deep-dive conversations with the people who know.

Why central beats a boiler in every house

Individual household biomass boilers are the mistake he keeps seeing. Each one is a permanent chore for its household (fuel deliveries, ash management, servicing) where a single community-scale plant concentrates that work in one place, run by whoever the community trains for it. The first move, though, is to need less: build to passive standard so the heat demand is small before you size any boiler at all.

The feedstock, and its hard limits

The fuel is forestry cutting waste (branches, crowns, bark off-cuts) which runs to 20–30% of everything a forest produces (Hungary alone has a theoretical 10–12 million m³ a year, though only about 6 million is currently harvested, as the timber industry has contracted). But it is heavy and low-value, so the economics die with distance: about 50 km is the practical radius, and past 100 km no project even tries. That locality is also its strength. When pellet prices spiked roughly tenfold in the 2021–22 crisis, because pellets trade across Europe, local wood chips barely moved. Wood chips are not traded over long distances.

How the machine actually runs

A biomass boiler is thermally slow: hours to reach temperature and hours to leave it. That makes it a baseload machine, with peaks handled by a supplementary source. Multi-source heat is now near-mandatory in Western European district heating. The buffer tank stores only an hour or two; the real energy store is the fuel stockpile itself, the woodpile as "eco-accumulator." And combined heat-and-power is dead in Europe: wind and solar now make electricity more cheaply, so biomass is heat-only.

The proof: Kaposvár

The Hungarian city of Kaposvár runs a 42 MW district-heating system with a 15 MW biomass boiler covering 80–85% of annual heat. When gas prices spiked in early 2022, its five-billion-forint biomass plant effectively repaid itself in about two months, because the gas it displaced would have cost six to eight billion. On the New Zealand fit: NZ timber grows 8–10 mm rings a year versus about 2 mm in Europe, and burns more like European softwood. Polytechnik adapted its boilers after two years of work.

His real warning: it isn't about heat

Asked what would actually break a self-sufficient community, he looked straight past energy. Heat, he thinks, is largely solved; food is the binding constraint. Without fossil fuel there are no tractors, and while growing grain is realistic, the chain from grain to fodder to meat is the one he doubts can be closed, not least because fuel crops and food crops compete for the same land.

The technology mix, in detail

Solar PV is the most scalable technology there is, one kilowatt to a gigawatt, and agrivoltaic tracker systems let farm machinery pass between the rows, cutting effective land use below 5%. Wind has largely answered its old environmental objections (bird and bat detection now auto-stops turbines) but it is far more complex: higher minimum scale, bird-migration studies of a year or more, turbine types fixed four years ahead. Storage is his strongest conviction: at under 1.5 cycles a day, lithium-iron batteries last fifteen years or more, and he is watching flow batteries (vanadium, iron-sodium) that could remove degradation entirely. Hydro he rules out on environmental impact, not cost.

Sizing a system for 20–100 families

The most directly usable part of the conversation, and it runs against intuition:

• Household-scale storage is not recommended, 2–3× costlier per unit than industrial scale, and hundreds of separate battery-management systems to coordinate.
• One central store instead, a single 1–2 MW off-the-shelf unit: transparent, one point of management, far cheaper.
• Household PV is fine, but it should feed the central store rather than a battery in every garage.
• A central 1–3 MW production unit gives the community its baseline stability.
• Keep the grid connection as backup. His analogy is road infrastructure: you keep it because some needs (his example: reaching help for an emergency birth) cannot be met locally. Isolated networks are the most vulnerable.

The Dubai model, and the dark-calm problem

Dubai reached 1 GW of baseload by oversizing panels eightfold and storing the daytime surplus to run the city overnight; IRENA's analysis showed it holds even in poor sun, a Stockholm case landing near €130–140/MWh. The counter-risk is Dunkelflaute (a windless, sunless spell) answered by massive storage, flexible gas, or hydrogen. All feasible; now an economic question, not a technical one.

Why renewables win, and what comes next

On the merit order, renewables already beat fossils economically. His own stance, put bluntly, is that he backs solar not out of environmentalism but because that is where the returns are. He sees AI-based grid management as the coming revolution (the Spanish blackout, he argues, was manual-regulation failure, not renewables) and drones as the bigger recent leap, for crop-disease detection and thermal PV-fault inspection.

The Wairoa pilot

A flood-resilience deployment on the East Coast: when the bridges go under, the council's three sensors keep reporting to a cloud nobody in the valley can reach. Mariko put in about ten lower-cost sensors, with roughly forty more imminent, at about a tenth of the council's price, dense coverage the council could never afford. They talk over a LoRa radio mesh (5–10 km line of sight), solar-and-battery powered with about two weeks of autonomy, feeding a public dashboard plus tailored views for emergency controllers. Local residents install and maintain the kit. The local welder makes the bridge brackets and gets paid from the project budget.

