Funds

The Race for Resources: Why Miners Fit Your Portfolio

Private Trade News trader edition (2026-11-03): As geopolitical conditions and AI integration rewire global supply chains, natural resources are becoming increasingly vital to economic resilience. This was… Primary source: original at ETF Trends (etftrends.com).

November 3, 2026 9:42 PM GMT · ETF Trends

Abstract

Global mineral extraction is expected to surge due to the growing demand for clean energy. While mining is critical to modern society, its environmental impacts, though increasingly studied, remain undocumented for half of the world’s mining areas and are rarely analysed at the commodity level. Here, we introduce a novel approach integrating remote sensing, machine learning, and cloud computing to classify approximately 70,000 mining sites by commodity. Using this newly detailed dataset, we quantify the nature loss associated with 20 extracted commodities, focusing on deforestation and habitat destruction. From 2001 to 2022, mining activities worldwide resulted in the removal of 16,268 km² of forest cover, with 65.64% occurring in tropical and subtropical regions. Notably, approximately half of this deforestation was attributed to the extraction of gold, coal, aluminium (bauxite), nickel-cobalt and copper, primarily in countries intersecting the Amazon, Southeast Asian, and Congo Basin rainforests. Our analysis also reveals that deforestation-to-mining area ratios and biodiversity risks vary by mining location, and conservation threats do not always scale with deforestation rates. By providing commodity-specific maps of mining-induced nature loss, our work equips companies and organisations with actionable insights to identify risks within their supply chains and implement targeted mitigation strategies.

Introduction

The global economy is advancing toward carbon neutrality, yet achieving a low-carbon future will remain impossible without a steady supply of minerals1. Clean energy technologies are expected to be significantly more material-intensive than traditional fossil fuel-based systems2,3. While overall mineral demand will continue to rise1,3, potentially increasing fivefold for some critical metals by 2050 compared to 2018 production levels1, current mineral extraction practices rank among the most socio-environmentally destructive human activities3,4,5. Decoupling the escalating demand for extractive resources from unsustainable outcomes requires a comprehensive and nuanced understanding of the mining industry’s diverse impacts.

Severe and enduring impacts from mining activities, including deforestation, biodiversity loss, pollution, water scarcity, greenhouse gas (GHG) emissions, disasters (e.g. tailings dam failures, fires, and ground collapses), and health and social risks, are widely documented in numerous case studies3,6,7,8. Forest clearing is a primary concern among these impacts, given that mineral deposits are frequently located beneath forested areas9. Mining-induced deforestation is particularly prevalent in tropical regions such as Amazonia10,11,12,13, West Africa14,15, the Congo Basin16,17, and Indonesia18,19, but it also occurs in non-tropical nations, including the United States20, Russia, and Canada21, to meet global demand for a wide range of minerals and metals. Water pollution resulting from the discharge of mining-related heavy metals and toxic chemicals has been reported since the early twentieth century22. Mercury contamination of streams arising from gold and silver extraction in Latin America23,24 and the release of rare earth elements and heavy metals into rivers, groundwater, soil, air, and food in China5,25,26,27 are notable examples. The pollutants, along with deforestation and even noise associated with mining, drive biodiversity declines across various ecosystems8, such as rainforests28,29,30, boreal forests31, rivers22,32, wetlands33,34, savannahs35, and grasslands36,37,38, through habitat fragmentation, degradation, and destruction. Furthermore, mineral and metal production emits significant GHGs39 and remains challenging to decarbonise5. Freshwater competition40, groundwater depletion41, and disasters42,43 further compound the industry’s impacts, ultimately threatening human health44,45,46 and well-being47. In addition, human rights violations often accompany both large- and small-scale extractive activities48,49.

In addition to findings derived from local and regional studies, recent advancements have illuminated several global-scale externalities of mining, including water conflicts50, river pollution51, tailings dam failures52, deforestation21,53, and biodiversity threats4,54,55,56,57. These global assessments extend the understanding of mining’s impacts beyond site-level case studies, revealing cumulative and transboundary patterns of environmental change. Such insights have been facilitated either by mineral deposit databases from S&P Global58 and the United States Geological Survey (USGS)59 or by manual identification using high-resolution satellite imagery60,61. While instructive, these efforts often focus only on the tropics51,53, cover a limited range of mining locations51,52,57, fail to distinguish individual commodities21,53,54, or consider only a few key metals4,50,55,56,57. Furthermore, methods relying on point data of mine coordinates frequently apply generalised buffer zones rather than actual land-use boundaries to estimate mining impacts4,54. Consequently, a systematic, global overview that integrates detailed land-use boundaries and commodity-specific classifications remains a critical gap in our understanding of the mining footprint.

Mapping commodity-level mining footprints requires labelled land-use data that specify which minerals or metals are extracted from each site. Unfortunately, such data is currently unavailable. Commercial records58 provide production information for only about 44% of mining sites62, and this coverage is even lower in open data sources63. However, recent advances in remote sensing, machine learning, and cloud computing offer new opportunities for automated mining land-use classification at a global scale. Remote sensing, in particular, has proven invaluable for various geological applications, including lithology mapping, hydrothermal alteration exploration, and mineral prospecting64,65. Among the available satellite datasets, ASTER (Advanced Spaceborne Thermal Emission and Reflection Radiometer) effectively identifies new mineral deposits65, such as gold66, copper67, other critical metals68, and lithologic features69, across multiple countries. Its global coverage, six SWIR (Shortwave Infrared) channels70, and numerous spectral band ratios71 tailored to enhance mineral detection, make it a critical tool for advancing mining mapping efforts.

