Climates are classified by sorting places into groups based on long-term averages of temperature and precipitation, then checking those averages against fixed thresholds. The most widely used scheme, the Köppen-Geiger system, assigns each location a letter code from five main groups and applies numeric cutoffs for temperature and rainfall. The result is a map any two climatologists can reproduce from the same station data.
That short answer hides a design choice. Every classification system decides in advance what counts: average temperature, seasonal swing, rainfall timing, or the vegetation a place can support. Change the deciding variable and the map changes with it. This explainer walks through the mechanics of the major systems, the exact thresholds that draw the lines, where the schemes disagree, and how to read a climate code without overtrusting it.
What a climate classification actually does
A climate classification compresses decades of weather records into a short label. The label answers a practical question: given the same inputs, would two trained analysts reach the same output? Köppen's system was built so they would. It uses monthly means of temperature and precipitation, compares them to fixed numbers, and stops. No judgment calls, no site visits.
The deeper mechanism is that climate types are proxies for something else. In Köppen's original 1884 design, the classes tracked the limits of plant communities. Temperature and rainfall thresholds stood in for vegetation belts that were hard to map directly. That is why the boundaries roughly follow biome edges: the tropical rainforest class ends near where frost becomes possible, not at an arbitrary rainfall total.
Modern systems keep the logic but change the target. Some classify by air mass source regions. Some classify by the statistics of daily weather. Some classify by the energy and moisture budget of the surface. Each choice embeds an assumption about what climate is for. A code optimized for agriculture will not match one optimized for building energy codes.
The Köppen-Geiger mechanism, step by step
Köppen sorts every place through a decision tree. The first split is temperature. If the coldest month averages at or above 64.4°F, the climate is tropical (A). If the coldest month stays above 26.6°F and the warmest month exceeds 50°F, it falls in the temperate group (C). If the coldest month drops to 26.6°F or below but the warmest month still exceeds 50°F, it is continental (D). If the warmest month never reaches 50°F, it is polar (E). A separate arid group (B) is defined by rainfall rather than temperature and cuts across the others.
Why temperature first? Because the coldest month controls whether water freezes, which in turn controls whether woody plants survive. A place with a mild winter supports different vegetation than one with a hard freeze, even at identical annual rainfall. The 64.4°F and 26.6°F cutoffs are not physical constants. They are calibrated lines that reproduce biome boundaries well enough to be useful.
The arid group uses a different test. A location is class B if precipitation falls below a threshold tied to temperature, because warmer air evaporates more water. The threshold is roughly 20 times the average annual temperature in Celsius, adjusted so rain falling in the warm season counts for less. A hot desert and a cold desert both qualify as arid despite wildly different temperatures.
After the first letter, a second letter adds seasonal detail. In the A group, f means rain in every month, m means a short dry season, and s or w mark dry summers or winters. In the B group, S means steppe and W means desert, with a third letter for temperature. In the C and D groups, the second letter marks the warmest season and the third marks summer heat. A full code like Cfa means temperate, no dry season, hot summer.
The thresholds reconcile explicitly across groups. The 50°F warmest-month line separates polar from everything else. The 64.4°F coldest-month line separates tropical from temperate. The 26.6°F coldest-month line separates temperate from continental. Read together, they form a consistent ladder: frost-free, mild-winter, cold-winter, and frost-all-year. A station that sits near two of these lines can flip class with a small data change.
The scale and the numbers behind the lines
Köppen-Geiger covers the whole land surface and has been mapped at fine resolution. The 2018 Beck et al. update, building on the 1961–1990 reference period, found that the temperate C group is the most widespread climate class on land, followed by the arid B group and the continental D group. Tropical A and polar E cover smaller shares. These shares shift with the reference period chosen, which is the first caveat for anyone citing them.
The numbers that draw the lines are simple enough to check by hand. Tropical requires every month above 64.4°F. Polar requires every month below 50°F. The dry-season rules compare monthly rainfall to fractions of the annual total, scaled so that a dry month is genuinely dry relative to the year. None of these thresholds are statistically derived. They are design choices matched to vegetation, which is why they survive revision.
The reference period matters more than most users realize. A 30-year window is the standard, but which 30 years changes borderline classifications. The 1961–1990 window is common for global maps; newer windows capture recent warming and push some cold-climate stations toward warmer classes. A station labeled Dfb under one window can become Cfb under the next. The code is only as stable as the window behind it.
Timeline table: how the classification systems evolved
| Year | Development | What changed |
|---|---|---|
| 1884 | Köppen's first vegetation-based scheme | Five groups tied to plant limits |
| 1900 | Köppen refines temperature thresholds | Monthly means replace annual means |
| 1918 | Köppen adds seasonal precipitation letters | Second and third letters appear |
| 1936 | Köppen-Geiger collaboration | The form still used today |
| 1950s | Thornthwaite's moisture-index system | Evapotranspiration replaces raw rainfall |
| 1960s | Trewartha revision | Cleaner cutoffs, closer to biome maps |
| 1996 | Köppen map digitized globally | Reproducible global grids |
| 2018 | Beck et al. high-resolution update | Finer grid, modern reference period |
The pattern is instructive. Early versions emphasized vegetation. Mid-century versions emphasized water balance. Recent versions emphasize resolution and reproducibility. Each step traded some simplicity for some accuracy, and none replaced the others outright.
