Global land rush concentrates potential zoonotic spillover risk in the tropics Authors: Chuan Liao, Seongmin Shin, Katie Rohrbaugh, Nabin Pradhan, Elizabeth Ludwig, Zhenci Xu, Manuel Ruiz-Aravena, Shuai Zhou, Arun Agrawal Published: 2026 ABSTRACT AbstractGlobal land-use change alters human-environment interactions in ways that can generate sustainability risks, including pathogen spillovers across species. Here we integrate global data on large-scale land transactions and zoonotic reservoir species distributions to assess their relationship with potential zoonotic spillover risk. Using a spatially explicit approach, we identify administrative units with elevated zoonotic spillover risk based on hazard-exposure overlap, compare high-risk and other units, and examine tele-coupled investor-host country relationships associated with these patterns. We find that Southeast Asia, Latin America, Eastern Europe, and Africa are major hosts, while spillover risks are more concentrated in Southeast Asia and Latin America. High-risk units tend to feature larger transacted areas and higher forest cover, but lower population and road network density. Domestic and intra-regional transactions account for over half of transacted areas and associated spillover risks. These results highlight the importance of ecological integrity and landscape immunity in sustainability-oriented land-use governance.Subject terms: Environmental studies, Sustainability Introduction Land-use change is a major driver of landscape configuration globally, with important consequences on ecosystem functioning and ecological interactions. In the past two decades, control over 40 million hectares of land has shifted, often consolidating large areas under fewer investors, as part of efforts to secure food, fiber, and biofuel production1. This phenomenon, often referenced as land grabbing or global land rush, is global in scope but especially widespread in lower- and middle-income countries2,3. Host country governments justify large-scale land transactions (LSLTs) as a means to boost agricultural yield. Large-scale clearing of existing vegetation cover, monocropping, and commercial farming, which relies on high chemical inputs and mechanization, characterize the implementation of LSLTs for agricultural production4. While this transformation of land use has the potential for positive benefits on neighboring smallholder farming systems5, there is increasing evidence of negative socioeconomic and environmental consequences, including compromised livelihoods, increasing social tensions, soil degradation, and accelerating deforestation and greenhouse gas emissions4,6–8. Changes in land cover that commonly follow LSLTs can also pose substantial risks for public health from local to global scales. Land-use transformation from natural landscapes into agricultural areas has been associated with increased opportunities for infectious disease emergence9. Emergence and re-emergence of zoonotic pathogens, especially in the last two decades, have been linked to land conversion10–13. Such transformations can shift the distribution of zoonotic reservoir hosts, intensify or create new animal-human interfaces, and are often accompanied by increased human mobility into newly developed areas. In conjunction, these phenomena increase opportunities for pathogen spillover from animals to humans, while also creating conditions for subsequent spread within human populations14–17. At broad scales, the expansion of agricultural frontiers further amplifies these dynamics by increasing the frequency and intensity of human-wildlife interactions. Conversion of forests and rangelands into farmlands has been associated with greater risks of inter-species pathogen transmission through the expansion of contact zones between humans and wildlife18–20. These land-use changes are embedded within wider processes of globalization, including international trade, capital flows, and migration, which can extend and accelerate pathways of disease emergence and transmission21,22. Meanwhile, the relationship between land investment and actual land-use change is complex and shaped by socioeconomic, political, and institutional factors23. Not all land-based investments lead to ecological transformation, and not all landscape changes result in spillover events. However, existing research points to general mechanisms through which land-use change can create conditions conducive to zoonotic emergence15,24,25, without predicting where spillover will occur. This motivates the need for spatially explicit approaches that identify landscapes where ecological conditions associated with zoonotic spillover risk may be elevated. Recent studies also reveal that landscape immunity, defined as ecological conditions that mitigate pathogen spillover risk from host species, can be compromised by large-scale disturbance of natural habitats26,27. In particular, some zoonotic reservoir taxa prosper more in disturbed ecosystems such as newly converted agricultural lands28. These taxa are more likely to forage on or near human-dominated landscapes, such as farmlands, when their natural habitats are disturbed. Existing research suggests that such land cover change is a key factor in zoonoses emergence in West Africa, Southeast Asia, and South Asia29,30. Among these taxa, passerine birds, bats, and rodents constitute a higher proportion of species abundance and richness in agricultural landscapes15. They serve as major reservoirs for diverse human-shared pathogens and parasites, and their diversity is