In the same manner, the Hyperspectral Imager Suite (HISUI), developed by the Japan Ministry of Economy, Trade and Industry and launched aboard the International Space Station in 2019, has provided hyperspectral data with 185 spectral bands ranging from the visible to shortwave infrared regions72. HISUI’s finer spectral resolution enables improved discrimination of mineralogical and lithologic variations compared with multispectral sensors such as ASTER, offering promising potential for high-precision mineral mapping and commodity-level classification.

We, therefore, develop an automated mining land-use classification method leveraging ASTER and HISUI imagery to assign 20 distinct commodities to approximately 70,000 mining extents (polygons). Using this newly detailed dataset, we assess where and to what extent each mineral and metal drives deforestation and threatens biodiversity through habitat loss on a global scale. The resulting maps provide a clear and systematic measure of the responsibility borne by specific commodities for the two significant environmental impacts of the extractive industry, highlighting their implications for Earth’s health and sustainability. Specifically, they lay the groundwork for informing regulatory measures such as the European Union Deforestation Regulation73 (EUDR) and voluntary initiatives like the Taskforce on Nature-related Financial Disclosures74 (TNFD) to develop mineral commodity-specific strategies/guidelines for mitigating nature loss and promoting sustainable mining practices.

Results

Key region-commodity links in mining-induced deforestation

Globally, mining activities removed approximately 16,268 km² of forest, of which 65.64% was tropical and subtropical forest, from 2001 to 2022 (Fig. 1a and Supplementary Table 2). Commodity classification shows that coal and gold mines accounted for over 49% of the total mining land analysed (Supplementary Table 1 and Supplementary Data 1). Correspondingly, these commodities caused the most significant deforestation, with coal mines contributing approximately 5474 km² (33.65%) and gold mines approximately 5125 km² (31.50%) of forest loss. Other non-metals (~2271 km², 13.96%), aluminium (bauxite) (~940 km², 5.78%), copper (~662 km², 4.07%), iron (~612 km², 3.76%), nickel-cobalt (~413 km², 2.54%), and tin (~260 km², 1.60%) follow as the next most deforested commodities (Fig. 1a and Supplementary Table 2). By contrast, tungsten, uranium, and chromium resulted in minimal deforestation because their mines are predominantly located outside forested areas. Notably, deforestation rankings are not fully aligned with mining area rankings; for instance, although titanium ranks 17th in mining areas (0.40%), the deforestation attributable to its mines ranks 10th (~80 km², 0.49%).

On average, mining causes approximately 739 km² of forest loss per year, but trends in deforestation within mining areas have changed over time. Between 2001 and 2015, deforestation generally rose, then began to decline after 2015 (Fig. 1a). This shift may reflect how mines expand horizontally during the initial phases of operation, followed by deeper excavation in subsequent phases, leading to less deforestation in later years. Tin exhibited the highest share of deforestation in the tropics and subtropics (93.96%), followed by manganese (87.03%), gold (86.50%), nickel and cobalt (78.35%), tungsten (76.85%), and aluminium (70.97%). However, certain commodities, such as lead and molybdenum, caused more deforestation in other biomes (Fig. 1a).

We further examined deforestation-to-mining area ratios for each site as a comparative indicator of direct deforestation impacts (Fig. 1b). It should be noted, however, that some deforestation may have occurred before 2001, which lies outside the temporal scope of our forest loss dataset. Although mining-induced deforestation areas vary significantly by commodity, the differences among individual mining sites are even more pronounced. Among commodities with high deforestation impacts, lithium and aluminium show relatively higher median deforestation-to-mining area ratios of 10.77% and 5.06%, respectively, suggesting that these commodities are associated with more intensified mining or are often located near or within forests. Despite its relatively smaller deforestation impact (260 km²) compared to major commodities such as coal and gold, tin mining has the third-highest median deforestation-to-mining area ratio (3.70%) among all commodities. Conversely, commodities with lower or near-zero median deforestation ratios tend to be associated with less forested or previously disturbed landscapes, smaller-scale or less recent mining activities, or sites where major forest clearing occurred before 2001. For instance, gold mining shows a relatively low median ratio of 2.10% despite its substantial overall contribution to deforestation, reflecting a spatially uneven distribution of impacts across mining sites.

Figure 1c presents the confidence intervals (CIs) for deforestation areas accounting for classification accuracy. The blue bars show the mapped deforestation areas derived from the predicted classes for the full dataset, while the black 95% CIs represent uncertainty estimates from Monte Carlo (MC) simulations based on the random-sampling confusion matrix, which aggregates minor commodity groups (Pb, Cr, Mn, Mo, Ni, Sn, W, Ti, Li, PGE, U, and Other metals) into a single Rare metals category (see Supplementary Note 2 for details). The figure shows that most commodities have mapped deforestation areas (blue bars; minor commodity groups are also aggregated as Rare metals for consistency) that fall within the CIs, and only Gold, Rare metals and Coal fall outside the intervals. For these three commodities, the deviations are small relative to their total deforestation magnitudes, with Gold differing by 141 km² (2.8%), Rare metals by 111 km² (10.8%) and Coal by 252 km² (4.6%). The overall ordering of commodities by deforestation magnitude is also preserved when comparing the mapped values with the MC intervals. Together, these results indicate that the deforestation derived from the predicted classes is robust to classification uncertainty.

