), Health Canada, utility field studies (S-007 precedent); then journalism; then advocacy and industry sustainability reports, labeled as such.

Provenance for quantitative estimates — 11 fields for R04. Every major quantitative estimate carries the standard 10-field record (geography, period, population, metric, functional unit, system, product boundary, method, directional bias, confidence) plus an eleventh field for this report: “impact category + characterization method + allocation rule” — which category the number belongs to, which LCIA method or measurement produced it, and how burdens were allocated across co-products (flower vs. trim vs. biomass) or across a multi-crop farm. The field is conditional: a directly measured quantity (a gauged withdrawal in L/day, a residue concentration in µg/L, a metered kWh) may have no characterization method or allocation rule, and for such figures the field reads “not applicable — [reason]” (e.g., “direct measurement, no impact characterization applied”). “Not applicable with reason” is a complete record. A blank eleventh field on a figure that should have one — any LCA output, any percentage-of-impact, any CO₂e — means the figure is not comparable to anything and is labeled boundary-orphaned.

Key questions

A. The LCA and measurement landscape (required output category)

1. Inventory every published cannabis-cultivation LCA and every direct-measurement study (metering, gauging, residue assays, land-cover mapping) located, as a table: system studied, population, geography, boundary, functional unit, impact categories, characterization method, data source (primary metering vs. modeled inventory), and the study’s own stated limits. Which systems have no LCA at all (R05-C009 found greenhouse subtypes are collapsed; is soil-grown outdoor, light-dep, or hemp-style field cultivation of drug-type cannabis covered anywhere)?
2. How independent is the literature? Map which studies share datasets, facilities, authors’ inventories, or emission-factor sources — the corroboration-independence question (rule 12) asked of the whole field.

B. Energy and carbon — update and contextualize (inherited axis)

Energy-update cutoff — shared rule for this section and M9. The January 1, 2025 cutoff applies to the update search (Q3) only: R05 already holds the pre-2025 energy/carbon literature. It does not apply to historical provenance tracing (Q4, which must reach back to the original sources of circulating claims), to comparator evidence (Q5, which may be any date if the boundary and unit are compatible), or to corrections of inherited R05 claims (any date). When a pre-2025 source is used under one of these exceptions, name the exception. 3. What has been published or revised from January 1, 2025 onward (the energy-update cutoff, per the shared rule above) that changes R05-C001–C010 or C033? Include any grid-decarbonization or LED-adoption scenario work and any embodied-carbon (construction, equipment) accounting the R05 corpus lacked. 4. Trace the provenance of the three most-circulated energy claims: the “1% of US electricity” figure and its descendants; per-unit equivalencies (“a gram/joint equals X km driven”); and any “cannabis is the most energy-intensive crop” statement. For each: origin, boundary, year, and whether later citations re-derived or merely repeated it. 5. On compatible boundaries, units, and methods, how does cannabis compare to (a) other controlled-environment crops (tomato, leafy greens, cut flowers), (b) other field crops, and (c) other consumer goods on a per-dose or per-$ basis? Where no commensurable comparison exists, say so.

C. Water

6. What is measured about cannabis water use — per plant, per kg, per m², per season — by system (indoor, greenhouse, outdoor container, outdoor in-ground), and what is assumed? Extend R05-C014 (the 22.7 L/plant/day ordinance-lore) with a full reconstruction chain for every widely cited water figure.
7. Water source and timing: what fraction of studied cultivation draws surface diversions vs. wells vs. municipal supply vs. stored/captured water, and how has that shifted under regulation (California’s cannabis water-rights and forbearance rules)? Update the low-flow coincidence finding (R05-C013, S-008, S-160) and the permitted-vs-unpermitted extraction contrast (R03-C024) with anything newer.
8. Indoor water: transpiration and condensate recovery (R05-C007), reverse-osmosis reject streams, and nutrient-solution discharge — what is measured, and are there any full facility water balances?

D. Land, habitat, and landscape

9. Land conversion and siting: what do land-cover and parcel studies (the Butsic/Brenner/Carah/Dillis Northern California family and any analogues elsewhere) establish about forest fragmentation, road building, steep-slope siting, erosion, and proximity to streams — separately for licensed vs. unlicensed vs. trespass populations, and before vs. after legalization?
10. How much land does cannabis actually occupy, by jurisdiction, against total agricultural land — the scale-contextualization requirement — and what do license/canopy data support vs. not support (R02-C022, R05-C031)?
11. Wildlife and habitat: consolidate the verified toxicology cluster (S-088, R03-C023) and any evidence on light pollution from supplemental-light greenhouses, noise, fencing, and water-diversion effects on aquatic species — each tagged to its population.