largely associated with pathogen richness. These groups are also generally characterized as disturbance-tolerant and widely distributed across human-modified landscapes, making them suitable indicators of ecological conditions associated with spillover hazard in areas undergoing land-use change31. Accordingly, our analysis focuses on these three taxa as indicators of relative zoonotic spillover risk. Land cover change driven by LSLTs and its spatial overlap with locations of high diversity highlight a major knowledge gap related to the externalized sustainability costs of LSLTs. Despite growing evidence linking land-use change to infectious disease emergence, there remains a limited spatially explicit understanding of where LSLT-driven land transformations coincide with ecological conditions associated with spillover risk. To address this gap, we adopt a tele-coupling perspective, which conceptualizes the spatial separation between investment decisions and the ecological consequences of land-use change in host countries32. Because LSLTs represent a political-economic pathway through which land-use change is organized and scaled, understanding where they overlap with ecological conditions associated with zoonotic hazards is important for anticipatory risk screening. In this paper, we integrate global data on LSLTs and zoonotic reservoir diversity to examine potential hotspots of zoonotic spillover risk, their location characteristics, and investor–host country relationships. By doing so, we provide a spatially explicit framework to inform sustainability governance, land-use planning, and cross-sector coordination aimed at anticipating and mitigating spillover-related risks. Results In this study, we estimate zoonotic spillover risk as a relative hazard-exposure index combining zoonotic reservoir species richness and land conversion associated with LSLTs. The exposure component accounts for variation in implementation by estimating the proportion of transacted land converted to cropland. The resulting values represent relative patterns of hazard-exposure overlap across administrative units rather than predictions of spillover events. Hotspots of LSLTs and potential zoonotic spillover risk Globally, 549 first-level administrative units host LSLTs with sizes ranging from 100 to 1.9 M ha. We identify 126 units as LSLT hotspots. Although these hotspots represent ~23% of units, they account for over 83% of total transacted area (42 M ha) (Fig. 1a). These LSLT hotspots cluster in four global regions: Africa (11.9 M ha), Southeast Asia (11 M ha), Eastern Europe (9.9 M ha), and Latin America (9.1 M ha). Overall, 64% of LSLTs occur on naturally vegetated landscapes (i.e., forests and rangelands), with substantial regional variation: 15% in Eastern Europe, but 74% in Africa, 81% in Latin America, and 83% in Southeast Asia (Supplementary Fig. 1). Global hotspots of large-scale land transactions and associated zoonotic spillover risk. a Spatial distribution of large-scale land transaction hotspots based on total transacted area. b–d Hotspot administrative units where potential zoonotic spillover risk reflects the overlap between zoonotic reservoir diversity and estimated land conversion for passerine birds (b), bats (c), and rodents (d). Circle size indicates the relative magnitude of total transacted area (a) or potential zoonotic spillover risk (b–d), calculated under the assumption of full implementation of large-scale land transactions for agricultural production. The color gradient in (b–d) indicates the estimated level of transaction implementation within each administrative unit. Shaded areas indicate first-level administrative units hosting large-scale land transactions. All values represent relative magnitudes across administrative units rather than absolute levels of risk. Our mapping of hotspots containing potential zoonotic spillover risk associated with passerine birds, bats, and rodents reveals convergent spatial distribution patterns (Fig. 1b–d). For bats (n = 120), all but two hotspots are in the tropical world. For passerine birds (n = 122) and rodents (n = 121), hotspots are also primarily located in tropical regions; however, we also identify administrative units with elevated risk in Russia, Ukraine, and China. Because most LSLTs are only georeferenced at the first administrative level, we further evaluated the sensitivity of the proportional land-cover allocation using a Monte Carlo uncertainty propagation analysis. The results produced nearly identical hotspot identification across simulations, with hotspot proportions differing by less than 1% across taxa (Supplementary Table 2). Regionally, although Eastern Europe is a major LSLT host region, it presents an apparently low spillover risk. In contrast, the findings reveal that the tropics have disproportionately higher risks, especially for administrative units in Latin America (Brazil and Argentina) and Southeast Asia (Indonesia). In Africa, the hotspots are distributed mainly along the West African coast and the East African Rift Valley. The level of LSLT implementation for agricultural development also varies substantially across regions. Transactions in administrative units in Latin America are nearly fully implemented (> 90%), while approximately half of the documented LSLTs in Southeast Asia exhibit implementation levels lower than 60%. Africa presents levels of implementation between Latin America and Southeast Asia, with relatively low values in major hosts of LSLTs such as Ethiopia, Mozambique, Nigeria, and Sierra