The deforestation-to-mining area ratios exhibit substantial spatial heterogeneity, regardless of commodity, as shown in the maps (Fig. 2 and Supplementary Data 2). Iron mining, for instance, has a widespread impact, with deforestation-to-mining area ratios reaching 30–70% in South America (e.g. eastern Brazil), and up to 40–90% in Southeast Asia (e.g. parts of Indonesia) (Fig. 2a). These findings align with a report75 indicating that a significant portion of tropical forests (nearly 25%) are now within active mining concessions. For example, the N4WS iron ore mine (owned by the Brazilian mining company VALE and part of the Carajás Mining Complex) is situated within the Amazon rainforest75. Similarly, gold mining exhibits widespread impacts, with higher deforestation-to-mining area ratios observed across countries and regions, particularly in tropical regions such as Southeast Asia (e.g. parts of the Solomon Islands and Indonesia; >70%), South America (e.g. parts of the Amazon Basin and Venezuela; 40–70%), and West and Central Africa (e.g. parts of Ivory Coast, Sierra Leone, and the Democratic Republic of the Congo; 40–90%) (Fig. 2b). Aluminium mining contributes to deforestation-to-mining area ratios exceeding 50–80% in tropical rainforests, particularly in parts of Indonesia, as well as in South America and Australia (Fig. 2c). While most commodities cause substantial deforestation, lithium mining generally has a more moderate impact (Fig. 2d). Brine-based extraction in arid regions like the lithium triangle (Chile) and Qinghai, China, results in minimal deforestation (<10%) due to sparse forest cover76. At the same time, hard-rock lithium mining, common in regions such as Australia and Canada, leads to moderate to high deforestation77 (20%–70%).

While global averages in deforestation-to-mining area ratios seldom rise above 10%, the examples above demonstrate that certain region-commodity combinations can produce pronounced deforestation hotspots. Notably, approximately 50% of total deforestation (~8156 km² out of ~16,268 km²) is concentrated in countries intersecting the three major rainforest regions: the Amazon, Southeast Asian, and Congo Basin rainforests. This deforestation is primarily associated with gold (20.8% of total deforestation ~16,268 km²), coal (20.2%), aluminium (2.5%), nickel-cobalt (1.7%), copper (1.4%), tin (1.4%), other non-metals (1.0%), and iron (0.7%) mining (Supplementary Table 3).

Expanded impacts of mining on biodiversity

Figure 3a shows the distribution of the Extinction Risk Index (ERI) across 20 mining commodities (Supplementary Table 4). The ERI measures biodiversity risk using species extinction risk scores based on the IUCN Red List78, ranging from 0 (least concern) to 1 (extinct). Notably, the ERI is a reversed version of the Red List Index (RLI). Tungsten and lithium exhibit the highest median ERI values (~0.30 and ~0.28), followed by chromium and PGE (both ~0.15), tin and aluminium (bauxite) (both ~0.10), gold, copper, and nickel-cobalt (all ~0.07). In contrast, other commodities have lower or near-zero median values. High median ERI commodities are typically associated with extraction in biodiversity-rich or ecologically sensitive regions, indicating greater potential risks to threatened species even when deforestation itself is limited. In contrast, low median ERI commodities may reflect extraction in less biodiverse areas or a high proportion of sites with minimal extinction risk (i.e. many zero values), such as iron and coal operations that are distributed across multiple biomes, many of which naturally have lower biodiversity outside tropical and subtropical regions. Figure 4 further examines this through each commodity’s global spatial distribution of ERI values.

Figure 3b shows the CIs for ERI values accounting for classification accuracy. The blue bars present the mean ERI values derived from the predicted classes for the full dataset, while the black 95% CIs show the uncertainty estimates obtained from MC simulations based on the random-sampling confusion matrix, which aggregates several minor commodity groups into a single Rare metals category (see Supplementary Note 2). The figure shows that the ERI values derived from the predicted classes generally align with the CIs, with a few commodities lying very near the CI boundaries (Zinc, Silver, Aluminium and Coal differ from the nearest CI bound by only about 0.001–0.01). The one relatively larger difference between the bars and CIs occurs for Other non-metals (0.019), which is expected because this category aggregates many heterogeneous non-metallic sites; this leads to higher misclassification rates and therefore greater variation in the ERI values included in the MC resampling.

In Fig. 4a, iron mining shows high ERI values in parts of Chile (>0.45), Southeast China (>0.30), and Southern Africa (e.g. parts of South Africa; >0.30). Copper mining exhibits high ERI values in South America (e.g. parts of Chile; >0.45), Africa (e.g. parts of the Democratic Republic of the Congo; 0.30–0.45), Southeast China (0.30–0.45), and eastern Australia (>0.45) (Fig. 4b). Nickel and cobalt mining impacts biodiversity across several regions, including Southeast China (0.30–0.45), South Africa (0.30–0.45), and parts of North America, Southeast Asia and Australia (e.g. Canada and Myanmar; >0.30) (Fig. 4c). Lithium extraction in Chile (the Lithium Triangle) shows the highest biodiversity impacts (>0.45), likely driven by habitat disruption in sensitive salt-flat and wetland ecosystems resulting from brine-based extraction79 (Fig. 4d).

It is important to note that deforestation levels do not necessarily correspond to areas of highest biodiversity risk. As illustrated by the lithium mining case, biodiversity risks from mining extend beyond forest-dwelling species. For example, the Lithium Triangle in Chile exhibits low deforestation yet high ERI values, as brine extraction fragments habitats and pollutes ecologically fragile salt-flat and wetland ecosystems characterised by naturally sparse vegetation. Specifically, the Andean mountain cat (Leopardus jacobita, Endangered) and the Andean flamingo (Phoenicoparrus andinus, Vulnerable), identified by the IUCN78 as threatened by mining-related seepage, occur in these areas. This comparison clarifies that deforestation was not used in this study as a proxy for biodiversity impact; rather, it highlights that deforestation data alone may underestimate risks in ecosystems with low vegetation cover but high ecological sensitivity8. At the same time, it is important to acknowledge the limitations of the Global Forest Watch dataset80, used in this study to quantify deforestation, which cannot distinguish between natural forests and plantations81. Consequently, areas showing high deforestation on our maps may correspond to plantation harvesting rather than loss of natural forests, and therefore may not represent equivalent impacts on biodiversity.

