Why coffee grows
where it does

01 the band

Everything in your cup came from one stripe of the planet.

Not a preference. A constraint, and it has three parts.

Look at the map behind the title. The filled countries are thirty-five of the countries that export coffee, and they all sit inside a band that the review behind most of the numbers here puts at twenty to twenty-five degrees north down to about twenty-four degrees south.¹ Nobody chose that line. It is what is left after three separate limits are laid on top of each other: the plant will not tolerate frost, it wants a temperature range that is cooler than most tropical lowlands, and it wants a long wet season broken by a dry one.

The tropics themselves are not a human invention either. The Earth spins on an axis tilted about 23.4 degrees away from straight up relative to its orbit, and the Tropics of Cancer and Capricorn are simply the two latitudes where the Sun can get directly overhead. Between them the year has no real winter, only a wet season and a dry one. That is the first condition coffee needs, and it rules out most of the planet before you have said anything about the plant.

The scale is worth holding in your head. The International Coffee Organization put world production at 168.2 million 60-kilogram bags in the 2022/23 crop year, and forecast 178.0 million for 2023/24.¹⁰ Of the 2022/23 total, 94.0 million bags were arabica and 74.2 million were robusta, so the split is roughly 56 to 44.¹⁰ Two species. One band. Everything else is detail.

Those countries are not variations on one theme. The Ethiopian highlands and the slopes around Mount Kenya and the Aberdares are equatorial, very high, and sitting on rift volcanics. Colombia and Central America are volcanic mountain country a few degrees off the equator. Sumatra and Java are a young island arc. The robusta lowlands of Vietnam and the plateaux of Brazil are neither high nor equatorial, and they are in the band for different reasons again. Any one origin makes one part of the argument; the argument only closes if you look at several.

This post is that detail, in the order I found it useful: the two plants and what they tolerate, the year a cherry takes to grow, the reason arabica has almost no genetic room to manoeuvre, the arithmetic that lets altitude stand in for latitude, what a volcano actually leaves in the ground, and then the part that changed how I think about the whole thing, which is what the climate models say happens to the band by 2050.

02 two plants

Arabica and robusta are not two grades. They are two different tolerances.

There are 124 known species in the genus Coffea.⁹ Two of them are farmed at scale. Coffea arabica is the one sold as arabica, and Coffea canephora is the one sold as robusta, which is strictly the name of its main cultivated group rather than the species.

The numbers that matter are narrow. A review of the field by DaMatta and Ramalho gives the optimum mean annual temperature for arabica as 18 to 21 degrees Celsius, and the optimum annual rainfall as 1,200 to 1,800 millimetres.¹ For robusta the optimum temperature is 22 to 26 degrees, and it copes better than arabica with heavy rain above 2,000 millimetres.¹ Four degrees of separation between the two species. That is the whole reason the industry has two crops instead of one.

"Optimum" here means the range in which the plant yields best, not the only range in which it lives. Coffee will survive outside those bands. It just stops paying for the land. Two thresholds above the optimum are worth naming, because they are where arabica starts going wrong in ways a farmer notices. Above 23 degrees, fruit development and ripening speed up, which usually costs quality.¹ Continuous exposure to 30 degrees depresses growth outright and produces abnormalities such as yellowed leaves and tumours at the base of the stem.¹

Why does a few degrees matter so much to a tree? Because of what heat does to the trade the leaf is making. A leaf takes in carbon dioxide through pores called stomata, and every time it opens them it loses water. Warm air holds more water vapour than cool air, so as temperature rises the air pulls harder on the leaf, and the plant has to choose between starving and drying out. Push it far enough and it closes the pores, stops fixing carbon, and starts spending reserves it needed for next year's crop. Arabica evolved in a place where it never had to make that choice often.

You can see the four degrees on a map. Vietnam and the Brazilian state of Espírito Santo are warm and low, and they are robusta country. The Kenyan highlands, the Colombian Andes and the Ethiopian forests are cool and high, and they are arabica country. Indonesia grows both, arabica on the volcanic uplands of Sumatra and Java and robusta on the ground below, which is the clearest demonstration there is that the two species are answering different questions about the same island.

That place is specific. Arabica is native to the tropical forests of Ethiopia at altitudes of 1,600 to 2,800 metres, where the annual average is about 20 degrees and rainfall runs from 1,600 to more than 2,000 millimetres.¹ Robusta, by contrast, is a lowland plant and can be grown from sea level up to about 800 metres.¹ Every farm in the world is an attempt to rebuild one of those two places somewhere else.

