The five pathways left out draw a different map: 144 companies in 16 member states, the first eligible suppliers in Czechia, Portugal and Romania, and most of the companies in Austria (10 of 11), Spain (7 of 8), Ireland (5 of 8) and Finland (6 of 8). Biochar alone is 113 companies in 15 states, 44 of them in Germany.

Eleven member states have no published removal company at all, and every one of them has carbon removal researchers, 512 between them out of 8,114 in the EU27; Italy has 960 researchers and four companies. The allowances come out of a cap that every installation in every member state pays into, and as drafted the revenue lands in Stockholm, Copenhagen and Rotterdam. That is an argument for letting the other sixteen states in, not against the Nordic projects.

The list backs the two pathways where Europe's companies are weakest, and skips the ones where they are strongest

Based on our data, in DACCS the EU27 has 31 companies against 50 in the United States, in bio-CCS 25 against 28. In biochar it is 113 against 77. In the two pathways the Commission chose, the EU27 has 56 companies and the United States 78. In the five it left out, the EU27 has 144 and the United States 138. Our researcher census says the same, with the EU27 ahead in every pathway it tracks except DAC.

Companies are not tonnes. DACCS has delivered only 1,186 tonnes worldwide since 2023, 81 % of them by Climeworks in Iceland; inside the EU27 it is pilots. No European bio-CCS plant has delivered a tonne yet, Ørsted's start has slipped to 2027 and Stockholm Exergi's is 2028, while US ethanol plants already deliver: 364,000 tonnes certified by Puro in 2025. Europe's 205 biochar plants removed about 170,000 tonnes of CO2 last year.

How does one explain the Commission's focus on DACCS and BECCS (with minimal or no delivered tonnes in Europe) over the pathways that have been delivering six-digit tonnages for years? In our blog post The Early CDR Certificate Is the Product we have already made the point that getting a CDR company to its first certified tonnes is the first goal to achieve. The EU must support its startups in this phase!

The current situation is strange: most "bought" tonnes are BECCS, but the "delivered" tonnes are 80 % biochar. In practice, delivered CDR in 2025 is biochar. And the EU wants to run its removal programme without it?

Permanence is the certification framework's job, not the Directive's

Yes, monitoring, reporting and verification (MRV) is not perfect yet for a number of pathways: for some methods we are still working out how to measure the tonnes of CO2 actually captured and stored, and for how long. But this is clearly a technical/scientific problem for the certification process, not a regulatory problem. The CRCF already does this job: a pathway enters only through a methodology adopted as a delegated act, with monitoring, liability and reversal rules attached, and Parliament and the Council can veto each act.

What we ask for, and what we have at stake

In Article 9c(1), Parliament should replace "generated by BioCCS and DACCS activities under Regulation (EU) 2024/3012" with "generated by activities certified as permanent carbon removals under Regulation (EU) 2024/3012". That is the wording of Amendments 1284, 1285, 1286, 1288, 1293 and 1300, and the first of the two options the five European removal associations put to the negotiators.

The ENVI committee votes on 1 December; if you run or work for a removal company in the EU, your MEP sits on that committee or knows someone who does.

For transparency, so nobody has to guess our interest: the Carbon Drawdown Initiative is funded by our family foundation, we are super-early investors in more than 30 removal companies and we run enhanced rock weathering research of our own; our CEO Dirk Paessler is co-founder and vice president of the Negative Emissions Platform; this post is the Initiative's view, not the Platform's. A wider gate in the ETS could help companies we hold shares in, and 144 other EU companies in 16 member states.

All numbers come from the Captain Drawdown data release 2026-11.4, open at captaindrawdown.com/data and on GitHub.

Release 2026-11.4, published on 5 November 2026, is the first public release. It holds three datasets.

• CDR Company Directory: 913 published companies out of 1,470 monitored, from 65 countries, 403 in Europe and 273 in the United States. Per company: pathway, country, a liveliness tier with a 0 to 100 score, EU SME size class, registry presence and evidence dates. Biochar 377, enabling technology 179, direct air capture 125, BECCS 84. Licence CC BY-NC-SA 4.0.

• CDR Researcher Census: 55,865 researchers from 15,896 papers and 11,015 institutions in 170 countries; a calibrated estimate of 51,946 corrects for author records that are still split. Six pathway top lists, each by activity and by citations. Licence CC BY 4.0.

• CDR Shops: 14 shops with a live checkout and third-party certification, prices per tonne, re-checked monthly. Licence CC BY 4.0.

The two maps above come straight out of these files, and they show two different geographies. China has 16,476 carbon removal researchers and 9 published companies, India 3,945 and 27. The United States has 6,606 researchers and 273 companies, Germany 1,728 and 73. Research is everywhere. The industry is not, yet.

Plot the two numbers against each other, one dot per country, and the gap is the picture: 106 countries have carbon removal researchers and not a single published company. Among countries with at least 100 researchers, Switzerland, Singapore, Sweden and the Netherlands have the most companies per researcher, China, South Korea and Saudi Arabia the fewest.

Every file is versioned and listed in a manifest with checksums. The files live in the public GitHub repository dirkpaessler/cdr-databases, mirrored at captaindrawdown.com/data and as datasets on Hugging Face; each dataset is archived at Zenodo with its own DOI (companies, researchers, shops). A JSON API and an MCP server serve the same release; Captain Drawdown's technical note walks through all five ways in.

A release every month

From now on we release an update every month, around the 15th. A release covers all three datasets and carries a version number: 2026-11.4 was the fourth cut of November, patches within a month count up. A published release never changes; corrections go into the next one, and v1/latest/ always points at the newest. The changelog lists what changed in each release, down to single corrections. Cite the version with the number, and the number stays reproducible whatever we fix later.

Quick start with Claude, ChatGPT or Gemini

Use the data where it lives, not a copy. A file downloaded in November is stale in December; every path below reads the current release at the moment you ask, and every answer names the version it comes from.

1. Hand the assistant the map to the files. The release manifest at captaindrawdown.com/data/v1/latest/manifest.json lists every file of the newest release with what it holds, its row count, size, licence and checksum: 37 CSV files, each also as JSON, from companies.csv (913 rows) and researchers_by_country.csv (171 rows) to the top-100 lists per pathway and shops.csv. Paste that one address into Claude, ChatGPT or Gemini, ask your question, and the assistant picks the file, fetches it and names the version. captaindrawdown.com/llms.txt is the same front door in plain words, and /api/v1/index.json lists the JSON endpoints for single companies, top lists, shops and headline numbers.

2. Connect the data as tools. For Claude Code, Cursor and ChatGPT in developer mode: pip install captain-drawdown-mcp, then claude mcp add captain-drawdown -- captain-drawdown-mcp. The server reads the public API when it is asked and keeps nothing older than six hours, so a question about enhanced rock weathering companies in Germany is answered from this month's release, with the version attached. The setup for ChatGPT and Cursor is on captaindrawdown.com/api.

3. Spreadsheets and code, always current. The newest release sits behind a fixed address, https://captaindrawdown.com/data/v1/latest/companies.csv (the same under v1/latest/ on GitHub). In Google Sheets, =IMPORTDATA("https://captaindrawdown.com/data/v1/latest/companies.csv") refreshes when the sheet opens. In Python, pandas.read_csv on that address, or load_dataset("carbondrawdowninitiative/cdr-company-directory") with the Hugging Face datasets library, pulls the current release on every run.

Two exceptions, on purpose. If a result must not change, say for a paper, use a version folder such as v1/2026-11.4/ and cite its DOI. And if your assistant cannot read the web, attach the file from v1/latest/, write the version into your notes, and replace it after the next release.

Please cite the data

The data is free to use, and we want it used. The researcher census and the shops are CC BY 4.0: use them for anything, commercial or not, and name the source. The company directory is CC BY-NC-SA 4.0, because the seed list we built on carries that licence: non-commercial use, share alike, name the source. A citation looks like this:

Captain Drawdown and Dirk Paessler (2026). CDR Company Directory, version 2026-11.4. Carbon Drawdown Initiative. https://captaindrawdown.com/data/companies/. https://doi.org/10.5281/zenodo.23957359

Replace the dataset, the DOI and the version as needed; every API answer and every dataset page carries its own citation line. If you publish something built on the data, a paper, a chart, a story, send us the link. We read everything.

Corrections and improvements

Nine hundred companies and fifty-six thousand people will not be right in every row. If a company is missing or misfiled, if a researcher's record is split or wrongly placed, if a shop has closed or changed its prices: write to info@carbon-drawdown.de or open an issue in the GitHub repository. Corrections go into the next monthly release and are listed in the changelog with the version they landed in. Researchers who want their entry corrected or removed write to the same address; the privacy notice describes what the census holds: authors, institutions and papers from OpenAlex and ORCID, nothing beyond the public record.

A year ago we tried to buy 200 tonnes of CDR to balance our own GHG emissions and found out that almost nobody would sell to us without a sales call. We built a small directory of the shops that would — that story is here. This month we rebuilt that directory from scratch, and this time we did not browse. Captain Drawdown, our AI agent, enumerated the market: every supplier on the Isometric registry (60 suppliers, 78 projects), every supplier on Puro.earth (70 suppliers, 119 projects), every durable-CDR project on Rainbow (51 suppliers), plus the 566 reachable supplier websites in Captain's own database of 1,814 CDR companies. Checkout pages, embedded payment widgets, storefront platforms, "buy" buttons hidden three clicks deep — if a company sells removal online, this sweep would have found it.

It found 14.

Fourteen shops, out of roughly 650 companies checked, where an individual can buy a few tonnes of durable, independently certified carbon removal with a credit card, today. That is 2 in 100. At the other 98 you will get a demo, a quote form, a "talk to our team" button, or — our favorite — a storefront that technically exists but requires an access code to see the buy button. This is an industry that spends every conference panel complaining about missing demand, and then locks its front door.

The 14 are not even the same 14

Three of last year's shops quietly stopped selling within twelve months: one vendor's purchase page simply vanished, one switched to contact-only, one webshop was shut down entirely. Three came in new. And one of the three new entries is not a supplier at all but a reseller — Remove Carbon Today, a founder-funded one-person shop from Finland that buys certified CORCs and retires them in your name within a month, because selling one tonne to one person is apparently a job the industry itself won't do.

When the best retail experience in a market is run by a volunteer-scale middleman, that tells you something about where the gap is: the unit economics of small sales (verification fees, payment fees, support) scare suppliers off, and the aggregator layer that should absorb them barely exists.

We even tried to buy the industry its customers

Before anyone answers "there just isn't any retail demand": we tested that, with money. In late 2025/early 2026 we spent about €8,700 on Facebook and Instagram ads pointing at this directory. Two months of ads bought 11,300 visitors at €0.76 each, and the directory passed 276 of them — 2.4% — straight to a CDR shop's door, at an effective €31.50 per delivered shop visitor. Under conservative assumptions that is a customer acquisition cost of roughly €315, a number plenty of subscription businesses pay happily.

Demand can be bought at ordinary ad-market prices. It is the supply side that answers the doorbell with a contact form.

What a year of traffic actually looks like

The full twelve months: 12,932 unique visitors, of which 91% came from that one paid campaign. After the ads stopped, the site settled at 22–109 visitors per month. Organic Google search delivered 84 visitors in an entire year — nobody types "buy carbon removal" into a search engine, because nobody expects it to be buyable. The one organic channel that did show up unprompted: AI assistants, with 23 referrals from ChatGPT alone. Vendors have started writing to us because "AI models are increasingly citing your site." Being purchasable online is quietly becoming the precondition for existing in AI-mediated buying advice. Companies without a checkout will not just lose the sale; they will not even be mentioned.

No shops means no prices

Across the 14 working checkouts, a tonne of durable CDR costs between €142 and US$1,000 — a 7x spread that nobody can arbitrage, because you cannot comparison-shop what you cannot buy. Price discovery for small buyers simply does not happen in this market. Case in point: our top-rated vendor cut its price from €400 to €300 per tonne this year, a 25% drop, and as far as we can tell nobody noticed. There was nowhere to notice it.

The fix is one widget away

Here is the uncomfortable part: the hard infrastructure already exists. Isometric ships an embeddable Stripe checkout; three of our 14 shops are exactly that widget — Inplanet, Planeteers and NULIFE all sell through it. If Isometric, Puro and Rainbow gave every validated supplier a "sell one tonne online" switch, the number of real CDR shops could grow tenfold overnight, with certification and retirement handled where they belong — in the registry. Until then, carbon removal remains un-amazonified: no one-click buying, hardly any visible prices, and a market that keeps asking where the buyers are while standing behind a locked door.