Discussion

The global deforestation estimate of this study (~16,000 km², 2001–2022) is slightly higher than the ~14,000 km² reported by the previous global assessment21. This difference reflects our longer temporal coverage and the use of gross tree-cover loss without applying a >30% canopy-cover threshold, which ensures the inclusion of all forested biomes, whereas the previous study focused primarily on humid tropical regions21.

The findings provide evidence to support nature-related risk regulatory measures, such as the EUDR73, in formulating commodity-specific mineral import restrictions in addition to its current focus on agriculture. This is particularly relevant for mineral commodities like iron, gold, aluminium (bauxite), and coal, which are strongly linked to substantial deforestation. The EU’s Carbon Border Adjustment Mechanism (CBAM)82 also regulates commodity sector-specific matters. However, unlike agriculture, mining commodities lack sufficient regulation due to limited commodity-specific environmental impact knowledge. This study addresses this gap by providing essential insights into deforestation and biodiversity loss for individual mining commodities, supporting the development of effective import regulations such as EUDR.

To operationalise these insights, governments could incorporate the derived nature loss indices into environmental due diligence, certification, and procurement frameworks, prioritising high-risk commodity–region combinations such as gold, coal, aluminium, nickel–cobalt, and copper extraction in rainforest areas. Even when companies do not have mine-level traceability, they can use these spatially explicit indices to estimate the potential biodiversity and forest loss impacts associated with sourcing specific commodities from particular countries or regions. This enables enterprises to benchmark the relative impacts of their sourcing portfolios against global averages and to prioritise risk mitigation accordingly. Furthermore, enterprises could integrate these datasets into supply chain traceability and disclosure systems, adopting supplier screening protocols that flag sourcing areas overlapping with high-biodiversity zones.

Similarly, the findings can support the TNFD74 by enhancing its LEAP guidelines and facilitating mining companies and companies directly or indirectly sourcing minerals in meeting sustainability requirements through tailored LEAP assessments. By integrating commodity-specific considerations, LEAP can effectively help mining companies address commodity-specific nature loss. For example, while Volkswagen requires its suppliers to maintain nature-loss-free supply chains83, suppliers of critical vehicle manufacturing materials, such as nickel, cobalt, and lithium sourced from Indonesia, could leverage a more tailored application of TNFD-LEAP assessment framework to address related nature loss and meet the requirement. In addition, financial institutions and investors could leverage these datasets to evaluate their exposure to biodiversity-related risks within mining portfolios and align investment strategies with TNFD-aligned nature-risk disclosure requirements.

Furthermore, our findings underscore the varying impacts of different commodities on nature loss, emphasising the need for tailored approaches to sustainable mining practices and site restoration. For example, the growing demand for cobalt and lithium, key inputs for electric vehicle batteries, requires strategies that account for distinct extraction methods, such as lithium derived from brines versus hard rock deposits, each posing unique environmental challenges84,85. International guidelines provided by organisations like the International Council on Mining and Metals (ICMM)86, which certifies responsible production practices, could be strengthened with commodity-specific guidance to address these complexities. Similarly, assurance processes offered by the Initiative for Responsible Mining Assurance (IRMA)87 could integrate these insights to refine criteria for site-specific compliance.

This study has three key limitations, primarily related to the data used. First, as the mining area data60,61 primarily reflects conditions in the 2020s and the deforestation data80 covers only the period from 2001 to 2022, this study does not capture dynamic changes in mining site expansions or closures over time, nor does it account for the potentially different share of deforestation associated with mines that began operations prior to 2000.

Second, as this study relies on data that include only mining-associated activities visible in satellite images60,61, it provides limited consideration of underground mining, which may pose unique challenges such as groundwater contamination. Existing industry analysis88, though excluding China and the Middle East, represents the most comprehensive assessment currently available. It indicates that underground hard-rock mines account for about 40% of global operations but contribute only 12% of run-of-mine production, suggesting that surface mining dominates and that this study remains broadly representative. Nonetheless, we acknowledge that the exclusion of underground mines means that the results do not capture all global extraction activities.

Third, this study focuses on the direct impacts of mining activities within the boundaries of mining extents, given the substantial challenges in accurately estimating indirect effects beyond these boundaries. Specifically, we do not include deforestation caused by mining-related land-use changes, such as the development of transportation corridors, energy infrastructure, agricultural expansion, or urbanisation in the vicinity of mines. Similarly, habitat destruction from these indirect activities and downstream pollution is beyond the scope of this study, which may lead to an underestimation of the overall mining footprint. Addressing these limitations in future research would enable a more comprehensive and holistic understanding of mining-related nature loss.

Despite its limitations, this study offers a global, commodity-level assessment of mining-induced deforestation and biodiversity risk. From 2001 to 2022, mining activities caused ~16,000 km² of forest loss, about two-thirds of which occurred in tropical and subtropical regions. Nearly half of this loss was associated with gold, coal, aluminium, nickel–cobalt, and copper extraction concentrated in the Amazon, Southeast Asian, and Congo Basin rainforests. Biodiversity risks, however, vary regionally and do not necessarily scale with deforestation extent. These findings provide practical guidance for managing nature-related risks in mineral supply chains and support frameworks such as the EUDR and TNFD through commodity-specific insights. Future work should consider temporal mining dynamics, underground operations, and indirect land-use effects to advance nature-positive resource governance.

Methods

We applied machine learning and satellite imagery to identify mining commodities and quantify their nature loss within delineated mining areas. The method comprises two steps for commodity identification: first, processing multiple datasets and utilising the open-pit locations identified by Cheng et al.89 to focus on key extraction areas, and second, classifying commodities based on these identified pits. In the third step, deforestation and biodiversity impacts are quantified by commodity type using the previously identified data. The overall modelling framework of this study is presented in Fig. 5.