03 the year of a cherry

The plant needs a dry season as much as it needs the rain.

An annual rainfall total hides the thing that actually governs a coffee farm, which is the order the rain arrives in. Coffee does not flower on a calendar. A flower bud forms, then stops and waits in a dormant state. A dry spell is what makes the bud competent to respond, and then the return of water is the trigger that opens it.⁵ Growers in the trade call it the flowering rain, and on a good farm the whole hillside opens white within a day or two of the same shower.

That single dependency explains a lot of the map. A place with 1,800 millimetres spread evenly over twelve months cannot synchronise its flowering, so the tree carries green, ripe and overripe fruit on the same branch and the picking crew has to walk the rows four times. A place with the same total delivered as a wet season and a clean dry season gets one flowering, one ripening, and one pass. The distribution is the asset.

After the flowers open the clock is long. For arabica the period from anthesis, which is the moment a flower opens, to a ripe cherry averages 220 to 243 days, and ranges from 204 to 266 days for the same cultivar across different environments and altitudes.⁵ C. canephora takes longer, around 284 days.⁵ Eight or nine months in which the farm is exposed to everything the weather does.

The other end of the temperature range is much harder than the top. Coffee is an evergreen perennial with no winter dormancy, so it has no strategy for sitting out a cold snap. The leaf lethal limit, the point at which ice forms and the tissue dies, is reached somewhere between minus 3 and minus 5 degrees.¹ There is no recovery within the season. A frost does not reduce this year's crop, it removes it, and if the wood is killed it removes next year's too.

Brazil is where this is not theoretical. Successive frosts hit the coffee regions in late June and July 2021. The ICO was still writing about it two crop years later, noting that the 2023/24 outlook would be shaped by "the impact of the July 2021 frost" as it continued to be resolved.¹⁰ A single week of cold in one country moved the world price of the whole crop. That is what having it concentrated inside a narrow band costs you.

04 a narrow genome

Arabica is a hybrid with almost no variation to breed from.

This is the part I did not expect, and it changes how the rest of the story reads. Most crops respond to a new climate the same way: you go through the diversity you already have, find the plants that cope, and breed from them. Arabica has strikingly little of that diversity to go through, and the reason is in how it was made.

Arabica is allotetraploid. A normal diploid plant carries two sets of chromosomes, one from each parent. Arabica carries four, and they come from two different species: a whole genome from C. canephora and a whole genome from C. eugenioides, combined in a single hybridisation event and then kept together rather than sorted out.³ It has 44 chromosomes where its parents each have 22. Lashermes and colleagues established the parentage in 1999 using restriction fragment markers and genomic in situ hybridisation, and noted that the two constituent genomes had barely diverged from their living relatives, which meant the event was recent.³

Recent has since been given a number. A chromosome-level assembly published in 2024 dates the founding polyploidy event to between 350,000 and 610,000 years ago, and finds several population bottlenecks after it, including a split between wild accessions and the ancestors of cultivated varieties around thirty thousand years ago.⁴ A bottleneck is what happens when a population is reduced to very few individuals: whatever variation those few did not happen to carry is simply gone, and no amount of later breeding brings it back. Arabica went through the founding one, then more, then a domestication one, then the handful of seeds that left Ethiopia and Yemen and became almost every tree grown in the Americas. The result is one of the lowest genetic diversities of any major crop.⁴

So when someone says the answer to a warming band is simply to breed a hotter-tolerant arabica, the honest reply is that the cupboard is nearly bare. The variation has to come from somewhere else: from C. canephora, which is where the disease resistance in most modern arabica varieties already came from, or from the wild species.

Which is the other reason the 2019 extinction assessment matters. Davis and colleagues produced formal IUCN risk assessments for all 124 wild coffee species and found that at least 60 percent are threatened with extinction: 13 critically endangered, 40 endangered and 22 vulnerable.⁹ The bank you would need to draw on is being closed while you are deciding whether to use it.

The plant cannot move and cannot vary much. Everything that has to give will have to give on our side.the sentence this post has to earn

05 altitude buys latitude

Air cools as it rises, at a rate you can measure and plan a farm around.

Inside the tropics the lowlands are too warm for arabica almost everywhere. The way out is upward. Air pressure falls with height, so a parcel of air that rises expands, and expanding costs it energy, so it cools. The rate at which temperature falls with height is called the lapse rate, and in real mountains you can go and measure it.