Leone. Contextual features of zoonotic spillover risks To assess contextual characteristics that could affect variation in spillover risk and subsequent spread of zoonotic pathogens across geographical areas with LSLTs, we investigate the location characteristics of 549 administrative units hosting LSLTs. Specifically, we compare high-risk units for all three zoonotic reservoir taxa (n = 95) with the remaining units. The results suggest the high-risk units harbor substantially larger-sized LSLTs with median forest cover nearly twice as large as these other units (Fig. 2a, b, P value < 0.01), while their human population and road network densities are lower (Fig. 2c, d, P value < 0.01). Contextual characteristics of high-risk and other administrative units. Comparison of high-risk units for all three taxa of interest (n = 95) and other units (n = 454) in terms of transaction size (a), forest cover (b), population density (c), and road network density (d). Violin widths indicate data density. However, it is crucial to note that there are five high-risk units with above-average population density (94 persons/km2) in Ghana, Nigeria, Malawi, Brazil, and Indonesia, and seven with above-average road density (348 m/km2) in Indonesia, Cambodia, Guinea, Liberia, and Sierra Leone (Supplementary Tables 3 and 4). These administrative units represent priority locations for further spillover-risk assessment and monitoring, as elevated spillover risk coincides with conditions that may facilitate broader societal exposure. Comparative analysis of LSLT intensity and risk hotspots To distinguish the role of LSLTs from the broader land-use context, we estimate logistic regression models to predict hotspot membership for each taxonomic group while controlling for forest cover, population density, road density, urbanization, and country-level differences. Across passerine birds, bats, and rodents, administrative units hosting larger LSLT footprints are significantly more likely to be identified as spillover-risk hotspots (Table 1). This association remains robust after accounting for land-cover composition and infrastructure conditions, which indicates that LSLT intensity is statistically associated with hotspot occurrence even after accounting for baseline land-use context. These results suggest that LSLTs represent a governance-relevant pathway through which land-use change is concentrated in landscapes associated with elevated spillover risk. Table 1Comparative analysis of large-scale land transaction intensity and spillover-risk hotspot occurrenceVariableRodentsBatsPasserine birdslog(LSLT natural area + 1)8.77*** (2.09)7.31*** (1.53)15.15*** (4.51)Forest cover0.01 (0.05)0.03 (0.04)0.18** (0.06)Population density0.29 (1.74)0.62 (1.08)1.71 (1.14)Road density−0.02* (0.01)−0.01 (0.01)−0.00 (0.01)Urban level−1.69 (5.63)4.55 (6.01)−1.17 (7.74)N532541544Note: Logistic regression coefficients are shown with standard errors in parentheses. All models include country fixed effects, which are not reported. *P < 0.05, **P < 0.01, ***P < 0.001. Comparative analysis of large-scale land transaction intensity and spillover-risk hotspot occurrence Note: Logistic regression coefficients are shown with standard errors in parentheses. All models include country fixed effects, which are not reported. *P < 0.05, **P < 0.01, ***P < 0.001. High-risk host countries Nine countries (Russia, Indonesia, Ukraine, Argentina, Brazil, Papua New Guinea, Ethiopia, Cambodia, and Myanmar) host over 1 M ha of LSLTs, accounting for 63% of the recorded transacted area globally. The overlap of LSLT and the geographical distribution of different groups of potential zoonotic reservoir species, however, is not uniformly distributed across these countries (Fig. 3a–c). Brazil and Indonesia consistently stand out as the top two countries with the highest spillover risk across taxa. In contrast, Russia and Ukraine demonstrate lower potential spillover risk relative to their transaction size, which reflects lower reservoir species diversity and a smaller proportion of LSLTs occurring in naturally vegetated landscapes. Argentina lies slightly above the diagonal line for passerine birds, indicating a marginally elevated zoonotic spillover risk relative to its transaction area for this taxon; but it falls near or below the line for bats and rodents, suggesting relatively proportional or lower risk for these groups. Country-level relationships between transacted area and potential zoonotic spillover risk. Relationships between transacted area percentage and potential zoonotic spillover risk for passerine birds (a), bats (b), and rodents (c). The red diagonal line indicates proportional contribution, where a country’s share of total transacted area equals its share of potential spillover risk. This line serves as a descriptive reference rather than a statistical threshold or fitted relationship. Countries above the line contribute disproportionately more potential spillover risk relative to their transacted area, while those below the line contribute less. All values are expressed as normalized shares of global totals. Tele-coupled nature of LSLTs The tele-coupled nature of LSLTs suggests the need to consider the interests of both host-country governments and investors. Circular flow diagrams show the relationships between top investors and host countries (Fig. 4; n = 22). Contrary to the existing literature that highlights the transnational nature of LSLTs, we find that over a third of the recorded transacted areas are contracted domestically, with