Tracing forest loss through global supply chains
Satellite observations show where forests are lost. We asked whether global economic data could help identify the consumers ultimately associated with those losses.
Forest loss is visible from space, but responsibility is harder to observe. A satellite image can locate a cleared area and, increasingly, distinguish whether the loss is associated with logging, agriculture, mining, or infrastructure. It does not show where the products linked to that loss are finally consumed. For us, this gap between ecological observation and economic responsibility became the starting point of the study.
We began with a practical question: could high-resolution forest-loss data be connected to global supply chains without reducing very different land-conversion processes to a single national total? To examine this, we linked a global forest-loss driver dataset at one-kilometre resolution with the REX3 multiregional input-output database. The resulting framework follows four activities - logging, permanent agriculture, hard commodities, and settlements and infrastructure - across 189 consumer countries and 163 economic sectors from 2001 to 2022.
The basic idea was straightforward: identify forest loss in producing countries, connect each driver to the relevant primary economic sectors, and use the input-output model to trace the production required by final consumption. In practice, the difficult part was making the two systems compatible. Remote-sensing data describe pixels and land-change processes, whereas economic accounts describe countries, sectors, and monetary transactions. Much of the work involved defining conservative links between the two, preventing double counting, and checking that the physical totals remained unchanged as responsibility moved through the model towards final consumers.
Domestic consumption remains dominant but international consumption is substantial
The first result was the scale of the international component. In 2022, foreign final consumption was associated with approximately 5.01 million hectares of the 13.12 million hectares of forest loss covered by the four drivers, or 38.15% of the total. Domestic consumption accounted for the remaining 61.85%. This balance was not uniform across drivers. Foreign consumption accounted for about 31% of the permanent-agriculture footprint, 49% of the hard-commodity footprint, and 47% of the logging footprint. Settlements and infrastructure were much more local, with domestic consumption accounting for roughly 86-90% of the footprint.
Figure 1 | Forest-loss footprints from logging, permanent agriculture, hard commodities, and settlements and infrastructure, divided into domestic consumption and three directions of international income-group flows from 2001 to 2022. Each panel uses its own vertical scale.
International flows became more multipolar
We next examined only the cross-border component. Cross-border forest-loss flows became more diversified between 2001 and 2022. The share flowing from lower-income producing countries to higher-income consumers declined from 46.71% to 39.22%, while flows in the opposite direction increased from 10.60% to 17.65%. The United States remained a major consumption hub, although its share of the global footprint fell from 27.21% to 17.95%. China’s share rose from 3.83% to 12.80%. For permanent agriculture, Brazil remained the largest producer within the international network, while production became more geographically distributed. Overall, the international footprint involved increasingly varied income-group flows and a broad network of bilateral producer–consumer links across countries.
Figure 2 | Changes in international forest-loss footprint networks between 2001 and 2022. The chord diagrams show all drivers combined and permanent agriculture; the lower panels compare the leading permanent-agriculture producers and consumers. Domestic flows are excluded.
Countries followed different outsourcing trajectories
To examine how countries’ dependence on external forest resources changed, we classified them by their average net-importer or net-exporter position and their trends from 2001 to 2022. This identified four trajectories. Russia remained a net exporter with increasing exports, while China and India were net importers with increasing import dependence. The United States, Japan, and the United Kingdom remained net importers but showed declining import trends; Brazil, Canada, and Indonesia were net exporters with declining export trends. National positions also varied by forest-loss driver. Hard commodities and logging contributed most strongly to polarising patterns, while permanent agriculture and settlements and infrastructure showed different trajectories. Thus, a country could approach balance for one driver while becoming more externally dependent on another.
Figure 3 | National trajectories of forest-loss footprint outsourcing from 2001 to 2022. The horizontal axis separates net importers from net exporters; the vertical axis shows whether import or export dependence intensified or weakened. The map applies the four resulting categories globally.
The correlates of forest loss differ among drivers
Finally, we used an Extra Trees model with SHAP values to examine how 13 characteristics of consumer countries were associated with their embodied forest-loss footprints. Across all drivers, urban population share, protected-area coverage, and manufacturing value added were among the most informative model features. Their relative importance changed when the analysis was separated by land-conversion activity.
For logging, higher urbanisation and manufacturing intensity were associated with larger imported footprints in the model, consistent with sustained consumption for timber and forest products. Permanent agriculture showed a more complex pattern. Protected-area coverage and manufacturing value added were important predictors, while variables such as income and cereal yield displayed non-linear relationships. The radial SHAP plot in Figure 4 represents predictive contributions, not causal effects. It helps identify recurring associations across country-year observations, but it cannot show that changing one national indicator would directly change forest loss elsewhere.
Figure 4 | SHAP analysis of consumer-country characteristics associated with forest-loss footprints for all drivers combined, logging, and permanent agriculture. Bars show relative feature importance; points show the direction and magnitude of each feature's contribution to individual model predictions.
What the framework can and cannot show
The analysis provides a consistent way to connect spatially observed forest loss with the final consumption represented in a global economic model. It does not trace a particular product from an individual forest-loss pixel to a specific buyer. The spatial maps are allocated using national driver totals and sectoral supply-chain relationships. The permanent-agriculture category may also include some rotational harvesting in existing plantations, so it should be interpreted as a broad measure of agriculture-related forest turnover rather than a strict estimate of export-driven conversion alone.
These limitations matter because the framework is intended to complement, not replace, national forest inventories, commodity-specific tracing, or local field evidence. Its value lies in applying one transparent accounting structure across countries, years, and multiple drivers. The results suggest that international responsibility for forest loss is substantial, increasingly distributed across a multipolar trade network, and different for logging, agriculture, extraction, and infrastructure.
Continuing the research
This study grew from a collaboration between researchers at Beijing Institute of Technology and Beijing Normal University working across environmental accounting, remote sensing, ecological conservation, machine learning, and public policy. We are now extending the framework to examine forest carbon loss, protected areas, landscape connectivity, biodiversity, wildfire, and mangrove systems, while improving the spatial matching and uncertainty analysis. We welcome discussion with researchers and students working on related data, methods, or regional case studies.