High School Earth & Space Science

Two white lumps on a sheet of paper

Put two white samples on a sheet of paper. One is glassy, about the size of your thumb, with flat faces meeting at sharp angles and a six-sided point at one end. The other is dull and chalky, and where it has broken it makes blocks like shoeboxes someone has leaned on: the corners are not square. Both are white. A photograph will not separate them, nor will your eye.

Four objects and five minutes will: a fingernail, a copper coin, a steel nail, an unglazed white tile, and a spoonful of white vinegar in a saucer. By the end of this lesson you will have run that kit on both samples and named them, and the naming will not be a guess.

Five conditions, and what each one is keeping out

It helps to know what you are testing for. The International Mineralogical Association, which decides what gets a name, holds a substance to five conditions: it occurs naturally, it is solid in its natural state, it has an ordered atomic arrangement (a crystal structure), its composition is reasonably well defined, and it is stable enough to be measured at all.

Each condition throws something out, and the discards teach more than the definition does.

• Steel and an ice cube both fail on natural occurrence, yet glacier ice passes all five. A glacier is a rock made of one mineral, flowing very slowly.
• Obsidian fails the lattice test. It is volcanic glass, cooled so fast the atoms froze where they stood. It is a rock, not a mineral.
• Coal fails the composition test: compressed plant remains whose carbon content varies from seam to seam, so no formula fits it.

One more definition. A rock is an aggregate of mineral grains: minerals are the alphabet, rocks the words. The alphabet is small. The IMA has approved over six thousand species, but only about 150 build rocks, and silicates are roughly 90 percent of the crust by volume.

Hardness: scratch it, and read the bracket

In 1812 the German mineralogist Friedrich Mohs lined up ten common minerals in order of which scratched which, and numbered them. Anything that scratches calcite and is scratched by fluorite sits at 3.5. That is still how it is done.

Your kit sits between those marks: a fingernail is about 2.5, a copper coin or copper wire 3.5, a steel nail or knife blade or piece of window glass 5.5, a hardened steel file 6.5, and an unglazed porcelain streak plate about 7.

The method has one trap in it. Drag the nail across the sample and you will often see a grey line, and the beginner calls that a scratch. Rub it with your thumb. If it wipes away, you smeared metal onto the mineral, which means the mineral is the harder of the two. A real scratch is a groove, and your fingernail catches in it.

Work upward and stop at the first tool that bites. If the fingernail does nothing, the coin does nothing, and the nail leaves a groove, the mineral sits between 3.5 and 5.5. That is a bracket, not a number, and a bracket usually leaves only one or two candidates standing.

Key idea: the Mohs scale is a ranking, not a measurement. Set talc at 1 on an absolute scale and calcite is 14, quartz 100, corundum 400 and diamond about 1500. The steps are not equal, and they grow wildly at the top.

Streak: the one colour that does not lie

Drag a mineral across the unglazed tile and you grind off a powder. The colour of that powder is the streak, and it beats the colour of the lump, because powder has no surface film, no tarnish and no play of light.

Hematite makes the case: a specimen can be silvery grey, black or brick red depending on how it grew, and all three leave the same red-brown streak. Pyrite is brass yellow and streaks greenish black; gold is yellow and streaks yellow. That one difference has settled a great many arguments at the side of a creek.

The plate is about 7 on the Mohs scale, so anything harder will not powder. Quartz leaves no streak and scores the tile instead, which is not a failed test but a hardness result you got for free. While the sample is in your hand, judge its lustre too: metallic, or one of the non-metallic looks (glassy, silky, pearly, earthy). That single cut halves the field before any tool comes out.

Cleavage and fracture: how a mineral chooses to break

Hit a mineral and it breaks along whichever planes hold its atoms together most weakly. If those planes exist, the mineral has cleavage, and the faces repeat: break the pieces again and the same angles come back. If no such plane exists, the mineral fractures instead, irregularly.

• Mica has one perfect cleavage. Slide a pin under an edge and it peels into sheets thin enough to see through, because the silicate layers are weakly bonded to each other.
• Halite, common salt, has three cleavages at 90 degrees and falls apart into cubes. Tip table salt onto dark paper and look: little cubes. Calcite also has three, but at about 75 degrees, so its blocks are leaning rhombs.
• Quartz has no cleavage. Its silicon-oxygen framework is equally strong in every direction, so it breaks in curved scoops called conchoidal fracture, the shape on a chipped bottle.

The vinegar test, and the reaction behind it

Put a drop of white vinegar on the chalky sample. If it is calcite, the drop crackles and a small head of foam builds. The gas is CO₂, released as the acid pulls the carbonate apart: calcium carbonate plus acetic acid gives calcium acetate, water and carbon dioxide.

Household vinegar is about 5 percent acetic acid and weak besides, so it is gentler than the 10 percent hydrochloric acid in a field kit. Two tricks recover the sensitivity: scratch off a little powder, which has far more surface, and warm the vinegar. Dolomite looks almost identical to calcite but barely reacts as a lump and fizzes only once powdered, which is the standard field test for telling limestone from dolostone.

The minerals that actually build the crust

Colour is the property a beginner trusts most and it is the least dependable thing about a sample. Pure quartz is clear; trace iron plus natural radiation makes purple amethyst, another trace makes pink rose quartz, aluminium makes smoky grey, and tiny fluid inclusions make milky white. Five colours, one mineral, one formula, SiO₂. Composition and structure are what you are really after, and seven minerals carry most of the crust. Feldspars alone are about 60 percent of Earth's crust and about 41 percent of the continental crust by weight.

Running the kit on the two samples

Start with the glassy one. Fingernail: nothing. Copper coin: only a smear that wipes off. Steel nail: nothing, and the nail has lost metal. On the tile: no streak, but a bright scratch in the tile. Vinegar: no reaction. Break: smooth curved scoops, no repeatable flat faces. Hardness at least 7, no cleavage, glassy lustre. That is quartz, and the six-sided point was the hint you could not yet use.

Now the chalky one. Fingernail: nothing, so above 2.5. Copper coin: a real groove your nail catches in, so below 3.5. Streak: white. Break: flat faces in three directions, corners leaning near 75 degrees. Vinegar: a steady crackle. Hardness 3, rhombic cleavage, fizzes. That is calcite.

Notice the third candidate you eliminated without meaning to: halite is also soft with three cleavages, but they meet at 90 degrees and it does not fizz. Half a minute of testing separated three minerals a photograph would have merged into one.

Common misconceptions

• "The minerals in my breakfast cereal are these minerals." Dietary minerals are elements such as iron and calcium. Calcium is an element; calcite is a mineral species.
• "Diamond is ten times harder than talc." Mohs numbers are ranks, not amounts. Set talc at 1 on an absolute scale and diamond lands near 1500.
• "Colour is the fastest way to identify a mineral." It is the fastest way to be wrong. Quartz comes in at least five colours, and hematite in three while always streaking red-brown.

What to carry forward

A mineral is natural, solid, ordered and reasonably fixed in composition. Rocks are made of minerals, silicates are about 90 percent of the crust, and roughly 150 species do nearly all the work. You identify one by testing it: bracket the hardness against a nail at 5.5 and a plate at 7; take the streak; read the break for cleavage planes and their angles; check the lustre; drop vinegar on any suspected carbonate.

The point: hardness, streak and cleavage come straight from the atomic lattice, which is why they repeat every time. Colour does not, which is why the expert checks it last.

Sources

1. Earle, S. (2019). Mineral properties. In Physical Geology (2nd ed., Section 2.6). BCcampus, via Geosciences LibreTexts. Geosciences LibreTexts
2. Wikipedia contributors. (2026). Mohs scale. Absolute-hardness table (talc 1, calcite 14, quartz 100, corundum 400, diamond 1500) and the 1812 origin. Wikipedia
3. Wikipedia contributors. (2026). Feldspar. Source of the 60 percent and 41 percent crustal figures. Wikipedia
4. U.S. Geological Survey. Collecting rocks (General Interest Publication). USGS. USGS

Three objects on a city street

Walk one block of an old city and you can touch all three families. The kerbstone is speckled pink, white and black, and the speckles are interlocking crystals two or three millimetres across, jammed together with no gaps and no layers. The step up to the front door is pale buff, faintly striped, and if you rub it hard a few sand grains come away on your thumb. The roof above you is covered in flat grey plates, each about three millimetres thick, that ring when tapped and split cleanly along one direction.

Three rocks, three textures, three completely different histories. You can read those histories off the surface, and this lesson is about how.

Igneous: crystal size is a clock

An igneous rock is one that froze from molten rock. That is the whole definition. Everything interesting follows from one variable: how fast it cooled.

Crystals need time. Atoms have to migrate through the melt and find the right seat in a growing lattice, so slow cooling grows large crystals and fast cooling grows small ones. Magma trapped ten kilometres down is insulated by the rock above it and may take a hundred thousand years to solidify, giving crystals you can see without a lens. Lava poured onto the surface chills in days or hours and gives crystals too small to resolve. Lava quenched in seconds gives no crystals at all, only glass.

That single relationship lets you read an igneous rock backwards:

• Visible interlocking crystals (coarse): it cooled slowly, underground. Intrusive. Granite is the common example.
• Crystals too small to see (fine): it cooled fast, at the surface. Extrusive. Basalt is the common example.
• Glassy, no crystals: quenched. Obsidian.
• Full of holes: gas bubbles frozen in place as the melt solidified. Pumice and scoria.

A second variable, silica content, sets the colour and will matter enormously in the volcano lesson. Silica-rich melts make pale rocks full of quartz and feldspar, such as granite. Silica-poor melts make dark rocks full of pyroxene and olivine, such as basalt. Pale and coarse means granite; dark and fine means basalt, and those two cover most of what you will meet.

Sedimentary: grains, cement and the record of a surface

A sedimentary rock is built at the surface out of pieces, and it is the only family that routinely preserves fossils, because it is the only one that forms cool enough and gently enough to leave a shell intact.

Clastic rocks are made of broken fragments, sorted by size: mud and clay give shale, sand gives sandstone, gravel gives conglomerate. Two properties of the grains record the journey. Rounding measures how far they travelled, because corners are knocked off in transport. Sorting measures the consistency of the current: a river that always flows at the same speed drops only one grain size and gives well-sorted rock, while a glacier drops everything at once and gives a jumble.

Chemical and organic rocks form differently. Limestone is calcium carbonate precipitated from sea water or built by shells and coral. Rock salt and gypsum are what is left when a shallow sea evaporates. Coal is compressed plant matter. These are the rocks that fizz, dissolve, or burn.

The making of a sedimentary rock has two steps worth naming. Compaction squeezes water out and packs grains closer as burial depth grows. Cementation then glues them: dissolved calcite or silica precipitates in the remaining pore space. That cement is why your front step sheds a few grains and not a handful.

Metamorphic: heat and pressure, but no melting

A metamorphic rock is one that was changed in the solid state. This is the definition students most often lose: if it melts, it is no longer metamorphism, it is the start of a new igneous rock. Minerals recrystallise, grow larger, or react to make entirely new minerals, all without the rock ever becoming a liquid.

Metamorphism begins around 150 degrees Celsius and runs up to the melting point. Geologists split that range into grades, and each grade turns mudrock into a different rock:

That layering is foliation, and it has a cause you can picture. Squeeze a rock hardest in one direction and its flat minerals, chiefly the micas, rotate until they lie perpendicular to the squeeze, the way pencils dropped in a box and then shaken settle flat. Foliation therefore records the direction of the pressure, not the direction of the original bedding, which is why slate splits at an angle to the layers it inherited from the shale.

Some parent rocks have no flat minerals to rotate, so they recrystallise without foliation. Limestone becomes marble, a mass of interlocking calcite that still fizzes in acid. Sandstone becomes quartzite at medium grade and above, so tough that it breaks straight through the sand grains rather than around them.

The three families side by side

What matters here: the four rows above are a decision procedure. Interlocking crystals with no layering means igneous; visible separate grains in layers means sedimentary; aligned bands of minerals with no gaps between grains means metamorphic.

Three real samples

El Capitan Granite, Yosemite. The 3,000 foot (914 metre) face of El Capitan is pale, coarse-grained granite about 100 million years old. Coarse means it crystallised kilometres down; the fact that you can stand at the bottom of it means about that much overlying rock has since been stripped away.

Navajo Sandstone, Utah. Quartz sand, up to 2,300 feet (700 metres) thick, deposited roughly 200 to 195 million years ago in the Early Jurassic. Its metre-scale cross-bedding is the frozen face of ancient sand dunes, and the direction those beds dip records the wind that built them.

The slate on the roof. Mudrock, buried, heated to somewhere between 150 and 300 degrees Celsius, its clay minerals recrystallised into micas all lying parallel to one another. That is the only reason it can be split into sheets thin enough and flat enough to keep rain out of a house.

The rock cycle is not a circle

Textbooks draw the rock cycle as a ring with three stations, and the picture is right about the destinations and wrong about the route. There is no required order. The arrows run between every pair of boxes and in both directions.

Take one quartz grain. It crystallises in granite ten kilometres down. Uplift and erosion expose it, weathering frees it, a river carries it to a beach, burial and cement lock it into sandstone. Heat and pressure at 500 degrees Celsius recrystallise the sandstone into quartzite. Uplift exposes the quartzite, a stream frees the grain again, and it is back on a beach. That grain has been igneous, sedimentary and metamorphic, and it never once melted. Melting is one route through the cycle, not the hinge it is usually drawn as.

The cycle also runs at very different speeds in different places. Sediment on a delta can become weakly cemented rock in a few thousand years, while a gneiss in the core of a continent may sit untouched for a billion.

Common misconceptions

• "Metamorphic rocks are partly melted." They are not. Metamorphism is change in the solid state; once melting begins, the product cools into an igneous rock instead.
• "The rock cycle goes igneous to sedimentary to metamorphic in order." Any box can lead to any other. A sedimentary rock can be eroded straight back into sediment without ever being metamorphosed.
• "Big crystals mean the rock took longer to form than a fine one, so it is older." Crystal size records cooling rate, not age. A coarse granite can be far younger than a fine basalt.
• "Marble and limestone are unrelated." Marble is limestone that has been recrystallised. Both are calcium carbonate and both fizz in dilute acid.

The short version

Igneous rocks freeze from melt, and their crystal size records how fast: coarse means slow and deep, fine means fast and shallow, glassy means quenched. Sedimentary rocks are assembled at the surface from grains or from dissolved material, preserve fossils, and record the currents that carried them in their sorting and rounding. Metamorphic rocks are cooked and squeezed while staying solid, gaining foliation if they contain flat minerals, and running through slate, phyllite, schist and gneiss as temperature climbs past 150, 300, 450 and 550 degrees Celsius.

In short: texture is the evidence and the three families are the conclusion. Look at how the grains meet each other before you look at anything else.

Sources

1. Earle, S. (2019). Classification of metamorphic rocks. In Physical Geology (2nd ed., Section 7.2, Table 7.1). BCcampus, via Geosciences LibreTexts. Source of the grade and temperature bands. Geosciences LibreTexts
2. Wikipedia contributors. (2026). Navajo Sandstone. Age of 200 to 195 Ma and thickness up to 2,300 ft (700 m). Wikipedia
3. Wikipedia contributors. (2026). El Capitan. El Capitan Granite, about 100 million years old, 3,000 ft (914 m) face. Wikipedia
4. Earle, S. (2019). Physical Geology (2nd ed.), Chapters 3 to 7. BCcampus, Victoria BC.

Standing at the bottom of a hole a mile deep

On the floor of the Grand Canyon, beside the Colorado River, the rock under your boots is dark, coarse and folded like stirred treacle. It is schist, part of the Vishnu Basement Rocks, formed between 1,840 and 1,660 million years ago. Look up. The pale cliff capping the rim, up to 1,857 metres (6,093 feet) above the river, is the Kaibab Formation: a limestone full of shells, 270 million years old.

About one and a half billion years of history sits on that wall, stacked in order, with nobody present to write any of it down. Here is the procedure that recovers the dates anyway, in five steps.

Step 1: put events in order using no numbers at all

Nicolas Steno, a Danish anatomist dissecting sharks in Tuscany, published the first rules in 1669, and they still carry most of the load.

• Superposition. In an undisturbed stack the bottom layer came first. Sediment falls downward; it cannot arrive underneath something already there.
• Original horizontality. Sediment settles flat, so a bed now tilted at 40 degrees was tilted after it hardened.
• Lateral continuity. A matching bed across a valley is one bed with the middle eroded out.
• Cross-cutting relationships. Anything that cuts something else is younger than what it cuts.
• Inclusions. A fragment trapped inside a rock is older than its host, because it had to exist to be picked up.

Those five solve a real wall. Picture a cliff 30 metres high: grey shale 8 metres thick at the base, brown sandstone 6 metres, white limestone 5 metres. A vertical fault cuts all three and drops the right side by 2 metres. A granite dyke 40 centimetres wide climbs through shale, sandstone, limestone and the fault plane itself, then stops at a flat surface 19 metres up, above which a horizontal conglomerate 11 metres thick holds rounded pebbles, some of them granite.

Order it. Shale, sandstone, limestone by superposition. Then the fault, which cuts all three. Then the dyke, which cuts the fault. Then erosion, because the dyke is beheaded rather than continuing. Then the conglomerate, where inclusions work twice over: those granite pebbles are pieces of the dyke, so it was emplaced, then exposed, weathered out and rolled in a river before burial. Seven events from three rock names and a hand lens.