The rebuilt directory is live at cdr-shops.com: 14 shops, each rated A–C on MRV, certification and delivery, every link and price checked continuously by Captain Drawdown. If you run a CDR company and we missed your working checkout, one email gets you listed, free. If you run a CDR company and you don't have a checkout: it is the cheapest demand-side intervention you will ever build.

We recently surveyed our portfolio companies on two questions: how many people do you employ, and how much CDR have you actually delivered?

The answers:

• 961 employees across the portfolio

• 806,000 tonnes of CO2 removed and delivered by end of 2025

• 55% of our companies have already issued verified CDR certificates to buyers

806 kilotonnes. The entire CDR industry globally was estimated to have delivered roughly 0.015 gigatonnes in 2025. Our portfolio accounts for a meaningful slice of that, about 5%. The 55% certificate figure matters most to me. More than half our portfolio has already transferred real, verified carbon removal to paying customers. It's in the ground, in the rock, in the ocean.

The 10% figure and where it comes from

To understand what 961 employees means, you need context for the whole industry.

Captain Drawdown, an AI assistant we built to track the CDR sector, puts the global CDR workforce ("Pure-play CDR" companies) at a hard lower bound of 10,000 people. Accounting for contracted field operators, CDR teams embedded inside major industrial firms, and companies not yet catalogued (particularly in the Global South), the defensible upper range is 20,000 to 25,000 people worldwide.

Let that sink in. The entire global carbon removal workforce fits inside a mid-sized university. For comparison: the global wind energy sector employs roughly 1.4 million people. Solar, around 4 million. The entire CDR industry, every company, every method, every geography, employs roughly the same number of people as a single large hospital system.

This is an industry in its infancy. And yet it's tackling one of the hardest problems in climate science.

A few players have a big reach

cdr.fyi mapped Microsoft's footprint. Microsoft is now the single most consequential CDR buyer on the planet, contracting with 39 suppliers and accounting for 78.5% of all publicly disclosed durable CDR tonnes as of April 2026. Using Captain Drawdown's CDR supplier database and Microsoft's list of suppliers (collected from the Internet) we can estimate that Microsoft's purchases co-finance an estimated 10 to 20% of the supplier-side CDR workforce (pure-play CDR companies only, i.e. without BECCS). Many people at purpose-built CDR companies depend on a primary commercial relationship with Microsoft. That's only possible because the industry is still this small.

Two sides of the same coin

The smallness is a risk. When this much depends on so few buyers, so few investors, so few early bets, the ecosystem is fragile. One bad market cycle, one shift in policy, one buyer pulling back, and you feel it across the entire sector. But the smallness is also an opportunity. In most industries, no single investor shapes the whole field. Here, they can. Individual companies, individual scientists, individual funders still move the needle in a way that simply isn't possible in mature markets. This is what concentrated, early-stage capital can do: not just back companies, but help determine what gets built at scale. Who shows up now writes the blueprint for what comes next.

What this actually tells us

I've watched a lot of industries grow from nothing. The early years always look the same: small numbers, scrappy teams, more believers than proof. CDR is in that phase right now. But the proof is coming. New companies are forming, buyers are writing real checks, and the science is advancing faster than most people realize. The companies that stick around are the ones that build something solid before the hype arrives. That's what we're trying to do.

Nearly a thousand people. 806 kilotonnes removed. 55% already delivering certified carbon removal. This isn't theory. It's working.

Precision Sampling tackles the biggest cost driver within an active ERW project. Leading certification standards already support stratified sampling to account for soil heterogeneity. In practice, that stratification has often relied on rough proxies like topography or broad soil type, not the specific parameters that actually drive CDR.

AEROC replaces this with data-driven soil zone mapping. Machine learning models trained on satellite data generate pixel-level predictions of key soil properties at 10x10 meter resolution. Those predictions are used to cluster fields into zones of similar weathering rates, allowing samples to be distributed representatively rather than randomly.

The outcome: up to 60% fewer soil samples required, with no loss in accuracy for CDR quantification.

Together, the two services can cut overall ERW project costs by up to 32%.

Why This Matters for the ERW Industry

Enhanced rock weathering is one of the most promising pathways for durable, scalable carbon removal. It builds on natural processes, works on existing farmland, and generates co-benefits for soil health. But it won't scale if the economics don't work.

Right now, they often don't. Not because the science fails, but because finding good sites and verifying results is too expensive. Every percentage point reduction in project costs makes more land viable, more projects fundable, and more carbon removals credible.

AEROC is solving a structural problem in the ERW value chain. Not by reinventing the method, but by making the existing one far more efficient.

The Team

The founding team brings together remote sensing scientists, geochemists, and soil scientists, an unusual combination that reflects exactly what this problem requires. The company includes Dr. Viktoria Kristollari (National Technical University of Athens), Dr. Rachael Marshal (University of Münster) and Casper Wijckmans (ETH Zürich), and Dr. Jonas Tusch (University of Cologne) as part of its core scientific team.

Why We Back AEROC

At Carbon Drawdown Initiative, we back solutions that make the whole ecosystem work better. AEROC doesn't compete with ERW project developers. It makes them more effective.

The combination of a founding team that has operated ERW projects themselves, a clear commercial model, and technology that addresses two of the most stubborn cost bottlenecks in the industry makes AEROC a company we're glad to support.

Better data means better projects. Better projects mean more carbon removed.

That's the logic we invest in.

To learn more about their work, visit:www.aeroc.earth

This series highlights the companies Carbon Drawdown Initiative has invested in.

There are more than 16,000 publicly-owned wastewater treatment plants in the United States and over 100,000 worldwide. Each one continuously produces CO₂ inside the same enclosed reactors where alkalinity enhancement can take place. That makes wastewater treatment one of the most practical platforms for durable carbon removal.

CREW Carbon integrates alkalinity enhancement directly into wastewater treatment plants to improve plant performance, while removing CO₂. Instead of building new carbon-removal infrastructure, the company optimizes the wastewater treatment process that already operates at massive scale in cities around the world.

Born out of research at Yale by co-founders Dr. Joachim Katchinoff and Professor Noah Planavsky, CREW applies alkalinity enhancement inside controlled wastewater reactors — enabling measurable, verifiable carbon removal alongside cleaner water and more efficient treatment facilities. By embedding carbon removal into essential infrastructure, CREW creates a scalable pathway that aligns climate impact with utility performance.

How It Works

Wastewater treatment plants grow trillions of microbes that digest organic waste. The process produces CO₂ inside large, controlled reactors. CREW enhances alkalinity within the treatment process, converting biologically produced CO₂ into stable dissolved forms, before it can return to the atmosphere, storing it in the water on a geological timescale.

Because wastewater facilities already:

• Continuously process biogenic carbon

• Operate in closed, monitored systems

CREW’s carbon removal is measurable and verifiable by design. At the same time, CREW’s patented wastewater alkalinity enhancement improves treatment efficiency and reduces pollution discharged into waterways — delivering operational benefits to utilities and ratepayers.

CREW’s model:

• Leverages existing infrastructure

• Avoids time- and capital-intensive projects

• Integrates into regulated public systems

• Generates co-benefits beyond CO2 removal

If even a small fraction of global wastewater plants adopt this approach, the impact scales quickly.

Key Milestones

CREW has already delivered the world’s first verified CO₂ removals from wastewater treatment facilities under the wastewater alkalinity enhancement protocol.

Since launching commercial operations in 2024, CREW has:

• Deployed at wastewater treatment facilities across the US and Europe, including leading utilities like HRSD (Hampton Roads Sanitation District)

• Delivered measurable performance improvements at wastewater treatment facilities, with one utility considering deferring $400 million in planned capital upgrades

• Secured $33M+ in carbon removal offtake agreements

• Captured 2,000+ tons of CO2 faster and more cost effectively than other carbon removal credit suppliers

• Won a $2.3 million award from the Colorado Energy Office as part of the state’s Clean Air Program

• Developed proprietary measurement, reporting, and verification (MRV) capabilities that enable high-confidence carbon removal quantification

• Raised $25M of funding across equity and non-dilutive capital sources in 2026

This is not a lab concept. CREW is operating inside real infrastructure.

Scaling Ahead

CREW plans to expand to a global network of wastewater partnerships — each capable of removing thousands of tons of CO₂ annually. The company is also expanding its MRV (Measurement, Reporting, and Verification) platform to ensure removals remain trusted, traceable, and audit-ready as scale increases. International expansion is already underway, with growing interest from Europe and Asia. Since closing its oversubscribed Series A round, CREW is adding to the talent on the team. Some of the high-priority roles to fill in 2026 include:

• Head of Wastewater Process & Technology

• Technology Team Manager

• Supply Chain and Logistics Manager

Why We Back CREW

At Carbon Drawdown Initiative, we prioritize solutions that combine:

1. Strong scientific foundations

2. Clear and credible MRV

3. Realistic scaling potential

CREW meets all three criteria. The science builds on leading enhanced rock weathering research. The MRV is embedded in wastewater chemistry. The scaling pathway is replication across utilities — not reinvention. Carbon removal will only reach climate-relevant scale if it integrates into systems that already exist.

CREW does exactly that. And that is why we are proud to support them.

When you spread finely ground silicate rock dust on a field, the material carries a theoretical capacity to pull CO₂ out of the air. As the rock weathers it releases base cations, those cations hold dissolved CO₂ in the form of bicarbonate, and that bicarbonate can eventually be stored away for a very long time. The interesting — and genuinely hard — question is how much of that theoretical capacity actually survives all the way to durable carbon removal. The diagram above follows one batch of rock dust from left to right and shows where the potential is spent along the way.

This is a concept map, not a scoreboard. The height of each block is illustrative. Most of these quantities vary enormously from one soil to the next, from season to season and from year to year, and several of them are difficult to measure at all. So the picture is meant to be read for the relationships and the sequence of events, not for the precise size of any single block. Where a block carries a percentage it is a wide “typical” range; the figures that can be pinned to the literature are taken from the recent 74-author systematic review of enhanced rock weathering by Suhrhoff and colleagues (2026), the current best synthesis of the field.

A quick walk along the flow.

• Making and moving the rock dust already costs some CO₂. Across the life-cycle studies the review compiles, these emissions “range from 1.1 to 215 kg CO2-eq t rock⁻¹ (mean = 51.6, median = 33.15),” with “grinding and transport … averaging 41% and 39.4% of total EW life cycle GHG emissions” — on the order of ten percent of a typical rock’s removal potential, and large enough that they “can even eliminate net removals altogether.”

• In the soil the rock then slowly dissolves. The dissolution stage of the figure shows this as one undivided block: the total amount mobilised from the rock so far, before any bookkeeping about where it went.

• That bookkeeping happens in the next stage, where the dissolved cations branch into their pathways: a small share is taken up by the plants, some is locked into secondary minerals and secondary carbonates (the loss channels below the line — more on those in a moment), and a large share is temporarily parked on the soil’s cation-exchange sites (delayed, but not lost: it can be released again later).

• Only what is still in solution after these branches reaches the leachate — and only there does the chemistry of the dissolution get attributed: the part dissolved by weak carbonic acid, formed from CO₂ and water, is the productive, bicarbonate-carrying form that shows up as alkalinity in the drainage water at the bottom of the root zone; the part dissolved by strong acids carries no CO₂ with it and is lost CDR potential.

• Further downstream another slice is lost again in rivers and the ocean, for which the review concludes that “an average value of ~15-20% cumulative loss (both rivers and ocean, 10-20% of the remaining CDR potential is meant) is more appropriate for long-term storage.” A separate bonus — extra carbon stored in the soil itself — can sit on top, but it lies outside of our current work in our greenhouse experiment (it is also genuinely hard to measure, too).

The most important message is below the line. Three permanent loss channels decide almost everything. The biggest is dissolution driven by strong acids — the nitric and sulfuric acid that arrive with fertiliser and acid deposition — because “protons derived from strong acids can drive mineral dissolution without consuming atmospheric CO2, thereby reducing net CDR efficiency,” and in croplands “nitric acid generated during fertiliser nitrification represents a major source of strong acidity.” The second biggest are the secondary minerals, since “the formation of secondary minerals—especially clays—hinders the efficiency of EW, such that less CDR is realised per unit of feedstock applied.” The third is secondary carbonates, and it deserves a word because it is tempting to count it as storage: chalk-like carbonate can form in the soil and does contain carbon, but it is not durable. In the review’s words, “the precipitation of solid carbonate minerals reduces the efficiency of EW by nearly half due to the re-release of CO2,” making it “a transient rather than permanent atmospheric CO2 sink.” Accordingly, in modern carbon accounting “secondary carbonate formation is mostly presented as a potential loss of CDR potential” — which is why it sits on the loss side of the diagram, next to the clays and oxides.