another 17% transacted intra-regionally (Fig. 4a). In this panel, the arrowheads point to investor countries, representing the virtual flow of land control resulting from LSLTs. Major shifts in land control occur within and between (1) Russia and Ukraine in Eastern Europe, (2) Malaysia, Singapore, Cambodia, Myanmar, Indonesia, and Papua New Guinea in Southeast Asia, and (3) Brazil and Argentina in Latin America. China and the US are the two largest investors outside of these regional hosts of LSLTs. Tele-coupled investor–host relationships in land transactions and associated zoonotic spillover risk. a Land area transacted through large-scale land transactions (LSLTs) among the top 22 host and investor countries. b–d Relative potential spillover risk contributions associated with each bilateral link for passerine birds (b), bats (c), and rodents (d). In all panels, the arrowhead points to the investor country, indicating the virtual flow of land control to investors through LSLTs in (a), and the association of potential spillover risk with investor–host land transaction relationships in (b–d). Ribbon width reflects the magnitude of transacted land area (a) or potential spillover risk (b–d), and ribbon color indicates the host country’s regional affiliation. Owing to variations in zoonotic reservoir taxa diversity, different host-investor relationships might have distinct implications for spillover risk. Southeast Asia and Latin America exhibit disproportionately higher risks relative to their transaction sizes (Fig. 4b–d). The arrowheads point to investor countries to illustrate tele-coupled relationships, which link external investment flows to local ecological conditions in host countries where spillover risk may be elevated. In Southeast Asia, the vast majority of risk is associated with domestic and intra-regional transactions, with China being the major trans-regional investor. In Latin America, areas of elevated spillover risk are largely linked to transactions in and between Argentina and Brazil, with the US and the Netherlands accounting for major trans-regional investment linkages. Although Africa is a major host of LSLTs at a regional scale, individual African countries host smaller-sized transactions compared to other regions. Among the top four African countries, LSLTs in Ethiopia and Nigeria are primarily domestic, while those in Ghana and Congo are largely transnational, linked to Indonesia, the UK, and the US. The largest transnational investors are China and the US, both involved in similarly sized trans-regional LSLTs. However, Chinese agricultural investments are associated with higher zoonotic spillover risk due to their concentration in biodiversity-rich regions, particularly in Myanmar and Indonesia. Discussion Our analysis offers a spatially explicit integration of zoonotic host species diversity and LSLTs to identify patterns of potential spillover risk globally. While recent research has highlighted the role of global finance in driving ecological and health risks21,22, our study adds value by operationalizing these links at the subnational level with ecological and spatial specificity. We highlight the necessity of considering zoonotic spillover as an externalized sustainability cost of implementing LSLTs for agricultural development. Our findings reveal broadly convergent spatial patterns of spillover risk across zoonotic reservoir taxa, while highlighting variation in the intensity of risk and the socioeconomic and environmental characteristics associated with high-risk units. These units represent areas where land-use change associated with LSLTs overlaps with elevated ecological hazard and thus serve as priority areas for further assessment and governance attention. The large spatial extent (> 40 M ha) of areas with elevated spillover risk poses a major challenge for governance systems seeking to anticipate and manage risk. Evidence from past outbreaks, such as Ebola virus in Africa33, Nipah virus in Bangladesh34, Hendra virus in Australia35, SARS-related coronaviruses in China36, and several infections in Southeast Asia28 illustrates general mechanisms through which land-use change can create conditions that facilitate zoonotic transmission, including altered wildlife dynamics and increased human-wildlife contact. Even for high-profile infections such as the Ebola virus, cases were detected only after transmission had been ongoing for weeks33, which underscores the limitations of reactive approaches. One proactive approach we suggest is that both host country governments and investors need to rethink the dominant model of high-input, chemical-intensive agricultural practices that underpin many LSLT-driven development strategies20,37. Potential risk mitigation strategies include enhancing agricultural yield in an ecologically sustainable manner38,39, such as requiring parts of the transacted area under original land cover40, limiting intensive farmland expansion in highly biodiverse landscapes41, and promoting diverse agroecological matrices that foster greater landscape immunity42. Broader food-system strategies, including circular nutrient management, may support sustainable intensification and help reduce pressure for land expansion43. Recent cost-benefit analyses demonstrate high returns from sustainable intensification when coupled with land-use zoning, market and credit restrictions, and state-of-the-art satellite monitoring, which can effectively balance the goals of boosting yields and reducing zoonotic spillover risks10. In addition to