Network Altitude — klymot labs

Why altitude changes temperature

The atmosphere is a gas, and gases follow a simple rule: pressure, volume, and temperature are linked. At sea level the full weight of the air column above presses down — air molecules are packed closely together and the pressure is high. Higher up, less air sits above you, so pressure drops, molecules spread further apart, and the air is less dense.

When an air parcel rises it moves into lower pressure and expands. Expansion requires work to be done against the surrounding atmosphere, and in an adiabatic parcel that energy comes from its internal energy, lowering its temperature. The long-term average decrease in temperature with altitude is called the environmental lapse rate.

This lab investigates how changes in the average altitude of the observing network would be expected to influence measured temperatures, considering only the effect of changing station altitude.

Environmental Lapse Rate

~6.5 °C / km

The observed average through the real atmosphere, accounting for moisture and clouds — used in this lab.

Dry Adiabatic Lapse Rate

9.8 °C / km

Theoretical upper bound for perfectly dry air rising without heat exchange — steeper than the real atmosphere.

Ideal Gas Law

PV = nRT

Pressure × Volume = moles × gas constant × Temperature. Lower P at altitude → expansion → lower T.

This means a thermometer on a hilltop will read colder than one in the valley below — even if both are at the same latitude and longitude, and even if the large-scale climate is identical.

A real example: Innsbruck, Austria sits at 574 m in a valley. The Zugspitze summit, about 30 km away, sits at 2,962 m. Their long-term mean temperature difference is close to 15 °C — almost exactly what the environmental lapse rate predicts for 2,388 m of altitude difference (2.388 × 6.5 ≈ 15.5 °C).

The physics in a little more depth

The ideal gas law (PV = nRT) relates pressure P, volume V, moles of gas n, the ideal gas constant R ≈ 8.314 J/(mol·K), and absolute temperature T. For a parcel of dry air rising adiabatically (no heat exchange), the First Law of Thermodynamics gives dT/dz = −g/c_p ≈ −9.8 °C/km — the dry adiabatic lapse rate. But in the real atmosphere air is never perfectly dry: condensation releases latent heat as water vapour forms cloud droplets, partially offsetting the cooling.

The result is the environmental lapse rate of ~6.5 °C/km — the observed average through the troposphere used by the International Standard Atmosphere. This is what we use in the model, because it reflects the conditions that surface thermometers actually measure rather than a theoretical dry-air limit.

How has the number of active stations changed?

This chart shows how the number of active stations contributing to the network has changed over time.

Active stations in the network

Loading…

How the network's altitude profile has changed

The global temperature measurement network has not been constant. Since the late 1800s, stations have opened and closed, networks have expanded into remote regions, and the mix of coastal, lowland, highland, and mountain stations has shifted year by year.

Annual network altitude from geo-gridded GHCNm stations

Loading…

Each month's value follows the same general geo-gridding approach used in many global temperature analyses: stations are first filtered to monthly observations with a value and no QC flag, then assigned to roughly 500×500 km grid cells so dense regions do not dominate. Within each occupied cell we average station elevation, then average the occupied-cell means equally to get the monthly network altitude. Yearly values are the average of those monthly network altitudes, and the incomplete final year is excluded.

Only stations with known elevation contribute to either series, so both are computed from the same underlying station set.

What counts as "active" in a given year?

For this lab, a station contributes in any month where its temperature value is present and the QC flag is blank. That monthly filter matches the way global anomaly products are usually assembled: observations are gridded first, then cells are averaged equally so dense station clusters do not dominate the result.

As the network's average altitude increases, the model predicts cooler temperatures; as it decreases, the model predicts warmer temperatures.

Why isn't this already corrected?

Different temperature analyses use different spatial and anomaly-aggregation methods. This lab isolates one sampling effect in the underlying station network so it can be examined independently.

What do you predict?

Now that you've seen how the network changed, predict what temperature signal you would expect from those altitude changes alone.

Question: Based purely on altitude changes in the network, what pattern would you expect to see in the raw global temperature series?

Options are shown in a random order to avoid nudging your choice.

Make a prediction above to reveal the model.

The altitude signal model

Using the environmental lapse rate we can estimate the temperature signal that would result from changes in the station network's average altitude.

ΔT_altitude(year) = −0.0065 °C/m × ( h̄(year) − h̄_baseline )

For each year we take the mean elevation of active stations, subtract the mean elevation during the baseline period, and multiply by the environmental lapse rate (0.0065 °C/m).

Model assumptions

The environmental lapse rate is fixed at 6.5 °C/km.
Annual mean station elevation represents the network's altitude.
Only changing station altitude is considered.
No other sampling effects are included here.

Loading observed anomaly from the same station network…

Compare the series

The chart below compares the predicted altitude signal with the global temperature anomaly computed from the same filtered station network. Use the legend to overlay the raw or adjusted observations on the model.

Baseline period:

What does the baseline change?

The baseline sets the reference period used to center both series. Changing it does not alter the underlying data; it shifts the zero point so you can compare how the same temperature signal looks against different reference windows.

Predicted temperature signal vs. observed global anomaly

Loading…

Observed series start hidden. Click the legend to overlay or remove them.

Showing both series relative to the same baseline makes their changes over time directly comparable.

Some features that may be worth examining:

the similarity or difference in long-term trends
periods where the two series move together
periods where they diverge
differences in year-to-year variability
whether similarities occur consistently or only during particular periods
how the relationship changes between the sparse early network and the modern network

The degree to which the two series agree or differ is left for the reader to assess.

Data sources & methodology

Station network altitude — NOAA GHCNm v4 (QCU): Station inventory (elevation, lat/lon) and monthly temperature data from the NOAA Global Historical Climatology Network Monthly v4 quality-controlled unadjusted dataset. For each month, stations are first filtered to values that are present and have no QC flag, then assigned to ~500×500 km geo-grid cells. Occupied-cell mean elevations are equally averaged to produce the monthly network altitude, following the general geo-gridding approach used in many global temperature analyses. Annual values are the average of monthly network altitudes, and the incomplete final year is excluded. Menne et al., 2018.
Observed global temperature anomaly — NOAA GHCNm v4 (QCU / QCF): Global mean temperature anomaly computed from the same GHCNm v4 station set, available as unadjusted (QCU / Raw) or adjusted (QCF / Adjusted). Geo-gridded global mean; annual averages; auto-decade anomaly baseline. Only stations with known elevation are included, so the anomaly and the altitude chart reflect the same network.
Environmental Lapse Rate: ~6.5 °C/km — the observed average temperature decrease with altitude through the troposphere, adopted by the International Standard Atmosphere (ISA). AMS Glossary of Meteorology.
Ideal Gas Law: PV = nRT underpins the atmospheric pressure–temperature relationship used here. See e.g. Wallace & Hobbs, Atmospheric Science (2nd ed., 2006).