Whatever ends up in these three channels is gone for the purpose of removal. The ranges on the figure — very roughly 10–40 % for strong acids, 10–40 % for clays and oxides, and up to about a quarter for secondary carbonates — are our own field-typical estimates informed by the review rather than single quoted values, and they are wide because these losses swing from almost nothing in a favourable soil to almost everything in an unfavourable one; in the worst case they can consume nearly the entire benefit. These, therefore, are the first numbers worth measuring well: if they are large, nothing else on the diagram can rescue the result.

Tellingly, the review flags exactly these as the weakest links: “Significant uncertainty remains regarding the magnitude, persistence, and timing of loss processes after initial CDR has occurred, including secondary phase formation and cation exchange.” The biggest levers on the outcome are also the ones we currently know least about — the core message of the picture.

Only once those are pinned down does the rest become interesting: how much of the dissolved rock ends up in the productive pathway (dissolution by weak carbonic acid rather than by strong acids), and how much of it is merely delayed. The delay can be long — “in large parts of the U.S. Corn Belt and Great Plains, less than 50% of CDR potential is realised within the first decade after EW deployment” — but delayed is not the same as lost.

So how much of a rock’s theoretical capacity survives all the way to durable removal at t = ∞? There is no single measured figure: the review reports removal as an annual flux, where “area-normalized CDR fluxes have a median value of 0.84 tonnes of carbon dioxide per hectare per year,” rather than as a fraction of the theoretical maximum. Working the losses through the chain gives a plausible long-run figure of very roughly 10–40 % of the rock’s chemical potential — a derived number, not a measured one, often nearer the lower end, and close to zero where the losses are large.

That, in the end, is what the figure is for: a map of where to point the instruments — and it marks two measurement windows. In the middle, telling apart the field pools — what is parked, what is locked into plants, clays and oxides, what has precipitated as carbonate — calls for advanced soil analysis, a demanding suite of techniques that is still partly in development. And at the outflow, the durable signal is the leachate alkalinity together with the drainage volume. That volume is deceptively hard to measure directly in an open field, though it can be derived from hydrogeological models that have been developed and refined over decades; in our greenhouse and lysimeter setups, by contrast, we can measure it directly and very precisely — which is exactly why controlled experiments remain so valuable for closing the balance. Between these two windows sit the losses that can quietly erase the whole exercise. Get those right first, and everything to the right of them is refinement.

The figure is schematic: it illustrates the concepts and the sequence, not exact numbers. Block heights are typical values with wide ranges that depend strongly on feedstock, soil, climate and time. Quoted figures are verbatim from Suhrhoff et al. (2026); the 10–40 % end-of-time estimate and the individual loss ranges are our own, derived from that review.

Reference: T. J. Suhrhoff, C. Dietzen, T. Kukla, et al. (2026), An Ecosystem of Carbon Dioxide Removal Reviews – Part 3: Enhanced Weathering, CDRXIV preprint #417 — cdrxiv.org/preprint/417.

Figure 1: Cumulative rainfall, on-site gauges vs official DWD station. Both on-site gauges under-catch — the tipping-bucket badly (~53 %), the Ecowitt piezo less so (~83 %).

Under-catch is normal for tipping-bucket gauges (wind, evaporation, missed light rain, the occasional fault), but a 47 % shortfall is large enough to distort a water balance if you trusted it blindly. The practical lesson: keep an official reference series, cross-check your on-site gauge against it, and know your gauge’s catch ratio before you use its numbers in a flux calculation. We ship both series in the dataset so the choice is explicit.

Ambient CO₂: the air the soil breathes into

Our ambient monitors (Seeed SenseCAP S2103) log CO₂ at two heights. Near the ground (20 cm) the mean is about 544 ppm; at 200 cm it is about 470 ppm. The near-surface air is enriched — soil respiration leaking upward — and that gradient is a small reminder that the “atmospheric” boundary condition for a field experiment is not a single global number. It matters mostly as context here, but for any scheme that infers fluxes from CO₂ concentrations, the reference level is a real choice.

The uncomfortable arithmetic of sensor survival

The hardest lesson is the simplest to state: sensors die. We started with a nearly full set — around 18 to 20 working channels for each buried variable. Four years later most of them are gone.

Figure 2: Active sensors over time, by variable. Most channels fail within two to three years; the soil-CO₂ sensors fall from a full twenty (one per pot, complete by October 2022) to a handful by late 2024, and to a single survivor by early 2025.

By the end of 2024 the moisture and 30 cm-EC channels were down from ~19 to 4, the 60 cm channels to about 8, and soil CO₂ from its full twenty to four by that autumn — and to a single sensor by early 2025. For the other buried variables the survivors thinned further through 2025. Batteries run out, connectors corrode, probes lose contact with drying soil — the ordinary attrition of instruments left outdoors for years.

This has two consequences we have been explicit about throughout the series:

• Late-period site means rest on few sensors. A “site average” in year four is an average of a handful of surviving probes, not the original twenty — less representative, and worth flagging rather than smoothing over. (It is also why, when part of the experiment was repurposed in 2025, we kept only the continuing pots in the released sensor record.)

• Survivorship is a confound to rule out, not ignore. Because sensors die non-randomly over time, a raw sensor↔leachate correlation could in principle be a time-trend artefact. We checked: after removing the common time trend, the core relationships hold (for example, 60 cm soil-EC ↔ leachate EC stays at r ≈ 0.61 partial for time, down from 0.70 raw but clearly real; soil CO₂ ↔ pH is unchanged). The signal is not an artefact of which sensors happened to survive.

What this means for MRV

None of this is a reason to distrust in-situ MRV — the earlier articles show how much these sensors reveal. It is a reason to engineer for the long run:

• Anchor every on-site gauge to an official reference, and carry the catch ratio.

• Over-provision sensors so the late record — usually the most valuable — still rests on several instruments, not one survivor.

• Report coverage plainly: say how many sensors a late-period mean is built on.

• Test time-trend and survivorship confounds explicitly before trusting a long correlation.

• Sample the solid phase, not just the drainage. Cation retention and exchange pools, secondary clay minerals, and the fate of soil organic carbon are invisible to a leachate sampler and a gas probe — yet they decide how much of any weathering becomes durable CDR.

The series, in short

Across 1,400 days at one well-characterised site, the macro-scale weathering behaved as the chemistry predicts, the dose–response we hoped to credit stayed inside the noise, and the most useful MRV signal came not from the leachate alone but from cheap, buried, continuous sensors — read carefully, corrected for moisture and season, and never trusted past what the survivors can support. Great macro-scale performance, messy micro-scale reality — measured, for once, over years. The full data, code and figures are in the accompanying dataset; take them apart yourself.

The XXL Lysimeter dataset (2022-2026) series

1. Intro article — why a 1,400-day, daily-instrumented experiment is worth the wait.

2. Dose–response — the initial flush (the Birch effect, not weathering), the long convergence, the n = 4 power problem, and the FINE-is-a-different-rock confound.

3. The soil breathes — how temperature and soil CO₂ drive a seasonal weathering engine visible in the buried sensors.

4. A buried EC sensor as a continuous MRV proxy — the 60 cm probe tracks leachate chemistry remarkably well (and the 30 cm one does not); the long-term, in-situ extension of our EC-as-alkalinity-proxy work.

5. What the soil tells us when it breathes — reading four years of buried CO₂ sensors: the soil as a living system of warmth, water and microbes — and why even a sensor in every pot is the weathering engine, not a carbon meter.

6. Four years of buried sensors — rainfall, ambient CO₂, and the reality that most field sensors die within two to three years.

Data, code & figures:doi.org/10.5281/zenodo.21216439 (CC-BY-4.0) · github.com/dirkpaessler/carbdown_xxl_lysimeter_2022_2026

Figure 1: Soil CO₂ by dose over four years (30-day rolling median on the daily record). A strong seasonal wave — summer peaks, winter lows — repeated in every treatment. (The five dose curves stay tangled; more on that below.)

What runs the breath: warmth, then water

Two things govern how hard the soil breathes, and the sensors let us see both at once. Temperature is the throttle: warmer soil respires more, and soil CO₂ climbs with it — up to a point. Above about 18 °C the trend reverses. The reason is in the second line on the same plot: as the soil warms into summer it also dries, and once the water runs short the microbes slow down and the breath falls despite the heat.

Figure 2: Soil CO₂ (black) and soil moisture (blue) against soil temperature, each a smooth trend with its 95 % confidence band; grey points are the daily readings. CO₂ climbs with temperature to ~18 °C (r = +0.73) then turns over (r = −0.32) exactly as the moisture trend slides away. On warm days it is moisture, not temperature, that sets the CO₂ (r = +0.66).

So the soil’s breath is a product of warmth and water — either one, missing, throttles it. That interaction is not an abstraction. We can watch it happen.

Rewetting wakes it up: a repeated Birch effect

When dry soil is rewetted, the dormant microbes roar back and the soil exhales a burst of CO₂ — the Birch effect. In article 2 we met it once, as the flush that followed the first big storm on our freshly built soil in August 2022. With daily CO₂ in every pot, we can now show it is not a one-off: it happens every time a dry spell breaks.

At this depth the soil dries deeply about once a year — late summer into autumn — so we found five such dry→wet transitions (2022 and 2023 in summer, 2023 and 2024 in autumn, and 2025, the driest of all — though by then nearly all the CO₂ sensors had died, so it survives only in the EC). We lined each record up at its moisture minimum and ran it to the next rewetting, so every curve is one full drought-to-drought cycle.

Figure 3: Soil CO₂ (left, per-pot median + IQR) and soil EC (right), aligned at each soil-moisture minimum (day 0) and run to the next rewetting (▼) — one full drought-to-drought cycle per event. Solid is the response to that rewetting (the burst, then the decay into winter); dashed is the following year’s warm season rising toward the next drought. The salt flush (EC) rises after every event, warm or cool. Labels: how dry the soil got and how warm the fortnight after.

Two signatures, behaving differently, and both instructive:

• The respiration burst (CO₂) needs warmth. The two summer rewettings (20–22 °C after) set off large bursts — CO₂ roughly doubling to quadrupling, the 2022 storm on the raw built soil more still. The two autumn rewettings (11–16 °C) barely stir: water alone is not enough; the microbes only throw a party when it is also warm. This is the temperature × moisture interaction of Figure 2, caught as a live event.

• The salt flush (EC) happens every time. Soil EC climbs after all five rewettings, warm or cool — including the 2025 drought that the CO₂ sensors never lived to see — because mobilising and dissolving salts is physical chemistry, not biology; it does not wait for warmth.

One pulse, two clocks. It is the clearest picture we have of the soil as a living system responding to its water supply.

The breath drives the weathering

This biology is not a sideshow to the carbon story — it is the engine of it. The CO₂ the soil breathes dissolves into the pore water as carbonic acid, and that acid is what attacks the rock. We can see the first half of that chain directly: more soil CO₂, lower leachate pH.

Figure 4: Soil CO₂ vs leachate pH (left) and vs leachate alkalinity (right), per pot. CO₂ drives the acid strongly (pH, median per-pot r = −0.68) but tracks the alkalinity that acid produces only weakly (TA, r = +0.36) — well below the leachate EC ↔ TA benchmark (r = +0.63).

But note the second panel, because it is the whole reason a CO₂ sensor cannot be a carbon meter. Soil CO₂ makes the acid (strong link to pH), yet it barely tracks the bicarbonate that finally leaves the pot (weak link to alkalinity). The gas is the cause at the top of the chain; how much carbon is actually exported depends on how much rock dissolves and, above all, how much water drains through to carry the ions away — and a gas sensor is blind to the water. Soil CO₂ is the engine of weathering, not the meter that counts the product.

There is a subtler trap, too. Even a clear CO₂ signal would be ambiguous. Spreading crushed rock also fertilises the soil, and well-fed microbes respire harder. That extra breath could mean more weathering (good) — or microbes burning through the soil’s own stored carbon, a CO₂ source that would eat into the removal (bad). The same reading can point either way, and the gas alone cannot tell you which. And the blind spot compounds over the years: with a burst on every rewetting (the repeated Birch effect above), those pulses could be steadily drawing down the soil’s own organic-carbon stock — and a probe in the pore space cannot see it. Only sampling the solid phase, the soil itself, would show whether that carbon is being lost; it is a gap this leachate-and-sensors series cannot close on its own.