Step 2: the time that is not there

That flat surface has a name. An unconformity is a gap: time during which rock was eroded away, or never deposited. Three shapes, told apart by eye. An angular unconformity shows tilted beds sliced flat with horizontal beds across their cut ends, so the lower set was deposited, hardened, tilted, planed and buried again; James Hutton took a boat crew to Siccar Point in Scotland in 1788 to show them one, arguing that no short chronology had room for that sequence. A disconformity shows parallel beds either side of an irregular, channelled erosion surface. A nonconformity shows sediment resting straight on igneous or metamorphic rock that was uplifted and stripped bare first.

The Grand Canyon has the famous nonconformity. In the inner gorge, Cambrian Tapeats Sandstone lies directly on basement rock whose top surface is about 1,660 million years old: the Great Unconformity. Where the older Unkar Group survives beneath it, the gap between the basement top at about 1,660 Ma and the Bass Formation at about 1,250 Ma is roughly 410 million years of missing record, produced by uplift and erosion that removed at least 25 kilometres of rock.

Worth holding on to: the rock record is mostly absent. A surface you could cover with your palm may stand for more time than every layer above it put together, which makes an unconformity the densest information on the wall rather than a blank in it.

Step 3: matching one wall to another

Steno's rules order one outcrop. They say nothing about whether a limestone in Arizona matches one in Wales. William Smith, a canal surveyor watching diggers cut through southern England in the 1790s, found that fossils came in a fixed order wherever he went and never reversed; his 1815 geological map of England and Wales was built on it. The principle is faunal succession.

An index fossil needs two properties that pull against each other: a short time range and a wide geographic spread. A species that lasted 200 million years dates nothing; one confined to a single lagoon correlates nothing. Swimmers and floaters win, because currents carry them worldwide. Trilobites serve for Cambrian rock, graptolites for the Ordovician, and ammonites for the Cretaceous, having changed shell design fast enough to beat the slow-evolving nautiloids.

Step 4: attaching a number, worked twice, then broken once

Every rule so far gives order, not years. Years come from radioactivity. An unstable parent isotope decays into a stable daughter at a rate no ordinary geological condition alters, quoted as a half-life: the time for half the parent atoms present to decay. After one half-life, half remain; after two, a quarter; after three, an eighth. As algebra, with N the parent atoms now, N₀ the parent atoms at the start, t the elapsed time and T the half-life:

N / N0 = (1/2)t/T

The USGS publishes the clocks geologists use.

Worked example one. A volcanic ash bed is crushed and its crystals measured. The laboratory reports 25 percent of the original potassium-40 remaining, the rest now argon-40 trapped in the crystal. Set N/N₀ to 0.25. That is one half squared, so t/T is 2. With T of 1.25 billion years, t is 2.50 billion years.

Worked example two. A second bed reports 30 percent remaining, which is no neat power of one half. It sits between 0.50 (one half-life) and 0.25 (two), so the age lies between 1.25 and 2.50 billion years, nearer the top. Take logarithms: t/T = log(0.30) / log(0.50) = 1.737, so t = 1.737 times 1.25 = 2.17 billion years. Five percentage points of measurement moved the age by about 330 million years, which is why every laboratory date carries an uncertainty beside it.

Now break it. Try that same ash with carbon-14. Its half-life is 5,730 years, so 2,500,000,000 divided by 5,730 is about 436,000 half-lives, and one half raised to that power is indistinguishable from nothing by any instrument that exists. This is why the USGS limits carbon-14 to roughly the past 50,000 years. The mistake runs the other way too: aim uranium-238 at a 3,000-year-old beam and the fraction decayed is under a millionth of a percent.

The point: choose the clock whose half-life is roughly the size of the age you expect. Too short and nothing is left to count; too long and nothing has happened yet.

One feature is easy to get wrong. A radiometric date records when a crystal stopped leaking its daughter product, not when the sediment around it settled. Date quartz grains in a sandstone and you get the age of the granite they eroded from, perhaps a billion years older than the beach. So geologists date ash beds and lava flows trapped inside sedimentary sequences: ash falls in a day, its crystals close at once, and its age brackets everything below as older and everything above as younger.

Step 5: the calendar everyone agreed on

Dates from thousands of outcrops are assembled into one chart by the International Commission on Stratigraphy, part of the International Union of Geological Sciences, reissued as boundaries are re-dated. That is why textbook numbers move: the base of the Cambrian was revised to 538.8 Ma in 2022.

Four eons. The Hadean opens at 4,567 Ma with formation and a magma ocean; almost no rock survives it, though Jack Hills zircon grains from Western Australia date to 4,404 plus or minus 8 Ma. The Archean opens at 4,031 Ma with the first preserved crust, the first microbial life and no free oxygen. The Proterozoic opens at 2,500 Ma: oxygen enters the air, complex cells appear, and the planet freezes over more than once. The Phanerozoic opens at 538.8 Ma with shells, skeletons and a dense fossil record, splitting into the Paleozoic, the Mesozoic from 251.9 Ma and the Cenozoic from 66 Ma. That last eon, effectively every fossil you have seen in a museum, is the final 12 percent of Earth's history, and the Holocene, which holds every city and every written word, is 0.00026 of it.

Five intervals when the record thins out

Read upward through Phanerozoic rock and five times the fossils change abruptly, with recovery taking millions of years.

The last of the five shows what it looks like when a claim arrives with its evidence. In 1980 Luis and Walter Alvarez and colleagues reported in Science that the thin boundary clay carried far more iridium than ordinary crustal rock. Iridium is scarce in the crust and abundant in asteroids, so they predicted an impact. A crater 180 to 200 kilometres across was later found beneath the Yucatan Peninsula in Mexico, dated to 66.043 plus or minus 0.043 Ma and consistent with an object about 10 kilometres wide. The prediction came first, the crater second. The largest of the five, the end-Permian, is instead blamed on the Siberian Traps: flood basalts that buried much of Siberia over roughly 60,000 to 100,000 years.

Reading the wall from the river to the rim

At the bottom, the Vishnu Basement Rocks: schist and granite, 1,840 to 1,660 Ma, once sediment and volcanic rock, buried deep enough to recrystallise and then intruded by granite. Above them the Great Unconformity. Above that the Tapeats Sandstone, a beach left by a Cambrian sea advancing across a continent planed flat, its grains coarse at the base and finer upward as the water deepened; then the Bright Angel Shale, mud settling further offshore; then the Muav Limestone, carbonate forming in clear warm water with almost no sediment arriving from land. Three formations, one story: a shoreline moving inland.

Higher up come limestones full of crinoids, red beds from river floodplains, the Coconino Sandstone whose steep cross-beds are frozen dune faces from a desert, the Toroweap, and finally the Kaibab Formation at 270 Ma, marine again, its shells recording a sea that came back over the sand. The canyon itself is the youngest event of all: the Colorado River took its present course about 5 to 6 million years ago and cut down through the pile. The rock is ancient; the hole is new.

Common misconceptions

• "Radiometric dating just assumes the decay rate never changed." It is tested, not assumed. One zircon holds uranium-238 at a half-life of 4.5 billion years and uranium-235 at 704 million. Two clocks at wildly different speeds give the same age; a drifting rate would pull them apart.
• "Carbon dating tells you how old a rock is." Carbon-14 dates carbon that was once alive, and its 5,730-year half-life confines it to about the past 50,000 years. Rocks are dated with potassium, uranium, rubidium or thorium.
• "The bottom layer is always the oldest." Only if the stack has not been turned over. Folding can invert a sequence, so geologists check which way was up using graded beds, ripple crests and mud-crack shapes.

Putting it together

Two systems, used together. Relative dating sequences events through superposition, original horizontality, cross-cutting, inclusions and faunal succession, and needs nothing but your eyes. Absolute dating attaches years through half-lives, using N/N₀ = (1/2)^t/T, and needs a laboratory and the right isotope: potassium-40 at 1.25 billion years for old volcanic rock, carbon-14 at 5,730 years for the last fifty millennia. Neither is enough alone. The ash bed supplies a number; only superposition tells you which beds that number brackets.

The core of it: a canyon wall is a document. Order comes from geometry, years come from physics, and a gap in the wall is not a blank page but a statement about time that nothing recorded.

Sources

1. U.S. Geological Survey. (1997). Geologic time: Radiometric time scale. USGS General Interest Publication. Source of the half-life table and the 50,000 year limit on carbon-14. USGS
2. National Park Service. Geology of Grand Canyon National Park. Source of the 270 million year Kaibab Formation and the 5 to 6 million years of carving by the Colorado River. National Park Service
3. Wikipedia contributors. (2026). Vishnu Basement Rocks. Source of the 1,840 to 1,660 Ma range, the 410 million year gap and the 25 km of rock removed. Wikipedia
4. Wikipedia contributors. (2026). Extinction event. Source of the dates and loss percentages for the five mass extinctions. Wikipedia
5. Wikipedia contributors. (2026). Chicxulub crater. Source of the 66.043 plus or minus 0.043 Ma age, the 180 to 200 km diameter and the 1980 Alvarez paper. Wikipedia

A meteorologist with a map, 1912

In 1912 a 32-year-old German meteorologist named Alfred Wegener published two articles arguing that the continents had once been joined in a single landmass, which he called Pangaea, and had since drifted apart. He was not the first to notice that South America and Africa look like two pieces of a torn page. Abraham Ortelius had said so three centuries earlier. Wegener was the first to treat it as a claim that could be tested against evidence, and for the next fifty years most geologists told him he was wrong.

They had a good reason. Watching that argument get settled is the best available lesson in how science actually decides things, because the winning side did not win by insisting. It won by making predictions that could have failed and did not.

What Wegener actually had

Four lines of evidence, and they were not weak.

• The fit. The coastlines of South America and Africa match, and they match better still if you fit them at the edge of the continental shelf rather than at the shoreline.
• Fossils. Identical fossil species turn up on both coasts, separated now by 4,000 kilometres of ocean. The seed fern Glossopteris is found across South America, Africa, India, Australia and Antarctica. Wegener's point was mechanical: these organisms could not have swum or blown across an ocean.
• Ancient climates in the wrong places. Coal, which forms from lush tropical swamp vegetation, occurs in Antarctica. Glacial deposits, scraped and dumped by ice sheets, occur in what is now arid South Africa, in places such as the Vaal River valley.
• Matching structures. Mountain belts and distinctive rock sequences run off the edge of one continent and pick up again on another.

Why the geologists said no

The objection was not stubbornness, and it is worth stating at full strength. Wegener had no mechanism. He proposed that continents were driven by the centrifugal effect of Earth's rotation and by tidal forces from the Sun and Moon, and physicists calculated those forces and found them far too small, by orders of magnitude, to shove a continent anywhere.

Worse, his picture required continents to plough through solid ocean floor like ships through pack ice. Ocean floor is basalt. Basalt is strong. A continent dragged through it would shatter long before the ocean floor gave way. The geologists were applying physics correctly to the model in front of them, and the model was wrong in exactly the way they said. Wegener died on the Greenland ice in 1930 with the argument still lost.

Why this matters: a correct conclusion supported by a broken mechanism is not yet science. The drift was real; the engine Wegener proposed was not. Fixing the engine took another thirty years and a war's worth of new instruments.

The sea floor turns out to be young

Wartime sonar and post-war ocean surveys produced two surprises. The deep ocean floor was not the flat, ancient plain everyone assumed; it carried a mountain range running for tens of thousands of kilometres. And the sediment on top of it was far too thin. If the ocean basins had been collecting dust and shells for four billion years, the layer should have been kilometres deep. It was not.

Harry Hess, a Princeton geologist and a Naval Reserve rear admiral, proposed the resolution in 1960 and published it in 1962. New ocean floor is made continuously at the ridge, spreads outward like a conveyor belt, and is destroyed at deep ocean trenches. Robert Dietz named the process seafloor spreading in 1961. The model explained the thin sediment, the young ocean rocks, and, crucially, why Earth does not swell: crust is created and destroyed at the same rate. Continents no longer had to plough through anything. They ride on the plate.

The stripes: a test that could have failed

Here is where the argument turns. Basalt contains magnetite, and when lava cools through a critical temperature its magnetite grains lock in the direction of Earth's magnetic field at that moment. Earth's field flips polarity at irregular intervals, a fact established from lavas on land.

Put those two facts together with Hess's conveyor belt and you get a prediction that is sharp enough to be dangerous. If new floor is made at the ridge and carried away on both sides, then the ocean floor must be striped with alternating bands of normal and reversed magnetisation, and, because the conveyor runs both ways at once, the stripe pattern on one side of the ridge must be the mirror image of the pattern on the other. A random or one-sided pattern would have killed the idea outright.

Frederick Vine and Drummond Matthews at Cambridge published the test in September 1963; Lawrence Morley reached the same conclusion independently and had both his papers rejected earlier that year. Magnetometers towed behind ships found the zebra pattern, symmetric about the ridge crest. The floor had been recording the planet's magnetic history like a two-headed tape recorder, and the tape could be read.

Earthquakes and hot-spot tracks

Two more independent lines arrived quickly. First, as seismograph networks spread worldwide, earthquakes turned out not to be scattered. They cluster along the ocean trenches and the spreading ridges, tracing plate edges. Along trenches the earthquakes get deeper as you move away from the trench, marking a slab of cold lithosphere sinking at an angle into the mantle. That is a plate boundary drawn from underneath.

Second, J. Tuzo Wilson proposed in 1963 that a chain of islands could be made by a plate sliding over a fixed hot spot in the mantle. The prediction is a line of volcanoes that ages steadily in one direction. The Hawaiian chain obliges: the oldest volcanic rocks on Kauai are about 5.5 million years old and deeply eroded, while on the Big Island, sitting over the hot spot now, the oldest exposed rocks are less than 0.7 million years old and new rock is still forming. The chain continues northwest as the drowned Emperor Seamounts, and the bend where the trend changes records a change in the direction the Pacific Plate was travelling.

From inference to measurement: GPS

For most of this story, plate speeds were inferred from the age and width of magnetic stripes. Since the 1990s they have simply been measured. Continuously recording GPS receivers bolted to bedrock give plate positions to a few millimetres, and the velocities they return match the geological estimates.

Do the arithmetic on the Atlantic. At 2.5 cm/yr, 25 km per million years, the 5,000 kilometres between Africa and South America takes about 200 million years to open, which is when Pangaea is independently dated to have begun breaking up, about 225 to 200 million years ago. Two completely separate methods, one magnetic and one geological, land on the same number.

What is actually moving, and on what

The moving layer is the lithosphere: the crust plus the rigid uppermost mantle welded to it, about 100 kilometres thick under oceans. Beneath it is the asthenosphere, hotter and weaker, which deforms by slow creep over geological time.

The asthenosphere is solid rock. We know this directly, because shear waves from earthquakes travel through it, and shear waves cannot pass through liquid. Solid rock held at high temperature and pressure for millions of years flows the way a glacier flows: it is not melted, it is creeping. The forces driving the plates come mostly from the plates themselves. A cold, dense slab sinking into the mantle at a trench pulls the plate behind it, and the raised ridge lets gravity slide the plate away downhill.

Common misconceptions

• "The plates float on a sea of molten magma." They do not. The asthenosphere transmits S-waves, so it is solid. Magma exists only in small pockets, chiefly where water lowers the melting point above a sinking slab or where pressure drops beneath a ridge.
• "Plates and continents are the same thing." The North American Plate carries the continent plus half the Atlantic floor. Most plate boundaries lie under the ocean, nowhere near a coastline.
• "Wegener was right and everyone else was just closed-minded." Wegener's conclusion was right and his mechanism was wrong, and his critics were correct about the mechanism. The theory only succeeded once seafloor spreading supplied an engine that worked.
• "Continental drift and plate tectonics are two names for one idea." Drift moved continents through the ocean floor. Plate tectonics moves the ocean floor and the continents together as one rigid sheet.

Where this leaves us

Continental drift had the fit, the fossils, the coal in Antarctica and the glacial scratches in South Africa, and it still lost, because it had no force capable of doing the job. Seafloor spreading supplied the missing engine, and then four independent measurements converged on it: the symmetric magnetic stripes, the thin ocean sediment, the belts of earthquakes deepening away from trenches, and the ageing chain of Hawaiian volcanoes. GPS turned the inference into a reading you can watch change.

The upshot: the theory was accepted when it made a prediction sharp enough to be wrong. Mirror-image magnetic stripes were not evidence anyone went looking for to confirm a favourite idea. They were a result that the conveyor belt required and that nothing else explained.

Sources

1. Kious, W. J., and Tilling, R. I. (1996). This Dynamic Earth: The Story of Plate Tectonics. U.S. Geological Survey. Sections on historical perspective, developing the theory, understanding plate motions, and hotspots. USGS
2. Wikipedia contributors. (2026). Vine-Matthews-Morley hypothesis. Dates for Hess (1960, published 1962), Dietz (1961), Vine and Matthews (September 1963) and Morley's rejected letters. Wikipedia
3. Wikipedia contributors. (2026). Oceanic crust. Crustal thickness of 7 plus or minus 1 km and the oldest large-scale ocean floor at about 180 to 200 million years. Wikipedia
4. Wegener, A. (1915). Die Entstehung der Kontinente und Ozeane. Vieweg, Braunschweig.

A trace, and a gap of twenty-five seconds

A seismogram is a line on a chart with time running left to right. For most of its length it is nearly flat. Then it develops a small, sharp jiggle. Twenty-five seconds later the trace jumps to a much larger, slower wobble. About a minute after that, the largest swings of all arrive and take several minutes to die away.

Three arrivals, three different waves, one earthquake. Everything in this lesson comes out of that gap of twenty-five seconds, because the gap is what tells you how far away the earthquake was.

What is actually shaking

Rock on either side of a fault is being pushed in opposite directions by plate motion, but friction locks the fault, so the rock bends instead of sliding. It stores elastic energy the way a bent ruler does. When the stress finally exceeds the friction, the fault slips in seconds and the bent rock springs back to its unstrained shape. That is elastic rebound, and the energy released as the rock straightens is the earthquake.