Where the systems disagree: comparison table
| System | Core variable | Strength | Weakness |
|---|---|---|---|
| Köppen-Geiger | Monthly temperature and rainfall | Reproducible, global, intuitive codes | Ignores evapotranspiration and extremes |
| Thornthwaite | Moisture index from evapotranspiration | Captures water stress | Needs more input data |
| Trewartha | Temperature and rainfall with biome logic | Cleaner biome match | Less globally standardized |
| Air-mass (Bergeron) | Source regions of air masses | Physically grounded | Hard to apply at fine scale |
| Holdridge | Biotemperature, precipitation, humidity | Links directly to vegetation | Sensitive to input estimates |
The disagreement is not a flaw. Each system answers a different question. Köppen tells you what a place feels like seasonally. Thornthwaite tells you whether water is scarce. Holdridge tells you what could grow there. Choosing a system is choosing which question matters.
Edge cases that break clean classification
Highland climates expose the biggest gap. Köppen has no dedicated highland class. A mountain station can qualify as tropical because its coldest month stays mild by the letter rules, even while frost occurs every clear night at altitude. Analysts often add an H modifier by hand, which is outside the standard code. Anyone reading a climate map should check whether highland zones were handled separately.
Coastal deserts and fog deserts are a second blind spot. Places like the Atacama coast receive almost no rain but stay humid from fog. Köppen classes them as arid deserts because the code counts liquid precipitation, not fog. The plant community there depends on fog water, so the biome and the code diverge. This is a known limitation, not an error in the data.
Borderline stations are the third case. A single station near the 50°F warmest-month line can classify as polar or temperate depending on the reference period and the interpolation method. Global grids smooth these lines, which hides the local uncertainty. The counterintuitive detail is that finer resolution can make a map less certain at the boundary, because sharp lines get drawn where the underlying data is thinnest.
Warming is the fourth case. As reference periods update, stations migrate between classes. A cold-winter continental station that becomes temperate under a new window reflects a real shift in averages, but the migration can also reflect a change in the averaging period alone. Distinguishing the two requires holding the period fixed and comparing underlying data, which most published maps do not show.
Misconceptions worth correcting
A climate code is not a weather forecast. It summarizes long-term averages and says nothing about what next Tuesday will bring. Two places sharing a code can have very different day-to-day variability.
A climate code is not a biome guarantee. The code approximates vegetation limits but does not inventory species. Agriculture, fire history, and soil determine what actually grows.
Tropical does not mean hot every day. It means no month falls below 64.4°F on average. A tropical highland station can be cool and still earn the A.
Arid does not mean hot. Cold deserts qualify for the B group through the temperature-adjusted rainfall threshold. Temperature is a modifier, not the defining test.
Köppen updates do not rewrite history. When a station changes class under a new reference period, the old records are still valid for the old window. Comparing codes across different windows is a common error.
Practical actions for reading and using climate codes
Start by identifying the reference period behind any map or table. A code without a stated 30-year window is incomplete. If two sources disagree, check whether the windows match before assuming a data error.
Confirm the system before comparing codes. Köppen Cfb and Trewartha Cfb are not identical. Check the reference period for every map you cite. Look for highland modifiers or separate highland layers. Treat borderline stations as uncertain, not decisive. Do not use a climate code to size building systems without local design data. Cross-check agriculture and water planning against evapotranspiration data, not rainfall alone.
For design and planning work, use the climate code as a first filter and then pull actual station statistics. A Cfa code tells you to expect humid summers and no dry season, but it does not tell you the design cooling load, the frost depth, or the extreme rainfall intensity. Those require the underlying distributions, not the letter code.
When two thresholds appear to conflict, reconcile them by reading the decision tree in order. Temperature gates come first in Köppen, so a station that qualifies as tropical never gets tested against the arid rainfall rule in the same step. The arid group is checked separately and can overlap with any temperature group. Reading the order resolves most apparent contradictions.
Limitations and what the code cannot tell you
Climate classifications compress distributions into categories, and compression loses information. A code cannot report the variance, the extremes, or the trend. Two stations with identical codes can have different heat-wave frequencies, different drought risk, and different future trajectories.
The systems also encode a mid-latitude bias. Köppen's thresholds were tuned to European and North American vegetation patterns, and some tropical and arid regions fit the letters less cleanly. Regional climatologists often maintain local schemes that outperform the global standard for their area.
Finally, classifications age. A map built on a past reference period describes that period. Using it to plan for the next 30 years assumes the averages hold, which recent warming makes unsafe in many regions. The code is a starting point for reading a place, not a prediction of what it will become.
The honest use of any climate classification is to ask what question it was built to answer, then check whether that matches yours. Köppen answers the seasonal-feel question well and reproducibly. Thornthwaite answers the water-stress question. Neither answers the extreme-weather question, and no letter code replaces the station record underneath it.