And the basalt dose barely touches any of it

Which brings us to the treatments. With a sensor in every pot we can finally ask whether more basalt changes the soil’s breathing — and the answer is no. It should not surprise us: the leachate alkalinity and the buried EC probe (articles 2 and 4) already show no dose effect, so a CO₂ dose signal would be the anomaly.

Figure 5: Season-adjusted soil CO₂ per treatment relative to control (dot = pot, diamond = mean ± 95 % CI). Every treatment overlaps control; 400 t/ha trends ~18 % lower but is not significant (p = 0.33), and the pots scatter ×0.5–×1.7 within each treatment.

The pots vary far more among themselves than the doses vary from one another. Whatever the basalt does to the soil’s biology here, it is smaller than the soil’s own pot-to-pot spread — the same verdict every other instrument returned.

What the sensors were good for

So the soil-CO₂ sensors did not measure our carbon removal. What they did was let us read the soil as a living thing: a factory breathing five-to-twenty times the outdoor air, its output set by the interplay of warmth and water, roaring back to life every time the rain returns, and quietly acidifying its own pore water — the precondition for the weathering we are after. For counting carbon, the instruments that see the alkalinity itself — the leachate, and the buried EC probe of article 4 — remain the ones to trust. A soil-CO₂ sensor is best used for what it is genuinely good at: watching the engine run.

Next, and last: four years of the plumbing behind all of this — rainfall and the water balance, ambient CO₂, and the uncomfortable arithmetic of sensor survival.

The XXL Lysimeter dataset (2022-2026) series

1. Intro article — why a 1,400-day, daily-instrumented experiment is worth the wait.

2. Dose–response — the initial flush (the Birch effect, not weathering), the long convergence, the n = 4 power problem, and the FINE-is-a-different-rock confound.

3. The soil breathes — how temperature and soil CO₂ drive a seasonal weathering engine visible in the buried sensors.

4. A buried EC sensor as a continuous MRV proxy — the 60 cm probe tracks leachate chemistry remarkably well (and the 30 cm one does not); the long-term, in-situ extension of our EC-as-alkalinity-proxy work.

5. What the soil tells us when it breathes — reading four years of buried CO₂ sensors: the soil as a living system of warmth, water and microbes — and why even a sensor in every pot is the weathering engine, not a carbon meter.

6. Four years of buried sensors — rainfall, ambient CO₂, and the reality that most field sensors die within two to three years.

References (Birch effect / drying–rewetting)

• Birch, H.F. (1958). The effect of soil drying on humus decomposition and nitrogen availability. Plant and Soil 10(1): 9–31. https://doi.org/10.1007/BF01343734

• Davidson, E.A., Belk, E. & Boone, R.D. (1998). Soil water content and temperature as independent or confounded factors controlling soil respiration. Global Change Biology 4(2): 217–227. https://doi.org/10.1046/j.1365-2486.1998.00128.x

• Kim, D.-G. et al. (2012). Effects of soil rewetting and thawing on soil gas fluxes: a review. Biogeosciences 9(7): 2459–2483. https://doi.org/10.5194/bg-9-2459-2012

• Barnard, R.L., Blazewicz, S.J. & Firestone, M.K. (2020). Rewetting of soil: revisiting the origin of soil CO₂ emissions. Soil Biology and Biochemistry 147: 107819. https://doi.org/10.1016/j.soilbio.2020.107819

• Miller, A.E. et al. (2005). Episodic rewetting enhances carbon and nitrogen release from chaparral soils. Soil Biology and Biochemistry 37(12): 2195–2204. https://doi.org/10.1016/j.soilbio.2005.03.021

• Liu, F. et al. (2025). Abiotic processes dominate short-term carbon emissions in sandy soils following rewetting. Communications Earth & Environment 6: 733. https://doi.org/10.1038/s43247-025-02733-z

Data, code & figures: doi.org/10.5281/zenodo.21216439 (CC-BY-4.0) · github.com/dirkpaessler/carbdown_xxl_lysimeter_2022_2026

We used the LSE01 — LoRaWAN Soil Moisture & EC Sensor from Dragino (and its sister product, the Dragino LSPH01 soil-pH sensor), a simple, low-cost sensor that transmits the data via the LoRaWAN protocol. The sensors sent one data point every 60 minutes. In total we deployed 120 probes — an EC/moisture probe and a pH probe at each of three positions (30 cm, 60 cm and in the drainage tank) for every one of the 20 pots (2 × 3 × 20 = 120). Each EC/moisture probe reports soil moisture, bulk electrical conductivity and temperature from the same three prongs; the EC reading is the one we test against the leachate.

The proxy chain, one link at a time

The open question: does a sensor in the soil see the same thing as the leachate?

For the 60 cm probe, it largely does. Pooled across all pots, the buried 60 cm EC tracks leachate lab-EC at r = 0.70 (n = 355), and it tracks the ions that carry the alkalinity charge just as well — Ca and Mg at r ≈ 0.70. Total alkalinity itself follows more loosely, at r ≈ 0.45: EC measures the total ionic strength of the water, of which bicarbonate is only one part, so the tightest match is EC-to-EC. But since leachate EC is itself a strong stand-in for TA (r ≈ 0.95 in our greenhouse data), the buried probe reaches alkalinity through that chain — soil-EC → leachate-EC → TA.

Figure 1: In-situ 60 cm soil-EC vs leachate EC, all treatments pooled (r = 0.70, n = 355). The buried probe follows leachate ionic strength across every dose.

Crucially, this is not carried by one treatment. Broken out per dose, the 60 cm probe tracks leachate EC in every one — control r ≈ 0.75, 100 t/ha 0.60, 200 t/ha 0.69, 400 t/ha 0.77, FINE 0.77. A proxy that only worked in the high-dose pots would be useless for MRV; this one works across the board.

Figure 2: Soil-EC vs leachate EC with all doses overlaid — 30 cm (left) and 60 cm (right); coloured line + r per dose. The 60 cm probe tracks leachate EC in every dose (pooled r = +0.70, n = 355); the 30 cm probe is consistently loose (pooled r = +0.39, n = 319).

Depth matters — the 30 cm probe does not work

Figure 2 also shows the failure. The 30 cm probe is much weaker: r = 0.39 pooled. The shallow soil dries out, warms and cools, and channels water unevenly; 60 cm sits closer to the draining pore water that actually becomes leachate. If you bury one EC sensor for MRV, depth is not a detail — it is the difference between a usable proxy and a poor one.

One caveat we deliberately do not act on: because the probe reads EC and moisture from the same prongs — and the LSE01 already temperature- and conductivity-compensates its EC — soil moisture is not an independent variable for us to filter or correct against. It would not matter anyway: whenever we pumped leachate the soil was wet, with only 3 of 355 paired readings below 10 % VWC. So we apply no dryness filter and no moisture correction — the raw buried EC is what we compare.

What this means for MRV

The picture is genuinely useful: a continuously logged 60 cm soil-EC probe tracks the ionic strength of the leachate — and hence its alkalinity — across every treatment at this site, at r ≈ 0.70. It cannot replace titration outright: it is a proxy, and (as our greenhouse work insists) a site-calibrated one, not a global formula. And it is a bulk-soil reading: what the probe registers depends not only on the pore-water chemistry but on how much water the soil is holding around the prongs, so soil moisture manipulates the value directly. Because soil texture and water-holding capacity — and the climate that drives them — differ from site to site, the soil-EC ↔ leachate relationship can shift drastically elsewhere; treat the calibration as site- and climate-specific and re-anchor it wherever you deploy, rather than assuming ours transfers. It also tracks all the dissolved load, not just the weathering-derived part — so it measures leachate chemistry, not weathering per se. But it can fill the long gaps between grab samples with continuous in-situ data, at a fraction of the cost — the continuous, buried extension of the EC-as-TA argument we made in the greenhouse.

The defensible recipe is a hybrid, the same shape as our greenhouse conclusion: bury a 60 cm EC probe for continuous data; take periodic leachate titrations to anchor and recalibrate the EC ↔ TA relationship for that site; and treat a sudden EC jump as a prompt to investigate, not as instant CDR.

Next: we chase a tempting third signal — a weathering fingerprint in the soil CO₂ itself.

The XXL Lysimeter dataset (2022-2026) series

1. Intro article — why a 1,400-day, daily-instrumented experiment is worth the wait.

2. Dose–response — the initial flush (the Birch effect, not weathering), the long convergence, the n = 4 power problem, and the FINE-is-a-different-rock confound.

3. The soil breathes — how temperature and soil CO₂ drive a seasonal weathering engine visible in the buried sensors.

4. A buried EC sensor as a continuous MRV proxy — the 60 cm probe tracks leachate chemistry remarkably well (and the 30 cm one does not); the long-term, in-situ extension of our EC-as-alkalinity-proxy work.

5. What the soil tells us when it breathes — reading four years of buried CO₂ sensors: the soil as a living system of warmth, water and microbes — and why even a sensor in every pot is the weathering engine, not a carbon meter.

6. Four years of buried sensors — rainfall, ambient CO₂, and the reality that most field sensors die within two to three years.
   
   Data, code & figures: doi.org/10.5281/zenodo.21216439 (CC-BY-4.0) · github.com/dirkpaessler/carbdown_xxl_lysimeter_2022_2026

Figure 1: The weathering engine in the sensors — soil temperature vs soil CO₂ (r = +0.58), soil CO₂ vs leachate pH (r = −0.74), and soil CO₂ vs leachate alkalinity (r = +0.58). Dots are coloured by basalt dose (each dose’s own soil-CO₂ sensor; temperature = site mean); the black line is the pooled fit across all doses.

This is the mechanism textbooks describe, caught in the act in a field soil over four years. And it is not a quirk of one treatment: split the same chain out by basalt dose and the signs and rough magnitudes repeat in the control, 100, 200 and 400 t/ha pots alike. The engine is a property of the soil and its seasons, not of the amendment.

With a soil-CO₂ sensor in every pot — four per dose — we can confirm the chain pot by pot as well as in the site average: temperature → soil CO₂ (median per-pot r = +0.78), soil CO₂ → leachate pH (−0.68) and soil CO₂ → leachate TA (+0.36). The same signs, in individual pots. Article 5 takes that per-pot record apart — including the point where the engine stalls: above about 18 °C the summer-dried soil goes quiet and CO₂ falls even as the temperature keeps climbing.

The seasonal wave

Because temperature drives it, the whole system oscillates with the year. Soil CO₂ peaks in summer and collapses in winter, and leachate pH moves in anti-phase — lowest when the soil is breathing hardest.

Figure 2: Soil CO₂ (daily sensor, site mean) and leachate pH over ~1,400 days. CO₂ peaks each summer; pH dips in anti-phase.

This seasonality is exactly the kind of structure you lose if you sample only occasionally. A grab sample in July and one in January are not two draws from the same distribution — they are two points on a wave. For MRV that is a double-edged message: the signal is real and repeatable, but its timing matters, and sparse sampling can alias the season into an apparent “trend.”

A complication: temperature and CO₂ travel together

Temperature and soil CO₂ are themselves correlated (r = +0.58) — both ride the same seasonal wave — so we should ask which one actually drives leachate pH. Partial correlations let us hold one constant and test the other:

• leachate pH vs soil CO₂, controlling for temperature: r = −0.56 (still strong);

• leachate pH vs temperature, controlling for soil CO₂: r = −0.69 (also still strong).

Both survive. Neither is a mirage of the other. The reading that fits the chemistry: CO₂ is the proximate lever on pH, temperature is the seasonal throttle on CO₂, and because they co-vary you see both linked to pH even after mutual control. The link from CO₂ to alkalinity weakens but does not vanish when temperature is held constant (r = +0.58 → +0.44), which is what you would expect if temperature acts through CO₂ rather than around it — leachate TA correlates with soil CO₂ (+0.58) more tightly than with temperature directly (+0.46).

One measurement caveat we keep in view: our soil-CO₂ probes are NDIR sensors, and NDIR readings can drift high when the pore space — and the sensor itself — is humid. Warm summer soil is often also moister at depth, so part of the measured temperature→CO₂ rise could be a humidity artefact rather than extra respiration. We therefore trust the direction of this chain more than its exact slope, and we do not over-read the temperature→CO₂ coefficient.

What this means for MRV

The mechanistic chain is legible in cheap, continuous sensors — and that is the opening for the next article. If soil CO₂ and temperature move leachate pH and alkalinity in a repeatable way, then continuously logged in-situ variables carry information about the weathering flux between the expensive monthly titrations. The soil’s breathing is not noise to be averaged away; it is signal, and it is measurable without a lab.