Córdova and colleagues did exactly that across the southern Ecuadorian Andes, over a station network running from 2,600 to 4,200 metres, which is above coffee country rather than in it and is the honest caveat on borrowing the number. They found an annual lapse rate of 6.9 degrees Celsius per kilometre for mean temperature, 8.8 degrees per kilometre for the daily maximum, and 5.5 degrees per kilometre for the daily minimum.⁷

Read those three numbers together, because the interesting thing is that they differ. Days cool faster than nights as you climb, so the daily swing narrows by roughly 3.3 degrees for every kilometre gained. A highland farm is not just cooler on average. It is cooler in the afternoon by more than it is cooler at dawn, which is precisely the shape a plant wants if what hurts it is the hottest few hours of the day.

Put the lapse rate together with the optima from section 02 and you can do the arithmetic yourself. Start with a warm tropical lowland at 28 degrees and ask where each species' window sits.

# lapse rate: Cordova et al. 2016, tropical Andes, annual mean [7]
# optima:     DaMatta & Ramalho 2006 [1]
LAPSE   = 6.9      # degrees C lost per 1,000 m climbed
LOWLAND = 28.0     # degrees C, a warm tropical lowland at sea level

for name, lo, hi in [("arabica", 18, 21), ("robusta", 22, 26)]:
    bottom = (LOWLAND - hi) / LAPSE * 1000
    top    = (LOWLAND - lo) / LAPSE * 1000
    print(f"{name:8s} {bottom:5.0f} m to {top:5.0f} m")

arabica   1014 m to  1449 m
robusta    290 m to   870 m

Real farms land where that arithmetic says they should. Arabica in Kenya, Colombia and Ethiopia is grown high, typically above a kilometre and often well above it, because those countries are close to the equator and the lowland is simply too warm. Brazil manages arabica lower than that because a good deal of its growing country is elevated plateau rather than mountain and sits well south of the equator, so latitude and altitude are doing the job together. Vietnam's robusta sits in the lowland range, and so does most of Indonesia's, with the arabica on the same islands moved up onto the volcanoes. One slope, one set of optima, and the whole distribution falls out of it.

That is why the two species stack vertically on the same mountain, and it is also the honest answer to "why not just grow it further from the equator instead". Moving poleward does cool the mean, and a few places genuinely gain from warming for that reason.⁸ But latitude and altitude do not buy you the same thing. Going up lowers the temperature and leaves the seasons alone. Going out adds a winter, and a winter means frost, and section 03 covered what frost does to a plant with no dormancy. It also changes daylength, which the flowering cycle is tuned to.

There is a quieter benefit to height as well. Cooler conditions stretch the ripening period, and the observed range of 204 to 266 days for the same cultivar across altitudinal gradients⁵ is that effect in the data. A longer fill means a denser seed. The trade has been selling that fact as "high grown" for a century, and for once the marketing and the physiology point the same way.

06 what the volcano left

Volcanic soil is not simply fertile. It is deep, light, and very good at hiding phosphorus.

Central America, the northern Andes, Sumatra and Java, the Ethiopian highlands, the flanks of Kilimanjaro. The best-known coffee origins sit on volcanoes with a regularity that stops looking like coincidence. The soils they sit on have a name: andisols, the soil order formed in volcanic ash and other volcanic ejecta. Globally they are the least extensive of the twelve soil orders, covering roughly 1 percent of the ice-free land surface.¹¹

What makes a soil andic is not vibes, it is three measurements. To qualify under the USDA classification, a soil material must contain less than 25 percent organic carbon by weight and then meet one or both of two sets of requirements. The first is all of the following, in its fine-earth fraction: aluminium plus half the iron, extracted with ammonium oxalate, totalling 2.0 percent or more; a bulk density measured at 33 kilopascals of water retention of 0.90 grams per cubic centimetre or less; and a phosphate retention of 85 percent or more. The second route asks for far less retention, 25 percent, and makes up the difference with a measured fraction of volcanic glass.¹² It is the first route that a weathered tropical andisol takes.

Take those one at a time, because each one is doing real work for the plant.

