A new global analysis has revealed that hunting pressure stretches across virtually the entire tropical belt, challenging the idea that the world’s remaining remote forests are broadly protected from human exploitation. The study, published in Nature Sustainability, uses machine learning to map where hunting is most likely to occur and how that probability changed between 2000 and 2015. Its results identify intense hunting hotspots in parts of Asia, South America and Africa, while also showing that areas once regarded as inaccessible refuges are increasingly exposed as roads, settlements and other forms of human access expand. The researchers describe the work as the first standardized, large-scale assessment of hunting probability across the tropics in both space and time. Their maps offer conservation planners a way to locate areas where wildlife is most vulnerable, even where direct hunting records remain incomplete or inconsistent.
Hunting is a major but notoriously difficult driver of biodiversity loss to measure. Unlike deforestation, which can often be detected from satellites, the removal of animals from a forest may leave little visible trace. A forest can remain apparently intact while its large mammals, birds and other hunted species disappear. This process, sometimes called “empty forest” syndrome, can disrupt seed dispersal, predator-prey relationships and nutrient cycling long before the forest canopy changes. Hunting can also push threatened species toward local extinction, especially when animals reproduce slowly or require large territories. Yet global comparisons have been hampered by differences in survey methods, reporting standards and geographic coverage. The new study addresses that problem by converting diverse evidence into a common prediction framework, estimating the probability that hunting occurs at locations across the pantropical region rather than relying only on confirmed observations.
To build the model, the researchers assembled data from 2,463 tropical sites classified as hunted or non-hunted. Each site was matched in space and time with ecological and socioeconomic information that could help explain why hunting pressure varies. These predictors included environmental conditions and indicators of human accessibility, allowing the algorithm to identify recurring relationships between landscape characteristics, human presence and hunting occurrence. In machine learning, a model is trained by presenting it with examples whose outcomes are already known. It then learns combinations of variables associated with those outcomes and applies the resulting statistical pattern to locations where observations are missing. The approach does not claim that every high-probability pixel contains hunters, but instead produces a geographically consistent estimate of risk. That distinction is crucial in remote regions, where field surveys are expensive, sporadic and often concentrated near roads or research stations.
The resulting maps indicate that hunting is not confined to a handful of isolated crisis zones. Its predicted spatial footprint extends throughout the tropical belt, including the forests of the Americas, Africa and the Indo-Malayan region. The strongest concentrations of high predicted occurrence probability appear in China, Sri Lanka and western India, as well as in the Brazilian Atlantic Forest and parts of West Africa. These regions differ greatly in ecology, population density and political history, but they share conditions that can make wildlife easier to reach or more heavily demanded. In some places, hunting may be linked to subsistence and local food security; in others, animals may be taken for commercial trade, cultural purposes or urban markets. The map therefore represents a broad pressure indicator rather than a single universal explanation for why hunting occurs.
The Indomalayan realm emerges as one of the most prominent areas of concern. Dense human populations, long histories of wildlife use, extensive transportation networks and severe habitat fragmentation can bring hunters and wildlife into close contact. In fragmented forests, animals may be concentrated in smaller remaining patches, while roads and settlements make those patches easier to enter. The Brazilian Atlantic Forest presents a different but equally urgent pattern. Although only fragments of this once-vast ecosystem remain, those fragments contain exceptional biodiversity and many species found nowhere else. Hunting in such landscapes can have effects disproportionate to the area affected because isolated populations have fewer opportunities for recovery. West African hotspots likewise underscore that hunting pressure can be high in regions where wildlife populations are already facing habitat conversion and other forms of human disturbance.
At the same time, the analysis identifies remaining refuges from hunting in remote parts of interior Borneo, Papua New Guinea, Central Africa and the western Amazon. These areas are not necessarily untouched, nor are they guaranteed to remain safe. Rather, their lower predicted probability appears to be associated with limited accessibility and distance from major human infrastructure. Such refuges can function as reservoirs for species that have vanished from more accessible forests, preserving ecological interactions and potentially supporting future population recovery. Their existence also offers a warning: remoteness is acting as a temporary barrier, not a permanent conservation strategy. If new roads, navigable routes, mines or settlements penetrate these landscapes, the conditions that currently suppress hunting may rapidly change.
The study’s time comparison provides evidence that this shift is already under way. Between 2000 and 2015, increasing human accessibility facilitated the geographic expansion of hunting, particularly into places historically considered undisturbed or remote. The pattern was especially notable in the Amazon basin, where rising access appears to have brought hunting pressure into areas previously buffered by distance. Expansion also occurred in regions that were already under substantial pressure, including China and Indonesia. Accessibility is a powerful ecological variable because it changes the cost of reaching wildlife and transporting carcasses or animal products to markets. A new road may therefore influence biodiversity far beyond the strip of forest cleared for its construction. By reducing travel time, it can transform previously isolated forests into hunting grounds connected to regional and international economies.
The researchers also found that hunting dynamics differed among tropical realms, meaning that a single global conservation response is unlikely to work everywhere. In one region, pressure may rise primarily as infrastructure expands; in another, high demand, dense settlement or long-established access routes may be more important. The machine-learning framework can help distinguish these broad spatial and temporal patterns, but it cannot by itself determine the motives of individual hunters or quantify the number of animals removed. Probability maps must therefore be combined with field surveys, interviews, wildlife monitoring, market data and local knowledge. They are best understood as tools for prioritizing attention: identifying where patrols, community programs, enforcement, alternative food initiatives or tighter controls on wildlife trade may have the greatest potential benefit.
The findings arrive as governments work toward international biodiversity targets that require measurable reductions in species loss and improved protection of ecosystems. Hunting is often underrepresented in global assessments because it is less visible than land-use change, yet the new maps show why that omission can be dangerous. A protected area may appear effective from satellite imagery while its wildlife community is being steadily depleted. Conservation assessments that measure only forest cover could therefore overestimate the real level of protection. Incorporating hunting probability into spatial planning would allow authorities to design more targeted interventions, such as focusing monitoring near newly opened roads, strengthening management in high-risk forest fragments and protecting remote refuges before access expands. The maps could also help direct scarce conservation funds toward locations where hunting and habitat loss overlap.
The broader message is both alarming and actionable: tropical forests are not simply divided into hunted and unhunted places, but exist along a shifting gradient of human access and wildlife pressure. As infrastructure spreads, that gradient can move rapidly, converting low-risk landscapes into areas of intense exploitation. Yet the persistence of hunting refuges demonstrates that prevention remains possible. Protecting remoteness, limiting damaging access routes and working with communities whose livelihoods depend on forests may preserve wildlife before emergency recovery measures become necessary. By bringing scattered observations into a standardized global framework, the study makes an invisible threat easier to see. Its maps cannot replace local conservation, but they can reveal where the next biodiversity crises are most likely to emerge—and where early action may still prevent them.
