Photograph: NASA · Public Domain (NASA) · source
Project Earth
Forty-three years of instruments pointed at the whole planet, and what they wrote down.
Forty-three years of instruments pointed at the whole planet, and what they wrote down.
Is the world growing more leaves than it used to?
Yes, but barely, and not everywhere, and the size of the yes depends on a judgement call about instruments.
Since 1982 a succession of weather satellites has measured how much leaf sits above every patch of land on Earth, every day. This page processed 43 years of that record, 51,255 patches of ground at a time.
The gap between those last two numbers is the most important thing on this page.
Fit the trend only to the single-sensor years and greening beats browning 5 to 1. Run the same fit across the whole record and it falls to a coin flip. The difference is not a new decade of vegetation. In 2014 the measurement moved to a new instrument, and a step at the handover pulls the fitted line toward flat.
This page cannot separate how much of that shift belongs to the sensor and how much to the plants. So it shows all three windows and lets you switch between them, rather than quietly picking the flattering one.
Why. Leaf cover rises when carbon dioxide fertilises plants, when farmland expands, when irrigation arrives, when a forest regrows after logging, when shrubs invade grassland, and when a decade simply happens to be wetter. All of those draw the same green pixel. A plantation replacing rainforest can raise leaf area while destroying almost everything else about that place.
The world is warmer, but how confident should anyone be in the number?
In recent decades, very. Four independent teams land within 0.06 of a degree of each other.
Four groups take overlapping piles of raw readings from weather stations, ships and buoys, and each processes them into a single global figure using its own methods. NASA, the UK Met Office with the University of East Anglia, Berkeley Earth, and NOAA.
For 2024 they report between 1.14 and 1.2 degrees above the 1961 to 1990 average.
One thing had to be fixed before they could be compared. The four teams publish against three different reference periods. Laying them on one chart without adjusting would be like comparing rulers whose zero marks are in different places. Each series here has been shifted onto a common 1961 to 1990 baseline, and the shift applied to each one is printed beside its name.
Four independent confirmations. These teams draw on overlapping station records, so their agreement means the processing is consistent, not that the answer has been checked four separate times. They differ most where stations are sparse: HadCRUT5 leaves those gaps empty where the others estimate across them, which is most of why the spread widens as you go back.
Is there less ice than there used to be?
At the north pole, dramatically. At the south, it depends entirely on which month you look at.
Sea ice extent counts ocean where at least 15% of the surface is frozen, quoted in millions of square kilometres. Australia is 7.7 million, for scale.
The two poles are not one story. The Arctic is an ocean surrounded by land; the Antarctic is land surrounded by ocean, with winds and currents that behave nothing like the north. Showing only the famous one would be a choice rather than a summary, so the chart below draws both and lets you change the month.
The usual glacier chart is an average, which hides that glaciers do not all do the same thing. This one is a headcount: how many of the monitored glaciers gained mass in each year, and how many lost it.
A random sample. The world has roughly 200,000 glaciers and 534 of them are monitored closely, chosen largely because somebody could get to them. The set leans heavily toward Europe and North America. The headcount also moves with how many glaciers were surveyed in a given year, so read the two lines together rather than either alone.
Where does the heat actually go?
Into the water, and the numbers look small until you work out what it takes to move them.
Since 2005 the top two kilometres of the ocean gained about 22.4 × 10²² joules of heat. That figure is meaningless as written, so here it is two other ways.
Both statements describe the same measurement. The gap between how trivial the first sounds and how large the second sounds is the entire point. Water is extraordinarily hard to heat, which is why the ocean can absorb an enormous amount of energy and show almost nothing for it on a thermometer.
It is also why ocean heat is a steadier measure of what the planet is doing than air temperature, which jumps around with weather.
What any particular coastline experiences. The altimetry record is the open ocean between 66 south and 66 north, and it misses the poles entirely. Local sea level also depends on whether the land itself is rising or sinking, and in many cities the land is sinking faster than the water is rising.
Every chapter above rests on a satellite getting a clear look at the ground, and over much of the planet it does not.
The rainforests are the cloudiest places on Earth, which is exactly why they are rainforests.
The places that come off worst are the ones a reader most wants to know about. That is why the map has a coverage view, why country rankings exclude anywhere below 10% coverage, and why the analysis here reports thin data as thin rather than colouring it in confidently.
An earlier version of this analysis read the satellite's land-cover codes as running from zero, when they run from one. That single offset made the code for evergreen broadleaf forest look like the code for water, and the processing threw the rainforests away. The Amazon was running on a fraction of a percent of its data, and the error was invisible in the output, because a map with no rainforest in it still looks like a map. It was caught by checking whether the Congo read like a rainforest. It did not.
Everything above, interactive. Same numbers, same caveats, more knobs.
Hover any patch for its history, how often it was actually seen, and whether its trend is strong enough to rule out chance. Switch the colour from trend to coverage to see which parts of the map are built on thin data.
Two different satellite instruments produced this record. An older one flew from 1982 to 2013, and a newer one took over in 2014. They are not identical, so a trend fitted straight through the changeover can pick up a step in the equipment rather than a change in the plants.
All 43 years uses everything and crosses that changeover. 1982–2013 only and 2014–2024 only each stay inside one instrument's lifetime. Comparing them is the honest way to see how much of the answer depends on the equipment. The panel further down does that comparison for you.
Ranked by absolute change in leaf layers, not by percentage. Percentage flatters low-vegetation places: a savanna starting near zero posts a huge percentage for a small real change, while a rainforest posts a small one for the same change. Limited to countries with at least 40 cells, 10% coverage, p < 0.05, which leaves 37 of them.
Leaf cover since 1982 alongside rainfall, forest area and burnt area. Each country shows how much of it was actually measured, because a small country covered by a handful of patches produces a noisy line.
12 sources, each linked to the page it was downloaded from.
All 14 images are public domain or openly licensed, credited individually below with a link to its source.
