How the Garden Planner plans — turning your inputs into a finished map

You draw your plot, nudge a few sliders for your preferences, click "Create plan" — and seconds later beds, trees, greenhouse and compost sit to scale on your map. It looks like magic. It isn't. It's sober calculation steps that anyone can follow. And you should follow them — because the plan belongs to you, not to the computer.

For a long time there were two steps; since the water update there are three: auto-distribution (how much of what), the waterscape (where the rain goes) and auto-placement (where exactly on the plot). This article explains all three, in that order — because internally the planner goes about it the same way.

First the planner reads your climate — not just your zone

Before the three steps even begin, the planner answers a question that paces everything else: when do you plant, and which variety will actually crop? Two plots in the same USDA zone can need completely different calendars — a mild Mediterranean climate plants in autumn, a temperate one in spring. So the planner reads not one number, but three layers.

The zone: how cold your winter gets

From several years of real weather data (the Open-Meteo archive) the planner computes your USDA hardiness zone (3–10) — the median of the coldest days each year, robust against outlier winters. The zone decides what survives winter: which vegetables overwinter, which perennial shrubs and fruit trees are hardy at all. It is the winter-hardiness filter — but no more than that.

The climate band: what rhythm your year has

Hardiness alone says nothing about the season's pacemaker. So the planner also classifies your climate with the Köppen-Geiger system (after Peel, Finlayson & McMahon, 2007) from monthly mean temperature and rainfall:

• Temperate — frost is the pacemaker: sow in spring, main season in summer. The classic Central-Europe / North-America calendar.
• Mediterranean (Cs) — summer is the bottleneck, not frost: main planting in autumn, an active mild winter, a high-summer pause. Spain, Italy, Greece, the US West Coast.
• Hot-arid / cold-arid (BW/BS) — heat and drought are the limit, with or without a frosty winter.
• Humid-subtropical (Cfa) — mild winter, humid summer, a strong second planting in autumn (US Southeast, eastern Australia).

The band shifts the entire calendar: at a Mediterranean site you plant garlic and broad beans in October, not in spring. The planner names this honestly as a curated seasonal model — a recommendation, not a yield guarantee; local timing shifts by a few weeks depending on your exact site, which you check against your own conditions before sowing.

The hemisphere: Christmas in high summer

If your plot is south of the equator, the planner mirrors the whole calendar by six months — in the Southern Hemisphere January is midsummer and July is winter. Sowing, care and harvest months move accordingly, and so do the frost dates (computed from the real southern winter season, not from a Northern-Hemisphere assumption). Australia, New Zealand, southern South America and South Africa are served correctly this way.

The varieties: not just zone-, but climate-tuned

Matching a variety to winter hardiness isn't enough — there are physiological gates that decide yield / no yield:

• Photoperiod: long-day onions (the Central-European default) never bulb below roughly 35° latitude — there you need short-day varieties (Rossa di Tropea, Babosa). Source: Texas A&M AgriLife, UF/IFAS.
• Chill hours: standard apple, pear and cherry need plenty of winter cold. Where it's missing (Mediterranean/subtropical) there is no fruit set — then fig, pomegranate, persimmon, almond, loquat or explicit low-chill varieties (apple “Anna”, peach “Flordaprince”). Source: UF/IFAS, UC Davis Fruit & Nut.
• Heat & bolting: above roughly 35 °C (95 °F) tomatoes and peppers set no fruit, and lettuce and brassicas bolt — in warm climates leafy crops move to autumn/winter, and we pick heat- and bolt-resistant varieties.

For warm sites the variety picker therefore also lists the fitting perennials — olive, fig, citrus, pomegranate, almond, loquat, persimmon, feijoa — while a temperate garden keeps seeing its chill-needing apple and pear varieties. The picker shows each region what actually bears there, not what grows elsewhere.

Step 1: Auto-distribution — how much of what

Before anything lands on the map, one question has to be answered: how many beds does each vegetable group actually get? Grow everything in equal parts and you'll harvest a tonne of lettuce and three beans in autumn — both unhelpful.

The basis: a balanced diet

The starting point is a researched split for a balanced vegetarian self-supply. Every plant family gets a base share of the growing area:

• 20 % legumes (beans, peas) — protein and nitrogen for the soil
• 15 % cucurbits — storage and calorie supplier for the winter
• 15 % brassicas (cabbage) — vitamins, hardy, long season
• 14 % nightshades (tomato, pepper)
• 12 % goosefoots (beetroot, chard, spinach)
• 10 % umbellifers (carrot, parsnip)
• 8 % composites (lettuce)
• 6 % alliums (onions)

Together that comes to 100 %. That's the default if you don't touch a slider — a sensible mix that won't leave anyone hungry.