Bulk density of 0.90 or less. Bulk density is the mass of a given volume of soil including the air in it. A litre of andisol weighs less than 900 grams, which is light for a mineral soil. Light means porous, and porous means two things a coffee tree cares about: roots can push down a long way without hitting a pan, and water drains through instead of sitting around the roots. Coffee is intolerant of waterlogging and it roots deep when it is allowed to. A measured Costa Rican coffee andisol came in at 0.74 grams per cubic centimetre with organic matter above 5 percent.⁶

Aluminium plus half the iron above 2.0 percent. This is a proxy for the thing that actually defines these soils, which is a family of minerals with no proper crystal structure: allophane, imogolite and ferrihydrite.¹² When volcanic glass weathers fast, as it does in a warm wet tropical highland, the products do not have time to organise into ordinary clay. They form tiny hollow particles instead, and because they are tiny and hollow they present an enormous amount of surface for their weight. Surface is what holds water and nutrients against gravity, which is why these soils manage to be free-draining and moisture-retentive at the same time.

Phosphate retention of 85 percent or more. Here is the twist, and it is the reason the phrase "rich volcanic soil" is only half true. The same surfaces that hold water grab phosphate ions and will not let go. Phosphate retention of 85 percent means that if you apply phosphorus fertiliser, the soil keeps most of it in a form the root cannot take up. Phosphorus is not optional: it is in every molecule of the plant's energy currency and in the backbone of its DNA. So an andisol hands a coffee tree a superb rooting medium, good water behaviour and a mineral supply from fresh ash, and then makes one of the six major nutrients expensive to deliver. In the Costa Rican study, yield tracked soil texture and phosphorus content rather than the nitrogen fertilisation strategy being tested, and the sector with 50 percent more clay and 47 percent more silt doubled the yield of its sandier neighbour.⁶

Now the geography closes the loop. Volcanoes are not scattered at random. They cluster where one tectonic plate dives beneath another and where continental crust is pulling apart, which is why the Smithsonian's Holocene catalogue puts more than half of the world's recently active volcanoes around the rim of the Pacific alone.¹³ Several of those arcs run straight through the tropics: the Central American arc, the northern Andes, the Sunda arc through Sumatra and Java, the island arcs of the Philippines and New Guinea. The East African Rift adds a second, non-subduction source down the spine of Ethiopia, Kenya and Tanzania, and Kenya is the cleanest single example of all three factors lining up at once: equatorial, so there is no winter; high, on the flanks of Mount Kenya and the Aberdares, so the temperature is in the arabica window; and volcanic, so the soil is deep and free-draining. Those are the coral chains on the map in the hero, and they are the same places as the filled countries because both are answers to the same question about where the planet is young.

I want to be careful about how far that goes. Coffee does not require volcanic soil. It requires a deep, well drained, slightly acid soil with organic matter in it, and volcanic highlands in the tropics happen to produce a great deal of exactly that. The stronger claim you sometimes hear, that volcanic soil makes the coffee taste better, is much less well evidenced than the geography is, and I have not found a study that isolates soil parent material from altitude, shade, variety and processing well enough to support it.

07 2050

Every serious model says the band shrinks. They disagree by a lot about how much.

This is the part I think readers should carry away, and it needs one definition first. These studies model climatic suitability: they learn which combinations of temperature and rainfall the crop is currently grown in, then ask how much land will still offer those combinations under a future climate. Suitable area is not production, and losing suitable area is not the same as coffee disappearing. It is a statement about where the crop can be grown without doing something new.

Globally. Bunn and colleagues trained machine learning models on a database of geo-referenced production locations and projected that climate change will reduce the global area suitable for coffee by about 50 percent, and that the figure holds across emission scenarios.² Their spatial pattern is the important part: impacts are highest at low latitudes and low altitudes, and still negative but less pronounced higher and further from the equator.² That is the lapse-rate figure from section 05 running in reverse. The band does not vanish, it climbs, and mountains get narrower as you go up.

Globally, but grading the land. Grüter and colleagues ran 14 global circulation models under three emission pathways and separated land by suitability class rather than treating it as a single yes or no. The highest suitability class falls by more than 50 percent globally under all three pathways, with declines in the main producing countries running from around half to almost all of it depending on country and scenario.⁸ Coffee came out the most vulnerable of the three crops they studied, with negative impacts dominating in every main producing region.⁸

Latin America. Imbach and colleagues modelled the largest growing region at higher resolution and got a much harsher answer: suitable area down 73 to 88 percent by 2050 across warming scenarios, which they describe as 46 to 76 percent greater loss than the global assessments had estimated.¹⁴ They also modelled the bees, since coffee yields improve with insect pollination even though arabica can self-pollinate, and found mean bee richness falling 8 to 18 percent inside the future suitable areas, though every one of those areas retained at least five bee species.¹⁴

