Three growing metros, Houston, Goiânia and Kelowna, and the climate they are growing into

The same tools in three countries: how much each metro grew, what the growth took, and what the climate scenarios indicate for heat, drought, fire and heavy rain. Read in scenarios, not in forecasts: the signal is broad, and it is the place that decides what is exposed.

Loading the numbers…

—inhabitants in the Houston metro area, 1 July 2025 measured

—inhabitants in the Goiânia metropolitan region, IBGE 2025 estimate measured

—inhabitants in the Kelowna metropolitan area, 1 July 2025 measured

—times Goiânia’s built-up area in 30 years, 25 km circle (1990–2020) measured

measured by an official table, a station or a satellite. estimated a reading that depends on a choice (a grid, a model, a threshold). missing a part that did not respond in this run.

Three metros, three countries, the same tools

Houston, in the United States, Goiânia, in Brazil, and Kelowna, in Canada, are metros that grow fast in each country’s official numbers, and each already has a climate story told by UmaTerra. In 2024–25, Houston was the metropolitan area that gained the most people in the United States. This story reads the three with the same satellite data and with UmaTerra’s tools: how much each grew, what the growth took, and what the climate scenarios indicate for heat, drought, fire and heavy rain. It reads in scenarios, not in forecasts.

What the official tables say

Each country defines its metropolitan area differently: in the United States a county-based statistical area; in Canada a census metropolitan area, built from municipalities linked by commuting; in Brazil a metropolitan region set by law, whose list of municipalities IBGE publishes. The three series below come from each country’s official tables and cover different periods; they are not the same ruler.

What the satellite sees

To the left of the bar, the 50 km area around the centre seen by Landsat 5 in the mid-1980s; to the right, by Sentinel-2 in 2025. The lines show each metro’s legal window and the fixed 10, 25 and 50 km circles, the story’s common ruler. Pick a city in the selector and drag the bar.

How much the built-up area grew

The Global Human Settlement Layer (GHSL) measures built-up surface in 100 m cells, from 1975 to 2030. This story uses only the observed epochs, up to 2020: the 2025 and 2030 ones are projections and are left out. The chart puts the three cities on the same base, 1990 equal to 100, to compare pace; the table gives the hectares.

Up or outwards?

Dividing population by built-up surface gives the density of the urban footprint. To have a continuous series we use GHSL’s modelled population; it is an estimate, and the table sets it against the official numbers where the years coincide. Density that falls while the city grows points to outward expansion; density that holds, to filling in.

Which ring took the growth

The map shows, by colour, the epoch in which each 100 m cell became built-up, up to 2020. The chart adds the new built-up surface between 1990 and 2020 in four rings of distance from the centre. It is the most direct reading of where the city grew: near the centre, in the middle, or on the far edge.

Land use, in each country, from 1985 to 2025

Each country has its own land-use map: MapBiomas in Brazil, the Annual NLCD in the United States and the Annual Crop Inventory in Canada. They are not the same product and do not add up. The table shows, in a 10 km circle, land use at the start and at the end of the series in two places of each city: an urban one and a rural one. In Canada the inventory is agricultural, starts in 2011 and changed its wild-land classes between versions; so only city, water and forest are read with confidence.

The canopy that went, and why we cannot say why

Hansen counts all tree-cover loss since 2001, from felling, fire, storm or pests. So it is not a deforestation map: next to the growing city it also records Kelowna’s 2023 fire and the felling of plantations. The map shows loss by period; the chart, hectares per year in the 25 km circle.

The scenario is broad; the place decides what is exposed

Climate models see in cells of about 25 km: a city’s whole circle gets the same signal, and in Kelowna the two places fall in the same cell. What changes from one place to the other is what is on the ground. The table crosses, at each place, the land use with what the models indicate for the cell and shows how much of the area is sensitive to each hazard. It is a way of saying that these broad numbers need local information to be read, and that none of them is a forecast for your street.

The property and the parcel, up close

Camposcope analyses a rural property from Brazil’s register (the CAR, self-declared) or a parcel from British Columbia’s register. Here it reads four examples: in Goiânia, one property on the urban edge and one rural; in Kelowna, one urban parcel and one rural, both owned by a local government, so that no one’s home is exposed. The United States is not in Camposcope yet: coverage is being expanded, and Houston is left out of this page.

Hot days since 1981, in what was measured

The ERA5-Land reanalysis gives the daily maximum temperature since 1950 in cells of about 9 km. For each year we count the days when the maximum passed 30 °C, which serves as a common ruler for the three cities, and those that passed the threshold each country uses. This is what has already happened, before any scenario.

Three countries, three thresholds

UmaTerra’s climate tool uses, for a hot day, 30 °C in Canada, 32.2 °C (90 °F) in the United States and 35 °C in Brazil: in Canada 35 °C almost never occurs, and the threshold has to mean something locally. The count of days by each threshold measures different things and cannot be compared between cities. Nor can it be compared between tools: the scenarios use a bias-corrected 25 km grid, the reanalysis a 9 km one, and the number of hot days from one does not hold in the other. What is compared is the change in degrees, and that is what the scenarios page uses.

The core against the ring, by satellite

MODIS measures land-surface temperature since 2000, in 1 km pixels, day and night. We compare the 5 km core around the centre with a 20–30 km ring, in each city’s warm season, without water pixels. It is a measured surface difference, not an air one, and the ring may have different terrain and vegetation: it does not attribute the heat to the city’s growth.