E. Chemical inputs and emissions

12. Pesticides: what do use records, residue surveillance, and watershed monitoring (S-020) show for licensed vs. unlicensed vs. trespass production, and which compounds are the documented problem set (carbofuran, anticoagulant rodenticides, myclobutanil, others)? What is the regulatory approval landscape (e.g., California DPR’s cannabis-legal list, Health Canada) and how does it constrain inputs relative to other crops?
13. Fertilizers and nutrient fate: application rates and forms by system, measured leaching/runoff, and greenhouse-gas contributions from N inputs (R05-C012 found nitrogen-heavy fertilization was a lever in the outdoor LCA — reconcile with R01’s N-asynchrony findings, R01-C008/C020). Is there any measured nutrient-loss study in cannabis?
14. Air: biogenic volatile organic compound (terpene) emissions and ozone-formation modeling (the Colorado studies), CO₂ enrichment as a direct emission, odor as a regulated nuisance — measured emission rates vs. modeled regional impact.

F. Waste and material streams

15. Growing media (peat, coco, rockwool, soil mixes), plastics (pots, film, drip line), plant waste under regulatory destruction mandates, and farm-gate packaging: what is quantified anywhere, what mitigation is documented (media reuse — R08-C027’s outdoor-media finding; peat alternatives — S-110 precedent), and where is the honest answer “no magnitude data exist”?

G. Mitigation levers and governance

16. Rank the mitigation levers within each impact category separately — carbon, water quantity, water timing, habitat/land, chemical fate, waste — by documented effect size against a stated baseline (which system, which region, which year’s grid or hydrology) on a stated boundary and unit: outdoor/greenhouse shift (R05-C010), LED conversion (R05-C004/C005), HVAC and dehumidification strategy (R05-C006), condensate recovery, renewable procurement, media reuse, irrigation practice, storage-and-forbearance water regimes. Show cross-category tradeoffs explicitly (e.g., an outdoor shift lowers carbon and may raise diversion-timing or habitat pressure; LED conversion may raise winter heating load, R05-C005). Do not produce a single overall ranking across categories unless an explicit weighting scheme is stated and its source named. For each lever: measured, modeled, or asserted?
17. What do regulatory instruments actually require and report: energy-use reporting and efficiency standards (Massachusetts, Illinois, others), Canadian energy disclosures, California water-rights and forbearance conditions, waste-destruction rules — and is there any evaluation of their effect?
18. Certification and marketing schemes (Sun+Earth, Clean Green, Envirocann, “carbon-neutral” claims): what do they measure, verify, and disclose? Evaluate as programs, not as evidence of impact.

H. Synthesis targets

19. The boundary-and-unit map: a table of every load-bearing figure in the report placed on its boundary, unit, category, and population — the artifact that makes the corpus’s numbers non-poolable by construction.
20. The three-class map (measured / modeled / asserted) by impact category, with named data gaps. Absence of measurement is itself a substantive finding: do not build a “total environmental footprint of cannabis” because readers expect one. Where the record supports only “not defensibly estimable,” say that.
21. The 5–10 most repeated public claims and what the evidence says about each (see the pre-registered list, M1–M10).

Pre-registered public claims to test

Word all verdicts so absence of evidence remains absence of evidence (“unsupported” ≠ “the opposite is true”; unknown is acceptable):

• M1. “Cannabis uses 1% of US electricity” — trace the chain; is any current version of this figure defensible, on what boundary and year?
• M2. “One joint = X kg CO₂ / X miles driven” — equivalency lore; which boundary, whose model, which year’s grid?
• M3. “Cannabis uses twice the water of wine grapes” (or any per-plant crop comparison) — align units before judging.
• M4. “Sun-grown is sustainable by definition” — test against R05-C011/C012: outdoor media, nitrogen, diversion timing, land conversion.
• M5. “Illegal grows are the environmental problem; legal grows are clean” — and the inverse. Population discipline (R05-C017, R03-C023) governs both.
• M6. “Cannabis / hemp is carbon-negative or sequesters carbon” — hemp fiber-biomass claims migrating to drug-type flower; check boundary and whether any sequestration is permanent.
• M7. “Cannabis terpene emissions are a significant regional air-quality problem” — measured emission rates vs. modeled ozone contribution.
• M8. “Packaging is the industry’s biggest environmental impact” — magnitude data or assertion?
• M9. “LEDs solve indoor cannabis’s carbon problem” — inherited R05-C004; governed by the §B shared cutoff rule: update with sources dated January 1, 2025 or later, and reach back further only under the stated exceptions (provenance tracing, comparators, corrections).
• M10. “Organic / living-soil cannabis has a lower footprint” — per-area vs. per-mass (S-045 precedent, R01), practices-not-labels (R01-C033).