Two words for two places. The focus (or hypocentre) is the point underground where the slip begins, typically 5 to 15 km down for a shallow continental quake. The epicentre is the point on the surface directly above it. News reports give the epicentre; the physics happens at the focus.

Three kinds of wave, three speeds

The energy leaves the focus as three kinds of wave, and the order they arrive in is fixed by their speeds.

The P wave is a sound wave in rock. It compresses and stretches the material along its path, and because liquids resist compression, it passes through them. The S wave shears the material sideways. A liquid has no resistance to shearing, so an S wave simply stops at a liquid boundary. That single fact, applied to waves crossing the whole planet, is how we know Earth's outer core is liquid: there is a zone on the far side of any earthquake where P waves arrive and S waves never do.

Surface waves are the slowest and the most destructive. They are confined to the top few kilometres, so their energy does not spread out into a whole sphere, and they arrive with large amplitudes and long periods that buildings find hard to ignore.

Turning a gap into a distance

P and S waves leave the focus at the same instant and travel the same path at different speeds, so the gap between their arrivals grows with distance. That makes the gap a ruler.

Work it out. If the distance is d, the P wave takes d/6.0 seconds and the S wave takes d/3.46 seconds, using standard crustal speeds. The gap is the difference:

t(S) - t(P) = d/3.46 - d/6.0 = d x (0.289 - 0.167) = d x 0.122 seconds per km

Turn it around and you get the rule seismologists carry in their heads: d = (S minus P time) / 0.122, which is about 8.2 kilometres for every second of gap. So the 25 second gap on our seismogram means the earthquake was about 205 km away.

Remember: the S minus P time gives distance, not direction. One station tells you the earthquake was somewhere on a circle 205 km in radius. It cannot tell you where on that circle.

Three circles, one point

This is the whole location procedure, and you can carry it out on graph paper.

1. At each station, measure the S minus P time in seconds and multiply by 8.2 to get the distance in kilometres.
2. On a map, draw a circle around each station with that radius.
3. The point where all the circles cross is the epicentre.

Why three? One circle leaves the earthquake anywhere on a ring. Two circles cross at two points, so you are down to a choice of two, which is not an answer. The third circle passes through only one of those two candidates, and that resolves it. Three is the minimum, which is why the method is taught with three stations.

Say station A reports a gap of 25 seconds, station B 50 seconds, and station C 75 seconds. Multiply each by 8.2: the epicentre lies 205 km from A, 410 km from B and 615 km from C. Draw those three circles and they enclose a single point.

In practice a seismologist uses far more than three stations and solves for four unknowns at once: latitude, longitude, depth and the origin time. Extra stations do not just confirm the answer, they shrink the uncertainty, and the reported location comes with an error estimate because of them.

Two different questions: how big, and how bad

People use one word, but seismologists ask two questions, and confusing them causes most public misunderstanding of earthquakes.

Magnitude answers how big the earthquake was. It is one number per earthquake, the same whether you are next to the fault or on another continent. Charles Richter's 1935 scale, ML, was built for southern California earthquakes recorded on one particular instrument within 600 km, and it saturates for very large events, over-reporting some and under-reporting others. The USGS now uses moment magnitude, Mw, which is computed from physical quantities: the rigidity of the rock, the area of fault that slipped, and how far it slipped.

The scale is logarithmic in a way worth stating exactly. Each whole number of magnitude is a tenfold increase in wave amplitude on the seismogram and about a 32-fold increase in energy released. So a magnitude 7 shakes the needle ten times as far as a magnitude 6 and releases roughly 32 times the energy. Going from 5 to 7 is a hundredfold in amplitude and about a thousandfold in energy.

Intensity answers how bad the shaking was at your address. It is not one number; it is a different number in every town, and it depends on distance from the fault, the depth, the direction the rupture travelled, and above all the ground you are standing on. The Modified Mercalli scale runs in Roman numerals from I, not felt, through VI, felt by all and plaster cracked, up to X and above, most masonry destroyed. Intensity is built from observations and damage reports, not instruments alone.

Why one building falls and its neighbour does not

Three effects do most of the sorting.

• Ground. Soft sediment and landfill amplify shaking, sometimes by a factor of several, compared with the same earthquake measured on hard bedrock a kilometre away. A city built on an old lake bed or a filled harbour will always shake harder than the ridge behind it.
• Resonance. Every building has a natural period at which it sways most easily, roughly one tenth of a second per storey. A ten-storey building swings in about one second. If the ground is delivering energy at that same period, the building absorbs it efficiently and the sway grows, the way a pushed swing does. Short stiff buildings are damaged by short-period shaking, tall flexible ones by long-period shaking, and which one dies depends on the frequency content of the waves that arrive.
• Liquefaction. Loose, water-saturated sand shaken hard behaves briefly as a liquid, because the shaking pushes grains together and raises the water pressure between them until they lose contact. Buildings tilt and sink, and buried tanks float upward.

The most lethal single building type is unreinforced masonry: brick or stone walls with no steel through them, which crack, separate and drop their floors. Steel reinforcement, bolting a house to its foundation, and bracing soft ground floors are the cheap engineering that keeps buildings standing.

What to do

Official guidance is three words. Drop, Cover and Hold On. Drop onto hands and knees before the shaking drops you. Cover your head and neck with one arm and crawl under a sturdy table if one is near, or move to an interior wall away from windows. Hold on until the shaking stops.

Two instructions are counter-intuitive. Stay inside and do not run outside, because most injuries come from falling objects and from glass and masonry at the entrance to a building. And avoid doorways: the advice to stand in one comes from photographs of collapsed adobe houses where the door frame was the only thing left, which is not how a modern building behaves. If you are in bed, turn face down and cover your head and neck with a pillow. Afterwards, expect aftershocks and be ready to do it again.

Common misconceptions

• "The ground opens up and swallows people." Faults slip sideways or vertically past each other, not apart. Open fissures do form in soft ground, but a gaping chasm that closes over a victim is a film invention.
• "A magnitude 8 is twice as big as a magnitude 4." It is four whole steps, so about 10,000 times the amplitude and roughly a million times the energy.
• "Earthquakes can be predicted a few days ahead." No method has ever passed a test. Long-term forecasts of probability over decades are reliable; short-term prediction of time, place and size is not available. Early warning systems are different: they detect the P wave and send an alert seconds before the S and surface waves arrive.
• "Stand in a doorway." Current guidance says avoid doorways and get under a sturdy table instead.

What to remember

An earthquake is stored elastic strain released in seconds when friction on a fault gives way. It sends out P waves that compress rock and pass through liquid, S waves that shear rock and cannot, and surface waves that arrive last and do most of the damage. The gap between the P and S arrivals is about 8.2 km per second of delay, so one station gives a circle, two give two candidate points, and three give the epicentre. Magnitude describes the earthquake once, on a scale where each step is ten times the amplitude and about 32 times the energy; intensity describes the shaking where you are standing and changes street by street with the ground beneath it.

So what?: the earthquake you cannot control is the magnitude. The intensity at your address depends on soil, on building period and on reinforcement, and those are choices. That is why identical quakes kill hundreds in one country and tens of thousands in another.

Sources

1. U.S. Geological Survey. Earthquake magnitude, energy release, and shaking intensity. USGS Earthquake Hazards Program. Source of the tenfold amplitude and roughly 32-fold energy relationship, and of the move from ML to Mw. USGS
2. Federal Emergency Management Agency. Earthquakes. Ready.gov. Source of the Drop, Cover and Hold On guidance, the instruction to avoid doorways and not to run outside, and the aftershock advice. Ready.gov
3. Wikipedia contributors. (2026). P wave. P-wave speeds from under 6 km/s in the crust to 13.5 km/s in the lower mantle, and the fact that S waves do not travel through liquids. Wikipedia
4. Richter, C. F. (1935). An instrumental earthquake magnitude scale. Bulletin of the Seismological Society of America, 25(1), 1-32.

8:32 on a Sunday morning

At 8:32 a.m. on 18 May 1980, a magnitude 5.1 earthquake struck directly beneath the north slope of Mount St Helens. The slope, which had been bulging outward for weeks, let go. About 2.8 cubic kilometres of mountainside slid away in the largest landslide in recorded history, and the pressurised gas and magma that the slope had been holding down were suddenly uncorked sideways. The lateral blast travelled at no less than 480 kilometres per hour. Fifty-seven people died. The summit dropped from 2,950 metres to 2,549 metres, and within fifteen minutes an ash column stood 24 kilometres high.

Two thousand miles southwest, people walk to within a few metres of moving lava on Kilauea and photograph it. Both are volcanoes. The difference is chemistry, and it comes down to two numbers.

Silica: why some magma will not flow

The building block of almost all magma is the silica tetrahedron: one silicon atom surrounded by four oxygens. What matters is whether those tetrahedra are joined to each other.

In a silica-poor melt, the tetrahedra drift as separate units with iron and magnesium atoms between them, and the liquid pours. As silica content rises, tetrahedra begin sharing oxygen atoms, linking into chains, then sheets, then a three-dimensional network. A liquid whose molecules are chained together resists flowing, and that resistance is viscosity.

The range is enormous. Basalt holds between 45 and 52 percent silica and flows like warm honey. Rhyolite is the most silica-rich volcanic rock, with high-silica varieties running from 75 to 77.8 percent SiO₂, and it is so viscous that it often cannot flow at all. Temperature works the same way: rhyolitic lava erupts at a relatively cool 800 to 1,000 degrees Celsius, well below basaltic lava, and cooler means stiffer still.

Gas: the variable that decides between a flow and a blast

All magma carries dissolved gas, mostly water vapour with carbon dioxide and sulphur dioxide. Under the pressure of several kilometres of rock the gas stays dissolved. As magma rises, pressure drops, and the gas comes out of solution as bubbles. That much is like opening a bottle of fizzy drink.

The analogy breaks at exactly the point that matters. In a drink the liquid is thin, so bubbles rise and escape. In basalt the same thing happens: gas escapes continuously, sometimes spectacularly as a lava fountain, but the pressure never builds. In rhyolite the liquid is so viscous that bubbles cannot rise. They stay trapped and keep growing as pressure falls, until the foam shreds itself apart and the magma is blown out as fragments.

What matters here: an explosive eruption is not magma that contains more gas. It is magma too stiff to let the gas out gently.

The two cases side by side

Mount St Helens, in order

Magma rose into the cone through March and April 1980 without reaching the surface. It pushed the north flank outward at up to about 2 metres a day, producing a visible bulge, and it kept its gas because the rock above held the pressure on. That is the loaded condition.

Then the earthquake removed the lid. Once 2.8 cubic kilometres of rock slid off, the pressure on the magma dropped almost instantly. The dissolved gas expanded all at once, sideways, through the new opening. The blast flattened forest over some 600 square kilometres, and it moved fast enough that nobody in its path outran it. The vertical column that followed put ash 24 km up and dropped it across eleven states.

The deaths matter for how they happened. Most of the 57 were killed by the hot ash-laden blast and the gases in it, not by lava. There was essentially no lava flow in the 1980 event.

Kilauea, in order

Kilauea's 2018 activity ran from 3 May into September. Magma drained out of the summit reservoir and moved into the lower East Rift Zone, opening a line of fissures through the Leilani Estates and Lanipuna Gardens subdivisions. Lava came out, spread, and buried what was in its way. The County of Hawaii recorded 716 dwellings destroyed.

At the summit, the withdrawal of magma left the roof of the reservoir unsupported, and the Halemaumau crater floor collapsed in a series of events that greatly enlarged the crater and drained the long-lived lava lake.

The property loss was severe and the death toll was not comparable to St Helens. Fluid basalt advances slowly enough that people can leave. What it destroys is everything that cannot.

The shape of a volcano is a record of its magma

• Shield volcanoes are built from fluid basalt that runs far before freezing, so the slopes are gentle, often only a few degrees, and the volcano is broad. Kilauea and Mauna Loa are shields.
• Composite volcanoes, also called stratovolcanoes, alternate viscous lava with layers of fragments from explosive eruptions. The stiff lava piles near the vent, giving the steep, symmetrical cone people draw when asked to draw a volcano. St Helens, Fuji and Vesuvius are composites.
• Cinder cones are small, steep piles of fragments thrown from a single vent, often built in weeks and never active again.

So the profile on the horizon is evidence. A gentle dome means fluid basalt and a long history of effusive eruption; a steep cone means viscous magma and a history that includes explosions.

What actually kills people

Lava is the hazard everyone pictures and rarely the one that kills, because it is usually slower than a walking pace. The lethal hazards are these. Pyroclastic flows, avalanches of hot gas and fragments at hundreds of degrees moving at highway speeds, cannot be outrun and are the main killer at explosive volcanoes. Lahars are mudflows of ash and water, from melted snow or heavy rain, that follow river valleys for tens of kilometres and arrive with the consistency of wet concrete. Ashfall is not fluffy; it is powdered rock, and 10 centimetres of wet ash on a roof is enough to collapse it. Volcanic gases, especially carbon dioxide, are denser than air and pool in hollows.

Common misconceptions

• "Explosive eruptions happen because the magma has more gas in it." It is the viscosity, not the gas budget. Stiff magma traps the gas it has until the pressure shatters it.
• "Lava is the main danger." Fluid lava destroys property and is usually survivable on foot. Pyroclastic flows, lahars and ashfall cause the great majority of volcanic deaths.
• "Volcanoes erupt without warning." Both of these volcanoes gave weeks of earthquakes and measurable ground deformation first. What cannot yet be forecast precisely is the day and the size.
• "A volcano is a hole that goes down to the molten centre of the Earth." Magma forms in local pockets in the upper mantle and crust, mostly where water from a sinking plate lowers the melting point or where rising rock decompresses. There is no open pipe to the core.

Pulling it together

Two numbers set a volcano's behaviour. Silica content, from about 45 percent in basalt to over 75 percent in high-silica rhyolite, decides whether the tetrahedra link into a network, and therefore whether the magma pours or resists. Gas content decides how much energy is available, but only viscosity decides whether that gas leaks out quietly or accumulates until the rock fails. Fluid basalt builds broad shields and effusive eruptions, as at Kilauea, where 716 dwellings were lost in 2018 but people could walk away. Viscous magma builds steep composite cones and, as at Mount St Helens on 18 May 1980, releases its energy in minutes: a 2.8 cubic kilometre landslide, a 480 km/h lateral blast, 57 deaths, 400 metres removed from the summit and ash to 24 kilometres.

In short: a volcano's shape, its eruption style and its hazards are all downstream of one property of its magma. Measure the silica and you can predict the rest.

Sources

1. U.S. Geological Survey. Mount St. Helens. USGS Volcano Hazards Program, Cascades Volcano Observatory. USGS
2. U.S. Geological Survey. Kilauea. USGS Volcano Hazards Program, Hawaiian Volcano Observatory. USGS
3. Wikipedia contributors. (2026). 1980 eruption of Mount St. Helens. Source of the 8:32 a.m. time, magnitude 5.1, 57 deaths, 2,950 m to 2,549 m summit change, 2.8 km³ debris avalanche, 480 km/h blast, 24 km ash column and VEI 5. Wikipedia
4. Wikipedia contributors. (2026). Rhyolite and Basalt. Source of the silica ranges (basalt 45 to 52 percent, high-silica rhyolite 75 to 77.8 percent) and the 800 to 1,000 degree Celsius rhyolitic eruption temperature. Wikipedia

A claim that sounds right

At 10:37 on the morning of 22 March 2014, a hillside above the Stillaguamish River near Oso, Washington collapsed. The slide ran for about two and a half minutes, covered roughly 2.6 square kilometres, destroyed 49 structures and killed 43 people. In the 45 days before it, the area had received up to 200 percent of its normal rainfall.

So the obvious explanation is: the rain did it. That claim is not wrong so much as it stops two steps too early, and tracing exactly where it fails will teach you weathering, soil and slope stability in one pass. We will come back to Oso at the end with a better answer.

First correction: weathering is not erosion

People use the two words interchangeably, and geologists do not, because they are different processes that can happen without each other.

Weathering is the breakdown of rock in place. Nothing moves anywhere. A granite headstone becoming crumbly and illegible has weathered and has not been eroded by a single millimetre. Erosion is the removal and transport of the pieces by water, wind, ice or gravity. A river carrying sand it did not produce is eroding without weathering.

The order matters for what follows. Weathering prepares the material; erosion takes it away; and a slope fails when the material has been prepared and gravity finally wins.

Physical weathering, and the nine percent that is not the whole story

Physical weathering breaks rock into smaller pieces without changing its chemistry. The textbook mechanism is frost wedging: water seeps into a crack, freezes, and expands. That expansion is a real number, nine percent by volume, and ice growing in a confined crack at minus 22 degrees Celsius can generate pressures up to 207 megapascals, enough to break any rock.

Here is the correction that most textbooks skip. Simple volumetric expansion was the accepted explanation until the 1980s, when researchers pointed out that it requires the crack to be almost completely sealed and almost completely full, which is unusual. The better model is ice segregation: unfrozen water is drawn through the rock toward a growing ice lens, which then grows and prises the rock apart, rather like frost heave lifting a road. The distinction matters because it predicts where frost damage actually happens, which is in damp rock at temperatures a little below freezing, roughly minus 3 to minus 8 degrees Celsius, rather than in the deepest cold.

Three other physical mechanisms matter. Exfoliation: a granite body formed under kilometres of rock is unloaded by erosion, expands slightly, and sheets off in curved slabs like the layers of an onion. Salt crystal growth: salt water enters pores, evaporates, and growing crystals push the grains apart, which is why coastal and desert stonework crumbles. Root wedging and burrowing: a tree root in a crack exerts steady pressure for decades.