Data preparation and open-pit integration

This study’s commodity classification builds on the delineation of global mining areas from Maus et al.60 and Tang and Werner61. The union of the two datasets, with overlapping polygons merged to avoid double counting, consists of 81,614 polygons covering 118,935 km² and represents a wide range of mining activities. However, as these polygon datasets lack detailed commodity information, we supplemented them with verified point data from sources such as the SNL Metals & Mining dataset58, USGS59, Jasansky et al.63, and Franks et al.90, all of which provide spatial coordinates and associated commodity attributes. A spatial join was then performed to integrate the point datasets with the polygons, linking points to polygons where spatial overlap was detected. To reduce potential misassignments, only polygons that fully contained exactly one verified point were retained as training polygons, ensuring a one-to-one correspondence between polygons and commodity types. As a result, 2212 polygons were labelled with commodities across 20 classes, serving as the training dataset for classification (Supplementary Table 5).

Because satellite imagery primarily captures surface features, open-pit locations (surface extraction areas) were explicitly used to better represent visibly extractive mining areas. Mining sites consist of various areas, such as pits, waste zones, and tailing dams91. To identify open-pit locations from mining polygons60,61, we used the global mining land-use dataset developed by Cheng et al.89, which provides land-use and land-cover classifications for over 80,000 recognised mining areas across 150 countries. The dataset was produced through the integration of Sentinel-2 imagery and TanDEM-X elevation-change data using a machine-learning approach to distinguish seven land-use categories: open pit (surface extraction area), waste dumping site (including waste-rock dumps and tailings dams), general disturbed land, water body, facility, bare soil, and vegetation. In this study, we focused on the open-pit class, which represents surface extraction areas formed through vertically downward digging or surface-level material removal, including large terraced structures, quarries, and shallow cavities commonly associated with artisanal and small-scale mining.

From Cheng et al.89, 50,093 polygons were found to contain open-pit pixels, covering an area of 112,980 km². These open-pit polygons represent ~61% of all 81,614 polygons and 95% of the total mining area (118,935 km²) from the union of the Maus et al.60 and Tang and Werner61 datasets. It is important to emphasise that these polygons typically include a combination of land-use types such as waste-rock dumps (WRDs), tailings dams, facilities, and vegetation, where only a portion of each polygon is classified as open pit. The term open-pit polygons, therefore, refers to polygons containing open-pit land-use pixels rather than polygons entirely covered by open pits.

Finally, mining polygons without open-pit pixels were linked to the nearest open-pit polygons within a 2-kilometre radius using a nearest-neighbour distance analysis, as they typically represent supporting areas of mining activity such as WRDs, tailings dams, or facilities. The 2000 m threshold was determined by analysing cumulative coverage and marginal increases in polygon connections and covered area at 500 m intervals, identifying where the growth began to level off (Supplementary Fig. 1). The incremental gains in both polygon count and covered area fall below 5% only after 2000 m, indicating diminishing returns. Therefore, 2 km was selected as the optimal threshold, effectively capturing the most relevant connections while limiting spatial spillover. As a result, 72,321 out of 81,614 polygons (~89%) and 115,930 out of 118,935 km² (~97%) were retained after spatial linkage, indicating that the vast majority of global mining polygons were successfully associated with open-pit or supporting areas.

Commodity classification model

After assigning open-pit pixel locations to each mining polygon, we applied a hierarchical random forest (HRF) model using ASTER and HISUI satellite datasets to classify commodities (Supplementary Note 1). The HRF builds on the random forest algorithm, which is widely regarded for its robustness in handling high-dimensional data, resistance to overfitting, and strong classification performance92. The hierarchical structure further enhances model interpretability and robustness by classifying commodities at multiple levels, effectively managing minerals with unique or overlapping spectral properties93.

HISUI data covered approximately 68% of all polygons to be classified, while ASTER was used for the remaining 32%, ensuring full global coverage. For ASTER, to enhance the classification, 26 band ratio indices from ASTER were used instead of the original 14 bands, as per the ASTER Mineral Index Processing Manual71 (Supplementary Table 7). For HISUI hyperspectral data, we evaluated multiple combinations of indices across the full 185 spectral bands, and the most effective configuration was obtained by averaging bands to mimic ASTER mineral indices (Supplementary Table 8). These HISUI-derived indices were then combined with ASTER thermal indices (since HISUI lacks thermal bands) to ensure completeness of the spectral features used in the model. To further utilise HISUI’s hyperspectral characteristics, which provide finer spectral discrimination and improved sensitivity to subtle mineralogical variations, all HISUI bands were initially included in the classification together with the indices. However, to minimise redundancy and improve model efficiency, a feature-importance analysis was conducted, retaining only the top 25% of features with pairwise correlations below 0.9 for the final model.

The HRF structure, illustrated in Fig. 5, divides Level 1 into non-metals, iron, chalcophile-related metals, and miscellaneous metals based on mineral spectral properties94,95,96,97. Subsequent levels further refine the classification: Level 2-1 separates coal from other non-metals, and Level 2-2 focuses on commodities in the chalcophile-related metals (copper, gold, silver, lead, zinc, and molybdenum) due to their co-occurrence and shared infrastructure98. Finally, Level 2-3 classifies miscellaneous metals such as aluminium (bauxite), nickel-cobalt, PGE, lithium, and uranium. Subsequently, 20 distinct mining commodities were classified. The decision to focus on these 20 commodities was primarily due to the limited training samples available for other commodity groups, which made further disaggregation impractical. In this context, two HRF models were built: one using ASTER data only (the ASTER model) and the other integrating HISUI and ASTER thermal bands (the HISUI model).

