A new CDC-led study identifies deforestation as a leading indicator of Ebola virus spillover. Using machine learning and two decades of satellite data, researchers found that forest loss and fragmentation were among the strongest predictors of where the virus might jump from animals to humans. The model doesn’t prove causation—but it does help identify environmental patterns that could guide preparedness in regions facing rising ecological pressure.
Published in Emerging Infectious Diseases, the study analyzed 22 independent Ebola virus disease (EVD) index cases reported between 2001 and 2021. These were instances where the virus is believed to have first spilled over into a human host, excluding cases traced to latent infections or human-to-human transmission. The team then used high-resolution data on forest cover, precipitation, elevation, and human population density to train a predictive model of spillover potential. The model was tested to see how well it predicted two spillover events that occurred in 2022, the year following the last year of data used in the model. The model distinguished between spillover and non-spillover locations with roughly 90% accuracy, highlighting key environmental and demographic factors.
Predicted risk is highly concentrated in the DRC, particularly in areas of recent forest loss. ... More Adapted from Telford et al. (2025).Telford et al. (2025)
Among the model’s most important predictors were forest loss and forest fragmentation, particularly when measured at small spatial scales. But what stood out was not just the strength of these variables, but their form. Spillover risk did not increase smoothly with forest loss. Instead, the model revealed threshold-like behavior, with risk remaining low until a tipping point in forest degradation was reached—at which point it rose sharply. These switch-like responses suggest that certain landscape changes may trigger spillover conditions rather than gradually increase them.
Statistical results from Telford et al. (2025) suggest that forest fragmentation and forest loss ... More contribute to Ebola spillover risk. The abrupt changes in these "partial dependence plots" suggest that there may be ecological thresholds that should not be crossed to avoid spillover. Image modified from Telford et al. (2025) to highlight switch-like behavior.Telford et al. (2025)
MORE FOR YOU
NSA Warning—Change Your iPhone And Android Message Settings
A Disappointing Update About ‘Solo Leveling’ Season 3
‘The White Lotus’ Season 3, Episode 7 Recap And Review: I’m Starting To Worry About This Show
This pattern is consistent with ecological observations elsewhere. In Australia, for example, habitat loss and subsequent changes in bat behavior have been linked to increased viral shedding in bats infected with Hendra virus—possibly due to stress or crowding. In central Africa, forest loss may increase human–wildlife interactions by opening up remote areas to hunting, driving bats toward cultivated fruit crops, or expanding the reach of bushmeat markets. These dynamics increase the opportunities for a zoonotic virus to make the leap into humans.
Importantly, the new study found that predicted spillover risk is not evenly distributed. Of all locations in the model’s top percentile of relative risk, nearly 80% were located in the Democratic Republic of the Congo (DRC). Other high-risk zones appeared in Cameroon, Gabon, and the Republic of the Congo. From 2021 to 2022 alone, the model estimated that spillover risk increased in 25% of the study area, largely driven by ongoing forest degradation and population growth.
According to Global Forest Watch, the DRC has lost almost 22,000 square miles of humid primary forest—an area slightly smaller than the entire state of West Virginia. While the DRC accounts for most of the predicted risk, it is not the only country where deforestation is intensifying the conditions linked to spillover.
While the model is not intended to forecast outbreaks in real time, it does offer a way to prioritize long-term investments in surveillance, ecological monitoring, and public health preparedness. As the authors note, conducting active surveillance everywhere would be inefficient—but focusing efforts in areas where spillover potential is rising could help public health systems stay a step ahead of future emergence.
This study adds to a growing body of research suggesting that spillover events are not random, but emerge from changing landscapes, shifting animal behavior, and evolving human-wildlife contact patterns.
Understanding where and when ecological changes elevate spillover risk can inform more strategic public health planning—and support ecological countermeasures aimed at preventing pathogens from crossing into human populations.