Your sliders shift the shares

Each preference slider sits at the middle for "neutral". Push it up and that family's share grows proportionally; pull it down and it shrinks. The maths behind it is deliberately simple:

effective share = base share × (slider value / 50)

Slider at 50 leaves everything as it is. Slider at 100 doubles that family's share, slider at 0 cuts it. Because that shifts the total away from 100 %, the planner then scales everything back to 100 % — whoever doubles tomatoes proportionally gets less of the rest. No conjuring trick, just the rule of three.

From percentages to whole beds

You can't plant percentages — you need whole beds. The planner distributes your total bed count according to the calculated shares and rounds honestly: first every group gets its full beds, then the remaining leftover beds go to the groups that came closest to the next whole number. That way the total stays exactly your bed count — no half beds, none vanished.

In doing so the eight families are paired into four companion-planting guilds (A: beans + squash, B: tomatoes + lettuce, C: cabbage + beetroot, D: onions + carrots), which travel together through the 4-year crop rotation.

How much fits on the plot at all?

In parallel the planner answers the second quantity question: how many beds, raised beds and planters realistically fit on your area? For that it first subtracts the permaculture essentials — water storage, trees, house, compost — and takes a share of the rest depending on ambition:

• Starter — 20 % of the area, the rest stays meadow and air
• Hobby — 50 %, a well-filled garden with paths
• Full — 75 %, the rest being paths, wild corners and buffer zones

Deliberately not 100 %: a garden where every square metre is bed isn't a permaculture garden, it's a field. The remaining 25 % at "Full" aren't wasted space — they're the space you walk through and where the beneficials live.

Step 2: The waterscape — water follows the land

Before the planner places a single bed, it plans the water. Not as an accessory, but first — for a simple reason: beds, paths and the greenhouse can be moved. Where rain flows cannot. In permaculture the principle is called Catch and store energy, and its oldest form isn't barrels but earthworks: slow the water down, spread it out, let it soak in — right where it falls.

Two taps instead of a survey crew

For that, the planner needs to know your slope. The input is deliberately simple: tap the highest spot of your plot on the map, then the lowest, and estimate the height difference. That's all it takes — for swales and a pond site only direction and difference matter, never the elevation above sea level. Where official elevation data exists, the planner pre-fills it: in Germany from the state surveys' laser terrain model (3.3 ft grid), in the USA from the USGS terrain model, in Switzerland from swisstopo. Your own input still always outranks the model — you know your plot; the aerial survey only knows how it looked back then.

A third pin is the most classic of all: "Water stands here after heavy rain." Observation beats calculation — where water demonstrably collects, it should be planned in, not modelled away. Set this sink pin and it becomes the preferred pond site.

What the planner makes of it

From slope, soil and the actual rainfall of your location comes the waterscape — that's the section's name in the plan. The logic behind it, in honest numbers:

• Below 2 % slope (practically flat) ditches aren't worth it. Instead the planner recommends a rain garden — a shallow, planted basin that takes the roof overflow and lets it soak away. It sizes it from the connected area and your soil: sand drains fast and needs little space, clay a lot — and on very clayey soil it recommends an infiltration test first.
• Between 2 and 15 % is swale country: swales, shallow ditches dug exactly on the contour line, with the excavated soil piled as a berm on the downhill side. They hold hillside water until it has soaked in. The planner places one per 5.2 ft–6.6 ft of height difference, at most three — the top one in the upper third, because you catch water before it picks up speed. The dimensions (1.3 ft–4.9 ft wide, 5.9 in–11.8 in deep) are calculated from the heaviest single day of rain of the past ten years at your location; a roof sheds 90 % of its rain, a lawn only 20 %.
• From about 15 % the planner turns honest instead of busy: that's terracing country, from about 10 % slope it recommends a professional — and above 20° it doesn't plan at all. A water-saturated steep slope can slide; that's not a weekend project.

A swale is a planting system

"Swales are tree growing systems," says permaculture teacher Geoff Lawton — a swale isn't a hole, it's a tree site. The berm on the downhill side is the moistest, loosest soil in the whole garden. That's why auto-placement deliberately pulls fruit trees and berry bushes there: onto the berm, not into the ditch.

The pond picks its own spot

For the pond there's a ranking that comes straight from the permaculture literature. First: the observed sink (your pin) — nature has already voted there. Second, with at least 6.6 ft of height difference: high in the terrain rather than low. A pond placed high lets its irrigation water work downhill, with no pump at all — Bill Mollison's old rule of storing water as high as possible. From about 9.2 ft of difference even regulated drip irrigation works; the planner calculates this gravity check for you. And only third, without any notable slope: the low point as the natural collector. The soil has a say too — clay soil often seals a pond without a liner, sandy soil needs one. But a dammed or elevated pond is still earthwork: anything above a low embankment, on a slope, or with a high water table belongs in a professional's hands, not read straight off the plan.