The training dataset was built from open-pit pixels within the labelled/training polygons. Given the class imbalances, the Synthetic Minority Oversampling Technique (SMOTE) was used to generate synthetic samples and oversample minority classes99. Additionally, clustering-based k-means undersampling was applied to majority classes to reduce their dominance in the dataset100 when there was a significant difference.

After training, the HRF models were applied to classify all polygons containing open-pit pixels without previously labelled commodities. For the ASTER model, the overall accuracy (OA) from 10-fold cross-validation using an 80:20 split across all HRF levels was 88% (Supplementary Tables 9–13), with strong accuracies for each level (85%, 86%, 90%, and 91%, respectively). For the HISUI model, the OA across all HRF levels was 86% (Supplementary Tables 14–18), with strong accuracies at each level (83%, 90%, 86%, and 90%). User’s accuracy (UA) and producer’s accuracy (PA) for each class in both models were all above 74% (Supplementary Table 19), indicating consistent and reliable classification performance.

Importantly, because SMOTE was applied, we report out-of-bag accuracy (OOB Acc) and balanced accuracy (OOB Bal) for both training and test sets to screen for potential overfitting (Supplementary Tables 9–18). Across all models, the gaps between OOB and test accuracies were below 0.025, well within the approximately 0.05 overestimation typically observed in random forests101, indicating no substantial overfitting.

After incorporating the results with trained and connected polygons within a 2-km radius, a total of 71,041 polygons covering 113,068 km² were identified (Supplementary Table 1). Overall, around 87% of the polygons and 95% of the mining area identified by the union of the Maus et al.60 and Tang and Werner61 datasets were covered.

Finally, the classification accuracy and validation procedures are detailed in Supplementary Note 2. We first report the external validation using independent mine datasets, which show accuracies above 74% for multiple commodities, followed by the stratified random-sampling accuracy assessment, for which MC–derived CIs were computed for accuracy estimates. Commodity-level PA and UA, along with the OA of 77%, are summarised in Supplementary Table 22. It is worth noting that, in this assessment, several rare-metal classes (Pb, Cr, Mn, Mo, Ni, Sn, W, Ti, Li, PGE, U, and Other metals) with small sample sizes in the random sampling were aggregated into a single group to ensure statistical robustness (Supplementary Note 2). For the commodity-level accuracies of the nine resulting groups (Fe, Cu, Zn, Au, Ag, Al, Rare metals, Coal, and Other non-metals) (Supplementary Table 22), most commodities exhibit moderate to high PA and UA with reasonable CIs, indicating stable performance despite the limited sample sizes. The one relatively lower value is the UA for Other non-metals (60%), which is expected because this category aggregates many heterogeneous non-metallic sites and therefore shows higher spectral confusion.

Collectively, the validations demonstrate that the classification performs robustly at the global scale, with overall accuracies (74% from external validation and 77% from random sampling) comparable to leading global land-cover products such as ESA WorldCover102 (~75–77%) and NASA MODIS MCD12Q1103 (~73.6%).

Mining-induced deforestation and biodiversity loss

We quantified the nature loss caused by mining commodities by overlaying mining polygons with deforestation maps and calculating the Extinction Risk Index (ERI) for 20 mining commodities.

First, deforestation maps from 2001 to 202280 were overlaid with mining polygons to calculate the total deforestation area (km²) and the deforestation-to-mining area ratio (deforestation area/mining area) for each mining site and commodity.

Second, biodiversity loss was quantified using the ERI, which incorporates extinction risk scores of species based on the IUCN Red List categories, ranging from 0 for least concern to 1 for extinct104. Notably, this index is a reversed version of the Red List Index (RLI). Using data78 for 10,685 species of mammals, birds, reptiles, and amphibians filtered by threats from mining and quarrying (811 species), we calculated ERI values resulting from mining at a 0.01-degree grid resolution (approximately 1 km). These ERI values were then averaged within mining polygons. Specifically, when multiple 1 km grid cells overlapped a single mining polygon, their mean value was assigned as the ERI for that polygon. This approach helps identify regions where mining-driven habitat loss coincides with high biodiversity risk.

Statistics and reproducibility

Sample sizes for the commodity classification validation were determined based on data availability and the stratified random-sampling design, with a minimum of 30 samples per class and increased sampling for major commodity groups to ensure adequate representation. No data were excluded from the analyses, except that randomly selected sites that could not be verified during the independent validation procedure were replaced through rerandomisation, as described in Supplementary Note 2. Because this study was an observational geospatial analysis rather than an intervention-based experiment, experimental randomisation and investigator blinding were not applicable.

Reporting summary

Further information on research design is available in the Nature Portfolio Reporting Summary linked to this article.

Data availability

Author-generated example datasets, together with grid-level commodity classification results in GPKG format, are publicly available through the Zenodo archive105 (https://doi.org/10.5281/zenodo.19648354). Additional publicly shareable materials are provided in the Supplementary Information and Supplementary Data 1–3, as cited in the main text. The example datasets and grid-level results are provided under CC BY-SA 4.0. The full training dataset cannot be publicly shared because part of it contains data derived from the licensed commercial S&P Global Market Intelligence SNL Metals & Mining Database, whose redistribution is restricted by the data provider’s terms of use. Previously published third-party datasets analysed in this study are available from their original sources, including the global mining area datasets60,61 available from PANGAEA and the Zenodo archive (https://doi.org/10.5281/zenodo.7894216), as cited in the manuscript.

Code availability

Example code used in this study is publicly available at Zenodo (https://doi.org/10.5281/zenodo.19648354) and is provided under CC BY-NC 4.0.

