Sources of Climate Data

The oldest and most direct source is the weather station: an instrument, or cluster of instruments, sitting in the real air at a real place, measuring temperature, humidity, wind, rainfall, pressure and often solar radiation, hour after hour, year after year. A well-run station is the closest thing we have to ground truth - it measures the actual atmosphere, not a model of it, and the long records from good stations are the bedrock on which almost everything else is calibrated. When a typical meteorological year is built for a city, it is usually a nearby station's decades of records that supply the raw material.

But stations have three limitations a designer must hold in mind. First, a station measures a single point, and the atmosphere varies over short distances - an airport on the edge of a city, where many official stations sit, can read several degrees different from a dense, paved, traffic-heavy neighbourhood a few kilometres away. The station's numbers are real, but they are the weather *there*, not necessarily *here*. Second, stations are unevenly distributed and often sparse: a wealthy region may have many, while vast areas have very few, so the weather across the gaps has to be interpolated - estimated by stretching the nearest stations across the space between them, which grows less reliable the wider the gaps. Third, records have holes: instruments drift, break, get recalibrated or relocated, stations open and close, and readings go missing, so even a good station's history has gaps and discontinuities that must be patched and corrected before use.

None of this makes stations untrustworthy - they remain the most accurate source and the anchor for the rest. It makes them *specific and incomplete*. The practical lesson is to ask, of any station-based file, which station it came from, how far that station is from your actual site, how similar their settings are, and how continuous the record is. A file built from a distant, differently situated station over an interrupted record is a weaker guide than one from a nearby, well-matched, well-maintained station - and in much of the world, including large parts of India, the nearest usable station may be far away and its record thin. That gap is not a detail; it is the difference between data you can lean on and data you must treat with real caution.

Stations leave gaps in space and time; reanalysis was invented to fill them. The idea is elegant. Take a modern weather-forecasting model - the same kind of physics-based simulation of the atmosphere used to predict tomorrow's weather - and instead of running it forward into the future, run it over the *past*, continually nudging it toward every observation ever recorded: station readings, weather balloons, ships, buoys, aircraft and satellites. The model supplies physical consistency and fills the gaps where no instrument looked; the observations keep it tethered to what actually happened. The result is a gridded reconstruction of the past climate - temperature, humidity, wind, radiation and much more, on a regular grid covering the entire globe, at hourly steps, often stretching back many decades. Datasets like ERA5 are the best-known examples, and they have become a workhorse of climate analysis.

The strengths are exactly the ones stations lack: reanalysis is complete and consistent. There are no missing places and no missing hours; every point on Earth has a value, produced the same way everywhere, so you can study anywhere, including the vast regions where stations are sparse or absent. For a designer working somewhere poorly served by ground stations, reanalysis is often the best available picture of the local climate.

But it carries its own signature weakness: resolution and model dependence. Reanalysis lives on a grid, and each grid cell can be tens of kilometres across, so a single value represents the *average* conditions over a large area - it cannot see a valley, a ridge, a lake shore or a dense city quarter within that cell, and it smooths away exactly the local variations that often matter most for a building. It also inherits the assumptions and imperfections of the model that produced it, and it can be less reliable for variables the model handles more crudely, or in regions where few real observations were available to anchor it - which, again, tends to be the data-poor global south. So reanalysis is not raw truth; it is a physically consistent, gap-free *estimate*, superb for broad patterns and comparisons across places, but coarse in space and always model-flavoured. Used well - to understand the shape of a climate where stations are thin - it is invaluable; mistaken for a precise local measurement, it misleads.

The third great source looks down from orbit. Satellites carry instruments that sense radiation coming off the Earth and its atmosphere, and from those signals a great deal about the climate can be inferred: cloud cover, land and sea surface temperatures, and - especially valuable for building design - the solar radiation reaching the ground, which satellites estimate well by watching clouds and the atmosphere from above. Their defining strength is reach: a satellite sweeps over the entire planet, including oceans, deserts, mountains and every data-poor region on Earth, giving a genuinely global, spatially continuous view that no network of ground stations could match.

But there is a fundamental catch in the word *infer*. A satellite does not stick a thermometer in the air you breathe; it measures radiation at the top of the atmosphere and works backward, through physical models and assumptions, to an estimate of conditions below. That indirection has consequences. Satellite estimates of ground-level air temperature - the number a building most cares about - are far more uncertain than a station's direct reading, because what the satellite most directly senses is surface and atmospheric radiation, not the air temperature in the shade at head height. Satellites are excellent for some variables, above all solar resource and surface temperature patterns, and much weaker or more indirect for others. Their records are also comparatively short - the satellite era spans decades, not the century-plus that some ground records reach - so they are less able to describe long-term change on their own.

The honest picture, then, is that none of the three sources is 'the truth', and the best climate datasets deliberately blend them - anchoring reanalysis and satellite estimates to station measurements, using satellites for the solar and spatial coverage stations cannot give, using reanalysis for physical consistency across the gaps. For the designer, the takeaways are practical. Know that a weather file's numbers may come from any of these, or a mix; expect solar data in particular often to lean on satellites; and understand that the strength of the data varies by *variable* as well as by *place*. Above all, treat the whole apparatus as a set of imperfect lenses on the atmosphere, not an oracle - and keep any binding result with qualified specialists working from verified, appropriate data, not from whichever file happened to download.

Behind the tidy interface that hands you a weather file lies a world of deeply uneven coverage, and this is where the lesson turns from technical to moral. Data density tracks wealth and infrastructure. The rich, temperate world - much of Europe, North America, parts of East Asia - is thick with well-maintained stations and long, carefully curated records, so its weather files are dense, current and reliable. Across much of the global south, including large parts of India, Africa, South and Southeast Asia and Latin America, stations are far sparser, records shorter and more broken, maintenance thinner, and the data leans more heavily on interpolation, reanalysis and satellite estimation, all of which grow less certain the fewer real observations there are to anchor them.

The cruelty of this is that the gaps fall exactly where the climate stakes are highest. It is the hot, humid, rapidly warming regions - the ones facing the deadliest heat, the fiercest monsoons, the greatest exposure - that tend to be the least well measured. And it gets more local still: within a data-poor country, the informal settlements and dense low-income neighbourhoods where the most vulnerable people live are typically the least instrumented of all. There may be no station in the slum, no record of the airless, heat-trapping conditions in tightly packed self-built housing under a metal roof; the official station sits at a leafy airport that reads cooler than the reality millions actually endure. So the people most exposed to dangerous heat, and least able to afford protection, are also the people whose actual climate is least captured by the data - a compounding injustice hiding inside a dataset.

For a designer, this demands humility and care rather than despair. The data still exists and is still useful - reanalysis and satellite products genuinely help where stations fail, and they are far better than nothing. But you must calibrate your confidence to the coverage: a weather file for a well-instrumented city can be leaned on more heavily than one for a sparsely measured region, and neither captures the microclimate of a specific dense, informal, or heat-trapping site without further local judgement. Ask how well your place is served; treat data from thin-coverage regions as a broad guide rather than a precise truth; lean on local knowledge, on-site observation and, where the stakes are high, qualified specialists and verified data; and never let a confident-looking file lull you into forgetting that, precisely where the heat is most dangerous, the numbers are often weakest. Good climate analytics begins with knowing how much to trust the data you have.