All of them do one thing that chemistry cares about. Split a cube of rock once in each direction and you get eight cubes with twice the total surface area. Physical weathering does not change the chemistry, but it multiplies the surface on which chemistry can act.

Chemical weathering: the reactions that destroy a mineral

Three reactions do most of the work.

• Dissolution. Rain absorbs carbon dioxide and becomes weak carbonic acid. Carbonic acid dissolves calcite, so limestone landscapes develop sinkholes, caves and fluted rock. This is the same reaction as the vinegar test from the first lesson, run for ten thousand years.
• Hydrolysis. Water attacks feldspar and converts it to clay minerals, releasing potassium, sodium or calcium ions into solution along with dissolved silica. This is the single most important weathering reaction on the planet, because feldspar is the most abundant mineral in the crust and clay is what most soil is made of.
• Oxidation. Oxygen reacts with iron in minerals such as olivine and pyroxene to form iron oxides. The red of a desert cliff and the rust brown of a weathered boulder are the same chemistry as a rusting nail.

Climate controls the rate, and the control is strong. Chemical reactions need water and speed up with temperature, so weathering is fastest in the hot wet tropics and slowest in cold deserts. A limestone gravestone in Scotland and one in Singapore, cut the same year, will not be in the same condition.

Soil is a structure, not a pile of dirt

Dig a metre down at the edge of a field and you will find layers, called horizons, that developed in place over centuries.

Five factors decide what soil you get: climate, organisms, relief (the slope and its drainage), parent material, and time. Change any one and the profile changes. The same granite gives a thin acidic soil in a cold wet upland and a deep red clay-rich soil in the tropics.

Worth holding on to: soil forms far more slowly than it can be removed. A profile that took centuries to build can be stripped by one season of bare ploughed ground on a slope, which is why contour ploughing, cover crops and terracing exist.

Second correction: slopes do not fail because they are steep

The plausible claim is that a slope fails when it gets too steep. The better statement is a ratio. Every slope has driving forces, essentially the weight of the material resolved down the slope, and resisting forces, essentially friction between grains plus any cohesion from clay and roots. The slope stands while resisting beats driving and fails when it does not.

That framing explains something the steepness rule cannot. Pour dry sand and it builds a cone at a fixed side angle, its angle of repose, about 34 degrees. Damp sand will stand at about 45 degrees, steeper, because surface tension in the water films pulls grains together and adds cohesion. Fill the pore spaces completely, though, and the angle collapses to somewhere between 15 and 30 degrees, because now the water carries part of the load and pushes the grains apart instead of pulling them together.

So water does two opposite things depending on how much there is, and the failure comes at the wet end, not the damp end.

Four ways a slope moves

• Creep. Millimetres a year, invisible while it happens, and detected by its results: fence posts leaning downhill, tree trunks curved at the base, walls bowed outward.
• Slump. A block slides along a curved surface and rotates backward, leaving a crescent-shaped scar at the top and a bulging toe at the bottom.
• Debris flow and mudflow. Saturated material behaving as a fluid, following valleys, moving at anything from walking pace to tens of metres per second.
• Rockfall. Free fall from a cliff, building a talus cone at the base that stands at its own angle of repose.

Back to Oso: what the rain actually did

Now the better answer. The hillside at Oso was not sound ground that rain destroyed. It was glacial sediment, sand and clay left by ice, that had been weathering and moving for a long time. The same slope, known as the Hazel landslide, had documented failures in 1937, 1951, 1952, 1967, 1988 and 2006. The resisting forces there had been losing slowly for at least 77 years.

What 45 days of rainfall at up to 200 percent of normal did was fill the pore spaces. Saturated sediment is heavier, which raises the driving force, and its grains are pushed apart by pore-water pressure, which cuts the resisting force. Both sides of the ratio moved the wrong way at once. The deposit that resulted was about 460 metres long and 1,300 metres wide, 9 to 21 metres deep.

The rain was the trigger. The cause was decades of weathering and previous movement in weak glacial material on a slope that had already announced, six times, what it was going to do.

Common misconceptions

• "Weathering and erosion are the same thing." Weathering breaks rock where it sits; erosion carries the pieces away. A crumbling gravestone is weathered and not eroded.
• "Rain caused the landslide." Rain is usually the trigger, not the cause. The cause is a long-running loss of strength in the material, which is why the same storm drops no other hillside in the county.
• "Wet ground is always weaker than dry ground." Damp sand stands steeper than dry sand, at roughly 45 degrees against 34, because water films pull grains together. It is full saturation that destroys the strength.
• "Soil is just crushed rock." Soil is a layered structure of weathered minerals, organic matter, water, air and living organisms, developed over centuries, with distinct horizons you can see in any road cut.

Summing up

Rock is broken in place by frost, unloading, salt and roots, and is chemically taken apart by carbonic acid dissolving carbonates, by hydrolysis turning feldspar into clay, and by oxygen rusting iron-bearing minerals. Physical and chemical weathering are partners: breaking rock into pieces multiplies the surface available for reaction. What survives, mixed with organic matter and sorted by percolating water, becomes soil with horizons that record the process. A slope holds while friction and cohesion beat the downslope component of weight, and it fails when saturation adds weight and removes friction at the same time.

The point: triggers are easy to see and causes take longer to accumulate. When you are told what made a hillside fail on a particular Saturday morning, ask what had been happening to that hillside for the previous seventy years.

Sources

1. Wikipedia contributors. (2026). 2014 Oso mudslide. Source of the 22 March 2014 date and 10:37 time, 43 deaths, 49 structures, 2.6 km² area, deposit dimensions, the 45 days at up to 200 percent of normal rainfall, and the earlier failures in 1937, 1951, 1952, 1967, 1988 and 2006. Wikipedia
2. Wikipedia contributors. (2026). Frost weathering. Source of the nine percent expansion, the 207 MPa figure at minus 22 degrees Celsius, the shift from volumetric expansion to ice segregation, and the effective range of minus 3 to minus 8 degrees Celsius. Wikipedia
3. Wikipedia contributors. (2026). Angle of repose. Source of the angles for dry sand (34 degrees), wet sand (45 degrees) and water-filled sand (15 to 30 degrees). Wikipedia
4. Earle, S. (2019). Weathering and soil. In Physical Geology (2nd ed., Chapter 5). BCcampus, via Geosciences LibreTexts. Geosciences LibreTexts

A question at the mortgage desk

You are buying a house. The flood map shows it sits just inside the hundred-year floodplain, and the lender wants insurance. You are told this means a flood of that size happens roughly once a century, and the last one was in 2019, so you reason that you are probably fine for a while.

Every step of that reasoning is wrong, and the arithmetic that shows why takes about a minute. Then we will work out where the number came from in the first place, and what the river was doing while the statisticians argued about it.

Working the probability

A hundred-year flood is not a schedule. The US Geological Survey defines it as a flood of a size with a 1 percent chance of happening in any given year. The phrase "hundred-year" is a simplification of that probability, and the USGS says plainly that hundred-year floods can happen two years in a row.

So work with the 1 percent. The chance of no such flood in one year is 0.99. Assuming years are independent, the chance of getting through 30 years untouched is 0.99 raised to the power 30, which is 0.74. The chance of at least one hundred-year flood during a 30-year mortgage is therefore:

1 - 0.9930 = 1 - 0.74 = 0.26, or about 26 percent

Roughly one chance in four. The same calculation gives about 10 percent over a decade and about 55 percent over an 80-year lifetime. And over any particular century, the odds break down as a 37 percent chance of no hundred-year flood at all, a 37 percent chance of exactly one, and a 26 percent chance of two or more.

The core of it: "hundred-year flood" describes a size, not a timetable. The river has no memory of 2019 and no obligation to wait.

Where the number comes from, and how much to trust it

The estimate is built from the record of annual peak discharge at a gauging station: the single highest flow of each year, one number per year. Rank those peaks from largest to smallest. In n years of record, the flood ranked m has an estimated recurrence interval of about (n + 1) / m years. Ninety years of record and the largest flood on the list gives about 91 years, so that event is called roughly a ninety-year flood.

Three honest limitations follow, and they are the reason flood maps get redrawn.

• You cannot estimate a hundred-year flood confidently from thirty years of data. You are extrapolating past the end of the record.
• The method assumes the river behaves the same way over time. Pave a catchment with roofs and roads and rain that used to soak in now runs straight to the channel, so the same storm produces a bigger peak. The old statistics describe a river that no longer exists.
• A single flood is not one recurrence interval. A flood peak is a different rarity at each point along the river, and news reports quote the largest figure from anywhere in the basin.

Discharge: the number a river is actually measured by

Discharge is the volume of water passing a point per second, and it is the product of the cross-sectional area of the flow and the average velocity:

Q = width x depth x velocity, in cubic metres per second

Take a stream 12 metres wide and 1.5 metres deep flowing at 0.8 metres per second. Q = 12 x 1.5 x 0.8 = 14.4 m3/s. Now put it in flood: it spreads to 30 metres wide, deepens to 3.5 metres, and speeds up to 2.5 metres per second. Q = 30 x 3.5 x 2.5 = 262.5 m3/s.

That is more than eighteen times the ordinary flow, and notice that all three factors grew at once. This is why rivers do most of their geological work in a handful of days per decade. A river at normal flow moves sand; the same river in flood rolls boulders.

How a river carries its load

• Dissolved load: ions in solution, invisible, mostly from chemical weathering upstream. It keeps moving even in still water.
• Suspended load: clay and silt held up by turbulence. This is what makes a flooding river brown.
• Bed load: sand, gravel and cobbles rolling, sliding and bouncing along the bottom. You can hear it as a grinding noise in a stream in flood.

Velocity decides which of these a river can carry. Slow the water and the largest particles drop first, then the smaller ones, which is why a flood leaves gravel near the channel and mud at the far edge of the floodplain.

Why rivers refuse to stay straight

Watch water go round a bend. On the outside of the curve the water has further to travel and is thrown outward, so it flows faster and deeper. On the inside it is slower and shallower. Fast water erodes and slow water deposits, so every bend attacks its outer bank and builds up its inner one.

That produces the two features you can identify from a bridge. The cut bank on the outside is steep, often undercut, and made of whatever the river is currently eating into. The point bar on the inside is a gentle shelf of fresh sand and gravel.

The consequence is that the bend gets bigger and migrates. Over years a meander swings sideways and downstream, and eventually the neck of a loop narrows until a flood cuts straight across it. The river abandons the loop, seals the ends with sediment, and leaves an oxbow lake: a curved pond with no river in it, which is a meander caught in the act of being deleted.

A perfectly straight channel is unstable for the same reason. Any slight irregularity sends the fastest thread of water to one side, that side erodes, and the bend grows itself.

Reading the valley

The shape of a valley tells you which process is winning.

The last row is worth dwelling on. Meanders form on flat ground; gorges form on steep ground. Finding meanders inside a gorge means the ground changed underneath a river that had already made up its mind, which is the standard evidence for uplift. The lowest level a river can cut down to is its base level, usually sea level, and when base level falls the whole river begins cutting downward again from the mouth upstream.

Common misconceptions

• "A hundred-year flood happens once a century." It has a 1 percent chance each year. Two can happen in consecutive years, and there is a 26 percent chance of two or more in any given century.
• "Last year's big flood used up the risk." The annual probability does not change because of last year. There is no reservoir of bad luck being drained.
• "Rivers erode most during ordinary flow, because that is most of the time." Most of the sediment moves in a few days of high flow. Discharge in our worked example rose eighteenfold in flood, and the size of particle the river can move rose with it.
• "Oxbow lakes were dug out by the river." They are abandoned meander loops, sealed off when the river cut through the neck and took the shorter route.

What you now know

A recurrence interval is a probability in disguise: one percent a year, 26 percent over a mortgage, 55 percent over a lifetime, and no memory of last year. The number is estimated by ranking annual peak discharges, so it is only as good as the record length and only valid while the catchment stays the same. Discharge itself is width times depth times velocity, and in flood all three rise together, which is when a river does nearly all of its cutting and carrying. Bends are self-amplifying because the outside of a curve runs faster, so rivers meander, migrate and cut off oxbow lakes, and the shape of the valley around them records whether the river is cutting down toward base level or spreading sideways across its own floodplain.

Remember: the flood map is a statement about probability that was calculated from a particular stretch of history. When the catchment or the climate changes, the map is describing a river that is no longer there.

Sources

1. U.S. Geological Survey. The 100-year flood. USGS Water Science School. Source of the 1 percent annual probability definition and the statement that hundred-year floods can happen two years in a row. USGS
2. Wikipedia contributors. (2026). 100-year flood. Source of the 10 percent per decade, 26 percent per 30-year mortgage and 55 percent per 80-year lifetime figures, and the 37/37/26 breakdown for a century. Wikipedia
3. Earle, S. (2019). Streams and floods. In Physical Geology (2nd ed., Chapter 13). BCcampus, via Geosciences LibreTexts. Geosciences LibreTexts
4. U.S. Geological Survey. National Water Information System: current conditions. Real-time gauge discharge and annual peak-flow records. USGS Water Data

A granite boulder in a limestone county

In parts of the American Midwest you can find a granite boulder the size of a small car sitting in a ploughed field, with every scrap of bedrock for hundreds of kilometres around it made of limestone and shale. Nothing in the local geology could have produced it. A river could not have rolled it that far. It is the wrong rock in the wrong place, and geologists call it a glacial erratic.

The boulder is a delivery note. Something picked it up off the Canadian Shield, carried it a thousand kilometres south, and set it down. This lesson follows that something from the first snowflake to the last melt.

Snow becomes rock that flows

A glacier begins where more snow falls in winter than melts in summer, so a surplus survives year after year. Fresh snow is about 90 percent air. Buried under the next year's fall, the crystals compact, the air is squeezed out, and the snow turns to granular firn. Under more years of burial the firn recrystallises into dense glacial ice, and remember from the first lesson that ice formed naturally this way is a genuine mineral. A glacier is a rock made of one mineral.

At a thickness of roughly 50 metres the ice at the bottom is under enough pressure to deform rather than fracture, and the whole mass begins to move downhill under its own weight. That threshold is the difference between a snowfield, which sits, and a glacier, which flows.

Every glacier has two zones. High up is the zone of accumulation, where the glacier gains more than it loses in a year. Low down is the zone of ablation, where melting and evaporation win. The line between them is the equilibrium line, and it is visible at the end of summer as the boundary between clean snow above and grey bare ice below. If accumulation exceeds ablation over several years, the terminus advances. If ablation wins, the terminus retreats, even though the ice inside the glacier keeps flowing downhill the whole time. A retreating glacier is not a glacier flowing backwards; it is a conveyor belt that is melting faster than it delivers.

Two ways ice moves, and one way it cracks

Ice travels by internal deformation, in which crystals slide over one another deep in the glacier, and by basal sliding, in which the whole mass skids on a film of meltwater at the bed. Basal sliding needs the base to be at the melting point, so it matters far more in temperate mountain glaciers than in the coldest polar ice.

The top 50 metres or so is too lightly loaded to deform. It is brittle, so it rides along on top of the flowing ice below and splits, which is exactly what crevasses are. Their depth is set by that brittle-to-ductile transition, which is why crevasses are dangerous and not bottomless.

What the ice does on the way down

Glaciers erode by two mechanisms. In plucking, meltwater seeps into cracks in the bedrock, refreezes onto the glacier, and the moving ice rips the block out. In abrasion, the rock fragments already frozen into the base are dragged across the bed like teeth on a rasp, polishing it and cutting parallel grooves called striations. Striations are a compass: they point in the direction the ice was travelling, and geologists mapped the ice sheets partly by reading them.

The shapes that result are unmistakable once you know them.

• Cirque: a steep-walled armchair hollow at the head of a glacier, where plucking eats backwards into the mountain.
• Arete and horn: when cirques on two sides of a ridge meet, they leave a knife-edge arete; when three or more meet, they leave a pyramid peak, a horn. The Matterhorn is the famous example.
• U-shaped valley: a river cuts a narrow V because it erodes at a point. A glacier fills the whole valley and erodes across its width, leaving a flat floor and steep walls.
• Hanging valley: a small tributary glacier cannot cut as deep as the trunk glacier, so when the ice goes its valley is left stranded above the main floor, and a waterfall drops out of it.
• Fjord: a glacial trough cut below sea level and flooded when the ice melted.

What the ice drops, and how to tell it apart

There is one test that separates glacial deposits from everything else: sorting. Water sorts sediment by size, because slowing water drops the big pieces first. Ice does not sort at all, because ice does not slow down selectively. It carries a boulder and a clay particle side by side and releases both at once.

So till, laid down directly by ice, is an unsorted jumble of every grain size mixed together. Outwash, laid down by the meltwater streams beyond the ice, is well sorted and layered like any river deposit. Standing in a gravel pit, that single distinction tells you whether you are looking at the work of the glacier or of its meltwater.

The last ice age, in numbers

The Last Glacial Maximum ran from roughly 26,500 to 20,000 years ago. The Laurentide Ice Sheet covered essentially all of Canada east of the Rockies and reached south roughly to the Missouri and Ohio Rivers and east to Manhattan. In Europe the southern edge of the ice ran through Germany and Poland, with permafrost beyond it as far south as southern Hungary.

All that water came from the ocean. About 21,000 years ago, sea level stood roughly 125 metres lower than it does now. Continental shelves that are seabed today were dry plains, which is why fishing boats trawl up mammoth teeth from the North Sea.

The ice sheets that are left

An ice sheet is glacial land ice covering more than 50,000 square kilometres, and only two survive.

Those last two numbers are the reason ice sheets appear in every climate report. They are the largest single stores of fresh water on the planet, and the ocean is where that water came from.