Next: we put that idea to the test with the one sensor that turns out to track leachate chemistry best — the buried electrical-conductivity probe.

The XXL Lysimeter dataset (2022-2026) series

1. Intro article — why a 1,400-day, daily-instrumented experiment is worth the wait.

2. Dose–response — the initial flush (the Birch effect, not weathering), the long convergence, the n = 4 power problem, and the FINE-is-a-different-rock confound.

3. The soil breathes — how temperature and soil CO₂ drive a seasonal weathering engine visible in the buried sensors.

4. A buried EC sensor as a continuous MRV proxy — the 60 cm probe tracks leachate chemistry remarkably well (and the 30 cm one does not); the long-term, in-situ extension of our EC-as-alkalinity-proxy work.

5. What the soil tells us when it breathes — reading four years of buried CO₂ sensors: the soil as a living system of warmth, water and microbes — and why even a sensor in every pot is the weathering engine, not a carbon meter.

6. Four years of buried sensors — rainfall, ambient CO₂, and the reality that most field sensors die within two to three years.

Data, code & figures: doi.org/10.5281/zenodo.21216439 (CC-BY-4.0) · github.com/dirkpaessler/carbdown_xxl_lysimeter_2022_2026

Figure 1: Cumulative bicarbonate export per treatment at day 1,011 (last date with the full n = 4); point = mean, bars = 95 % CI across the four replicate pots. Every treatment’s interval overlaps the control band (grey). Control 1.58 ± 0.14, 100 t/ha 1.52 ± 0.07, 200 t/ha 1.58 ± 0.03, 400 t/ha 1.49 ± 0.07, FINE 1.69 ± 0.13 tCO₂e/ha. Volume model avgBD.

The control is not at the bottom. It is in the middle — level with 200 t/ha, above both 100 and 400 t/ha. The dose ladder we expected is simply not there. Every error bar overlaps every other. Whatever the basalt is doing to alkalinity export at this site, over these four years, it is smaller than the pot-to-pot scatter of the soil itself.

Figure 2: The same export as a time series, per treatment (tCO₂e/ha = local CDR), volume model avgBD. Solid band: 95 % CI across replicate pots (n ≤ 4); the lines stay bunched throughout.

The early peak is disturbance, not weathering

The obvious objection is that the basalt surely showed its hand early, before the doses converged. The data say no — and the reason is worth spelling out, because the first big signal in this experiment is not weathering at all.

The soil was built in spring 2022 — sieved, homogenised, shovelled into the pots — and then stayed relatively dry until a heavy thunderstorm in August 2022 delivered the first real drainage to the tanks. You can see it in the volumes: the pumped leachate jumps from ~2–5 L per sampling to ~40 L on 30 August. That first big soaking of freshly built soil sets off several disturbance processes at once, none of them weathering: the loose soil is still settling, so it drains more freely than the compacted structure it becomes later; a flush of pre-existing salts and exchangeable cations is washed out; and dormant microbes revive in a burst of mineralisation — the Birch effect. Together these drive the initial alkalinity and EC peak (concentrations climb to ~5,200 µmol/L by September–October, then fall back), and the fingerprint is a salt flush as much as a respiration one: TA, Ca, Mg and Na all rise and fall together, with sodium — which has no weathering-carbonate source — tracking alkalinity most tightly of all (r ≈ 0.90). It is a short-term response to disturbance, not the long-term weathering signal, and it appears in the controls just as strongly as in the amended pots. We treat the first ~90 days as unrepresentative, exactly as in our greenhouse work — not every EC increase is weathering; sometimes it is just salts (greenhouse series, Part 3). Our own companion column experiment shows the same on a longer record: an intense rainfall after a dry spell produced a flushing event that contributed roughly a quarter of the total alkalinity export over an 850-day run (Hammes et al., 2025) — again a disturbance signal, not weathering.

So we tested for a dose effect the right way, comparing each treatment against control at day 90, day 180, day 400 and day 1,011. At no horizon is any dose significantly different from control (Welch’s t, n = 4, all p > 0.28). The early scatter is the largest of all — at day 90, FINE sits +48 % and 200 t/ha +25 % above control — but with p = 0.45 and 0.29, that is the Birch flush and pot-to-pot noise, not a dose signal.

Figure 3: Difference of each treatment vs control over the day 800–1,011 window; none is significant after Holm correction (n = 4 pots per treatment).

The whole-period picture agrees. A log-linear mixed model over all samplings puts FINE at +6 % in concentration and +9 % in export versus control, and 100/200/400 t/ha within a few percent of control — none of it significant.

Figure 4: Mixed-model treatment effects vs control (%), for concentration and three volume models; point = mean effect, bar = 95 % CI. Almost every interval crosses zero.

There is one nominally significant result worth naming precisely so it is not over-read: under the avgBD volume model, 400 t/ha comes out −14 % versus control (p = 0.034) — a negative effect that vanishes under the other two volume models (p = 0.39 and 0.30). More basalt producing less alkalinity is not a mechanism we believe; it is what a single volume model plus pot heterogeneity plus multiple comparisons will occasionally hand you. It is a caution, not a finding.

Why the signal hides

Three reasons, and all matter for anyone planning field MRV.

The baseline is large — and part of it is not weathering at all. An unamended control pot at this site “exports” ~1.6 tCO₂e/ha of bicarbonate over four years. But that gross number is not four years of weathering: it includes the Birch-effect flush, carbonate traces and organic-matter turnover — the disturbance baseline we met in the early peak. The basalt signal we want to credit has to stand out above that moving baseline — and here it does not clear the noise.

Two things follow. For the dose comparison, none of this is a problem: every pot got the same disturbance and the same shared baseline, so these common terms cancel exactly when we net against control — the “no significant dose effect” result is robust to them. For the absolute CDR number, they matter: the real weathering-attributable removal is lower than the gross figures above.

n = 4 is not much power. With four pots per treatment and pot-to-pot scatter of order ±0.1 tCO₂e/ha, only a large dose effect would show up as significant. A modest but real effect — say 10–15 % — would sit inside the error bars, exactly where FINE sits now. Absence of significance here is not proof of absence of weathering; it is proof that this design cannot resolve the difference. That is worth stating plainly.

And some weathering never reaches the drainage. Our metric counts only the inorganic carbon that leaves dissolved in the leachate. But base cations released by weathering can be held on the soil’s exchange sites or built into secondary clay minerals before they ever drain out — and where secondary carbonate precipitates, it even re-releases CO₂. Those retained cations, and the alkalinity that would have charge-balanced them, stay locked in the solid phase, invisible to a leachate sampler. So a real dose effect could be partly hidden in the soil rather than absent — which is exactly why the solid phase, not just the drainage, belongs in serious weathering MRV (article 6).

The FINE trap

FINE is the one treatment that trends highest, and it is tempting to read that as “finer grinding works.” It is not that simple. By XRD the fine batch shares almost none of the coarse basanite’s minerals: it is 44 % K-feldspar and 34 % natrolite (both essentially absent from the coarse rock), with 5 % calcite, and it lacks the leucite, nepheline and plagioclase that define the coarse batch — a different rock, not a finer grind. Its calcite alone can manufacture an early alkalinity surplus. FINE therefore confounds grain size with mineralogy, so its +6 % tells us nothing clean about fineness. We flagged this in article 1 and we hold to it.

What this means for MRV

At a single, well-characterised, well-replicated site, over four years, we could not demonstrate a dose–response in alkalinity export. That is not an argument against enhanced rock weathering — our greenhouse work shows strong, rock-specific weathering across many soils. It is an argument against a specific, common shortcut: treating dose–response as a settled, universal rule you can assume rather than measure. The soil’s own flux can rival the signal; small plots have limited power; and control-netting (comparing against unamended pots on the same soil) is not optional. Any MRV scheme that credits “more rock = more CDR” without measuring against a real control on the real soil is crediting an assumption.

Next in the series: What sensor data told us.

The XXL Lysimeter dataset (2022-2026) series

1. Intro article — why a 1,400-day, daily-instrumented experiment is worth the wait.

2. Dose–response — the initial flush (the Birch effect, not weathering), the long convergence, the n = 4 power problem, and the FINE-is-a-different-rock confound.

3. The soil breathes — how temperature and soil CO₂ drive a seasonal weathering engine visible in the buried sensors.

4. A buried EC sensor as a continuous MRV proxy — the 60 cm probe tracks leachate chemistry remarkably well (and the 30 cm one does not); the long-term, in-situ extension of our EC-as-alkalinity-proxy work.

5. What the soil tells us when it breathes — reading four years of buried CO₂ sensors: the soil as a living system of warmth, water and microbes — and why even a sensor in every pot is the weathering engine, not a carbon meter.

6. Four years of buried sensors — rainfall, ambient CO₂, and the reality that most field sensors die within two to three years.

References

• Hammes et al. (2025), Soil processes govern alkalinity and cation retention in enhanced weathering for carbon dioxide removal, EGUsphere preprint, https://doi.org/10.5194/egusphere-2025-5402 (the Birch-effect / initial-flush framing).

• Greenhouse series, Part 3 — Transient disturbances: initial flush and fertilizer event: https://www.carbon-drawdown.de/blog/2026-1-23-49-transient-disturbances-initial-flush-and-fertilizer-event

Data, code & figures:doi.org/10.5281/zenodo.21216439 (CC-BY-4.0) · github.com/dirkpaessler/carbdown_xxl_lysimeter_2022_2026

Twenty outdoor lysimeter pots, each 0.406 m² of surface (0.70 × 0.58 m), filled in spring 2022 with a single soil (“Fürth 2022”) and planted. Four to six weeks later we mixed crushed rock into the top 15 cm (except for controls). Five treatments, four replicate pots each.

Water percolates through; roughly monthly we pump out and analyse the leachate — the Sickerwasser — for total alkalinity, major ions, pH and electrical conductivity. Alongside the chemistry, each instrumented pot carries buried probes logging daily: soil moisture, temperature, electrical conductivity and pH at 30 and 60 cm, a soil-CO₂ sensor at ~20 cm in every pot, plus on-site rain gauges and ambient-CO₂ sensors.

For scale: the site received roughly 533, 775, 691 and 548 mm of rain in 2022 to 2025 (official DWD station Nürnberg 03668). Our on-site gauges consistently under-catch — a story in itself for article 6.

The feedstock — and a caveat we plant now

The basalt is “Eifelgold” (Basalt-Union, from the Eifel) — geologically a basanite, silica-poor and rich in fast-dissolving feldspathoids (leucite, nepheline). That mineralogy is why we would expect these rocks to weather quickly. The coarse batch used for 100/200/400 t/ha has a median grain size around 39 µm, a BET surface area of 2.3 m²/g, and a theoretical maximum of 478 kg CO₂ per tonne of rock.

One caveat we plant here and return to later: the “FINE” treatment is not a clean grain-size experiment. By X-ray diffraction the fine batch is a different rock — same supplier, same name, but it shares almost none of the coarse basanite’s minerals. Where the coarse batch is dominated by clinopyroxene, plagioclase, leucite and nepheline, the fine batch is 44 % K-feldspar and 34 % of the fast-weathering zeolite natrolite (both essentially absent from the coarse rock), plus 5 % calcite, whose rapid dissolution alone can produce an early alkalinity surplus. So FINE is not a finer grind of the same rock; when it looks like the front-runner later in this series, remember that it confounds grain size with mineralogy. The full XRD, XRF and grain-size tables travel with the dataset.

What we measure, and how we turn it into CDR

The signal is total alkalinity (TA) in the leachate, interpreted as our bicarbonate (HCO₃⁻) signal. Dissolving silicate or carbonate minerals releases base cations that charge-balance bicarbonate, and exporting that bicarbonate in the drainage is a direct, measurable proxy for carbon captured at the source. We convert measured bicarbonate export to CO₂ at 1 mol CO₂ per mol HCO₃⁻ and normalise per unit surface area, so our headline quantity is the cumulative export of inorganic carbon in the leachate, in tCO₂e/ha — our measured CDR. We name it that way deliberately: it counts only the carbon that leaves dissolved as bicarbonate. Other, unmeasured pathways — chiefly the build-up of stable soil organic carbon — could add to the total; conversely, soil processes that lock base cations into secondary clay minerals or exchange sites (or precipitate secondary carbonate, which re-releases CO₂) remove cations from the drainage and are a loss, not a gain. So this inorganic-leachate number is best read as an upper bound on that one pathway, not a full carbon budget.

