A primary forest cannot be replanted in any meaningful sense. It can only be lost. What distinguishes primary forest from a tree plantation, a regenerating woodlot or even a forest that was selectively logged a century ago is not simply the presence of large, old trees — it is the absence of significant human disturbance to the ecological processes that built the place: the undisturbed soil fungal networks, the deadwood at every stage of decay, the multi-generational canopy structure that took centuries to assemble[1]. Once that structure is broken, it does not return on a human timescale.
That distinction matters because it changes what "forest loss" means in practice. A commercial plantation harvested and replanted on a thirty-year cycle can show up in satellite data as tree cover gain almost as fast as it shows up as loss. A primary forest cleared for cattle pasture or palm oil does not. Tracking the second kind of loss — separating it from the ordinary churn of managed forestry — is one of the more difficult problems in environmental monitoring, and it is largely a satellite problem now.
What the data actually measure
The most widely used global dataset comes from a partnership between the University of Maryland, Google and the U.S. Geological Survey, built on imagery from NASA and USGS's Landsat satellites. Each pixel in a Landsat scene corresponds to a patch of ground roughly the size of a baseball diamond — about 30 metres across[2]. To build the first global forest-change map at that resolution, researchers processed 654,000 Landsat images — 143 billion pixels — comparing each one across time to flag where tree cover appeared or disappeared[2]. Google Earth Engine's cloud infrastructure reduced what would have been years of computation to a matter of days, and the resulting maps, published annually through Global Forest Watch, are what most contemporary reporting on deforestation is built on.
In 2023, the tropics lost 3.7 million hectares of primary forest — roughly ten football pitches every minute — releasing an estimated 2.4 gigatonnes of CO₂, close to half the annual fossil-fuel emissions of the United States.[3]
This kind of pixel-level change detection is powerful, but it answers a narrower question than it might appear to. Landsat can tell you, with high confidence, that tree cover disappeared from a given patch of ground in a given year. It cannot, by itself, tell you whether that patch was primary forest, a tree plantation nearing harvest, or land recovering from a fire two years prior. That distinction has to be layered in separately, using forest-type maps built from ground surveys, national reporting and other satellite products — which is why figures for "primary forest loss" and "tree cover loss" from the same year can differ substantially, and why researchers are careful to specify which one they mean.
The current scale of loss — and the exceptions
According to the FAO's Global Forest Resources Assessment, roughly 34 percent of the world's remaining forest is primary forest, and it has shrunk by an estimated 47 million hectares since 2000[4]. Three countries — Brazil, Canada and Russia — together hold 61 percent of what remains[4]. The rate of that decline, however, is not fixed. It slowed by more than half in the 2010s compared with the 2000s, and the 2023 tropical data showed a 9 percent year-on-year drop, driven substantially by steep reductions in Brazil (down 36 percent) and Colombia (down 49 percent) following changes in enforcement policy[3].
Those national-level declines were not universal. Over the same period, primary forest loss rose sharply in Bolivia, Laos and Nicaragua, a reminder that global totals can mask regional reversals running in opposite directions at the same time[3]. Roughly a third of all forest loss recorded between 2003 and 2018, tropical and otherwise, was fire-related — and the great majority of forest fires are started by people, whether through land clearing, agricultural burning or accident, not lightning[5].
Abstract illustration of a satellite grid passing over a canopy line, representing pixel-based forest monitoring Illustrative — each cell represents one Landsat pixel: roughly 30 metres of ground, tracked annually since 2000.
What monitoring can and cannot tell us
The rise of near-real-time satellite monitoring has changed the politics of deforestation as much as the science of it. Enforcement agencies, journalists and conservation groups can now identify clearing events within days rather than years, which is part of why national policy shifts — like Brazil's 2023 change in enforcement — show up in the data so quickly. But the technology has limits worth stating plainly: cloud cover routinely obscures tropical regions for months at a time, forest degradation that thins a canopy without fully clearing it is much harder to detect than outright clearance, and the distinction between "natural forest" and "primary forest" is itself a matter of definition, not just measurement[4].
Where the record is uncertainForest degradation — canopy thinning through selective logging, drought stress or edge effects, short of outright clearance — is not reliably captured by tree-cover-loss datasets, which are built to detect the disappearance of cover, not its gradual decline. The FAO itself notes that fewer than 40 percent of countries reporting to its most recent assessment monitor degraded forest area at all, and those that do use inconsistent definitions[4]. Global degradation is therefore understood to be significant, but it cannot currently be quantified with the same confidence as outright forest loss.
None of this diminishes what the data does show clearly: primary forest is being lost, the loss is measurable to within a hectare at a global scale for the first time in history, and — encouragingly — the rate of loss is not fixed. It responds, sometimes quickly, to enforcement, land tenure and policy. Mapping the canopy is only the first half of the story. What happens in response to the map is the half still being written.