The land is still coming back up

Three kilometres of ice is a serious weight, and it pressed the crust down into the mantle beneath it. When the ice melted, the crust began rising again, slowly, because mantle rock has to flow back in underneath. This is post-glacial rebound, and it is still happening twenty thousand years later.

The northern Gulf of Bothnia is rising at a peak rate of about 11 millimetres a year. Finland gains roughly seven square kilometres of new land every year as its coast lifts out of the Baltic. Studies suggest the rebound will continue for at least another 10,000 years. Old harbours in Sweden now sit inland, and the relevant paperwork about who owns the new land is a live legal question.

Back to the boulder

Now the delivery note reads. The boulder was plucked from Canadian Shield granite, frozen into the base of the Laurentide Ice Sheet, and dragged south for perhaps a thousand kilometres, scratching striations into the bedrock as it went. When the ice front stalled, the boulder was released with everything else the ice was carrying, into unsorted till. Meltwater streams beyond the front sorted the finer material into outwash plains, buried ice blocks melted out to leave kettle ponds, and the ground beneath the whole system began, slowly, to rise.

So what?: the Great Lakes, the soils of the American Midwest, the fjords of Norway and the shape of half the northern hemisphere's coastline are all the same event. The boulder in the field is the part of it you can put your hand on.

Common misconceptions

• "A retreating glacier flows uphill." The ice always flows downhill. Retreat means the terminus is melting back faster than the ice is delivered to it.
• "Glacial and river deposits look the same." Ice deposits are unsorted, with boulders and clay mixed together. Water deposits are sorted and layered by size. It is the most reliable field test there is.
• "Crevasses are bottomless." They form only in the brittle upper ice, roughly the top 50 metres. Below that, pressure makes the ice deform instead of crack.
• "The land under an ice sheet is permanently crushed." It springs back. Scandinavia is rising at about 11 mm a year and will keep rising for at least another 10,000 years.

Looking back

Snow that survives the summer is compacted into firn and then into glacial ice, and at about 50 metres thick it begins to flow. It gains mass above the equilibrium line and loses it below, and the terminus advances or retreats depending on which wins, while the ice inside always travels downhill. On the way it plucks blocks out of bedrock and abrades what is left, carving cirques, aretes, horns, U-shaped valleys, hanging valleys and fjords, and scratching striations that record its direction. It deposits unsorted till as moraines and drumlins, while its meltwater deposits sorted outwash as eskers and plains. At the Last Glacial Maximum, 26,500 to 20,000 years ago, ice reached Manhattan and the Ohio River and sea level was about 125 metres lower. What remains is Antarctica, with 58 metres of sea level locked up, and Greenland with 7.4 metres, and the crust beneath the old ice is still rising at about a centimetre a year.

Sources

1. National Snow and Ice Data Center. Ice sheets. NSIDC. Source of the 50,000 km² definition, the Antarctic figures (almost 14 million km², nearly 4.9 km thick, about 58 m of sea level) and the Greenland figures (1.7 million km², over 3 km thick, about 7.4 m). NSIDC
2. Wikipedia contributors. (2026). Last Glacial Maximum. Source of the 26,500 to 20,000 years ago range, the sea level about 125 m lower at 21,000 years ago, and the ice limits at the Missouri and Ohio Rivers, Manhattan, Germany and Poland. Wikipedia
3. Wikipedia contributors. (2026). Post-glacial rebound. Source of the peak rate of about 11 mm per year in the northern Gulf of Bothnia, the roughly seven square kilometres per year gained by Finland, and the expectation of at least 10,000 further years of rebound. Wikipedia
4. National Snow and Ice Data Center. Glaciers. NSIDC. NSIDC

A well in western Kansas

The High Plains aquifer underlies about 174,000 square miles across eight states, from South Dakota to Texas, and it is the reason the southern Great Plains grow corn. At the height of extraction, the water table under parts of it was measured dropping more than 5 feet, about 1.5 metres, in a single year. Since 1950, irrigation has removed an estimated 9 percent of the aquifer's saturated volume, an overdraft of 332,000,000 acre-feet, which is 410 cubic kilometres, or about 85 percent of the water in Lake Erie.

Whether a well runs dry is not a mystery. It is a subtraction, and this lesson works it twice: once for an aquifer that holds, and once where a single changed input flips the answer.

The inventory: where the water actually is

There are about 332.5 million cubic miles of water on, in and above Earth. Almost none of it is available to you.

Read the last row again. Every river on the planet, at any instant, holds six thousandths of one percent of the fresh water. Groundwater holds five thousand times as much. The visible part of the water cycle is the smallest part of it.

Porosity and permeability are not the same property

Porosity is the percentage of a rock's volume that is empty space. Permeability is how well those spaces connect, which is what decides whether water can travel through.

Clay is the example that makes the distinction stick. Clay can have a porosity of 50 percent, higher than most sandstone, and it is nearly impermeable, because the pores are microscopic and water clings to the enormous surface area inside them. Sand and gravel have lower porosity and enormous permeability. So clay is a soaking sponge that will not let go, and gravel is a sieve.

A rock permeable enough to supply water to a well is an aquifer; an impermeable layer such as clay or shale is an aquitard. The two together build every groundwater system there is.

The water table, and the two kinds of aquifer

Dig down in most places and you pass through the zone of aeration, where pores hold both air and water, and then reach the zone of saturation, where every pore is full. The boundary is the water table. It is not flat. It is a subdued copy of the land surface above it, higher under hills and lower under valleys, and where it intersects the ground you get a spring, a lake or a river. That last point matters: in dry weather a river is not fed by rain, it is fed by groundwater leaking in through its bed.

An unconfined aquifer has its water table open to the air above, so it recharges directly from rain. A confined aquifer is sandwiched between aquitards and recharges only where it outcrops, often far away and much higher. Because that recharge area is elevated, the water in a confined aquifer is under pressure, and a well drilled into one can flow without pumping. That is an artesian well.

Procedure: the water budget, worked

The method is a subtraction, done in consistent units.

1. Recharge in, per year, in cubic metres.
2. Withdrawals out, per year, in cubic metres.
3. Subtract. A negative balance is being taken out of storage.
4. Convert the deficit into a drop in the water table using the drainable porosity.

Case one. An irrigation district sits on 100 square kilometres of aquifer. Average recharge is 50 millimetres per year.

Recharge = 0.050 m x 100,000,000 m2 = 5,000,000 m3 per year

Six hundred farms each pump 8,000 cubic metres a year.

Withdrawal = 600 x 8,000 = 4,800,000 m3 per year

The balance is positive by 200,000 cubic metres a year. Storage is very slightly increasing. This district is living within its income, with almost no margin.

Case two: change one input. A drought cuts recharge to 20 millimetres a year, and because it is dry the farms irrigate 25 percent more.

Recharge = 0.020 x 100,000,000 = 2,000,000 m3
Withdrawal = 4,800,000 x 1.25 = 6,000,000 m3
Balance = 2,000,000 - 6,000,000 = -4,000,000 m3 per year

Now convert that to a measurable fall. Take the drainable porosity of the sand and gravel as 0.15, meaning each cubic metre of aquifer yields 0.15 cubic metres of water when drained. The volume of aquifer that must be emptied is:

4,000,000 / 0.15 = 26,700,000 m3 of aquifer

Spread over 100,000,000 square metres, that is a water-table drop of 0.27 metres, about 27 centimetres, in one year.

Notice what flipped the answer. Demand rose 25 percent and recharge fell by 60 percent, and the two together turned a sustainable district into one losing a quarter of a metre a year. Drought does not cut supply and leave demand alone; it moves both in the wrong direction at once.

How fast does groundwater move?

Slowly, and the number is worth having. Darcy's law says the flow depends on how permeable the material is and how steeply the water table slopes:

velocity = K x hydraulic gradient

Take a sand and gravel aquifer with a hydraulic conductivity K of 10 metres per day, and a water table that drops 2 metres over 1,000 metres, a gradient of 0.002.

10 x 0.002 = 0.02 metres per day

That is the flow averaged over the whole cross-section. The water is only travelling through the pores, so divide by the porosity, say 0.25, to get the speed of an actual water molecule: 0.02 / 0.25 = 0.08 metres per day, about 29 metres a year.

Bottom line: a contaminant spilled on the ground today may take a human lifetime to travel a kilometre underground. That cuts both ways. Pollution arrives long after the spill, and it keeps arriving long after the spill is cleaned up.

What over-pumping does

• Cone of depression. Pumping lowers the water table into a funnel around the well. A deep well next door can lower your shallower one below its intake and dry it out without taking a drop from you directly.
• Subsidence. Removing water from between grains lets the sediment compact, and the ground surface sinks. Parts of California's Central Valley have dropped by metres. Compaction of clay is largely irreversible, so that storage does not come back even if the water does.
• Saltwater intrusion. On a coast, fresh groundwater floats on denser salt water. Pump the fresh water down and the salt water moves inland and upward into the wells.
• Lost streamflow. Pumping near a river intercepts the groundwater that was feeding the river, so the river runs lower in dry weather.

Karst: where the aquifer is a cave system

Where limestone dominates, carbonic acid dissolves the rock along joints and turns them into conduits and caves. This is karst, and it breaks the slow-flow rule. Water moves through open channels at stream speed instead of seeping through pores, sinkholes swallow surface streams whole, and there is almost no filtration. Contamination entering a karst aquifer can appear in a spring kilometres away within days.

Common misconceptions

• "Groundwater flows in underground rivers." Except in karst, it seeps through pores and fractures at centimetres per day. The image of a river under your feet is wrong nearly everywhere.
• "Rain refills an aquifer as fast as we pump it." Recharge across the High Plains aquifer ranges from 0.024 inches a year in parts of Texas and New Mexico to 6 inches in south-central Kansas, and the aquifer would take over 6,000 years to refill naturally once depleted.
• "Clay is impermeable because it has no pore space." Clay can be 50 percent pore space. The pores are simply too small and too poorly connected for water to travel through.
• "A river in a drought is running on recent rain." In dry weather rivers are fed by groundwater leaking in through the bed. That is why pumping near a river lowers the river.

The takeaway

Of all Earth's water, 96.54 percent is saline and 2.5 percent is fresh, and of that fresh water 68.7 percent is locked in ice and 30.1 percent is underground, leaving rivers with six thousandths of one percent. Groundwater lives in the pore space of aquifers, and what matters is not how much space there is but how well it connects, which is why clay holds water and gravel delivers it. A water budget is recharge minus withdrawal, and dividing the deficit by the drainable porosity converts it into a fall in the water table you can measure with a tape. Darcy's law puts the speed of that water at centimetres a day in ordinary sediment, so aquifers respond slowly to everything, including our attempts to fix them.

What matters here: groundwater is a bank account, not a spring. The deposits are measured in millimetres a year and the withdrawals in metres, and the statement arrives decades late.

Sources

1. U.S. Geological Survey. Where is Earth's water? USGS Water Science School. Source of the 332.5 million cubic miles total, 96.54 percent saline, 2.5 percent fresh, and the shares for ice (68.7 percent), groundwater (30.1 percent), lakes, soil moisture, atmosphere and rivers. USGS
2. U.S. Geological Survey. High Plains aquifer. USGS Water Resources. Source of the 174,000 square mile extent and the eight states. USGS
3. Wikipedia contributors. (2026). Ogallala Aquifer. Source of the 9 percent reduction in saturated volume since 1950, the 332,000,000 acre-feet (410 km³) overdraft compared with Lake Erie, the recharge range of 0.024 to 6 inches per year, the 5 feet per year maximum drawdown, and the 6,000 year natural refill estimate. Wikipedia
4. U.S. Geological Survey. The water cycle. USGS Water Science School. USGS

Two high tides, and a question that breaks the usual answer

At Burntcoat Head in the Bay of Fundy, the water level between low tide and high tide has been measured through a range of 16.3 metres, recorded in November 1998, with a highest predicted extreme of 17 metres. That is a five-storey building of water arriving and leaving twice a day.

Twice. That word is the problem. The usual explanation is that the Moon's gravity pulls the ocean toward it, and if that were the whole story there would be one high tide a day, on the side facing the Moon. There are two, and the second one is on the side facing away from the Moon. Explaining that second bulge takes three sentences of physics and explains almost everything else about tides once you have it.

Working the tide from the Moon

Gravity weakens with distance. The Moon therefore pulls the near side of Earth slightly harder than it pulls Earth's centre, and pulls Earth's centre slightly harder than it pulls the far side. Three different pulls on three parts of one planet.

Now watch from Earth's point of view, which moves with the centre. Relative to the centre, the near-side water is pulled toward the Moon, so it bulges toward the Moon. Relative to the centre, the far-side water is pulled less than the centre is, so it is left behind, which from Earth's frame looks exactly like being pushed outward. Two bulges, on opposite sides, produced by one difference in pull.

The rest is bookkeeping. Earth rotates through both bulges each day, which is why high tides are semidiurnal with a period of about 12 hours and 25.2 minutes. That extra 25 minutes exists because the Moon has moved along its orbit while you rotated, so a lunar day, the time for the Moon to return to the same place in your sky, is about 24 hours and 50 minutes rather than 24 hours. The tide is late by about 50 minutes a day, and tide tables are built on that slippage.

Why the Sun matters, but less

The Sun is 27 million times more massive than the Moon, so the intuition is that it should dominate the tides. It does not, and the reason is that tides depend on the difference in gravitational pull across Earth's diameter, not on the pull itself. That difference falls off much faster with distance than gravity does, and the Sun is 400 times further away. The Moon's tidal effect ends up roughly 46 percent larger than the Sun's.

Both still act, and they add or cancel depending on geometry:

• Spring tides occur at new moon and full moon, when Sun, Moon and Earth line up and the two effects reinforce. The range is largest; at these times the Moon contributes about 69 percent of the effect and the Sun about 31 percent. Nothing to do with the season, despite the name.
• Neap tides occur at first and third quarter, when the Sun and Moon are 90 degrees apart and partly cancel. The range is smallest.

So the tidal range at your beach has a rhythm of roughly two weeks, tracking the phase of the Moon, and you can predict it by looking up.

Why the Bay of Fundy, and not everywhere

The tide-raising force is nearly the same everywhere, yet ranges run from a few centimetres in the Mediterranean to 16 metres in Fundy. The difference is the shape of the basin. Fundy narrows and shallows toward its head, funnelling the same volume of water into a smaller cross-section, and the natural sloshing period of the bay is close to the tidal period, so each tide arrives in step with the water already moving. That is tidal resonance, the same effect as pushing a swing at the right moment, and it is why Fundy averages about 16 metres while the open ocean tide is under a metre.

The shape of the sea floor

The ocean covers 70.8 percent of Earth's surface, some 361 million square kilometres, and holds 97 percent of Earth's water. Its average depth is 3,688 metres, and the deepest measured point, the Challenger Deep in the Mariana Trench, was revised by a 2021 study to 10,935 metres plus or minus 6 metres.

Leave any coast and you cross the same sequence. The continental shelf is a gently sloping drowned edge of the continent, flooded when the ice melted, where most fishing happens. At the shelf break the continental slope drops steeply, often cut by submarine canyons. The continental rise is the apron of sediment at its base. Beyond that the abyssal plain is the flattest surface on Earth, sediment blanketing old ocean crust. Rising from it is the mid-ocean ridge from the tectonics lesson, and plunging below it, at subduction zones, are the trenches.

Waves: what moves is not the water

Watch a gull sitting on a swell. The wave passes; the gull goes up, forward, down and back, and ends where it started. Water in a wind wave moves in circles, not forward. What travels is the energy.

Those circles shrink with depth and become negligible at about half the wavelength down. That depth is the wave base, and it explains two things at once. A submarine below wave base sits in calm water during a storm. And when a wave runs into water shallower than its wave base, the bottom of the circle drags on the sea floor. The wave slows, shortens and steepens, the crest outruns the trough beneath it, and the wave breaks. Surf is the geometry of a circle meeting a sloping floor.

Three things set how big waves get: wind speed, how long the wind blows, and fetch, the distance of open water it blows across. A gale on a pond makes ripples because the fetch is tiny.

Surface currents: wind, turned by rotation

Wind dragging on the sea surface drives currents, but they do not travel in the direction of the wind, because Earth is rotating underneath them. The Coriolis effect deflects moving fluids to the right in the northern hemisphere and to the left in the southern. Combine that with continents in the way and the surface ocean organises into great loops called gyres, turning clockwise in the north and anticlockwise in the south.

The western edge of each gyre carries a fast, narrow, warm current. The Gulf Stream is the Atlantic's, and it transports water at a rate of 30 million cubic metres per second through the Florida Straits, rising to about 150 million cubic metres per second south of Newfoundland. For comparison, every river on Earth combined delivers a small fraction of that.

Where wind pushes surface water away from a coast, deeper water rises to replace it. This upwelling brings nutrients from below into the sunlit layer, which is why a handful of upwelling coasts, off Peru, California, northwest Africa and Namibia, produce a share of the world's fish catch out of all proportion to their area.

Deep circulation runs on temperature and salt

Below the wind-driven layer, water moves because of density, and density depends on temperature and salinity. Cold water is denser; saltier water is denser. In the North Atlantic near Greenland, surface water is chilled, and sea ice formation leaves its salt behind in the water below, so the water becomes both cold and salty. It sinks, and flows south along the ocean floor.

That sinking is one end of a global circulation linking all the ocean basins, and its timescale is measured in centuries. A water molecule that sinks off Greenland may not see daylight again for hundreds of years. This is also the ocean's memory: the deep ocean stores heat and carbon dioxide on timescales far longer than the atmosphere's, which is why it appears again in the climate lesson.