highlighting hotspots with elevated spillover risk, we identify 95 administrative units as having high zoonotic spillover risk that may warrant priority attention for precautionary governance and further spillover-risk assessment. In those areas, environmental ministries and land-use planners could benefit from maintaining high ecological integrity (i.e., habitat intactness, connectivity, generative tropic system) to foster landscape immunity26, which contributes to sustaining the positive effects of biodiversity by reducing transmission of infections via processes such as dilution effects. Among the 95 high-risk units, the 12 characterized by high human population and road densities may warrant particular attention, given their potential to facilitate rapid post-spillover spread. Accordingly, governance responses need to be tailored to local conditions rather than applied uniformly across LSLT settings. To support this, we outline several precautionary governance considerations. First, host country environmental agencies and land-use planning authorities could assess the overlap between proposed LSLTs and known zoonotic reservoir species distribution prior to project approval and implementation, particularly in identified hotspot units. Second, regulatory bodies could consider requiring and enforcing risk-mitigation plans during the implementation phase, including measures to alleviate stress on wildlife populations and establish biosecurity protocols for communities living in and around LSLTs, particularly during early stages of land conversion. Third, investor-country institutions and development finance actors, including private investors, agribusinesses, and financial intermediaries such as commercial banks and asset managers, should incorporate zoonotic spillover risk into environmental and social due diligence frameworks. In regions where LSLTs are prevalent, such as South America and Southeast Asia, coordination across sectors can strengthen local capacity for monitoring and response, particularly in areas with high population density and connectivity44. These measures can be most effective when designed and implemented in a context-specific manner to reflect heterogeneity in land-use dynamics and governance arrangements and are intended to support anticipatory screening and prioritization rather than prescribe uniform interventions. The relationships between host and investor countries further demonstrate that addressing landscapes of elevated spillover risk is a governance challenge that extends across administrative jurisdictions and sectors, which is consistent with the One Health approach. We highlight three general considerations for governance to move towards landscape immunity as part of national and global security agendas. First, the large share of domestic LSLTs reveals opportunities for host countries to coordinate across agricultural, environmental, and public health sectors. Alliances among NGOs, communities, and public and private actors can help conserve biodiversity and monitor human-wildlife interactions relevant to spillover risk45. Second, given the substantial role of intra-regional LSLTs, regional intergovernmental organizations, such as the Association of Southeast Asian Nations (ASEAN), Latin America Integration Association (LAIA), and Commonwealth of Independent States (CIS), are well-positioned to facilitate regional alignment between agricultural development and biodiversity conservation objectives. Third, the dominance of a small number of investor countries, including China, the US, Malaysia, and the Netherlands, suggests that regulatory action and public advocacy in these countries can play a critical role in shaping global land-use outcomes. Cross-sectoral collaboration among international organizations, including WHO, FAO, WOAH, and UNEP, supported by commitments from major investor countries, may help realign economic decision-making with planetary health protection46. Several limitations of this study warrant consideration. First, our analysis does not explicitly incorporate vulnerability dimensions, such as variation in public health infrastructure, socioeconomic conditions, or baseline population health. While this enables global comparison for prevention-oriented risk screening, it constrains inference regarding outbreak amplification following spillover. Future work could integrate vulnerability indicators to produce more nuanced zoonotic spillover risk assessments. Second, the spatial resolution of LSLT data is coarse, with most transactions georeferenced only at the first administrative level. Consequently, land-cover change is estimated using administrative-level proportions, which may obscure finer-scale heterogeneity. Third, we rely on proxies to infer both LSLT implementation status and associated land conversions, recognizing that not all LSLTs result in immediate or complete landscape transformation. More broadly, land-use frontiers are shaped by contested and nonlinear processes involving governance, tenure systems, and local resistance25, and by multiple pathways beyond LSLTs, including smallholder agricultural expansion, infrastructure development, and other forms of land-based investment, which are not explicitly captured in our analysis. Fourth, due to limited spatially and temporally explicit pathogen data, we rely on host species richness and land-conversion pathways as proxies for spillover risk rather than direct measures of pathogen prevalence or transmission. Species richness captures broad ecological conditions but does