One qualifier still matters: this is local CDR — the carbon captured at the pot. The bicarbonate still has to travel through soils, groundwater, rivers and finally the ocean, and a fraction is lost or re-emitted along the way, so a discount applies before this becomes certified, durable CDR. And, as noted above, the leachate captures only the inorganic carbon that drains out — an upper bound on that one pathway.

The dose question, first look

This is where a tidy story would show the highly amended 400 t/ha line pulling clear of the pack. It does not.

Figure 1: Cumulative bicarbonate export as CO₂-equivalent (tCO₂e/ha = local CDR) per treatment, volume model avgBD. Solid band: 95 % CI across replicate pots (n ≤ 4); hatched band: last measured CI carried forward for the single pot per treatment that continues past the 2025 refill.

All four replicates of every treatment ran side by side until early 2025 (day 1,011). At that point — the last date with the full n = 4 — cumulative local CDR lands at:

The unamended control (1.58) sits squarely in the pack — above 100 and 400 t/ha, level with 200. FINE noses ahead at 1.69, but its error bar overlaps every other treatment, and (see above) it is a different rock.

There is a strong early alkalinity peak in the first months — but it is not a basalt signal: it is disturbance, not weathering. Several things happen at once when the freshly built, disturbed soil first gets a proper soaking (the heavy thunderstorm of August 2022 that delivered the first real drainage): the loose soil settles and its permeability changes, a flush of pre-existing salts and exchangeable cations is washed out, and the Birch effect adds a burst of microbial mineralisation. It shows up in the controls just as strongly. Once that disturbance flush fades, the treatments simply converge, and by whole-period statistics no dose differs significantly from control. With only four replicate pots per treatment, the experiment does not have the statistical power to resolve the modest differences that remain.

So, in essence, even after almost 1,400 days we do not see any significant “signal” that would tell us: “CDR has happened”. Instead we have proven that we were able to build 20 large pot experiments that behave exactly the same way, even though we put enormous amounts of rock dust on them.

As we have found out in the follow-up greenhouse experiment, we had made a very unfortunate choice of soil and feedstock (this combination behaved similarly in our greenhouse experiment, while other combos worked fine).

But there are still several interesting insights to be gained from this experiment, so we wrote a whole blog series about it:

The XXL Lysimeter dataset (2022-2026) series

1. Intro article — why a 1,400-day, daily-instrumented experiment is worth the wait.

2. Dose–response — the initial flush (the Birch effect, not weathering), the long convergence, the n = 4 power problem, and the FINE-is-a-different-rock confound.

3. The soil breathes — how temperature and soil CO₂ drive a seasonal weathering engine visible in the buried sensors.

4. A buried EC sensor as a continuous MRV proxy — the 60 cm probe tracks leachate chemistry remarkably well (and the 30 cm one does not); the long-term, in-situ extension of our EC-as-alkalinity-proxy work.

5. What the soil tells us when it breathes — reading four years of buried CO₂ sensors: the soil as a living system of warmth, water and microbes — and why even a sensor in every pot is the weathering engine, not a carbon meter.

6. Four years of buried sensors — rainfall, ambient CO₂, and the reality that most field sensors die within two to three years.

All of it is reproducible: the data, code and figures behind every article ship as an open, versioned dataset (our XXL Lysimeter Experiment, Fürth-2022 soil), released alongside this series under CC-BY-4.0 — doi.org/10.5281/zenodo.21216439 (GitHub: dirkpaessler/carbdown_xxl_lysimeter_2022_2026).

References and further reading

• Setting up the Carbdown Greenhouse Experiment 2023 (With Recipe And Shopping-list) https://www.carbon-drawdown.de/blog/2023-3-16-setting-up-the-carbdown-greenhouse-experiment-2023-with-recipe-and-shopping-list

• Hammes et al. (2025), Soil processes govern alkalinity and cation retention in enhanced weathering for carbon dioxide removal, EGUsphere preprint, https://doi.org/10.5194/egusphere-2025-5402 (the Birch-effect / initial-flush framing).

• Our greenhouse ERW series — MRV Proxies for ERW: a guided tour (Part 1 of 9): https://www.carbon-drawdown.de/blog/2026-1-23-19-mrv-proxies-for-ew-a-guided-tour-through-our-data-from-our-two-year-greenhouse-experiment

• Total Alkalinity and Electrical Conductivity as MRV Proxies for Enhanced Weathering (PDF): https://www.carbon-drawdown.de/s/Total-Alkalinity-and-Electrical-Conductivity-as-MRV-Proxies-for-Enhanced-Weathering_-Insights-from-a-jtnx.pdf

• XXL Lysimeter project introduction (2022-02): https://www.carbon-drawdown.de/blog/2022-2-22-introducing-the-carbdown-xxl-lysimeter-project

• Photo logbook — building the XXL lysimeters (2022-06): https://www.carbon-drawdown.de/blog/2022-6-13-photo-logbook-building-the-xxl-lysimeter-weathering-experiment-part-1

• Biomass assessment of weathering in the XXL lysimeters (2022-10): https://www.carbon-drawdown.de/blog/2022-10-9-biomass-assessment-of-weathering-in-the-xxl-lysimeters

• Whitepaper — What We Learned from the World’s Largest Greenhouse Experiment (Jan 2025), Tables 5–7 (feedstock characterization): https://www.carbon-drawdown.de/blog/2025-1-29-what-we-learned-from-the-worlds-largest-greenhouse-experiment

Averaged over three years and four sensor zones, our greenhouse runs at a mean of 22.5 °C with a ~10 °C summer-to-winter swing. The water is the striking part. We aimed for 2,000 mm of irrigation a year; the collection buckets show we actually applied ~3,300 mm, more in summer, to offset the intense evaporation and still leave enough water to sample. Of that, only about 1,100 mm actually drains through the pots; the other two thirds evaporate.

That last number is the important one. The alkalinity (the carbon-removal signal) leaves dissolved in the water that drains through the soil. So the number that matters for "where is this climate real" is not the 3,300 mm we pour in, nor the temperature alone, but the ~1,100 mm of drainage.

It has to be warm and actually drain ~1,100 mm

Match on temperature alone and you get a broad warm band, plus, absurdly, deserts (they have the steadiest temperatures on Earth). So we added the real constraint: a place must also naturally drain about 1,100 mm a year. Crucially, we do not match our inflated 3,300 mm of irrigation, our greenhouse over-evaporates because of warm pots (no cooling by ground) and small pot sizes. We match the drainage, and we estimate each location's natural drainage from climate (rainfall minus evaporation; see Methods).

The result is a much smaller, and much more honest, map. Look at the red stars, these are the closest twins to our greenhouse. The next 8 “almost best” matches are shown with orange markers.

Only a narrow band qualifies on both axes (essentially the humid subtropical coasts and islands of East Asia).

• Okinawa, Japan: our closest twin of all: mean 22.9 °C, a matching ~12 °C seasonal swing, and ~1,200 mm of natural drainage. Almost literally our greenhouse, set outdoors.

• Hong Kong: mean 22.4 °C, ~1,170 mm drainage.

• Southern Taiwan (around Kaohsiung): a touch warmer, ~940 mm drainage.

One further match sits just outside East Asia, the windward side of Hawaii (warm and very wet, ~1,340 mm drainage), though its almost seasonless climate makes it a looser fit than the three above.

Could somewhere be even better?

A fair question: if warmth and water both help, why not the planet's rainiest, hottest places, Quibdó in Colombia (one of the wettest inhabited spots on Earth), Singapore, Monrovia, the Amazon? On paper they are: warm all year, no cold season, often wetter than us, so for uninterrupted weathering they sit even deeper in the favourable zone than our greenhouse does.

And our own back yard proves the point

A few kilometers from the greenhouse we run the XXL lysimeters: a selection of the same soils and rocks, outdoors, rain-fed, for four years. Cool, unremarkable Bavaria (Germany) drains only a few hundred mm a year (and it still removes real, measured CO₂, just slowly). (In fact our free-draining lysimeter drains ~360 mm/yr, about double what a natural Fürth field would recharge, a reminder that even our outdoor twin is on the generous side.) The lesson holds across the whole range: more drainage, faster removal, Okinawa fast, Rio slower, Bavaria slowest. Warmth is not the master dial; water is.

That is worth one honest admission. When we turned everything to "11" in 2023, the dial we trusted most was temperature: the textbook rule that weathering slows roughly ten-fold per 10 °C of cooling. Three experiments later, our own data say the carbon-relevant alkalinity barely moves with temperature between ~1.5 °C and 33 °C; temperature's real influence runs through the water (evaporation, and freezing). And the field literature quietly agrees: when Kantzas et al. (2022) ranked the UK for ERW, the deciding variable was the site water balance, and geochemists find natural weathering in the warm tropics "not temperature limited" (Edwards et al. 2017; Moore et al. 2024). The best-region maps point the same way, warm and wet (Strefler et al. 2018; Beerling et al. 2020; Baek et al. 2023).

For enhanced rock weathering: follow the water!

Methods. Greenhouse climate: daily air-temperature logs from four in-house sensor zones (2023–2026) give the monthly profile (mean 22.5 °C, amplitude ~10 °C). Irrigation: litres caught in the greenhouse's control collection buckets (empty buckets that only collect water), per table per ~4–6-week collection, converted to depth via the pot opening area (POT_AREA_M2 = 0.0415475 m², so 1 L ≈ 24.1 mm) and annualised per interval; time-weighted mean ≈ 3,300 mm/yr. Drainage (leachate): per-pot leachate volumes from the routine sampling workbooks, same conversion; ≈ 1,100 mm/yr. Global climate match: WorldClim 2.1 monthly normals at 10 arc-minutes (~18 km). Temperature similarity = the RMSE between our 12 monthly means and each land pixel's, minimised over a calendar shift so Southern-Hemisphere sites are compared season-for-season. Natural drainage everywhere: we do not match our (inflated) irrigation but each site's natural drainage, estimated with the standard Budyko water balance: potential evapotranspiration from the Hargreaves equation (WorldClim tmin/tmax/tavg + FAO-56 extraterrestrial radiation), actual evapotranspiration via the Budyko–Fu relation (ω = 2.6), drainage = precipitation − actual evapotranspiration. This is a spatially varying estimate, not a single global rainfall-to-drainage factor; it validates against known values (e.g. our Fürth site ≈ 173 mm/yr, matching the German national groundwater-recharge range of ~135–200 mm/yr). Ready-made global datasets (WaterGAP recharge, GRUN runoff, TerraClimate, GLEAM) give equivalent pictures if higher authority is wanted. Twin criterion: temperature RMSE < 2.6 °C and natural drainage between 800 and 1,600 mm/yr (bracketing our measured ~1,100). Data: the underlying experimental data are not public yet; WorldClim is public. All figures © Carbon Drawdown Initiative.

1. Claude Chat: you think with it

This is the one everybody knows as started by ChatGPT. You open a chat in your browser, you paste something in, you go back and forth. You are in the loop on every single turn, no progress without you.

It is brilliant for thinking through a problem, a quick translation, drafting and then tearing apart a piece of text, or a quick one-off analysis you watch happen in front of you. I personally use this AI-option less and less while moving to the more context-stable other options (“new school”).

But here is the part scientists need to internalise: when you upload a data file into a chat, you send it to the cloud, the model reads the actual values and interprets them. It is looking at your numbers and telling you what it thinks they mean. That process is non-deterministic: There is a hard-coded randomizing element built into LLMs and this can’t be worked around. Drift happens because AI labs actively retrain, patch, and update the models behind the scenes. Ask the same question tomorrow and you may get a different reading. For brainstorming, fine. For "what is the mean CO₂ flux in treatment group B," that is not how I want my science done. The model is a marvellous interpreter and a terrible calculator-of-record.

Memory here is short: it lives and dies inside the single chat.

2. Claude Cowork: you delegate a project to it

Cowork is where you stop driving every turn. You hand over a goal - "synthesise these 40 papers," "draft the methods section," "build me a clean figure set," "untangle this messy dataset" - and it makes a plan, works across your folder and connected apps (PDFs, spreadsheets, notes, drafts, all at once), and comes back with a finished knowledge product. You review the plan, you check the deliverable, and you are mostly hands-off in the middle.

You give it your data by compiling files (CSVs, XLS, PDFs, etc.) that describe your project into a folder on your disk. The “what happens to the data?” question gets interesting here, because Cowork can work either way. It can read your files directly (non-deterministic, same caveat as Chat), or it can write code to process them properly (locally on your computer). The crucial bit: if the numbers matter, you have to explicitly ask it to compute, not eyeball. I now say this out loud in almost every Cowork session - "do this in code, show me the script." It is a habit worth building.