Common misconceptions

• "There should be one high tide a day, on the side facing the Moon." There are two, because tides come from the difference in pull across Earth. The far side is pulled less than the centre and is left behind, producing a second bulge.
• "Waves carry water toward the shore." Water moves in closed circles; the wave carries energy. Only when the circles hit the bottom and break does water actually move forward.
• "Spring tides happen in spring." They happen twice a month, at new and full moon, in every month of the year.
• "The Sun barely affects tides because it is far away." It affects them substantially, contributing about 31 percent of the effect at spring tide. It is smaller than the Moon's contribution because tidal force depends on the difference in pull across a diameter, which falls away with distance much faster than gravity itself.

Recap

Tides are a consequence of differential gravity, which is why there are two bulges and two high tides in every 12 hours 25 minutes, sliding about 50 minutes later each day as the Moon moves on. The Moon dominates, with roughly 46 percent more tidal effect than the Sun, and the two combine into spring tides at new and full moon and cancel into neap tides at the quarters. Local range is set by basin shape, which is how the Bay of Fundy reaches 16 metres. The ocean itself covers 70.8 percent of Earth at an average depth of 3,688 metres, waves move energy through circling water down to a wave base of half a wavelength, wind and Coriolis drive gyres whose western boundary currents carry 30 million cubic metres a second past Florida, and beneath all of it density-driven circulation turns over on a timescale of centuries.

Why this matters: everything in this lesson follows from differences, not absolutes. A difference in gravitational pull makes the tide, a difference in density drives the deep circulation, and a difference in pressure drives the wind that raises the waves.

Sources

1. Wikipedia contributors. (2026). Tide. Source of the 12 hours 25.2 minute semidiurnal period, the 24 hour 50 minute lunar day, the Moon's tidal force being about 46 percent larger than the Sun's, the 69 to 31 percent split at spring tide, and the 16.3 m Burntcoat Head measurement of November 1998. Wikipedia
2. Wikipedia contributors. (2026). Ocean and Challenger Deep. Source of 70.8 percent surface coverage, 361 million km², 3,688 m average depth, 97 percent of Earth's water, and the 2021 revision of Challenger Deep to 10,935 plus or minus 6 m. Wikipedia
3. Wikipedia contributors. (2026). Gulf Stream. Source of the 30 sverdrup transport through the Florida Straits and about 150 sverdrups south of Newfoundland. Wikipedia
4. National Oceanic and Atmospheric Administration, National Ocean Service. Tides and water levels and Ocean currents (education tutorials). NOAA, Silver Spring, MD. Consulted for the standard treatment of tidal bulges, spring and neap tides, and wind-driven circulation; the host refuses automated requests, so no link is given.

A wrong answer that almost works

Ask why the sky is blue and a common answer comes back: the sky reflects the ocean. It has everything a good explanation seems to need. The ocean is blue, the sky is blue, the sky is above the ocean, and the two even change together on a grey day.

Test it and it collapses in one step. The sky is blue over Kansas, over the Sahara and over the middle of Antarctica, a thousand kilometres from any ocean. It is blue looking straight up from the bottom of a valley. And the ocean is largely blue because it reflects the sky, so the explanation has the arrow pointing the wrong way. Finding the right answer means climbing through the atmosphere first, because the real cause is the air itself.

What the air is made of

Dry air, by number of molecules, is 78.08 percent nitrogen, 20.95 percent oxygen and 0.93 percent argon. Everything else together makes up about a tenth of one percent, including carbon dioxide at a little over 0.04 percent. Water vapour is excluded from that list because it varies enormously, from almost nothing over a polar desert to around 4 percent in a tropical afternoon, and that variability is most of what weather is.

Notice how small the trace gases are. Carbon dioxide is four hundredths of one percent of the air, and it will turn out to matter out of all proportion to its abundance, for a reason the climate lesson will make precise: what a gas does to infrared radiation has nothing to do with how much of it there is.

Four layers, and why the temperature keeps changing its mind

The atmosphere is divided into layers by what temperature does with height, and each reversal marks a different heat source.

Two of those rows repay attention.

The troposphere holds nearly all the weather, and the reason is its temperature profile. Warm air at the bottom and cold air on top is unstable: warm air rises, so the layer convects, and convection is what builds clouds and storms. The stratosphere above has the opposite arrangement, warm air sitting on cold, which is stable and suppresses vertical motion. You can see the boundary from the ground. A thunderstorm grows upward until it hits the tropopause, cannot rise further into the stable layer, and spreads sideways into the flat anvil top that says a storm has reached the ceiling. The stratosphere holds 19 percent of the atmosphere's gases and almost no water vapour, which is why airliners cruise in its lower part: above the weather, in smooth air.

The thermosphere is the one that sounds alarming and is not. Temperature measures the average kinetic energy of molecules, not the amount of heat available. At 600 kilometres the molecules are moving fast enough to correspond to 2,000 degrees Celsius, and there are so few of them that a surface up there would feel bitterly cold, because almost nothing hits it. Temperature and heat are different quantities, and the thermosphere is the cleanest demonstration of the difference on the planet.

Pressure: the weight of what is above you

Air pressure at any altitude is the weight of all the air above that point. Standard pressure at sea level is 1013.25 hectopascals, also written as 1013.25 millibars, and hectopascals and millibars are the same size, which is why weather maps use them interchangeably.

Pressure falls with altitude because there is less air left above you. It does not fall in a straight line. Air is compressible, so the air near the bottom is squashed by everything above it and is therefore denser, while the air at the top is thin. The result is that pressure falls fast at first and then more slowly, roughly halving for every 5.5 kilometres you climb. Sea level is about 1013 hPa; at 5.5 km you are near 500 hPa with half the atmosphere already beneath you; at 11 km, near cruising altitude, you are near 250 hPa with three quarters of it below.

Key idea: half the mass of the atmosphere lies below 5.5 kilometres. The layer that contains almost everything you care about is thinner than the width of an average city.

Now the sky

Sunlight is a mixture of wavelengths, from violet at about 400 nanometres to red at about 700. When it meets molecules far smaller than its own wavelength, it is deflected by Rayleigh scattering, and the strength of that scattering is inversely proportional to the fourth power of the wavelength.

The fourth power is what does the work. Compare blue at 450 nm with red at 700 nm:

(700 / 450)4 = 1.5564 = about 5.9

Blue light is scattered nearly six times as strongly as red. So when you look at any part of the sky away from the Sun, what reaches your eye is sunlight that was scattered sideways out of a beam heading somewhere else, and that scattered light is overwhelmingly blue. No ocean required.

The same law explains sunset, and the explanation is the other half of the same sentence. At sunset the light reaches you through a much longer slant path through the atmosphere. Along that path the blue has been scattered out of the direct beam, again and again, leaving what continues to you dominated by the red and orange that scatter least. A blue sky and a red sunset are one phenomenon seen from two angles: you are looking at the light that was removed, or at the light that survived.

One honest loose end. If shorter wavelengths scatter more, the sky should be violet, and it is not. Two reasons: sunlight contains less violet than blue to begin with, and human eyes are far more sensitive to blue than to violet. The physics gives violet a head start and the light source and the eye take it away.

The ozone layer is not the greenhouse effect

These two get merged constantly, and they are different problems with different chemistry in different layers.

A hole in the ozone layer lets more ultraviolet in. It does not warm the planet in any significant way, and fixing it did not fix climate change. They are separate stories that happen to involve the same sky.

Common misconceptions

• "The sky is blue because it reflects the ocean." The sky is blue over deserts and ice caps, and the ocean looks blue partly because it reflects the sky. The cause is Rayleigh scattering by air molecules.
• "The thermosphere at 2,000 degrees would burn you." Temperature there measures molecular speed, not available heat. There are so few molecules that you would freeze.
• "The ozone hole causes global warming." Ozone depletion in the stratosphere lets ultraviolet in; the greenhouse effect in the troposphere keeps infrared from getting out. Different layers, different radiation, different gases.
• "Air pressure drops steadily, so halfway to space is half the pressure." Pressure roughly halves every 5.5 km, so it falls off far faster near the ground. Half the atmosphere's mass is below the height of a large mountain.

What to carry forward

Air is 78.08 percent nitrogen, 20.95 percent oxygen and 0.93 percent argon, with the traces that matter most making up the last tenth of a percent. It is layered by what temperature does with height: falling through the troposphere because the ground heats it from below, rising through the stratosphere because ozone formation heats it from above, falling again through the mesosphere, and rising through the thermosphere where the few remaining molecules absorb the hardest radiation. That first reversal creates the ceiling that flattens thunderstorms into anvils. Pressure is the weight of the air above, 1013.25 hPa at sea level, halving about every 5.5 kilometres. And the colour of the sky comes from a single exponent: scattering that goes as the inverse fourth power of wavelength sends blue sideways into your eye by day and leaves red in the beam at sunset.

The upshot: the wrong answer about the ocean failed because it was a correlation with no mechanism. The right answer is an exponent you can evaluate on a calculator, and it predicts the sunset as well as the midday sky.

Sources

1. National Oceanic and Atmospheric Administration. Layers of the atmosphere. NOAA JetStream. Source of the layer altitudes and temperature ranges, the 19 percent of gases in the stratosphere, and the anvil-top explanation. NOAA
2. National Oceanic and Atmospheric Administration. Air pressure. NOAA JetStream. Source of the 1013.25 millibar standard sea-level pressure and the millibar to hectopascal equivalence. NOAA
3. Wikipedia contributors. (2026). Rayleigh scattering. Source of the inverse fourth-power wavelength dependence and of the explanation for red and orange sunsets. Wikipedia
4. Wikipedia contributors. (2026). Atmosphere of Earth. Source of the dry-air composition figures of 78.08 percent nitrogen, 20.95 percent oxygen and 0.93 percent argon. Wikipedia

Twelve hours in one place

Here is a sequence NOAA uses to describe a textbook frontal passage. Clouds lower and thicken over several hours and light to moderate rain sets in; the temperature sits in the 50s Fahrenheit and the wind is from the east. The rain stops, the sky breaks into patches of blue, the temperature climbs into the mid 70s and the wind turns gusty from the south. A few hours later a line of thunderstorms sweeps through. Behind it the wind swings to the northwest, the temperature falls into the 40s and the sky clears.

Nothing in that description mentions a map. But every change in it was drawn on a weather map hours earlier, and by the end of this lesson you will be able to look at such a map and write that paragraph yourself, before it happens.

Where the air came from

An air mass is a large body of air with roughly uniform temperature and humidity, and it gets both from the surface it sat over. Sit over the Gulf of Mexico for a week and you are warm and wet. Sit over northern Canada in January and you are cold and dry. Meteorologists label them with two letters: a lower-case letter for moisture and a capital for temperature.

The labels are not fixed for life. An arctic air mass that crosses a warm ocean in winter picks up heat and moisture from below and arrives as maritime polar: still cold, but now wet enough to make cloud. The letters describe where the air has been, and air keeps moving.

The line where two air masses meet

Where air masses converge they do not blend. Cold air is denser, so it wedges underneath and lifts the warm air, and lifted air cools, condenses and rains. That boundary is a front, and the four kinds are drawn differently because they behave differently.

Two details make the table usable. Cold fronts nearly always run south and west from the centre of a low-pressure area, and warm fronts run east from it; fronts never extend from a high. And the words cold and warm are relative. A summer cold front can have a 32 °C air mass behind it, if the air ahead is 35 °C.

Why this matters: the slope of the front, not its temperature, decides what the weather looks like. A steep cold front lifts air fast over a short distance and gives you a violent hour. A gentle warm front lifts air slowly over hundreds of kilometres and gives you a grey afternoon.

Decoding one station

A surface map is not drawn from nothing. It is drawn from thousands of stations, each reporting the same set of numbers, each plotted as a station model: a small circle with figures around it. Learn the six positions and the map opens up.

• The circle itself is sky cover, shaded by how much of the sky is cloudy: empty for clear, half black for half covered, fully black for overcast.
• Upper left is temperature; lower left is dew point, the temperature to which the air must cool before water condenses out of it.
• Upper right is sea-level pressure, written as three digits.
• Below it is the pressure tendency, how the pressure has changed in the last three hours.
• A shaft out of the circle is the wind. It points toward where the wind is coming from. Barbs on the end give the speed: a half barb is 5 knots, a full barb 10, a solid pennant 50.

The pressure code is the part that catches people. Only the last two whole digits and the first decimal are printed, so 984 and 148 are both three digits with very different meanings. The rule is to put a 9 or a 10 in front, whichever lands you nearer 1000 hectopascals, and then insert the decimal point before the last digit. So 984 becomes 998.4 hPa and 148 becomes 1014.8 hPa. Sea-level pressure almost never leaves the band from about 950 to about 1050 hPa, which is why two digits and a decimal are enough.

Now a worked station. The circle is half filled. Upper left, 71. Lower left, 66. Upper right, 984. A shaft runs out to the south-west with one full barb and one half barb. Read it: the sky is half covered; the air is 71 °F with a dew point of 66 °F, so the spread is only 5 degrees and the air is close to saturated; the pressure is 998.4 hPa, on the low side; the wind is from the south-west at 15 knots. Warm, humid, breezy, and not far from cloud. This is maritime tropical air.

A second station 200 kilometres to the north-west reads: circle fully black, 54, 51, 1002.6 hPa, wind from the north-west at 10 knots. Cold, damp, overcast, higher pressure, wind from the opposite quarter.

Finding the front without being told where it is

Put those two stations side by side and you have already found a front, because three things change together across the gap.

1. Temperature. 71 °F against 54 °F over 200 km is a steep gradient. Within one air mass, temperature drifts slowly.
2. Wind direction. South-west on one side, north-west on the other. Winds converge at a front, which is what forces air upward.
3. Pressure. 998.4 against 1002.6 hPa. Fronts sit in troughs of low pressure, so the pressure reaches a minimum at the line and rises on both sides.

Any one of the three can mislead you. All three changing along the same line, at the same place, is a front. Draw the line through the stations where the change is sharpest, and decide which symbol to use by asking which air mass is winning: if the cold air is advancing into the warm, it is a cold front.

Forecasting the next twelve hours, and why the next twelve days are different

With the map read, the short-range forecast is mostly geometry. The cold front lies to your west and low-pressure centres in mid-latitudes generally travel eastward, so the front will reach you. Ahead of it you stay in the warm, humid air: muggy, south-west wind, falling pressure. When it arrives you get the narrow band of thunderstorms, because the steep slope lifts that humid air fast. Behind it the wind swings to the north-west, the temperature drops, the pressure rises and the sky clears. That is the NOAA sequence from the opening of this lesson, and you have now derived it.

Push the same reasoning out a fortnight and it fails, for a reason discovered in 1963. Edward Lorenz found that the equations describing a fluid atmosphere are chaotic: a very small error in the starting temperatures or winds does not stay small. In current models such errors roughly double every five days. Start with a thermometer reading that is wrong by a tenth of a degree in one place, and after ten days the error is four times larger and spreading. That is the butterfly effect, and it is a property of the atmosphere, not of the computer.

Two numbers put a boundary on the whole enterprise. Numerical weather prediction models currently show useful skill out to about six days, and the theoretical ceiling, with perfect data and a perfect model, is about fourteen. Part of the gap is resolution: model grid boxes have sides between about 5 km and 300 km, while a single cumulus cloud is under 1 km across, so entire storms live inside one box and have to be approximated rather than simulated.

The working response, used operationally since 1992 at the European Centre for Medium-Range Weather Forecasts and at the National Centers for Environmental Prediction, is to stop pretending you know the starting state exactly. An ensemble forecast runs the model dozens of times from slightly different starting conditions. If all the runs agree, confidence is high. If they scatter, it is low, and that scatter is where the phrase 40 percent chance of rain comes from. The forecast is not hedging; it is reporting a measurement of its own uncertainty.

Common misconceptions

• "A cold front means it will be cold." It means colder air is replacing warmer air at that spot. In July a cold front can drop the temperature from 35 °C to 32 °C and nothing about the day feels cold.
• "The wind barb points where the wind is going." It points where the wind is coming from. A shaft to the south-west is a south-west wind.
• "A 40 percent chance of rain means the forecaster is guessing." It is a number produced by running the model many times from slightly different starting states and counting how many produced rain.
• "Better computers will give us month-long forecasts." The six-day skill limit can be improved, but the fourteen-day ceiling comes from the atmosphere amplifying its own measurement errors, and no computer removes that.

The short version

Air masses take their temperature and humidity from the surface they formed over, and carry two-letter labels: cA, cP, mP, mT, cT. Where two of them meet, the denser cold air undercuts the warm air and lifts it, producing a front: cold fronts steep and violent, warm fronts gentle and drizzly, stationary fronts stuck, occluded fronts old. Each station on a map reports temperature and dew point on the left, three-digit pressure on the upper right, sky cover in the circle and wind as a shaft pointing upwind with 5-knot half barbs. A front is where temperature, wind direction and pressure all change along one line. From there the next twelve hours are geometry, and the next twelve days are not, because initial errors double about every five days and cap useful skill near six.

The point: a weather map is not a picture of the future. It is a set of measurements, and the forecast is what you get by applying physics to them and then honestly reporting how fast that physics loses its grip.

Sources

1. National Oceanic and Atmospheric Administration. Air masses. NOAA JetStream. Source of the cA, cP, mP, mT and cT classification, the modification of air masses in transit, and the steep cold front versus gentle warm front slopes. NOAA
2. National Oceanic and Atmospheric Administration. How to read surface weather maps. NOAA JetStream. Source of the four frontal symbols and colours, the rule that cold fronts extend south and west of a low, the relative meaning of cold and warm, and the twelve-hour frontal passage sequence. NOAA
3. Wikipedia contributors. (2026). Station model. Source of the plotting positions, the three-digit sea-level pressure convention and the 5, 10 and 50 knot barb values. Wikipedia
4. Wikipedia contributors. (2026). Numerical weather prediction. Source of the six-day skill figure, the fourteen-day chaos limit, the five-day error doubling, the 5 to 300 km grid spacing and the 1992 start of operational ensembles. Wikipedia

Two numbers that cannot come from the same process

Air trapped in Antarctic ice from the last glacial maximum holds about 30 percent less carbon dioxide than the air of 1750. Air sampled on top of Mauna Loa averaged 315.98 parts per million in 1959 and 427.35 ppm in 2025.