not reflect host competence or transmission dynamics and may be influenced by dilution or dominance effects. In addition, our focus does not capture other important spillover pathways, such as those involving primates, livestock-wildlife interfaces, or vector-borne systems, and thus represents a partial assessment of ecological hazard rather than total zoonotic risk. The COVID-19 pandemic illustrates the global consequences of pathogens emerging locally and spreading unchecked through human populations. Beyond the trillions of dollars in economic losses, broader costs in lives lost, disrupted education, and diminished well-being are incalculable. Recovery will take years, if not decades. Importantly, COVID-19 will not be the last pandemic; on average, major outbreaks impose economic costs exceeding USD 500 billion annually47. These realities underscore the urgency of integrating zoonotic spillover risk into sustainability governance and land-use decision-making, particularly in the context of the ongoing global land rush. Methods We use LSLT data from Land Matrix, which documents 2134 cases of domestic and transnational land transactions in lower- and middle-income countries since the early 2000s48. Land Matrix publishes data collected from media reports, company resources, research and policy papers, and government sources with a well-defined protocol, and data sources are triangulated by different sources. Among all deals, 312 have exact locations, 1777 have approximate location information at least at the subnational level (first-level administrative unit), and 47 only have location information at the country level (excluded in our analysis due to low spatial accuracy). The dataset reports the size, investor, and host country of each transaction. It also aims to track the implemented area of each land deal for agricultural development, with 961 cases containing information on the area converted into farmlands. The dataset used in this study reflects the most recent version publicly available at the time of download (2020); therefore, more recent transactions may not be captured. To assess zoonotic spillover risk from passerine birds, bats, and rodents, we obtain their diversity data from BiodiversityMapping.org49. The dataset includes the total species richness of major taxonomic groups, which we interpret as a proxy for potential ecological hazard for zoonotic spillover that reflects the presence of ecological conditions that may facilitate spillover. The map of passerine birds is derived from BirdLife International. Maps of bat and rodent diversity include total species richness of major taxonomic orders, and the data are derived from IUCN. The spatial resolution of species diversity maps is 10 km. We define zoonotic spillover risk as a relative, prevention-oriented risk index that captures the potential for disease emergence based on the spatial overlap between zoonotic reservoir species diversity and the estimated extent of land conversion from natural to agricultural landscapes. Accordingly, we calculate risk as: where risk refers to relative hazard-exposure index representing potential ecological conditions associated with spillover, hazard refers to zoonotic reservoir species diversity, and exposure refers to the areal extent of land conversion from natural to agricultural landscape50. This formulation is not intended to represent a mechanistic or linear relationship between land conversion and spillover risk. Rather, it serves as a parsimonious metric to identify locations where two necessary components of spillover landscapes, namely reservoir host presence and anthropogenic disturbance, co-occur. Species richness is therefore interpreted as a proxy for broad ecological hazard landscapes in large-scale comparative analysis, rather than a measure of host competence, pathogen prevalence, or transmission dynamics. This approach is consistent with prior global-scale studies that use proxy-based indicators to identify relative spillover-risk landscapes under data constraints12,17. While some risk frameworks incorporate vulnerability as a third component in addition to hazard and exposure, we follow a widely used two-factor approach that estimates zoonotic spillover risk as the product of ecological hazard and land-use exposure26,28. This formulation aligns with prior studies that quantify the relative spatial risk of zoonosis emergence based on environmental drivers, particularly in data-sparse or global contexts. Our approach is therefore designed to capture where zoonotic spillover is most likely to originate due to ecological disturbance, rather than to quantify the potential severity or spread following emergence. Once natural habitats are disturbed and fragmented, these host species can become reservoirs of pathogens for other species15. Therefore, for hazard, we assume that locations with higher zoonotic reservoir species diversity are more hazardous under the condition of land cover change, and use the relative species diversity calculated as: to represent the level of hazard for disease spillover from passerine birds, bats, and rodents. The relative diversity takes values between 0 and 1. For exposure, we consider land-use change pathways associated with LSLTs. While LSLTs typically occur in naturally vegetated landscapes (forests and rangelands), they also take place on existing croplands51. However, transactions involving cropland-to-cropland conversions are excluded from our risk calculation, as these are unlikely to substantially alter local biodiversity or increase