Brazil. Koh and colleagues looked backwards as well as forwards. Temperatures in the Brazilian coffee regions rose about 0.25 degrees per decade between 1974 and 2017, rainfall fell during the blooming and ripening windows, and yields in southeast Brazil dropped by more than 20 percent.¹⁵ That is not a projection. That already happened. Looking ahead, they cite work putting the loss of climatically low-risk area in São Paulo state at around 20 percent under modest warming and as much as 75 percent under more severe scenarios.¹⁵

Ethiopia, and the one hopeful line. Moat and colleagues found that by the end of this century 39 to 59 percent of Ethiopia's current growing area could become climatically unsuitable in the absence of significant intervention, leaving 41 to 61 percent of it.¹⁶ Then they modelled the intervention. Relocating coffee areas uphill, combined with forest conservation and re-establishment, could deliver at least a fourfold increase in suitable farming area against that no-action case.¹⁶ That is the yellow bar running off the right of the chart, and it is the only row that goes the other way.

Global suitable area, 2050

−50%Bunn 2015, across scenarios ²

Latin America, 2050

−73 to 88%Imbach 2017 ¹⁴

Ethiopia by 2100, no intervention

−39 to 59%Moat 2017 ¹⁶

Ethiopia, uphill vs no action

×4Moat 2017, with forest ¹⁶

Read the figure as a spread and not as a set of competing point estimates. Every study points the same direction; the one that does not is the one that assumes the crop is allowed to move. The honest summary is that the direction is settled and the magnitude is not, and section 08 is about why.

08 the evidence is thin

What those percentages do not know.

I found the range in the last section uncomfortable, so I went looking for why it is so wide. Six reasons, and the first two are arguments that the losses are overstated.

Five levers, and the thing none of them move.

09 for your agent

Ask your agent to read this, and it can say it back at whatever level you need.

Coffee is the most universal subject on this site, so this post has the widest possible range of readers, and one piece of prose cannot be pitched at all of them at once. The page therefore carries four read-only tools. They run in your browser, answer from the page itself, and call nothing: there is no key here and no backend.

The one that matters is explain_section. Every section of this post has three hand-written explanations attached to it, at the level of a curious child, a secondary-school student and someone who already knows the field. Ask for the one you want. A plain-text copy of the whole post lives alongside it for agents without a browser.

10 what I take from it

Three things I did not know a week ago.

I started this because I kept reading an altitude on the side of a bag and could not actually defend what it meant. What I found changed the shape of the question.

The soil story is backwards from how it is sold. I had absorbed "volcanic soil, therefore fertile" without ever asking what the mechanism was. The mechanism turns out to include a property that is actively unhelpful: a phosphate retention of 85 percent or more is part of the definition of the soil order.¹² What volcanic highlands really give coffee is depth, drainage and a low bulk density, which is a structural gift rather than a chemical one, plus the altitude that comes free with a volcano. I now think the altitude is doing more of the work than the ash is.

The genetics are the quiet emergency. Everyone writes about the climate projections. Very few people write about the fact that the crop facing them was assembled from a single hybridisation between 350,000 and 610,000 years ago and squeezed through several bottlenecks since,⁴ and that 60 percent of the wild species you would breed from are threatened with extinction.⁹ Those two facts belong in the same paragraph more often than they appear in one.

The disagreement is the finding. My first instinct with the projections was to average them, which is exactly the wrong move: they have different scopes and different definitions of the thing being lost. The useful reading is that five independent studies point the same direction, that the one optimistic result is conditional on being allowed to plant on land that is currently forest, and that the range between 50 percent and 88 percent is a measure of how much we do not know rather than a menu to choose from.

What I will be watching is not the headline percentage. It is whether the wild species survive long enough to be useful, and whether the countries that can move their crop uphill are allowed to do it without felling the forest that keeps the hillside cool in the first place. Moat's fourfold figure and the deforestation it would take are the same sentence.¹⁶

So here is what an altitude on a bag actually says, whichever country it comes from. It says the farm is sitting inside an 18 to 21 degree window that the lowland below it has not had for a very long time. It says the afternoons there are cooler by more than the mornings are. And it says that the window is moving up the hill faster than the hill is getting taller.