The first number took thousands of years to build. The second took sixty-six. Both are real measurements from the same planet, and between them sits the question this lesson has to answer: what normally moves carbon dioxide and temperature, on what timescale, and whether the present change resembles the normal thing at all.

Why one place is colder than another

Start with the local question, because climate is a place before it is a planet. Five things dominate.

Latitude is the largest. Sunlight arrives as a beam of fixed intensity; what varies is the area that beam is spread over. If the Sun stands directly overhead, a beam of cross-section A lands on area A. If it stands 60 degrees from vertical, the same beam lands on A divided by cos(60), which is 2A. Identical energy, half the heating per square metre. That single cosine is why the tropics are hot and why a summer afternoon beats a winter one.

Altitude works through the lapse rate established in the atmosphere lesson: the troposphere is heated from below, so temperature falls with height. Quito sits almost on the equator and is cool all year because it sits 2,850 metres up.

Distance from the ocean shows up in the warming record itself. Between 1850 to 1900 and 2011 to 2020, global surface temperature rose 1.09 °C, but land rose 1.59 °C while the ocean rose 0.88 °C. Water takes far more energy to change temperature than rock does, and it mixes that energy downward. So maritime places have mild winters and cool summers, and continental interiors swing hard in both directions.

Ocean currents move heat along coasts, which is why the module on the ocean spent a lesson on the Gulf Stream. Mountains force air upward, wring the moisture out of it on the windward side and leave a dry rain shadow on the other.

What matters here: climate is not one variable. Two towns at the same latitude can differ by 20 degrees in January because one has an ocean upwind and the other has two thousand kilometres of frozen continent.

The clock in the orbit

Over tens of thousands of years, the biggest changes come from the shape and tilt of Earth's path. Milutin Milankovitch worked out the arithmetic in the 1920s and 1930s, and the three Milankovitch cycles carry his name.

None of these changes the total energy Earth receives by much. What they change is where and when it lands, and the trigger for an ice age is summer sunshine at high northern latitudes: if summer is too weak to melt the winter's snow, the snow survives, reflects more sunlight, and the cold deepens itself.

The theory has a famous loose end worth stating honestly. From 3 million to 1 million years ago the glacial cycles followed the 41,000-year obliquity beat, exactly as the theory predicts. Over the last million years they have followed a roughly 100,000-year rhythm that matches eccentricity, which is the weakest of the three forcings. Why the weakest driver came to dominate is called the 100,000-year problem and is still argued over.

Reading a cylinder of ice

Where do the glacial cycles come from? Not from thermometers. Snow falling on Antarctica never melts; it compresses into ice and seals air into bubbles. Drill down and you recover the atmosphere itself, layer by layer.

The EPICA core at Dome C reached bedrock at 3,260 metres and reaches back 800,000 years. The Vostok core went to 3,769 metres and gives a reliable record for 420,000 years, down to about 3,310 metres, below which the layers are disturbed.

Two measurements come out of each sample. The air in the bubbles is a direct sample of ancient atmosphere, so its carbon dioxide and methane are measured, not inferred. The ice itself gives the temperature, through the ratio of oxygen-18 to oxygen-16 and the ratio of deuterium to ordinary hydrogen in the water molecules: heavier molecules evaporate less readily and condense more readily, so the isotope ratio in the snow records how cold the air was when it fell.

One complication matters, and it is the sort of detail that separates a real measurement from a slogan. An ice core has two chronologies, one for the ice and one for the gas. Near the surface the snow is porous firn through which air still circulates, so the bubbles seal only after the snow is buried. The gas at a given depth is therefore younger than the ice around it, by over a thousand years at a low-snowfall site such as Vostok. Any claim about whether carbon dioxide led or lagged temperature has to handle that offset before it means anything.

What the ice actually shows

Across the last 800,000 years, temperature and carbon dioxide rise and fall together, cycle after cycle, with carbon dioxide about 30 percent lower at the depth of a glaciation than in the air just before the industrial era. The cycles are paced by the orbit, but the orbit alone is far too weak to produce the temperature swings observed. What amplifies it is feedback: a colder ocean dissolves more carbon dioxide, ice reflects sunlight that open water would have absorbed, and both changes push the climate further in the direction it was already going.

Bottom line: the orbital cycles are the pacemaker and the carbon and ice feedbacks are the amplifier. That is why carbon dioxide is not a bystander in the ice-core record. It is part of the machinery.

The instrumental record

Thermometer records begin in the mid-nineteenth century. What they show, as assessed by the IPCC Sixth Assessment Report, is this. Global surface temperature in 2011 to 2020 was 1.09 °C above 1850 to 1900, with an assessed range of 0.95 to 1.20 °C. Each of the last four decades has been warmer than any decade before it since 1850. Global mean sea level rose 0.20 metres between 1901 and 2018, and the rate itself accelerated, from 1.3 mm per year over 1901 to 1971, to 1.9 over 1971 to 2006, to 3.7 over 2006 to 2018. Arctic sea ice area in September fell by about 40 percent between 1979 to 1988 and 2010 to 2019.

Notice the form of every one of those statements: a quantity, a baseline period, a comparison period, a number and a range. That is what an observational claim looks like, and it is worth insisting on the same shape from anyone who tells you something about the climate.

How the cause is established

Warming happening is one claim. Warming caused by people is a second claim, and it needs its own evidence. The IPCC assessment does the accounting explicitly for the period 1850 to 1900 against 2010 to 2019, and the numbers are worth reading as a budget.

Read the last two rows against the first. The natural drivers and the internal wobbles together cannot account for more than about 0.3 °C in either direction, and the observed change is about 1.07 °C. There is no natural budget line large enough to pay for it. Meanwhile greenhouse gases alone would have produced more warming than we have seen, and industrial aerosols have masked part of it.

The gas concentrations themselves are direct measurements, not models: 410 ppm of carbon dioxide, 1,866 ppb of methane and 332 ppb of nitrous oxide as annual averages in 2019, and their increase since about 1750 is assessed as unequivocally caused by human activity. Land and ocean have absorbed a near-constant share, about 56 percent per year, of human carbon dioxide emissions over the past six decades. The remainder is what stays in the air, which is what Mauna Loa has been counting since 1958.

Common misconceptions

• "The climate has always changed, so this is natural." The first half is true and is why ice cores exist. It does not license the second half: the assessed contribution of natural drivers to the observed warming is between minus 0.1 and plus 0.1 °C, against an observed change near 1.07 °C.
• "It is just the Sun." Solar output is measured by satellite and appears in the natural-drivers line of that budget. It is also the wrong shape: a brighter Sun warms the whole atmosphere, whereas greenhouse warming heats the troposphere while the stratosphere cools.
• "Carbon dioxide lagged temperature in the ice cores, so it cannot be a cause." In a glacial cycle the orbit starts the change and carbon dioxide amplifies it, so a lag is expected. Today the sequence is reversed: the gas is being added first.
• "Weather forecasts fail after a week, so century projections are worthless." Different questions. Nobody can say whether it will rain in Lyon on a date next month, and everybody can say Lyon will be warmer in July than January. Climate is the statistics of the system, not the trajectory of one day.

Pulling it together

Latitude, altitude, distance from the sea, currents and mountains set the climate of a place, and the cosine of the Sun's angle does most of the work. On the scale of tens of thousands of years, eccentricity near 100,000 years, obliquity at about 41,000 and precession at about 21,000 move sunlight around the globe and pace the ice ages, amplified by carbon and ice feedbacks. Ice cores at Dome C and Vostok record 800,000 and 420,000 years of that history, giving air directly from bubbles and temperature from oxygen and hydrogen isotopes, with the gas always younger than the ice around it. The thermometer record shows 1.09 °C of warming into 2011 to 2020 and four successively warmer decades. And the attribution budget leaves no natural line item large enough to explain it.

In short: the same planet runs a slow clock and a fast one. The slow clock is astronomical and takes tens of thousands of years. The fast one has moved carbon dioxide from 315.98 to 427.35 ppm in a single human lifetime.

Sources

1. IPCC. (2021). Summary for Policymakers. In Climate Change 2021: The Physical Science Basis. Contribution of Working Group I to the Sixth Assessment Report. Cambridge University Press. Source of statements A.1.1, A.1.2, A.1.3, A.1.5 and A.1.7: the 1.09 °C warming and its land and ocean split, the attribution budget, the 2019 gas concentrations, the 56 percent uptake, the sea-level figures and the Arctic sea-ice decline. IPCC
2. National Oceanic and Atmospheric Administration, Global Monitoring Laboratory. Trends in atmospheric carbon dioxide: Mauna Loa annual means. Source of the 1959 annual mean of 315.98 ppm and the 2025 annual mean of 427.35 ppm. NOAA GML
3. Wikipedia contributors. (2026). Milankovitch cycles. Source of the cycle periods, the obliquity range of 22.1 to 24.5 degrees, the current eccentricity of 0.0167 and the 100,000-year problem. Wikipedia
4. Wikipedia contributors. (2026). Ice core. Source of the EPICA Dome C depth of 3,260 m and 800,000 year span, the Vostok figures, the isotope thermometry and the separate ice and gas chronologies. Wikipedia

A cloud 65 light years across

Take the oldest solid objects anyone has ever weighed: inclusions inside meteorites, dated at 4,568.2 million years. That is the moment solid matter first condensed in this part of the galaxy, and everything in this lesson descends from what happened next.

Before it, there was a cold molecular cloud roughly 20 parsecs across, about 65 light years, made mostly of hydrogen with traces of everything heavier. Clouds like that are not stable. A fragment about one parsec across became dense enough to fall inward under its own gravity, and as it fell, something the cloud had all along became decisive: it was rotating, slightly.

Angular momentum is conserved, so a shrinking rotating body spins faster, the way a spinning skater speeds up by pulling in her arms. The collapsing fragment spun up as it contracted, and spin flattens things. Material falling along the rotation axis met nothing and fell straight in; material falling in the equatorial plane was held up by its own orbital motion. Within roughly 100,000 years the fragment was a disc with a dense, hot centre. The centre became the Sun. The disc became everything else.

Why the disc came out in two halves

The inner disc was hot, and heat sorts materials by what can survive it. Close to the young Sun, water, methane and ammonia stayed as gas and were blown outward; only metals and silicates, which condense at high temperatures, could exist as solid grains. Beyond a boundary called the frost line, near 5 astronomical units, between the present orbits of Mars and Jupiter, it was cold enough for those volatile compounds to freeze into solid ice.

That one temperature boundary explains the shape of the solar system.

Notice what the explanation does. It does not simply describe two families of planets; it predicts that the boundary between them must lie where ice can first survive, and in this system it does. Terrestrial embryos stopped growing roughly 100,000 years after the Sun formed, and the giant planets were assembled over the next few million.

So what?: the split between rock and gas is not a coincidence of our solar system. It is a thermometer reading from 4.5 billion years ago, still visible in the arrangement of the planets.

The collision that made the Moon

One problem remained. Earth has a moon a quarter of its diameter, 3,474 kilometres across at a mean distance of 384,399 kilometres. Nothing else in the inner solar system has anything like it, and no ordinary formation story produces one.

The giant-impact hypothesis says that around 4.4 to 4.5 billion years ago a Mars-sized body, called Theia, struck the young Earth at over 9.3 kilometres per second and at an oblique angle near 45 degrees. The collision threw a ring of vaporised and molten rock into orbit, and the Moon assembled out of it.

Three independent observations support it, and it is worth seeing what each one does.

1. The Moon has almost no core. Its iron core has a radius under about 25 percent of the Moon's own radius, where most terrestrial bodies run near 50 percent. A body that formed on its own from the disc would have collected its share of iron. A body assembled from material blasted off the outer layers of an already differentiated planet would not, because the iron had already sunk to the centre of the Earth.
2. The Moon's density is much lower than Earth's, which is the same observation expressed as a bulk property: it is made of mantle material, not of a planet's full inventory.
3. The oxygen isotope ratios of lunar rock and terrestrial rock are identical. Isotope ratios vary measurably with distance from the Sun, so this rules out capture of a body formed elsewhere. The Moon is made of Earth, or of something so thoroughly mixed with Earth during the impact that the two cannot be told apart.

Take the three together and they exclude the alternatives one at a time: not a captured asteroid, because the isotopes match; not a twin formed alongside Earth, because the iron is missing; not a piece spun off a rapidly rotating Earth, because the energy and angular momentum do not work out. What is left is the impact.

One face, always

The Moon turns on its axis once every 27.321661 days and orbits Earth in exactly the same 27.321661 days. That is not coincidence: tidal forces from Earth have raised a permanent bulge in the Moon's rock and dragged on it until the rotation matched the orbit. The state is called tidal locking, and it is why the far side was unseen by anyone until a spacecraft photographed it.

A second number belongs beside it. The Moon returns to the same phase every 29.530589 days, not 27.32. The difference is Earth's own motion around the Sun: by the time the Moon has completed one orbit relative to the stars, Earth has moved along its own path, and the Moon must travel a little further to line up with the Sun again. The first number is the sidereal month, measured against the stars; the second is the synodic month, measured against the Sun, and it is the one that sets the phases.

Why there is not an eclipse every month

If phases repeat every 29.5 days, and a solar eclipse needs a new moon while a lunar eclipse needs a full one, why are there not two eclipses a month?

Because the Moon's orbit is tilted by about 5 degrees to the plane of Earth's orbit. Most new moons pass above or below the Sun in the sky, and most full moons pass above or below Earth's shadow. Eclipses happen only when a new or full moon falls near one of the two points where the tilted orbit crosses Earth's orbital plane. Those points are the nodes, and the result is between two and five solar eclipses per year, about 240 per century.

The alignment also repeats. After 6,585.3 days, a little over 18 years, the geometry returns closely enough to produce a practically identical eclipse. This is the saros, and it is how Babylonian astronomers predicted eclipses without knowing what caused them: they had noticed the period without needing the mechanism. Each saros series runs 1,226 to 1,550 years and contains 69 to 87 eclipses.

A coincidence you can stand inside

Total solar eclipses exist because of an accident of scale. The Sun's diameter is about 400 times the Moon's diameter, and the Sun is about 400 times further away. Those two factors of 400 cancel, so the two objects appear almost exactly the same size in the sky: about half a degree across, roughly the width of a pencil eraser held at arm's length.

The cancellation is not perfect, which is why eclipses come in kinds. Both orbits are slightly elliptical. When the Moon is near its closest and the Sun near its furthest, the Moon more than covers the disc and totality can last up to 7 minutes 29 seconds; the next eclipse exceeding seven minutes is on 25 June 2150. When the Moon is near its furthest, it falls short, leaving a bright ring around a black disc: an annular eclipse, where nothing goes properly dark.

A lunar eclipse is the reverse arrangement and behaves differently. Earth's shadow at the Moon's distance is far wider than the Moon, so the Moon can sit inside it for over an hour, and the event is visible from the entire night side of Earth at once rather than along a narrow track. That asymmetry is why most people have seen several lunar eclipses and may never see a total solar one from home.

Common misconceptions

• "Moon phases are Earth's shadow falling on the Moon." Phases are the changing angle from which we see the Moon's own sunlit half, and they cycle every 29.5 days. Earth's shadow reaches the Moon only during a lunar eclipse, a few times a year at most.
• "The Moon does not rotate, since we always see one face." It rotates exactly once per orbit, every 27.321661 days. A non-rotating moon would show us every side in turn.
• "The far side of the Moon is the dark side." The far side receives just as much sunlight as the near side; it is dark at new moon locally when the near side is fully lit, and vice versa.
• "The Moon was captured by Earth's gravity." Identical oxygen isotope ratios rule that out, because isotope ratios differ with distance from the Sun. Captured bodies carry their birthplace's chemistry with them.

What to remember

A rotating fragment of a molecular cloud collapsed 4,568 million years ago, spun up as angular momentum was conserved, flattened into a disc in about 100,000 years, and lit a star at its centre. Heat sorted the disc: inside the frost line near 5 AU only rock and metal could condense, giving four small dense planets, while outside it ice was abundant, cores grew to about 4 Earth masses in some 3 million years, and captured hydrogen and helium made four giants. A Mars-sized body then struck Earth at over 9.3 km per second, and the debris became a Moon with almost no iron core and Earth's exact oxygen isotope signature. That Moon is now tidally locked, showing one face, cycling phases every 29.530589 days, and crossing the plane of Earth's orbit twice per circuit, which is why eclipses come two to five times a year rather than monthly.

The upshot: every fact in this lesson is a consequence of two things, gravity and the temperature of a disc. The arrangement of the planets, the composition of the Moon and the shadow that crosses a continent are all the same story at different scales.