Memory is mid-term: it builds up inside a "project" over time, which is exactly what you want for a multi-week paper.

3. Claude Code: you build with it, and the code persists

This is the one I trust with the real measurements, and the reason is the single most important box on the whole cheat sheet:

Again you give it a folder with the necessary data on your hard drive. Your data stays local and is processed locally. The code does the analysis. The model sees your code and your result summaries at best, but it never reads the raw values. Your data isn’t uploaded to the cloud.

Think about what that buys you. The analysis is deterministic and re-runnable. Run it today, run it in a year, get the identical answer. When a reviewer asks "how did you get this p-value," the answer is a script, not a vibe.

People hear "deterministic" and assume the non-determinism magically disappeared. It did not - it moved. The model still writes the analysis stochastically; it might phrase the same statistical test three different ways on three different days. But that stochastic step is now sitting in front of you as readable code that you review and freeze. The randomness moved out of "interpreting my data" and into "drafting my method" - where it belongs, because that is the part a scientist is supposed to check anyway.

You interact at a high level - you discuss how to solve the problem with a very capable coder, and then it does research if necessary, writes and runs the code. What you get back is committed code that runs again tomorrow: pipelines, notebooks, a working model, a deployed dashboard. Memory is genuinely long-term, because your codebase, models and files are maintained over months.

Don’t get scared by the work “code”: The great thing is that you do not need to know how to program to work successfully with Claude Code (but it sure helps a lot).

4. OpenClaw agent: it runs without you

The fourth one is a different animal entirely, and regular readers know I have written about my OpenClaw adventures before. You configure it once, and then triggers - a cron schedule, a webhook, a Telegram message - run it without you. It pulls from live and scheduled sources: sensor streams, database tables, APIs, incoming files. Out comes a steady drip of artifacts on schedule: refreshed plots, summary emails, updated database records.

For a scientist this is the "watch my experiment for me" tool. Every morning at 06:00, scan the sensor data, flag anomalies, plot the trend, email it to the team.

The trade-off is real and I will not sugar-coat it, because I have the scars: an agent flows your data through a mixed LLM-and-code pipeline, and its behaviour drifts over time - what we politely call "learning." Its long-term memory is long but lossy. It is a workhorse, but a high-maintenance one that needs herding. (My evangelist agent once posted five replies to the same post instead of one reply to five posts. Charming. Not what I asked for.) Use it for the recurring, the unattended, the "I don't want to think about this every day" - but not for the analysis you are going to publish. BTW, this option creates the by far highest cost for tokens, our shared OpenClaw agent burns through ca. 1000 Euro in tokens each month, this is down from 8000 Euro/month after I talked it into running a lot of recurring tasks in Python code instead of using LLMs prompts.

The one row that matters most

If you take nothing else from the cheat sheet, take the "How your data is handled" row. It maps cleanly onto a spectrum every scientist already understands intuitively:

• Chat: The model reads it directly

• Cowork: model or code, your choice

• Code: code reads it, model reads only summaries

• OpenClaw: pipeline code reads it, model reads only slices.

From top to bottom, you get more automation. Chat is the most fluid and the least auditable. Code is the most rigid and the most trustworthy. There is no "best" - there is only "right for this task."

This is still early: but the map already helps

None of these tools is finished. They overlap at the edges (Chat can now run code too, which smudges the left column), they change every few weeks, and they will all be different by the time you read this. But the underlying axes - who drives the loop, where the work lives, who touches your data, how long it remembers - those are stable, and they are what the cheat sheet is really about.

Four years in

Back in 2021, I expected (rather naively) meaningful measurement results within the first year. That didn't happen. What followed was a long process of trying to understand what actually occurs in rock-dust-applied soil and how difficult it is to measure.

Standing in front of the room giving the welcome address, I felt less frustrated about that than I expected. The naivety of year one didn't slow us down – it got us started. The open questions feel like evidence of progress now, not failure.

Three observations

1. The people drawn to this work are good people. Everyone in the room was collaborative, constructive, and motivated by the problem rather than personal credit. Two days with that group was genuinely energizing.

2. New data produced new questions. From the greenhouse results, we collectively developed better explanations for the mechanisms at play. And in doing so, it felt like each answer surfaced three new follow-up questions that will need further investigation. I now accept this is just what science looks like up close.

3. The field is getting better at predicting where ERW works. Which rock types, which soil conditions, which geographies. The discrimination is improving. That matters enormously for where new projects should – and shouldn't – be launched.

What comes next

We'll reconvene once the second greenhouse experiment yields new samples. Another year or two away, most likely.

Session recordings from the Symposium are now available on Youtube and on our event website (which also hosts the slides as PDFs).

Thank you to everyone who came to Erlangen and those who joined online. You made two days of science genuinely enjoyable.

PS: Here is the recording of the intro session

Herzlich willkommen, alle zusammen. ‍

Ich bin Dirk Paessler, Gründer und CEO der Carbon Drawdown Initiative. Ich bin ein lebenslanger Software-Unternehmer, der zum Klima-Aktivisten geworden ist.

Ich möchte Ihnen heute nichts verkaufen. Was ich möchte, ist, Sie mit meiner Begeisterung für Klimaprojekte und Klimafirmen anzustecken — und Sie zu meinen Komplizen zu machen, damit wir gemeinsam unseren Themen die gesellschaftliche Aufmerksamkeit verschaffen, die sie verdienen.

Wir stehen vor enormen Herausforderungen. Aber diese Herausforderungen sind zugleich enorme Chancen.

‍Für mich fühlt sich das oft an wie 1995, als ich schon voll vom Siegeszug des Internets überzeugt und begeistert war und viele Menschen um mich herum sagten: „Ach Dirk, Du bist ja schnell von was begeistert. Aber das wird doch alles nicht so groß." Wir alle wissen, was dann kam.

Beim Klima, und bei CDR im Besonderen, sehe ich eine starke Welle kommen — sie ist schlicht unvermeidbar. Hier liegen großartige Chancen für den Einzelnen: akademisch oder um eine Karriere aufzubauen. Und für die Unternehmertypen, die sich an diese Themen heranwagen.

Auf nationaler Ebene sind Klima und insbesondere CDR eine große Chance für Deutschland — für Arbeitsplätze, für Wirtschaftswachstum und für den Schutz des Klimas.

‍Und wir sind tatsächlich ganz gut dabei. Wir führen eine eigene Datenbank über die CDR-Firmen weltweit. Gerade einmal etwa 900 Firmen — nur „Pure-Play CDR", wie wir das nennen, also ohne Airbus mitzuzählen, wenn die intern mit ein paar Leuten an DAC arbeiten — mit über 9.000 Mitarbeitern arbeiten weltweit an diesem Thema. 10 Leute pro Firma — das heißt, wir sind so unglaublich früh auf dieser Reise!

‍Deutschland liegt bei der Zahl der Firmen und bei der Zahl der wissenschaftlichen Veröffentlichungen bereits auf Platz 3, direkt nach UK und den USA. Das ist eine gute Ausgangssituation. Die sollten wir halten und ausbauen.

‍Lassen Sie mich ein wenig erzählen, wie ich hierher gekommen bin.

‍Ich habe mehr als ein Dutzend Firmen gegründet oder mitgegründet. Die erfolgreichste war die Paessler AG — 500 Mitarbeiter, Internet-Infrastruktur-Software. Vor knapp zehn Jahren bin ich als CEO ausgestiegen und habe mir eine Frage gestellt: Wo kann ich nützlich sein?

‍Diese Frage führte mich zum Klima und dann in die Nische CDR. Die Carbon Drawdown Initiative ist ein philanthropisches Projekt unserer Familienstiftung. Wir haben sie 2020 gegründet, mit dem Fokus auf nichts als CDR.

‍Unsere Mission ist es, negative Emissionen schneller zu machen. Und dafür drücken wir auf drei Hebel.

Der erste Hebel ist Impact-Kapital. Wir geben Geld, um neue Ideen auszuprobieren — und ziehen so die wichtigen ersten Schritte ein Stück aus der Zukunft in die Gegenwart. Wir geben super-frühes Kapital an CDR-Firmen: über 30 Investments in 14 Ländern auf der ganzen Welt. Das macht uns zum größten privaten CDR-Fonds der Welt (zumindest nach Anzahl). Alle wichtigen Verfahren sind vertreten — darunter Climeworks, InPlanet, Octavia, Syncraft und so weiter — plus die Wegbereiter im Ökosystem, wie Carbonfuture.

‍Der zweite Hebel ist die politische Arbeit. CDR ist ein bisschen wie die Müllabfuhr: Ohne Regulierung wird es keine negativen Emissionen im großen Maßstab geben. Staatliche Beteiligung ist unerlässlich. Deshalb sind wir Mitgründer von NEP, DVNE und CRIA. Und ich bin seit 2021 stellvertretender Vorsitzender (Vice Chairman) von NEP.

‍Der dritte Hebel ist die wissenschaftliche Arbeit. Wir wollten verstehen, wie beschleunigte Gesteinsverwitterung eigentlich funktioniert — was passiert da unten im Boden wirklich? Und das haben wir gelernt: Wir haben versucht, der akademischen Welt Geld zu geben. Aber das war wahnsinnig langsam (also wirklich langsam!). Finanzierte Wissenschaft erwies sich — für mich persönlich wie auch fürs Klima — als zu langsam. Also haben wir uns stattdessen selbst tief in ERW eingearbeitet. Wir haben uns buchstäblich die Hände schmutzig gemacht. Weil wir im Startup-Tempo denken.

Und lassen Sie mich Ihnen einfach die Liste der Universitäten vorlesen, die hier mit uns an diesem Gewächshaus-Experiment zusammenarbeiten, weil wir echt stolz darauf sind: Universität Erlangen-Nürnberg, ETH Zürich, Universität Mainz, Universität Hamburg, Universität Bonn, GFZ, das Alfred-Wegener-Institut, Universität Wageningen, Universität Antwerpen, Universität Kopenhagen, Newcastle University, James Cook University in Australien, das UFZ und das ELGO-Dimitra-Institut in Larissa, Griechenland (in keiner bestimmten Reihenfolge!).

Unser Wissenschafts-Fokus ist die beschleunigte Gesteinsverwitterung (Enhanced Rock Weathering, ERW). Wir haben sieben ERW-Investments in Europa, Afrika sowie Nord- und Südamerika — darunter InPlanet, Flux, Lithos, Silicate, Eion, Silica und Tropicarbon. Plus das Ökosystem drumherum: Everest in Berlin baut Feldsensoren, die die Gesamtalkalinität im Sickerwasser messen. AEROC in Köln nutzt Satelliten- und Drohnendaten, um die Kosten für Messung, Berichterstattung und Verifizierung (MRV) zu senken — was uns erlaubt, das Problem der Feld-Heterogenität anzugehen. Syntopa versucht, ERW mit Bakterien zu beschleunigen. Und die Rock Flour Company liefert das feinste Material, das es gibt: Gletscherstaub.

Wie sind wir also zum Gewächshaus gekommen?

• 2021 haben wir unsere erste Tonne Basalt — buchstäblich von Hand — auf drei Kreisen auf meinem eigenen Grundstück in Fürth verteilt. Dann haben wir das ganze Jahr lang wie verrückt gemessen und überwacht. Und wir haben nichts gemessen. Kein Unterschied zwischen Kontrolle und Behandlung. Null. Nada. Nichts.

• 2022 haben wir die nächste Generation des Experiments gebaut: zwanzig XXL-Lysimeter aus 300-Liter-Wassertonnen, im Boden vergraben. Wir haben dem Fürther Boden noch mehr Basalt hinzugefügt und wieder gemessen. Und wieder fanden wir kein Signal für eine CO₂-Entnahme. Nichts.

• 2023 sind wir noch weiter aus der Natur heraus in eine kontrollierte Umgebung gegangen und haben das weltgrößte Gewächshaus-Experiment für ERW gebaut.

Und das haben wir herausgefunden: Der Boden auf meinem eigenen Grundstück, der „Fürther Boden", macht einfach kein CDR — zumindest nicht, wenn man es über die Gesamtalkalinität im Sickerwasser misst. Aber im Gewächshaus haben wir auch jede Menge Boden-Gesteins-Kombinationen gefunden, bei denen ERW richtig gut funktioniert.

