fields/climate-weather/model families
Linear models in weather, climate and oceans
How simple fitted relationships, where a result is treated as a weighted sum of measured inputs, are used in research on weather, air quality, climate and the oceans.
How it is used
Weather, climate and ocean science studies the atmosphere, the seas and the way heat, water and gases move between them. Much of its data is a long record of measurements: temperature, wind, humidity, pollutant concentrations, sea surface temperature. A linear model is the plainest way to connect such measurements. It assumes the thing you want to know is a sum of the things you measured, each multiplied by a fitted weight, plus an allowance for error. Fitting the model means choosing those weights so the sum comes as close as possible to the observations. Generalised versions allow the response to be counted, bounded or skewed rather than freely varying, and some allow each input to bend along a smooth curve instead of a straight line.
In this field the inputs are usually either weather variables or laboratory measurements of a sample, and the output is a quantity that is hard to measure directly. Researchers use the fit in two ways. One is prediction: given the weather, say what the pollution should be. The other is accounting: once the weather's contribution has been absorbed by the model, whatever is left over can be examined as a possible effect of something else, such as a change in human activity. Regularised variants add a penalty that keeps the weights small, which helps when many inputs are measured and most of them say much the same thing.
Why researchers use it
The appeal is that the fitted weights can be read. Each one says how much the output moves when that input moves, which lets a researcher check the result against known physics and chemistry rather than accepting it on trust. In a field where the aim is often to explain a signal rather than only to anticipate it, that matters. A linear fit also needs comparatively little data, trains quickly, and behaves predictably when the record is short or patchy, which is common for laboratory samples and for instruments that run in campaigns rather than continuously.
In the record
Limits
A linear fit assumes effects add up and do not interact, which the atmosphere and ocean rarely respect: chemistry speeds up non-linearly with temperature and sunlight, and winds mix sources in ways no weighted sum captures. Weather variables are also strongly related to each other, so individual weights can become unstable and invite causal readings the data does not support. Fits hold only over the range of conditions observed, and extrapolating beyond them, as a changing climate requires, is unreliable. Smooth and regularised versions ease some of this, but not the underlying assumption.
Written by AIxSci: the background from general knowledge, the examples from checked records.
Related
Every article
aix-00289
Machine learning points to filter loading as main driver of levoglucosan desorption temperature
aix-00281
Bacteria-derived aerosol matter made to produce oxidants, then modelled from its optical fingerprints
aix-00290
Machine learning separates weather from policy in China's winter heating air pollution
aix-00291
Statistical model estimates what Makkah's air would have held without lockdown
aix-00292
Neural network forecasts daily sea surface temperature up to 90 days ahead
aix-00293
Machine learning used to predict fine particle emissions at a Greek regional airport
aix-00294