human-wildlife contact. This distinction is supported by evidence showing that zoonotic spillover risk is most strongly associated with transformations from natural vegetation into agricultural land, rather than with shifts within already cultivated lands50. To operationalize exposure, we estimate the area of land conversion, specifically the transition from natural land cover to agricultural use within each LSLT. Because 85% of LSLTs are only georeferenced to the first-level administrative unit, we estimate pre-conversion land cover by applying the proportional distribution of cropland, forest, and rangeland within each unit to the reported transaction area. This approach allows us to capture the ecological disturbance associated with each LSLT and distinguish between transactions that result in substantive land cover change versus those that do not. While we know LSLTs are intended for crop production, prior land cover is unknown from the data source. Therefore, we determine proportions of land cover at the administrative level and assume equivalent proportions at the land deal level. To estimate proportions of land use for each administrative unit, we first create a baseline map for 2000 based on the global land cover dataset52. Specifically, using the MODIS Land Cover Type (MCD12Q1) dataset in 2000, we reclassify land cover types into three groups (forest, rangeland, and cropland), and excluded urban/built-up areas and water bodies where LSLTs are unlikely to occur (Supplementary Table 1). Then, we calculate the proportions of forest, rangeland, and cropland in each administrative unit, and apply the estimated proportions as proxies of the land cover types on transacted lands prior to implementation for agricultural development. Based on the reclassified land cover types, we estimate the distribution of croplands and naturally vegetated landscapes (including forests and rangelands) in four major regional hosts of LSLTs. Because most LSLTs are only georeferenced at the first administrative level, allocating land-cover composition proportionally within administrative units may introduce uncertainty in the estimation of exposure. To evaluate the sensitivity of this assumption, we conducted a Monte Carlo uncertainty propagation analysis. For each administrative unit hosting LSLTs, the land-cover composition of the estimated LSLT footprint was randomly drawn from a Dirichlet distribution centered on the observed proportions of forest, rangeland, and cropland. The resulting exposure estimates were propagated through the hazard-exposure risk calculation and hotspot identification procedures. We repeated this process across 1000 simulations and compared the resulting hotspot classifications with the baseline proportional allocation approach. Details of this procedure and robustness results are reported in the Supplementary Materials. After rescaling species diversity and determining the area of LSLT under natural vegetation cover, our calculation of risk is performed below: Under the assumption that unimplemented LSLT areas present substantially lower zoonotic disease risk, it is necessary to determine the level of implementation for each case of LSLT. With available implementation information on 961 cases, we infer the level of land conversion for cases with such information. Specifically, we estimate the average ratio of implemented LSLTs in each country unit using the 961 LSLTs. Then we assign the average level of implementation to each LSLT without information on the implementation level according to its host country. Subsequently, we estimate the total area of land cover changes in each administrative unit by summing LSLTs with known implemented areas and LSLTs with inferred implemented areas under the average level of implementation. At last, we calculate the percentage of implementation for each administrative unit by dividing the total implemented area under natural vegetation cover by the total transacted area under natural vegetation cover. We follow the method by Darrouzet-Nardi (2025)53 to identify hotspots, defined as administrative units with disproportionately high values based on statistical deviance from a reference distribution. Specifically, we use the hotspots package in the R statistical software54, which flags outliers in a univariate distribution. In this study, hotspots are interpreted as distributional extremes rather than spatially contiguous or spatially independent clusters. Accordingly, LSLT hotspots refer to administrative units where the total size of land transacted exceeds the identified outlier threshold. Similarly, we identify zoonotic spillover risk hotspots using the same method, based on our calculated risk index. The hazard component reflects the rescaled species diversity of zoonotic hosts (bats, rodents, and passerine birds), while the exposure component represents the extent of natural vegetation cover in LSLTs. This distribution-based approach enables consistent cross-regional comparison of hotspot units across heterogeneous administrative and ecological contexts, without imposing assumptions about spatial autocorrelation. Administrative units with zero estimated exposure are retained in the analysis with risk values equal to zero and contribute to the reference distribution used for hotspot identification. To distinguish the role of LSLTs from the broader land-use context, we conducted a comparative analysis examining whether variation in LSLT intensity explains hotspot occurrence beyond baseline environmental conditions. Specifically, we