Reality 2: a stack organised by proximity

Underneath is Roy's own platform, built around proximity, location and context rather than the classic hardware-to-application stack. It meshes Bluetooth, Wi-Fi and LoRa (and can bolt on satellite or GSM), runs edge processing on anything from an ESP32 up to small Linux machines with onboard AI, and auto-configures: "turn it on and it will just form a network by itself." Hardware runs to tens or hundreds of dollars a node, "not thousands."

Data sovereignty via trust groups

Ownership is the question he says the project has actually solved: the river owns its data, administered by kaitiaki, the people who live there. Each device belongs to an encrypted trust group, so information stays local by default and is shared only on the community's own terms. This is the lever that makes the economics possible.

The novel idea: a community data marketplace

His strongest non-obvious idea is a marketplace for community sensor data: communities set the rules, the marketplace serves AI-generated insights to buyers (insurers, researchers, planners), and revenue percolates back to fund hardware replacement and a small local economic loop. The intent is a generational solution rather than a project that dies when the grant ends. Not yet operational. This is the trajectory.

The hard part is social

He returned to this twice: designing the radios and software is "the easy bit"; the real challenge is multi-stakeholder coordination between councils, iwi, individuals and businesses who often don't trust each other. Mariko employs someone locally just to manage those relationships, and reuses existing structures (church, marae) as durable backup nodes that already carry community trust.

Where satellites genuinely help, and where they don't

They earn their place spotting landslides and ground movement, tracking how rivers and streams change before and after storms, watching crop and soil health, and monitoring biodiversity and waterway pollution. It is possible to use satellite data for long range climate modelling and early warning, however this is usually a highly technical endeavour. Local communities might find their needs served better by existing national services.

The translator problem

A community on its own won't have the expertise, the tools, or the bandwidth to pull raw satellite data and turn it into something useful. Closing that gap might need a third party (a partner organisation, a productised tool, or an embedded specialist) that takes the raw signal and converts it into community-relevant information. Without that intersection layer, it is difficult to extract maximum value from Earth observation data.

AI: a cheaper start, the same ceiling

New tools can build an analysis on demand from a plain-English prompt (vibe coding), which makes getting started dramatically cheaper. But asking the right question, choosing the right data, and knowing when an answer is wrong still requires deep specialisation. Used without that grounding, AI increases the potential for mistakes for communities with limited resources.

Resilience is relational, not a property of your systems

The spade line is the whole argument. You can grow all your own food and still be stopped by a broken tool, an illness, a skill you don't have. Real resilience comes from relationships with people who have the skills and resources you don't. He takes it somewhere uncomfortable: in a relatively low-income area, visible self-sufficiency "just makes us a target", the security that matters is being worth more to your neighbours intact than looted. And Earthsong, he notes, was never conceived as a closed community; the goal was to live below the average New Zealand resource footprint and share what could be shared.

The energy shift, live

The community's solar panels sit on the carport roofs, while the houses themselves still run solar hot water; there are no batteries yet, so the surplus is exported. And the export is a bad deal: "we sell it back to the grid, we get very little for it, and then when we want to buy it back again, we have to pay the same price as everybody else." That is what has the energy review group moving toward battery storage and EV car batteries for balancing, and toward more panels by shifting the houses' solar-thermal cylinders to heat pumps, all decided collectively, not house by house.

How the community actually works

The common house is the affective centre: a resident's child called the whole neighbourhood "home," not their own unit. Cohesion was damaged by COVID and has since healed. And the way the community makes decisions has evolved in recent years: now it emphasises seeking out and engaging with diverse views early in the process, minimising unexpected tensions later in the process.

Founding: four years, held together by a written vision

Earthsong took four years from idea to land (bought with a 20% deposit and nine-month settlement), settling in November 1999 with 26 households and building from the end of 2000. Finance and development were challenging, as were group dynamics. What held the group together was a written shared vision every member signed. The size was set deliberately: reading and visits pointed at 15–30 households as the range that works, enough diversity without being so large it fractures into factions. They started by designing 25–26 houses and added a few to make the finances close. The whole thing was built as a published demonstration model, and she wrote a book about its process and systems.

The resale-to-commons leak: her main regret

Their development company built the houses and sold them to members at cost. But a large share of every house price had funded the common house and the shared land, so when the first residents sold at market, the difference became private profit rather than returning to the commons. The obvious fix is a clawback clause in the constitution, and here is the sharp institutional finding: other New Zealand communities that tried to write one were refused bank finance (a bank would not lend against it), which forces them back into ordinary individual ownership.