References

Hund, K. L., Arrobas, D. L. P., Fabregas, M. T. P., Laing, T. J. & Drexhage, J. R. Minerals for Climate Action: The Mineral Intensity of the Clean Energy Transition (World Bank, 2023).
Arrobas, D. L. P., Hund, K. L., Mccormick, M. S., Ningthoujam, J. & Drexhage, J. R. The Growing Role of Minerals and Metals for a Low Carbon Future (World Bank, 2017).
Ayuk, E. et al. Mineral Resource Governance in the 21st Century: Gearing Extractive Industries towards Sustainable Development (International Resource Panel, United Nations Environment Programme, 2020).
Luckeneder, S., Giljum, S., Schaffartzik, A., Maus, V. & Tost, M. Surge in global metal mining threatens vulnerable ecosystems. Glob. Environ. Change 69, 102303 (2021).
Sovacool, B. K. et al. Sustainable minerals and metals for a low-carbon future. Science 367, 30–33 (2020).
Bridge, G. CONTESTED TERRAIN: mining and the environment. Annu. Rev. Environ. Resour. 29, 205–259 (2004).
Mancini, L. & Sala, S. Social impact assessment in the mining sector: review and comparison of indicators frameworks. Resour. Policy 57, 98–111 (2018).
Sonter, L. J., Ali, S. H. & Watson, J. E. M. Mining and biodiversity: key issues and research needs in conservation science. Proc. R. Soc. B Biol. Sci. 285, 20181926 (2018).
Maddox, T., Howard, P., Knox, J. & Jenner, N. Forest-Smart Mining: Identifying Factors Associated with the Impacts of Large-Scale Mining on Forests (World Bank, 2019).
Sonter, L. J. et al. Mining drives extensive deforestation in the Brazilian Amazon. Nat. Commun. 8, 1013 (2017).
Asner, G. P., Llactayo, W., Tupayachi, R. & Luna, E. R. Elevated rates of gold mining in the Amazon revealed through high-resolution monitoring. Proc. Natl. Acad. Sci. USA. 110, 18454–18459 (2013).
Alvarez-Berríos, N. L. & Aide, T. M. Global demand for gold is another threat for tropical forests. Environ. Res. Lett. 10, 014006 (2015).
Asner, G. P. & Tupayachi, R. Accelerated losses of protected forests from gold mining in the Peruvian Amazon. Environ. Res. Lett. 12, 094004 (2017).
Schueler, V., Kuemmerle, T. & Schröder, H. Impacts of surface gold mining on land use systems in Western Ghana. AMBIO 40, 528–539 (2011).
Akiwumi, F. A. & Butler, D. R. Mining and environmental change in Sierra Leone, West Africa: a remote sensing and hydrogeomorphological study. Environ. Monit. Assess. 142, 309–318 (2008).
Ladewig, M., Angelsen, A., Masolele, R. N. & Chervier, C. Deforestation triggered by artisanal mining in eastern Democratic Republic of the Congo. Nat. Sustain. 7, 1452–1460 (2024).
Debus, A. et al. A labelled dataset to classify direct deforestation drivers from Earth Observation imagery in Cameroon. Sci. Data 11, 564 (2024).
Abood, S. A., Lee, J. S. H., Burivalova, Z., Garcia-Ulloa, J. & Koh, L. P. Relative contributions of the logging, fiber, oil palm, and mining industries to forest loss in Indonesia. Conserv. Lett. 8, 58–67 (2015).
Lo, M. G. Y. et al. Nickel mining reduced forest cover in Indonesia but had mixed outcomes for well-being. One Earth 7, 2019–2033 (2024).
Palmer, M. A. et al. Mountaintop mining consequences. Science 327, 148–149 (2010).
Stanimirova, R., Harris, N., Reytar, K., Wang, K. & Barbanell, M. Mining is Increasingly Pushing into Critical Rainforests and Protected Areas (World Resources Institute, 2024).
Carpenter, K. E. A study of the fauna of rivers polluted by lead mining in the Aberystwyth District of Cardiganshire. Ann. Appl. Biol. 11, 1–23 (1924).
Smith, N. J. H., Alvim, P., Homma, A., Falesi, I. & Serrão, A. Environmental impacts of resource exploitation in Amazonia. Glob. Environ. Change 1, 313–320 (1991).
Almeida-Filho, R. & Shimabukuro, Y. E. Digital processing of a Landsat-TM time series for mapping and monitoring degraded areas caused by independent gold miners, Roraima State, Brazilian Amazon. Remote Sens. Environ. 79, 42–50 (2002).
Liang, T., Li, K. & Wang, L. State of rare earth elements in different environmental components in mining areas of China. Environ. Monit. Assess. 186, 1499–1513 (2014).
Liu, W.-S. et al. Water, sediment and agricultural soil contamination from an ion-adsorption rare earth mining area. Chemosphere 216, 75–83 (2019).
Li, Z., Ma, Z., van der Kuijp, T. J., Yuan, Z. & Huang, L. A review of soil heavy metal pollution from mines in China: pollution and health risk assessment. Sci. Total Environ. 468–469, 843–853 (2014).
Owusu, E. H., Ofori, B. Y. & Attuquayefio, D. K. The secondary impact of mining on primates and other medium to large mammals in forest reserves in southwestern Ghana. Extr. Ind. Soc. 5, 114–121 (2018).
Junker, J. et al. Threat of mining to African great apes. Sci. Adv. 10, eadl0335 (2024).
Alvarez-Berríos, N. et al. Impacts of small-scale gold mining on birds and anurans near the Tambopata Natural Reserve, Peru, assessed using passive acoustic monitoring. Trop. Conserv. Sci. 9, 832–851 (2016).
Venier, L. A. et al. Effects of natural resource development on the terrestrial biodiversity of Canadian boreal forests. Environ. Rev. 22, 457–490 (2014).