estimated logistic regression models predicting hotspot membership for each taxonomic group. The models include LSLT intensity (log-transformed area of LSLT under natural vegetation) as the primary explanatory variable and control for forest cover, population density, road density, urbanization level, and country-level fixed effects. These models allow us to assess whether provinces hosting larger LSLT footprints are systematically more likely to be classified as spillover-risk hotspots after accounting for land-cover composition and infrastructure conditions. To inform governance-relevant interpretation, we investigate characteristics of locations where the documented LSLTs occurred that likely facilitate initial animal-human transmission and subsequent spread of a pathogen in the human population. Accordingly, we obtain two covariates (in our baseline year 2000), which are forest cover and human population density. We also obtain the global road network density data from the Global Roads Inventory Project. The spatial resolution is 5 arc minutes, and the value represents the distance of roads in each grid cell (m/km2). For each raster image, we calculate the mean value within each administrative unit. We then use the two-sided Kolmogorov-Smirnov (K-S) test to statistically assess whether the distributions of these variables differ between high-risk administrative units (i.e., zoonotic spillover risk hotspots) and all other units. The K–S test is a nonparametric method suitable for comparing the distributions of continuous variables across two independent groups, without requiring assumptions about the underlying data distribution. At the country level, we categorize each LSLT according to investor (n = 118) and host (n = 92) countries and calculate the total number and area of land transactions in which each country is involved. For host countries, we investigate the proportional contribution to spillover risk by considering the size of LSLTs in each country. We use the circlize package55 to create circular flow diagrams to examine the complex and dynamic nature of LSLTs among the top investor and host countries (n = 22), which illustrates how spillover risk is distributed across domestic, intra-regional, and trans-regional investment pathways. Supplementary information Transparent Peer Review file Supplementary Information Acknowledgements This research was supported by the National Science Foundation (grant number 1617364), the National Aeronautics and Space Administration (grant number NNX15AD40G), and Cornell Center for Social Sciences. Author contributions C.L. designed the research, conducted the analyses, and led manuscript writing. S.S. contributed to the result interpretation and manuscript writing and editing. K.R. contributed to the result interpretation and manuscript writing and editing. N.P. contributed to the result interpretation and manuscript writing and editing. E.L. contributed to the result interpretation and manuscript writing and editing. Z.X. contributed to the result interpretation and manuscript writing and editing. M.R.-A. contributed to the result interpretation and manuscript writing and editing. S.Z. contributed to the result interpretation and manuscript writing and editing. A.A. contributed to research framing, result interpretation, and manuscript writing and editing. Peer review Peer review information Communications Sustainability thanks Yusuf Amuda Tajudeen, Soushieta Jagadesh and the other, anonymous, reviewer(s) for their contribution to the peer review of this work. Primary Handling Editors: Hamid Rabiei and Yann Benetreau. A peer review file is available. Data availability All datasets used in this study are publicly available. Large-scale land transaction data are available at https://landmatrix.org/. Biodiversity data for passerine birds, bats, and rodents are available at https://biodiversitymapping.org/. Land-use and land-cover data are available through Google Earth Engine at https://developers.google.com/earth-engine/datasets/catalog/MODIS_061_MCD12Q1. Additional covariate datasets and access information are described in the Supplementary Information. Processed data necessary to reproduce the analyses and figures are available on Zenodo at 10.5281/zenodo.20332275. Code availability R code used to process the data, conduct the analyses, and generate the figures is available on Zenodo at 10.5281/zenodo.20332275. Competing interests The authors declare no competing interests. Footnotes Publisher’s note Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations. Supplementary information The online version contains supplementary material available at 10.1038/s44458-026-00102-1. References Associated Data Supplementary Materials Transparent Peer Review file Supplementary Information Data Availability Statement All datasets used in this study are publicly available. Large-scale land transaction data are available at https://landmatrix.org/. Biodiversity data for passerine birds, bats, and rodents are available at https://biodiversitymapping.org/. Land-use and land-cover data are available through Google Earth Engine at https://developers.google.com/earth-engine/datasets/catalog/MODIS_061_MCD12Q1. Additional covariate datasets and access information are described in the Supplementary Information. Processed data necessary to reproduce the analyses and figures are available on Zenodo at 10.5281/zenodo.20332275. R code used to process the data, conduct the analyses, and generate the figures is available on Zenodo at 10.5281/zenodo.20332275.