Exposure, not prediction

Nine CMIP6 climate models, downscaled by NASA, compare 2041–2070 with a 1985–2014 baseline under two emission scenarios: the intermediate (SSP2-4.5) and the high (SSP5-8.5). A hazard counts as present when the model median passes a threshold and at least two-thirds agree on its sign; as strong, at a higher threshold with at least 80%. The result describes what the models indicate for a cell, never a given year’s weather nor anyone’s street.

How much warmer, and how many more days

The change in mean temperature and in the hottest day of the year, in degrees, is the measure comparable across the three cities. The table also gives each country’s hot days, which are read only within their own city. The global frame is the IPCC’s: 2.0 °C above 1850–1900 under the intermediate scenario and 2.4 °C under the high one, in 2041–2060.

Less rain, more evaporation, more dry days

Drought is read through three numbers: dry days per year, annual rainfall, and the water balance, which is rain minus potential evaporation. A warmer year with the same rain ends up drier; so the balance can worsen even where rain does not change. The hazard appears as present or strong by the rule on the previous page.

Fire seen by satellite and fire weather in the scenarios

The map shows the years each 500 m pixel burned between 2001 and 2025, from MODIS. The chart adds what burned per year in the 25 km circle. For the future, the Canadian Fire Weather Index (FWI) counts days of high and extreme danger in the scenarios. Fire weather is a condition, not a fire, and heat and drought drive it in many regions.

The water that has stayed and what the models indicate

The map shows the water the satellite saw in at least a quarter of images since 1984 and, in Houston, the radar reading of Harvey’s 2017 flood. For the future, the models give the change in the heaviest rainy day and in heavy-rain days; a warmer atmosphere holds more water vapour, and extreme rains tend to intensify. Houston’s and Kelowna’s events have their own stories, linked at the end.

Which place today each city’s 2041–2070 climate resembles

To picture a future climate, the tool looks for the cell whose present climate is closest to what the models indicate for the city: mean temperature, annual rainfall and dry days. The reading compares three annual means, not extremes, seasons or soil; it says which place the climate resembles, not that the land will look like that place.

A strong El Niño, and what past seasons show

NOAA’s Climate Prediction Center tracks an El Niño under way, strong by the relative index (RONI). For each city, the table groups the October–March seasons since 1981 by ENSO phase and shows the mean rain and how many were below normal. There are few strong events, and a single one can carry the mean. The models also do not agree on how ENSO itself will change in a warmer climate.

Heat, drought, fire and heavy rain, in both scenarios

The matrix summarises, for each city’s cell, the level of each hazard (0 no signal, 1 present, 2 strong) in the two scenarios, with how many of the nine models agree on the sign of the change. It shows where the signal is robust and where it is weak; it does not say what will happen in a given year or where inside the cell.

Drawing places prone to each hazard

Inside each city’s 25 km circle, three strata: lowland (height above drainage up to 3 m), repeated fire (burned in at least two years between 2001 and 2025) and heat (the hottest tenth of the daytime surface temperature). From each stratum a draw with a fixed seed picks four points, and Naturametrics’ buffer service reads a 1 km circle around each: land cover, built-up area, temperature, years burned and height above drainage. An empty stratum stays empty.

Three cities, the same questions, different answers

The final table gathers what the story measured, city by city, without ranking them: growth, the shape of growth, the heat that has already come, the intermediate scenario’s signal and the heat island. The differences come from the place, from the definition of a metro and from the data; the story does not claim any city is better or worse prepared.

What this story does not say

Not a forecast. The scenarios describe what nine models indicate for a 25 km cell in 2041–2070, not a given year’s weather nor what will happen in a neighbourhood. A city’s two places are told apart only by what is on the ground, and in Kelowna they fall in the same cell.

The three metros are not defined the same way. A county-based statistical area, a commuting-based census region and a legal metropolitan region have different sizes and logics; that is why the story’s common ruler is the fixed 10, 25 and 50 km circles.

GHSL built-up and population are modelled. Only the epochs up to 2020 are used; earlier ones come from lower-resolution imagery and built-up is a surface per cell, not a map of buildings. Population is an estimate and differs from the official count.

Hot-day thresholds change by country. The day count is by each country’s threshold and cannot be compared between cities nor between the scenarios’ grid and the reanalysis; the change in degrees is what is compared.

Surface temperature is not air temperature. MODIS measures the ground and the canopy top in 1 km pixels; the 20–30 km ring has its own terrain and vegetation (in Kelowna, hills and forest). The difference does not attribute the heat to the city’s growth.

Canopy loss is not deforestation, and each country has its own land-use map. Hansen counts fire, felling, storms and pests; MapBiomas, the NLCD and AAFC’s inventory are different products, and the last is agricultural and changed classes between versions: in Canada only city, water and forest are read with confidence.

El Niño has few strong events in the sample. Five strong El Niños since 1981: a single one can carry the mean. The models do not agree on how ENSO will change.

Camposcope and the draws are examples, not samples. They are four properties, two per city; the CAR is self-declared and Camposcope does not yet cover the United States. The draws have a documented seed and design, but four points per stratum do not represent the city; in Houston the up-to-3 m lowland covers most of the circle and does not discriminate, and the repeated-fire stratum is empty.

What is not here. People’s vulnerability, infrastructure, adaptation and costs: the story measures growth and climate exposure, and leaves the question of who is most exposed, and why, to local work.

Houston: Harvey’s rain, 2017 →

Kelowna: the West Kelowna fire, 2023 →

Kelowna: the thinning winters →

The Cerrado: the 2024 fires →

Written with Anthropic’s Claude Sonnet 5.5, from the tables and papers cited and the numbers Earth Engine and the UmaTerra tools computed: no fact or number comes from the model. Versão em português.