The overflow chain: every drop has a plan B

Every water element gets a defined overflow: roof → storage → swale → rain garden → pond → infiltration. The last link always sits on your own plot — and not just out of politeness: in most places you may not redirect runoff to the disadvantage of downhill neighbours (in Germany that's § 37 of the federal water act). The plan shows the whole chain as a diagram.

And what the planner deliberately doesn't claim

All of this is recommendations with sources behind them, not a build manual with a warranty. Two things stay with you: you level the exact line of a swale before digging, with a water level or an A-frame — the ditch floor needs an accuracy of a few centimetres, and no elevation model in the world delivers that. And the satellite-based soil data is coarse; the hand test on site always beats it. Whether a pond needs a permit is a matter of local law and can apply even to small basins — check with your local authority first; for larger ponds (roughly from 26,417 gal) almost always. (The slope analysis is shown to everyone; the concrete dimensions, the chain and the checks are part of the full plan.)

Step 3: Auto-placement — where exactly on the map

Now the planner knows how much of everything and where the water wants to go. That leaves the question: where does everything else go? Here classic permaculture knowledge comes in, translated into geometry.

The plot thinks in zones

Permaculture orders a garden into five zones — rings around the point you're at most often (your house):

• Zone 1 (closest): herbs and lettuce you need daily
• Zone 2: the main beds, greenhouse, compost
• Zone 3: storage crops that want attention less often
• Zone 4: fruit trees and berry bushes
• Zone 5: the wild edge for nature

The planner lays these rings as distance bands over your plot's shape. Each element type has a preferred band: compost stays close to the kitchen in Zone 2 (a compost heap at the far fence never gets used), fruit trees belong on the outside with their crowns.

And in compass directions

On top of that, each element gets a preferred compass sector. The greenhouse wants south (maximum sun), the compost north (shade, out of the line of sight), the insect hotel southeast (morning sun), the deadwood pile northeast (shady and moist). The four bed guilds spread out classically: heat-loving beans-and-squash to the southeast, sun-hungry tomatoes to the southwest, shade-tolerant cabbage to the northwest, undemanding onions-and-carrots to the northeast.

The order: big things first — and water before everything

The planner doesn't place at random. It places in a fixed order — the water right at the front (pond, swales, rain garden: they follow the terrain and don't negotiate), then the big, immovable things (cistern, trees, berry bushes, greenhouse), then the beds, last the small gap-fillers like planters and insect hotel. The reason is the same as packing a suitcase: pack the hiking boots last and they won't fit. Big things first, then the rest fills in around them.

The swales themselves aren't found by any search algorithm: their position is geometry. They lie where the contour lines of your slope run — across the hillside, with clearance from house and boundary. Everything else arranges itself around them.

Straight rows instead of scatter

For each new bed the planner first tries to dock it directly onto the previous bed of the same guild — with a narrow path between. It remembers the direction of the first successful dock, so the following beds keep growing in the same direction. That produces straight, walkable rows instead of a random scatter. If docking fails (edge reached, something in the way), it instead searches in a spiral from the inside out for the next free spot.

Berry bushes use the same trick, only packed tighter — that becomes a living hedge. If your plot is in the wind and you're planning several bushes, the planner even deliberately moves them to the windward side: a hedge that protects beds, instead of just bearing berries. And if there's a swale, its berm takes precedence — the moistest planting spot goes to the woody plants that use it best.

What the planner always keeps clear

Four things the placement respects without compromise: nothing overlaps (every element gets its buffer), nothing lands in the house, no bed moves into a tree's root zone — the planner keeps one and a half times the crown radius clear there — and no bed lands in a swale. A bed under the walnut tree looks nice on paper and harvests nothing in reality.

And the AI? It distributes finer, never more

With Premium a fourth ingredient is added. The AI reads your site details and can refine the distribution of beds across the zones — say, storage crops further out, kitchen-near guilds further in. Important here, and built that way on purpose:

• The AI never changes the number of your beds. It's fixed from Step 1 and stands.
• If the AI proposes a distribution that doesn't exactly fit your bed counts, it gets automatically corrected — it can't invent or swallow beds.
• If the AI is missing (free version) or returns nonsense, the planner falls back silently to the standard distribution. You notice nothing, it just works.

So the AI is a consultant who rearranges the furniture in the room — but tears down no walls and buys no extra furniture.

The most important sentence at the end

Everything described so far is a proposal. A considered, to-scale, permaculturally grounded one — but one you can overrule with the mouse in the detail planner. Move beds, rotate them, draw your own areas — and you can grab the swales there too, shorten them or add one by hand. The algorithm thinks the plan through in seconds so you don't have to start from a blank sheet. The last metre you walk yourself.

The tool thinks the plan. The human carries it out.

Curious how this looks on your plot? The Garden Planner runs all three steps through live as soon as you've drawn your area.