How to read the climate exposure of a place, one layer at a time

A point, a circle and a climate cell

It all starts with a click and a radius, from 1 to 20 km. Land use is read inside the circle, pixel by pixel. Projected climate comes at another scale: the climate models' 0.25° cell (about 28 km) that holds the point, the dashed square on the map.1

In the app: search by postcode, address or coordinate, a click on the map, the radius selector, and the direct link /app/?lat=…&lon=…&r=….

What is in the circle today

In Canada the land use comes from the AAFC Annual Crop Inventory, in Brazil from MapBiomas Collection 11: both map the latest year in 30 m pixels. climate-shift groups their classes by how they respond to climate: forest, grassland and shrub, pasture, field and perennial crops (orchards, vineyards), urban areas and water.

In the app: the Land use tab, and the land-use layer on the map.

Where today's land came from

Today's land use is the starting point, but it has changed too. Pixel by pixel, the land-cover record says what each part of the circle was in the comparison year: the group then on the left, the group now on the right, each band the area that took that path.

In the app: the sidebar's Land use section, Compare with a year — the map gets the divider and the Land use tab shows these bands under today's bar.

Nine climate models, each compared with itself

Projected climate comes from NASA's NEX-GDDP-CMIP6: nine models, one per modelling centre (from EC-Earth3 to INM-CM5-0), each compared with its own past, 1985–2014 against 2041–2070. What matters is the change, not any model's absolute value.2

A hazard appears only when at least two-thirds of the models agree on its sign, and is called strong only when at least 80 % agree and pass a higher threshold.

In the app: the Climate tab, with the range across models for each indicator.

How to read a projection

A projection does not say what 2050 will be like: it compares two 30-year climates, 1985–2014 and 2041–2070, each simulated by the same model. A model's change is its future minus its own past, so each model's own errors cancel out.

Each dot below is one model; the dark band holds two-thirds of them (IPCC's "likely"); the tick is the median. climate-shift speaks of a hazard only when at least two-thirds agree on the sign, and of a strong one when 80 % agree and the change passes a higher threshold. A band that crosses zero is a warning: the models do not even know which way it goes.

In the app: the Climate tab, each indicator's band and the per-model table.

Heat: the signal everyone agrees on

Mean temperature and days above a threshold (30 °C in Canada, 35 °C in Brazil). Each bar is one model.

In the app: the Climate tab, mean temperature and hot days indicators.

Rain: where the models most often split

On annual rainfall the models often disagree, sometimes even on the sign. The screen shows each model's change and says when there is no agreement — disagreement is information too.

In the app: the Climate tab, annual rainfall, dry days and heavy-rain days indicators.

Middle or high emissions: still close by 2050

The scenario selector swaps SSP2-4.5 (intermediate emissions) for SSP5-8.5 (high emissions). Until mid-century the two paths still run together on most indicators; the big split comes later.

In the app: the Scenario selector, or &s=ssp585 in the address.

Which world, and where it already exists

Each scenario is a warmer world: by the IPCC, SSP2-4.5 at mid-century is about +2 °C above pre-industrial. And the climate projected here may already exist somewhere today: the climate analog is the cell whose 1985–2014 climate best matches this point's future — temperature, rainfall and dry days.

In the app: the In essence box and the dotted orange cell on the map.

What has already been measured, and when two records disagree

Before looking ahead, it pays to see what has already happened. Measured rainfall comes from two independent products, and a trend is called clear only when both agree: TerraClimate everywhere, with ERA5-Land in Canada (CHIRPS, used in Brazil, stops at 50° latitude). Drought comes from TerraClimate's water deficit. Every trend goes through a Mann–Kendall test: without p < 0.05 the screen says "no clear trend".3

In the app: the Climate tab, Already measured section.

Wind: only what was measured

The CMIP6 daily archive has no gusts, so climate-shift shows only measured wind: each year's peak gust in the ERA5 reanalysis. In Brazil the nearest INMET weather station is added; in Canada, for now, ERA5 stands alone.

In the app: the Climate tab, Wind gusts card; in the Story tab, the Strong wind (observed) hazard.

Terrain decides where heavy rain does harm

The same heavy rain passes harmlessly over a plateau and wrecks a ploughed slope or a valley floor. The Terrain tab measures slope (SRTM), low ground near drainage (MERIT Hydro, up to 5 m above the channel) and water seen by satellite (JRC).

"Low ground" is a terrain index, not a flood map: on very flat plateaus it flags a lot and says little.

In the app: the Terrain tab, and the Low ground, Slope and Water layers on the map.

A decade of fire

Burn scars from 2016 to 2025: how much burned, how much burned more than once, and how much burned in the last two years. In Canada they come from MODIS (MCD64A1, 500 m pixels), so small fires go unseen; in Brazil, from MapBiomas Fire at 30 m.

In the app: the Terrain tab, fire section, and the Fire — times burned layer.

The fire danger ahead

Burn scars say what has burned; the Canadian Fire Weather Index (FWI) says how many days will have fire weather. NASA computed it from the same models (NEX-GDDP-FWI). Two counts: days above FWI 30, and days above the place's own 95th percentile — extremes for there, where FWI 30 is rare.

In the app: the Climate tab's FWI indicators, and the Fire weather hazard.

The climate class inside the circle

The only climate layer that varies inside the circle: the 1 km Köppen-Geiger maps of Beck et al. (2023), for 1991–2020 and 2041–2070, from another 42 models — a second opinion. The Sankey shows how much of the circle changes class.

In the app: the Climate tab, at the top.

Hazard × land use = exposed area

This is where the two scales meet. For each hazard present, climate-shift adds up the land-use groups sensitive to it — crops and pasture to heat, cropland and bare soil to heavy rain, and so on. The result is the probable exposed area, with the reason for each group.4

In the app: the Story tab, Probable exposed area box, one card per hazard.

A written story, and the data's fingerprint

Finally the app writes up the result in paragraphs, always closing with the caveat on scale and limits. Each analysis carries a provenance hash computed over the source versions: same point and same sources, same hash.5

In the app: the story in the Story tab; the sources and the Provenance hash in the Method tab.