A practitioner-record passage is withheld from the public copy.

Standing source & citation requirements (PROTOCOL v1.4 — verbatim)

• Every citation supporting a substantive claim must carry a resolvable DOI, stable URL, or independently retrievable identifier — an unresolvable citation makes the claim unsourced.
• Citation-identity integrity: for every substantive source, the authors, title, publication year, venue, volume/pages where applicable, and DOI or stable identifier must resolve to the same bibliographic object — a DOI that resolves to a different real paper fails identity verification. If you cannot fully resolve the tuple, flag it explicitly as a partial citation rather than completing it from memory.
• Exact claim support: a source supporting the general direction of a finding does not automatically support its stated magnitude, population, geography, time period, boundary, or impact category — preserve these distinctions in candidate claims.
• Corrections inherit the complete verification burden: a replacement citation or corrected statistic remains unverified until its existence, identity, and claim support are checked.
• Source-identity deduplication and corroboration independence (rules 11–12): one paper is one source however many times it is cited; “multiple studies agree” requires demonstrably distinct studies or datasets.
• When testing public claims, “the claim is unsupported” and “the opposite is true” are different findings — do not overcorrect; unknown is an acceptable verdict.
• For each major claim, tag the population and system it covers, the jurisdiction and period, and — for magnitudes — the boundary, functional unit, impact category, and measured/modeled/asserted status.
• Report ranges and disagreements between sources, not averaged-away conclusions; list data gaps explicitly.

Candidate claims for editorial review

Conclude with 15–30 concise, testable candidate claims with supporting citation(s), each carrying its population/system/jurisdiction/period tags and, for magnitudes, the boundary, unit, impact category, and measured/modeled/asserted tag. Do not assign final evidence status or confidence; the editor will.

Search-stopping rationale

End with an explicit, bounded statement of why searching concluded: the search cutoff date, the databases and query strings covered, the inclusion criteria applied (language, publication types, date range, system/population scope), saturation observed, and named question areas where no eligible studies were located within the documented search — distinguishing “not searched” from “searched and none located.” Do not write that the literature is “genuinely absent”; absence is a property of the search, not a property of the world, and the corpus records it as such.

What “done” looks like

A structured report answering A–H with cited, population- and boundary-tagged evidence; the LCA/measurement inventory (A) and the boundary-and-unit map (H19) as distinct outputs; the measured/modeled/asserted map by impact category (H20); calibrated verdicts on the pre-registered public claims (M1–M10); a top-10–15 strongest-sources list with one-line reliability and institutional-position notes; a per-impact-category mitigation-lever table with stated baselines, effect-size provenance, and cross-category tradeoffs; an explicit data-gap list; candidate claims; and the bounded search-stopping rationale. Uncertainty findings are as valuable as positive findings.

Editor’s note (repo workflow): outputs land in raw-inputs/chatgpt-YYYY-MM-DD.md and raw-inputs/gemini-YYYY-MM-DD.md; then verification & critique (critique.md), claim extraction (ledger.md, IDs R04-C001+), synthesis, and content extraction per PROTOCOL v1.4. Corpus anchors: R05-C001–C010 (energy/carbon spread — inherited, not re-litigated), R05-C011 (outdoor is not automatically sustainable), R05-C012 (outdoor media up to ~75% of GWP; N-heavy fertilization), R05-C013 (diversion timing vs. low flow), R05-C014 (22.7 L/plant lore), R05-C017 (contamination is population-specific), R05-C031 (no production-by-environment breakdown), R05-C032/C033 (functional-unit and boundary incompatibility — the keystones R04 extends), R08-C027 (media as outdoor carbon lever; the audit-corrected claim), R03-C001 (which population, measured how), R03-C023 (P3 toxicology cluster, S-088), R03-C024 (permitted vs. unpermitted extraction), R02-C022/C030, R01-C033 (practices-not-labels), R01-C008/C020 (N asynchrony), S-034/S-041/S-045 (land-use carbon opportunity cost; regenerative-sequestration bounds; per-area vs. per-mass).