Sources

1. Wikipedia contributors. (2026). Formation and evolution of the Solar System. Source of the 4,568.2 million year age, the 20 parsec cloud, the 100,000 year disc formation, the frost line near 5 AU and the 4 Earth mass core threshold. Wikipedia
2. Wikipedia contributors. (2026). Giant-impact hypothesis. Source of Theia, the 4.4 to 4.5 Ga impact, the 9.3 km per second speed and 45 degree angle, the core radius under 25 percent against about 50 percent elsewhere, and the identical oxygen isotope ratios. Wikipedia
3. Wikipedia contributors. (2026). Moon. Source of the 3,474 km diameter, the 384,399 km mean distance, the 27.321661 day sidereal and 29.530589 day synodic periods and tidal locking. Wikipedia
4. Wikipedia contributors. (2026). Solar eclipse. Source of the 400 by 400 size and distance ratios, the 5 degree orbital tilt, the two to five eclipses a year, the 7 minute 29 second maximum totality and the 6,585.3 day saros. Wikipedia
5. National Aeronautics and Space Administration. Eclipses. NASA Science. Background on eclipse types and observing safety. NASA

Off by a factor of nearly three, and right anyway

In 1923 Edwin Hubble measured the distance to the Andromeda nebula and got 285 kiloparsecs. The accepted value today is 770 kiloparsecs. He was wrong by a factor of 2.7.

He was also right about the only thing that mattered. Even 285 kiloparsecs put Andromeda far outside the Milky Way, which settled a long argument about whether the spiral nebulae were clouds inside our own galaxy or galaxies of their own. This lesson is about how such distances are measured at all, what the objects at the far end are doing, and why every rung of the measuring system depends on the rung below it.

What a star is doing

The Sun formed about 4.6 billion years ago, has a mass of 1.988 x 10³⁰ kilograms, a photosphere at 5,777 kelvin and a core close to 15.7 million kelvin. At that core temperature and pressure, hydrogen nuclei fuse through the proton-proton chain: four hydrogen nuclei end as one helium nucleus, and the helium weighs slightly less than the four hydrogens did.

That missing mass is the energy source, and you can check the arithmetic yourself. About 600 billion kilograms of hydrogen fuse into helium every second, and of that, 4.26 billion kilograms are converted into pure energy. Einstein's relation gives the output:

E = mc2 = 4.26 x 109 kg x (3.00 x 108 m/s)2 = 3.8 x 1026 joules per second

The Sun's measured luminosity is 3.846 x 10²⁶ watts. The calculation and the measurement agree to two significant figures, which is the sort of check worth doing at least once: the mass lost and the light emitted are the same fact stated twice.

Key idea: a star is a balance. Gravity pulls its mass inward; the pressure of fusion energy pushes outward. Every stage of a star's life is a new arrangement of that standoff, and the end of its life is the moment one side wins.

The diagram that organises every star

Around 1911 Ejnar Hertzsprung, and independently in 1913 Henry Norris Russell, plotted stars on a graph with luminosity up the vertical axis and surface temperature along the horizontal. There is one oddity to remember: temperature runs backwards, hot on the left and cool on the right, for historical reasons to do with spectral classification.

The result, the Hertzsprung-Russell diagram, is not a scatter. Stars fall into distinct regions, and the regions are stages of life rather than kinds of object.

The upper right and lower left are the informative corners. A star that is cool yet luminous must be enormous, because a cool surface emits little per square metre and only vast area can make up the difference. A star that is hot yet faint must be tiny for the same reason in reverse. The diagram measures temperature and brightness, and gives you size for free.

How this one ends

The Sun is a G2V star, meaning a G-type star on the main sequence. In four to seven billion years its core hydrogen will run out. Fusion pressure fails, the core contracts and heats, hydrogen begins burning in a shell around it, and the outer layers swell: the Sun becomes a red giant, moving up and to the right on the diagram. Eventually it sheds those outer layers and leaves the hot dense core behind as a white dwarf, which will glow from stored heat for perhaps trillions of years without any fusion at all.

Notice that the Sun's path across the diagram is not a line of similar stars. It is one object moving between regions, which is why the diagram is a life history rather than a filing system.

Climbing the ladder

None of this is measurable without distance. Brightness alone tells you nothing: a faint star may be dim and near or brilliant and far. Astronomers solve it with a ladder in which each rung is calibrated by the one beneath it.

The word ladder is exact and slightly alarming. An error low down propagates upward through everything above it, which is precisely what happened to Hubble in 1923: his Cepheid calibration was off, so his Andromeda distance was off, and so was every distance built on it. Correcting a lower rung silently rewrites the entire scale of the universe.

Two definitions worth pinning down. A parsec is the distance at which a star would show a parallax of one arcsecond, about 3.26 light years. A megaparsec is a million of them. Andromeda at 770 kiloparsecs is therefore 0.77 megaparsecs, comfortably inside the Cepheid rung, which is how the measurement was possible in 1923 at all.

Everything is receding, and the far things faster

In 1929 Hubble published the relationship that carries his name. Measure the distances to galaxies, measure their redshifts, and the two are proportional:

v = H0 x D

with v the recession speed in kilometres per second, D the distance in megaparsecs, and H₀ the Hubble constant. Hubble's own value was 500 km/s per Mpc, badly wrong for the calibration reason above. Modern measurements are near 70, and they run the clock backwards to a universe about 13.8 billion years old.

The interpretation is the part that trips everyone. Galaxies are not flying outward through space from a centre. Space itself is expanding, so every observer anywhere sees the same pattern of recession, and there is no location that counts as the middle. The standard image is a loaf of raisin bread rising: every raisin moves away from every other, and the more distant pairs separate faster, without any raisin being the origin.

The honest complication is current. Two independent methods give different answers. The cosmic microwave background, as measured by Planck in 2018, gives 67.4 plus or minus 0.5 km/s per Mpc. The distance ladder, from Hubble Space Telescope observations and confirmed by James Webb in 2023, gives 74.03 plus or minus 1.42. Each method is precise enough that the two disagree by more than five standard deviations, and improved technique has not closed the gap. This is the Hubble tension, and nobody yet knows whether it comes from an unfound systematic error or from physics missing from the model.

Common misconceptions

• "The Sun is burning." Burning is chemistry, and a ball of chemical fuel this size would last a few thousand years. The Sun converts 4.26 billion kilograms of mass into energy per second through nuclear fusion, which is why it has lasted 4.6 billion years.
• "Red stars are hot and blue stars are cool." The opposite. Hotter objects peak at shorter wavelengths, which is why the H-R diagram puts blue-white stars on the left and red ones on the right.
• "The universe is expanding into something, from a centre." Space itself is expanding, and every observer sees the same recession pattern. There is no centre inside it and no outside for it to expand into.
• "A white dwarf is a young star." It is the leftover core of a dead one: no fusion at all, cooling slowly, Earth-sized but with a star's worth of mass packed inside.

Looking back

The Sun runs on the proton-proton chain at 15.7 million kelvin, converting 4.26 billion kilograms of mass to 3.846 x 10²⁶ watts every second, and the two figures reconcile through E = mc². Plot luminosity against surface temperature and stars organise themselves: a main sequence running hot and bright to cool and faint, giants in the cool luminous corner, white dwarfs in the hot faint one. The Sun will cross from the first region to the second in four to seven billion years and end in the third. Distances to all of it come from a ladder: geometry to 1,000 parsecs, Cepheids to 29 megaparsecs, Type Ia supernovae past 1,000, redshift beyond that, each rung calibrated by the one below. And at the top, Hubble's law gives a universe expanding everywhere at once, about 13.8 billion years old, with two precise measurements of its expansion rate that still refuse to agree.

In short: astronomy measures almost nothing directly. It measures brightness, colour and wavelength, and builds everything else on a chain of calibrations, which is why knowing where each rung comes from matters as much as knowing the answer.

Sources

1. Wikipedia contributors. (2026). Sun. Source of the 4.6 billion year age, 5,777 K photosphere, 15.7 million K core, 1.988 x 10³⁰ kg mass, the 600 billion kg per second fusion rate, the 4.26 billion kg per second mass-to-energy conversion, the 3.846 x 10²⁶ W luminosity and the red giant to white dwarf sequence. Wikipedia
2. Wikipedia contributors. (2026). Hertzsprung-Russell diagram. Source of the axes and their orientation, the 1911 and 1913 discovery dates, and the positions of the main sequence, giants, supergiants and white dwarfs. Wikipedia
3. Wikipedia contributors. (2026). Cosmic distance ladder. Source of the rung ranges, Leavitt's period-luminosity relation, the 1.4 solar mass Chandrasekhar limit, the standard supernova magnitude near minus 19.3, and Hubble's 1923 Andromeda distance of 285 kpc against the modern 770 kpc. Wikipedia
4. Wikipedia contributors. (2026). Hubble's law. Source of the 1929 publication and Hubble's original 500 km/s per Mpc, the Planck 2018 value of 67.4 plus or minus 0.5, the distance-ladder value of 74.03 plus or minus 1.42, the greater than five sigma tension and the 13.8 billion year age. Wikipedia
5. National Aeronautics and Space Administration. Stars. NASA Science. Background on stellar classification and life cycles. NASA

Two files, no commentary

The first line of NOAA's annual carbon dioxide file reads: 1959, 315.98, 0.12. The last line at the time of writing reads: 2025, 427.35, 0.12. Three numbers on each line, nothing else, no argument attached.

The second file is a query, not a download: ask the USGS catalogue how many earthquakes of magnitude 5.0 or greater occurred worldwide between 1 January 2025 and 1 January 2026, and it answers 2,129.

Every lesson in this course has handed you conclusions with their evidence. This one reverses the direction. You take the raw numbers and produce the conclusions, including the ones the numbers refuse to support. That last part is the harder half.

Dataset one: where 427.35 comes from

C. David Keeling of the Scripps Institution of Oceanography began measuring carbon dioxide on Mauna Loa in March 1958, on a bare lava slope in the middle of the Pacific, thousands of kilometres from any continent and above the local boundary layer. That siting is the point: a sensor in a city measures traffic, and a sensor in a wheat field measures the wheat. Mauna Loa measures the air of the Northern Hemisphere.

NOAA's Global Monitoring Laboratory now publishes the record as plain text and CSV: daily, weekly, monthly and annual means, plus annual growth rates. It is free, it needs no account, and it opens in any spreadsheet.

Step 1: read the header before the numbers. Every good dataset begins with comment lines explaining itself. This one states the units (parts per million by mole fraction in dry air), the columns (year, annual mean, uncertainty) and a warning that values may change slightly under continuing quality control. The uncertainty column reads 0.12 ppm. That single figure tells you how many digits are worth arguing about: a difference of 0.05 ppm between two years is noise, and a difference of 2 ppm is not.

Step 2: the trend, in one subtraction

The annual means for the first and last complete years are 315.98 ppm in 1959 and 427.35 ppm in 2025. Subtract:

427.35 - 315.98 = 111.37 ppm over 66 years

Divide and the average is 1.69 ppm per year. Now do the thing that separates reading a dataset from quoting one: check whether that average describes anything real.

Step 3: the rate is not a constant

NOAA also publishes the year-to-year growth rate directly. Four values from that file:

The average of 1.69 sits between a 1960 rate of 0.50 and recent rates above 3, so it describes no actual year well. The curve is not a straight line; it bends upward. This matters because an average is a summary, and summarising a curve with its mean slope throws away the one feature that makes it a curve.

The 2022 value is worth a second look. It fell to 1.84 after 2.35 the year before and rose to 3.32 the year after. Year-to-year wobbles of this size come mostly from how much carbon the land biosphere takes up, which depends on drought and on El Nino. An individual year down is not a trend reversing; it is variability on top of a trend. Three consecutive years would be a different conversation.

Remember: a single point never establishes a direction. Two points establish a direction and nothing about its reliability. You need the scatter before you can say whether a change is real.

Step 4: the wiggle inside the trend

Open the monthly file and take one recent full year. Here is 2025 at Mauna Loa, in ppm:

The peak is 430.51 in May and the minimum is 424.37 in September: a swing of 6.14 ppm within one year, which is larger than three years of trend. If you looked only at May 2025 and September 2025 you would conclude that carbon dioxide was falling fast.

The cause is biology. Most of the world's land is in the Northern Hemisphere, so when northern forests leaf out in spring they draw carbon out of the air, and when the leaves fall and decay they put it back. You are watching a hemisphere breathe, once a year.

The way to separate the two signals is to compare the same month in different years, which cancels the seasonal term. August 2025 read 425.48 ppm; August 2026 read 427.55 ppm. The difference, 2.07 ppm, is trend with the season removed.

Step 5: what this file cannot tell you

Three limits, stated plainly.

• It is one station. Mauna Loa is not the globe. NOAA publishes a separate globally averaged series, and it differs from the Mauna Loa numbers by a small amount in both level and seasonal amplitude.
• It measures concentration, not cause. The file contains no information about where the carbon came from. Attribution needs the isotope work and the emissions accounting summarised in the climate lesson; it is not in these three columns.
• It measures gas, not temperature. No line in this file is a thermometer reading. Linking carbon dioxide to temperature requires physics and other datasets, which is exactly why the previous lesson kept the two evidence chains separate.

Dataset two: a year of earthquakes

Now a different kind of file. The USGS earthquake catalogue is a searchable database rather than a fixed table: you specify a time window, a magnitude threshold and a region, and it returns every matching event, with a plain count available as well.

The query used here was worldwide, from 1 January 2025 to 1 January 2026, at four magnitude thresholds. Here is what came back, with the base-10 logarithm of each count in the third column.

Look at the last column before doing anything clever. Going from magnitude 5 to 6, the logarithm falls by 1.17. From 6 to 7, it falls by 0.96. The drop is close to 1 per magnitude unit, which means each whole step up in magnitude brings roughly ten times fewer earthquakes.

Step 6: fitting the law

That pattern has a name. The Gutenberg-Richter law states:

log10 N = a - bM

where N is the number of earthquakes of magnitude M or greater, a measures how seismically busy the region is, and b is the slope. In seismically active regions b is commonly close to 1.0, with a usual range of about 0.5 to 2.

Get b from the data by taking the drop in the logarithm across a magnitude interval and dividing by the width of the interval. Over the widest clean interval, magnitude 5 to magnitude 7:

b = (3.33 - 1.20) / (7.0 - 5.0) = 2.13 / 2 = 1.06

A textbook value, recovered from a year of real events with two logarithms and a subtraction. And it has a practical reading. Predicting from the 2,129 events at magnitude 5 with b exactly 1 would give about 213 at magnitude 6 and 21 at magnitude 7. The observed counts, 145 and 16, are somewhat lower, which is what b slightly above 1 means: the largest events are a little rarer here than a strict tenfold rule would say.

Step 7: what this dataset cannot tell you either

• Completeness. The magnitude 4.5 row is the weakest. Small earthquakes are recorded only where instruments are dense, so global catalogues under-count them and the line rolls off at the low end. That is a property of the seismometer network, not of the Earth. The fitted b from the 4.5 row alone, 1.21, is inflated for exactly this reason.
• No timing. Gutenberg-Richter is a statement about how many, not about when. It gives you the long-run proportions and says nothing whatever about next Tuesday.
• One year is a small sample at the top. Sixteen events of magnitude 7 or more is a small number, so the count for that row carries real statistical noise. A decade of data would tighten it; a month would be worthless.

Set that beside the energy result from the earthquake lesson, where each magnitude step is about a 32-fold increase in energy. Combine the two and a genuine conclusion appears: events get ten times rarer and about thirty times more energetic per magnitude unit, so almost all the energy released by earthquakes in a year comes from the handful of largest ones, not from the thousands of small ones.

Common misconceptions

• "Raw data speaks for itself." The May to September fall of 6.14 ppm is raw data, and read alone it says carbon dioxide is plummeting. Data needs a question, a comparison and a stated time window before it says anything.
• "More earthquakes are being recorded, so there are more earthquakes." Catalogue counts at low magnitudes track the number and sensitivity of seismometers. Compare like with like, at a magnitude threshold where the network was complete throughout.
• "A b-value near 1 means the next big one is due." The law is a frequency distribution, not a schedule. Nothing in it makes a large earthquake more likely because recent ones have been small.

Where this leaves us

You have now run the whole procedure twice. Read the header and the units before the numbers. Subtract to get a change, divide to get a rate, then check whether that rate describes any actual year. Separate the signal that repeats and cancels, the 6.14 ppm annual breath of the northern forests, from the signal that accumulates, the 111.37 ppm rise since 1959. Take logarithms when counts fall by factors rather than by amounts, and read the slope: 1.06 earthquakes-per-magnitude, straight out of a year of the USGS catalogue. And then write down what the file cannot support, because a dataset that measures gas concentration on one Hawaiian mountain is not a thermometer, and a catalogue of events is not a forecast.

Bottom line: the skill this course has been building is not the memorising of numbers. It is the habit of asking, of every number you meet, where it was measured, over what interval, with what uncertainty, and what it does not cover. Two free public files were enough to practise all four.

Sources

1. National Oceanic and Atmospheric Administration, Global Monitoring Laboratory. Trends in atmospheric carbon dioxide: Mauna Loa. Source of the August 2025 value of 425.48 ppm, the August 2026 value of 427.55 ppm and the description of the record begun by C. David Keeling in March 1958. NOAA GML
2. National Oceanic and Atmospheric Administration, Global Monitoring Laboratory. Mauna Loa CO2 data files: annual means, monthly means and annual growth rates. Source of the 1959 annual mean of 315.98 ppm, the 2025 annual mean of 427.35 ppm, the 0.12 ppm uncertainty, the 2025 monthly values and the growth rates of 0.50 in 1960 and 3.32 and 3.33 in 2023 and 2024. NOAA GML
3. U.S. Geological Survey. Earthquake catalog search. Earthquake Hazards Program. Source of the 2025 worldwide counts of 8,558, 2,129, 145 and 16 events at magnitude thresholds of 4.5, 5.0, 6.0 and 7.0, queried for 1 January 2025 to 1 January 2026. USGS
4. Wikipedia contributors. (2026). Gutenberg-Richter law. Source of the equation, the meaning of a and b, the typical b near 1.0 with a range of 0.5 to 2, and the roll-off caused by catalogue incompleteness at low magnitudes. Wikipedia