Im Kern haben uns zwei spektakuläre Fehlschläge dazu gebracht, immer genauer hinzusehen. Das Ergebnis von vier Jahren Gewächshaus-Arbeit war unser CDI-Symposium vor drei Tagen, bei dem wir uns mit 50 Wissenschaftlern von den eben genannten Universitäten zusammengesetzt und diskutiert haben, was uns unsere Gewächshaus-Daten und -Proben sagen.

‍

Lassen Sie mich also damit schließen, wie wir die Dinge sehen. ‍

Die beschleunigte Gesteinsverwitterung ist in bestimmten Umgebungen, zum Beispiel auf tropischen Böden, praktisch so gut wie gelöst. Wir haben kürzlich die ersten Zertifikate weltweit gesehen, die ausgestellt wurden. Firmen wie InPlanet, MATI, Alt Carbon und Lithos haben ihre ersten Zertifikate für CO₂-Entnahme verifiziert bekommen.

Auf komplexeren Böden wie in Mitteleuropa, zum Beispiel dem Fürther Boden, ist ERW noch nicht vollständig verstanden. Drei Gründe: Erstens, welcher Stein passt auf welchen Boden? Zweitens, wie verrechnen wir CDR, wenn manche Kationen eine Zeit lang im Boden geparkt bleiben? Und drittens macht moderne Intensiv-Landwirtschaft das Messen nicht gerade einfacher.

Wir vermuten außerdem, dass der Zusatznutzen von ERW für Ernte-Qualität und Nahrungsmittel-Produktion wichtiger ist, als viele denken.

ERW braucht — wie alles CDR — den Support durch die Politik. CDR ist eine Chance für Wachstum und Wohlstand für ein Ingenieursland wie Deutschland.

Und als wohlhabendes Land und als einer der größten kumulativen Emittenten ist es unsere Pflicht, voranzugehen. ‍

Hier also meine eine Bitte. Bitte tragen Sie die Begeisterung für CDR und ERW, die Sie hier bei uns erleben, weiter — ins System hinein — damit wir mehr Unterstützung bekommen, und zwar bald.

Danke an den DVNE für die Organisation dieser Tour. Und… danke Ihnen allen, dass Sie hier sind und dass Ihnen negative Emissionen am Herzen liegen.

PS: Hier ist ein Zeitraffer-Video der eher wissenschaftlichen Vorträge im Gewächshaus, die der Begrüßung folgten.

Hier ein Video vom DVNE über die 2 tägige Tour:

I'm Dirk Paessler, founder and CEO of Carbon Drawdown Initiative. I'm a lifelong software entrepreneur turned climate activist.

I'm not here to sell you anything today. What I want is to infect you with my enthusiasm for climate projects and climate companies — and make you my accomplices, so that together we can get our topics the public attention they deserve.

-----

We're facing enormous challenges. But those challenges are also enormous opportunities.

For me, this often feels like 1995, when I was convinced and excited about the internet, and a lot of people around me said: "Ah, Dirk, you do get excited quickly. But this won't ever be that big." We all know what came next.

In climate, and in CDR specifically, I see a strong wave coming — it's simply unavoidable. There are great opportunities here for the individual: academically, or to build a career. And for the entrepreneurial types who dare to step into these topics.

On a national level, climate and especially CDR are a big opportunity for Germany — for jobs, for economic growth, and for protecting the climate.

And we're actually off to a good start. We keep our own database of CDR companies worldwide. Just about 900 companies — “pure-play CDR” only, as we call them, not counting Airbus when they internally work on DAC with a few people — with more than 9,000 employees are working on this globally. 10 people per company means we are so super early in this journey!

Germany already ranks third, both in the number of companies and the number of publications, right after the UK and the USA. That's a good position. We should hold it and build on it.

-----

Let me tell you a bit about how I got here.

I've founded or co-founded more than a dozen companies. The most successful was Paessler AG — 500 employees, internet infrastructure software. Almost ten years ago I stepped out of the CEO role, and I asked myself one question: where can I be useful?

That question led me to climate, and then to the niche of CDR. Carbon Drawdown Initiative is a philanthropic project of our family foundation. We founded it in 2020, with a focus on nothing but CDR.

Our mission is to make negative emissions happen faster. And to do that, we push on three levers.

The first lever is impact capital. We give money to try out new ideas — pulling the important first steps a little bit out of the future and into the present. We provide super-early capital to CDR companies: over 30 investments in 14 countries around the world. That makes us the largest private CDR fund in the world (at least by count). All the important pathways are represented, among them darunter Climeworks, InPlanet, Octavia, Syncraft, plus the ecosystem enablers, like Carbonfuture.

The second lever is policy work. CDR is a bit like waste collection: without regulation, there will not be negative emissions at scale. Government involvement is indispensable. So we are co-founders of NEP, DVNE, and CRIA. And I've served as Vice Chairman of NEP since 2021.

The third lever is scientific work. We wanted to understand how enhanced rock weathering actually works — what is really happening down there in the soil? Here's what we learned: we tried giving money to the academic world. But it was incredibly slow (like, really slow!). Funded science turned out — for me personally as well as for the climate — too slow. So instead we dug deep into ERW ourselves. We literally got our hands dirty. Because we think in startup speed.

And let me just read you the list of universities working with us here on this greenhouse experiment, because we're genuinely proud of it: University of Erlangen-Nürnberg, ETH Zürich, University of Mainz, University of Hamburg, University of Bonn, GFZ, the Alfred Wegener Institute, Wageningen University, University of Antwerp, University of Copenhagen, Newcastle University, James Cook University in Australia, the UFZ, and the Larissa ELGO Dimitra institute in Greece (in no particular order!).

Our science focus is enhanced rock weathering. We have seven ERW investments across Europe, Africa, and North and South America — among them InPlanet, Flux, Lithos, Silicate, Eion, Silica, and Tropicarbon. Plus the ecosystem around them: Everest in Berlin builds field sensors that measure total alkalinity in the leachate. AEROC in Cologne uses satellite and drone data to cut monitoring, reporting and verification (MRV) costs — which lets us tackle the problem of field heterogeneity. Syntopa is trying to speed up ERW with bacteria. And the Rock Flour Company delivers the finest possible material there is: glacial dust.

So how did we end up with a greenhouse?

• In 2021 we spread our first tonne of basalt — literally by hand — across three circles on my own property in Fürth. Then we measured and monitored like crazy all year long. And we measured nothing. No difference between control and treatment. Zero. Nada. Nothing.

• In 2022 we built the next generation of the experiment: twenty XXL lysimeters made from 300-litre water barrels, buried in the ground. We added even more basalt to the Fürth soil and measured again. And again we found no carbon removal signal. Nothing.

• In 2023 we went even further out of nature and into a controlled environment, and we built the world's largest greenhouse experiment for ERW.

And here's what we found: the soil on my own property, called “Fürth soil”, simply doesn't do CDR — at least not when you measure it via total alkalinity in the leachate. But in the greenhouse we also found plenty of soil-and-rock combinations where ERW works really well.

In essence, two spectacular failures led us to look closer and closer. The result of four years of greenhouse work was our CDI Symposium three days ago, where we sat down with 50 scientists from the universities I just named and discussed what our greenhouse data and samples are telling us.

-----

So let me close with how we see things.

Enhanced rock weathering in certain settings, e.g. on tropical soils, is, for practical purposes, pretty much solved. We recently saw the first certificates being issued around the world. Companies like InPlanet, MATI, Alt Carbon, and Lithos have had their first carbon removal certificates verified.

On more complex soils like in Central Europe, for example the Fürth soil, ERW is not yet fully understood. Three reasons: first, which rock fits which soil? How do we account CDR when some cations stay parked in the soil for some time. And third, modern intensive agriculture doesn't exactly make measuring easier.

We also suspect the co-benefits of ERW for crop quality and food production are more important than many people think.

ERW — like all of CDR — needs the support of policy. CDR is an opportunity for growth and prosperity for an engineering country like Germany.

And as a wealthy country, and as one of the biggest cumulative emitters, it is our duty to lead the way.

So here's my one ask. Please carry the enthusiasm for CDR and ERW that you experience here with us further — into the system — so that we get more support, and soon.

Thanks to DVNE for organizing this tour. And…. Thank you all for being here and for caring for negative emissions.
-----

PS: Here is a time lapse video of the more scientific talks in the greenhouse the followed the welcome message

Here is a video from DVNE about the 2 day tour:

I remember the first time I heard about climate change at school. It has influenced me ever since. Learning that it is human-induced and unavoidable was both frustrating and worrying. However, I also remember that at first, I could not believe that climate change really existed. The reason for this was that I would have expected to have learnt about it earlier. If the world was about to 'die', I would have expected it to be the number one topic because people would be doing everything to prevent it. Nevertheless, I had to recognise that most people seem to accept the situation and continue living their lives as usual. At least I have started doing my very best to reduce my personal footprint. But why aren't most people and politicians joining in? We are running out of time, and I began to lose hope that we can reduce climate change.

But then, a few years ago, I watched the documentary “Kiss the Ground”, directed by Joshua Tickell and Rebecca Harrell Tickell. It was the first time I had ever learnt about carbon capture and storage via bio sequestration. The fact that there is a way to reduce the concentration of CO₂ in the atmosphere gave me hope. It was a hope I had not had before. I realised that we could not only reduce global warming, but even reverse it by offsetting past emissions and those that are hard to abate, e.g. such as those from water treatment plants, shipping or steel industry. The idea of negative emissions fascinated me so much that it manifested my dream of studying sustainable agriculture.

When I talked to my former boss about my plan and intentions, I learned about the Carbon Drawdown Initiative, which aims to achieve negative emissions through enhanced rock weathering. He offered to put me in touch with his friend who works there. Now I am here at the CDI and wrapping up my two-month internship. Through this experience, I have realised that enhanced rock weathering is more complex than the underlying geochemical processes alone.

In general, ERW offers huge potential for negative emissions. Natural silicate weathering is considered a very stable geochemical process for long-term carbon sequestration in oceans and soils, and ERW aims to accelerate this process. But not just that, the application of rock dust could bring positive side-effects for agriculture like reduced uses of fertilizer, increased yields, higher crop quality and resilience of the plants.

What I really like about the idea of using rock weathering as a natural process is that we already have all the tools we need to put it into practice and that you can use agricultural areas for dual purposes. You can in principle rely on natural processes with a lower risk of facing new unknown consequences, which sometimes happens with new technologies. However, getting deeper into the CDR topic reveals that it is not as easy as expected to understand all processes. There are many different interactions between soil, feedstock, surroundings, and livestock for ERW, which make its direct implementation more difficult.

So far, it is not clear which feedstock each soil responds best to in terms of ERW for CDR. The current state of research in this field is that the ERW community is still learning about the optimal conditions and underlying principles for maximising carbon sequestration and developing guidelines. In addition, some rock dusts have the potential to improve soil quality and plant growth, for example by raising the pH level and providing mineral nutrients. Here it is important to consider the maximum amount of rock you add to the soil for enhancing the process of ERW while not doing harm to the soil system. By adding large amounts of rock dust to soils you want to make sure not to change the soil too drastically in a negative way and overshoot dosage limits regarding different nutritions and heavy metals. For this reason, care must be taken to ensure that any deployment complies with regional guidelines, to guarantee that it does not result in any danger or long-term environmental damage.

Another important key challenge for ERW is to make it more attractive and affordable by improving MRV strategies. Accessing finely-ground rocks, spreading it on fields and measuring the CDR by ERW requires costly infrastructure and labour. In particular, the irregularities in the soil make it still difficult to identify an universal parameter that can be used to calculate the amount of stored carbon in an easy and cheap way. For scaling up the sector, affordable CO₂ certificates are urgently needed but without funding and policy guidelines this will be challenging.

However, we must first solve the aforementioned issues before we can realise all its benefits. Most of the environmental concerns can be managed, such as high concentrations of heavy metals, by ensuring compliance with threshold limits. The largest real uncertainty is the associated costs of MRV, which may not allow ERW to be a commercially viable climate solution for helping to achieve net-zero emissions. This is why more research in this field is needed! We need scientists to work out how to measure ERW cheaply, establish when it is most effective and prove how it can be made safe for the environment. Then, we need people — especially farmers — to apply rock dust to their land. Throughout this process, we need policies that support them.

I really admire the CDIs' patience and endurance. They tirelessly explore new ideas to solve the puzzle! Thank you for all your efforts! I am sure the breakthrough will come. In the meantime, I am going to study agriculture in order to learn more about the miracles of the soil and its interactions within the ecosystem. My internship showed me that there is still much to discover, and confirmed that this is the subject I want to study.

Dear reader, please do your bit to fight global warming!

We thank Tina for her valuable help in spring 2026 and for sharing her thoughts here!