The four resource systems: demand-reduction, not autonomy

Energy: PV was too expensive at build and would have priced out poorer members, so they pre-ran cabling for future PV and chose cheap solar-thermal hot water instead; PV is now on some carports. Water: all roof water is collected and reticulated for non-potable use. Rainwater for more than one household is a public-health matter, so drinking water comes from the mains. The tanks would need to be four to five times bigger for real autonomy, and when the tanks run dry over summer, the dual system allows mains water to compensate. Wastewater: they engineered on-site treatment, but Watercare refused consent. Pipes were located to allow a future switch to on-site wastewater. Food: about 180 fruit trees on common land, strong on fruit and greens, weak on grains and root veg. Full food self-sufficiency was never the aim, "because we're part of the wider society."

Governance, contribution, and COVID

Member-equality governance has run largely unchanged for twenty-five years: monthly full-group meetings plus focus groups that shift as needs change. Contribution runs on expectation, invitation and mild peer pressure rather than a formal quota, though inevitably a smaller group does much of the work. COVID was a major shock. The community had to act as individual households, the common house shut, meals ended, some vulnerable residents left, and two lockdowns brought turnover in a disruptive wave. It took about two years to settle, and the community is now more connected than before.

What biodynamic actually is

He describes it as formalised pre-industrial farming, displaced by monocrops and efficiency: minimal external inputs, heavy composting, rotational grazing, crop rotation, cows kept with their horns, planting timed to the moon. He is refreshingly candid that the biodynamic "preparations" have no proven mechanism. Practitioners simply observe a difference, and the thing actually doing the work, he says, is attention. The closed circle is the ideal: cows feed the soil, the soil feeds the plants, the plants are eaten or sold, and it all returns, kept moving so the soil is never over-asked.

The land numbers

A community of about twenty can be fully food-self-sufficient (bar things that can't grow locally) on roughly 3–8 hectares with animals, fruit trees and some grain for bread; a one-hectare market garden covers a year's vegetables for twenty. And there's no inherent limit to the method: the largest biodynamic farm he knows runs to 1,500 hectares, worked by a whole town, while a backyard also works, since quail need only about two square metres.

Rewilding as infrastructure

His most novel argument is that biodiversity isn't a cost you pay for virtue; it's a productive input. Flower strips pollinate and house the predators that eat your pests; owl and bat boxes are free pest control; runner ducks out-lay chickens and eat the slugs; and beaver-dam analogues filter farm runoff before it reaches the river, raise the water table and cool the stream. Set against that, his critique of industrial farming is precise: synthetic fertiliser plus monocrops overload the soil with nitrogen, and the handpicking his method needs is where the real cost sits. Solar weeders and laser-weeding rigs are closing the gap, but not yet enough.

The two futures, and the ideal settlement

He sees the knowledge as already free (for example, Justin Rhodes and Joel Salatin on YouTube), with AI and an app about to lower the barrier further ("make a picture and it will say what you need to do"). Farming, he thinks, is splitting into full machine automation on one side and a low-input return to horses and minimal fossil fuel on the other; he's qualifying in draught-horse work and building a work-with-nature advisory app himself. His ideal settlement: a managed forest cut at ~5% a year for wood and heat, a compost pile piping its heat to the houses, a flower ring for pollinators, fruit and herbs to pick along the paths, and a stream with a water wheel driving a grain mill.

Separate the three things

His central analytical point: the off-grid framing wrongly merges three different questions. First, the switch from diesel to electric, the enormous, obvious win, turning a $50k diesel bill into $25k of export revenue. Second, why grid electricity is expensive: because of delivery (transmission, distribution, retail), not generation; he calls the network an "electricity landlord." Third, whether to self-generate and store, a separate, second-order question. Jam them together and the off-grid argument gets confused.

Why he keeps the grid

Going off-grid would forfeit his export revenue and demand roughly a million dollars of battery to survive a frost night, when the orchard pulls 90 kW for ten hours and his 300 kWh battery covers about three. So he keeps the grid deliberately: as export earner, frost backstop and crop insurance. The batteries he did buy roughly break even on money; he bought them for resilience, to keep farming through a power cut or an earthquake, and doesn't pretend otherwise.

The cost stack, and a ruthless tech mix

His numbers: grid about $0.40/kWh, self-plus-battery about $0.25, mortgage-financed rooftop solar about $0.11, diesel $1.50–2.00. On the mix he is unsentimental: no stream, so no hydro; wind uneconomic at his scale (you'd need it under about $0.18/kWh); biomass never considered. His prediction for a small New Zealand community site: it "will not include wind." His irrigation, notably, runs 24/7 rather than being shifted to daylight: the value is utilisation, not time-matching.

The real barrier: finance and rules

The binding constraint, he insists, isn't the price of the hardware: it's access to finance and the financial literacy to argue lifetime cost against upfront price. His electric tractor was self-imported and its maker later went bankrupt, so the supply chain, not the performance, is the real risk. His regulatory asks are blunt: "you shouldn't need consent to generate energy," and the avoided-network value of batteries should be recognised.