Koehnken, L. et al. Impacts of riverine sand mining on freshwater ecosystems: a review of the scientific evidence and guidance for future research. River Res. Appl. 36, 362–370 (2020).
Purdy, B. G., Ellen Macdonald, S. & Lieffers, V. J. Naturally saline boreal communities as models for reclamation of saline oil sand tailings. Restor. Ecol. 13, 667–677 (2005).
Sievers, M., Hale, R., Parris, K. M. & Swearer, S. E. Impacts of human-induced environmental change in wetlands on aquatic animals. Biol. Rev. 93, 529–554 (2018).
Skirycz, A. et al. Canga biodiversity, a matter of mining. Front. Plant Sci. 5, 653 (2014).
de Castro Pena, J. C. et al. Impacts of mining activities on the potential geographic distribution of eastern Brazil mountaintop endemic species. Perspect. Ecol. Conserv. 15, 172–178 (2017).
Faucon, M.-P. et al. May rare metallophytes benefit from disturbed soils following mining activity? The case of the Crepidorhopalon tenuis in Katanga (D. R. Congo). Restor. Ecol. 19, 333–343 (2011).
Buisson, E. et al. Resilience and restoration of tropical and subtropical grasslands, savannas, and grassy woodlands. Biol. Rev. 94, 590–609 (2019).
Norgate, T. & Haque, N. Energy and greenhouse gas impacts of mining and mineral processing operations. J. Clean. Prod. 18, 266–274 (2010).
Romero, H., Méndez, M. & Smith, P. Mining development and environmental injustice in the Atacama Desert of Northern Chile. Environ. Justice 5, 70–76 (2012).
Howladar, M. F. Coal mining impacts on water environs around the Barapukuria coal mining area, Dinajpur, Bangladesh. Environ. Earth Sci. 70, 215–226 (2013).
Kossoff, D. et al. Mine tailings dams: characteristics, failure, environmental impacts, and remediation. Appl. Geochem. 51, 229–245 (2014).
Howe, L., Johnston, S. & Côte, C. Mining-related environmental disasters: a High Reliability Organisation (HRO) perspective. J. Clean. Prod. 417, 137965 (2023).
Grandjean, P., White, R. F., Nielsen, A., Cleary, D. & de Oliveira Santos, E. C. Methylmercury neurotoxicity in Amazonian children downstream from gold mining. Environ. Health Perspect. 107, 587–591 (1999).
Xiao, R., Wang, S., Li, R., Wang, J. J. & Zhang, Z. Soil heavy metal contamination and health risks associated with artisanal gold mining in Tongguan, Shaanxi, China. Ecotoxicol. Environ. Saf. 141, 17–24 (2017).
Zhang, X. et al. Impacts of lead/zinc mining and smelting on the environment and human health in China. Environ. Monit. Assess. 184, 2261–2273 (2012).
Kitula, A. G. N. The environmental and socio-economic impacts of mining on local livelihoods in Tanzania: a case study of Geita District. J. Clean. Prod. 14, 405–414 (2006).
Hilson, G. An overview of land use conflicts in mining communities. Land Use Policy 19, 65–73 (2002).
Hilson, G. Child labour in African artisanal mining communities: experiences from Northern Ghana. Dev. Change 41, 445–473 (2010).
Northey, S. A. et al. The exposure of global base metal resources to water criticality, scarcity and climate change. Glob. Environ. Change 44, 109–124 (2017).
Dethier, E. N. et al. A global rise in alluvial mining increases sediment load in tropical rivers. Nature 620, 787–793 (2023).
Islam, K. & Murakami, S. Global-scale impact analysis of mine tailings dam failures: 1915–2020. Glob. Environ. Change 70, 102361 (2021).
Giljum, S. et al. A pantropical assessment of deforestation caused by industrial mining. Proc. Natl. Acad. Sci. USA. 119, e2118273119 (2022).
Sonter, L. J., Dade, M. C., Watson, J. E. M. & Valenta, R. K. Renewable energy production will exacerbate mining threats to biodiversity. Nat. Commun. 11, 4174 (2020).
Durán, A. P., Rauch, J. & Gaston, K. J. Global spatial coincidence between protected areas and metal mining activities. Biol. Conserv. 160, 272–278 (2013).
Kobayashi, H., Watando, H. & Kakimoto, M. A global extent site-level analysis of land cover and protected area overlap with mining activities as an indicator of biodiversity pressure. J. Clean. Prod. 84, 459–468 (2014).
Murguía, D. I., Bringezu, S. & Schaldach, R. Global direct pressures on biodiversity by large-scale metal mining: spatial distribution and implications for conservation. J. Environ. Manage. 180, 409–420 (2016).
S&P Global Market Intelligence. SNL Metals and Mining Database (S&P Global Market Intelligence, 2023).
United States Geological Survey (USGS). Mineral Resources Online Spatial Data (U.S. Geological Survey, 2024).
Maus, V. et al. An update on global mining land use. Sci. Data 9, 433 (2022).
Tang, L. & Werner, T. T. Global mining footprint mapped from high-resolution satellite imagery. Commun. Earth Environ. 4, 1–12 (2023).
Maus, V. & Werner, T. T. Impacts for half of the world’s mining areas are undocumented. Nature 625, 26–29 (2024).
Jasansky, S., Lieber, M., Giljum, S. & Maus, V. An open database on global coal and metal mine production. Sci. Data 10, 52 (2023).
Sabins, F. F. Remote sensing for mineral exploration. Ore Geol. Rev. 14, 157–183 (1999).
van der Meer, F. D. et al. Multi- and hyperspectral geologic remote sensing: a review. Int. J. Appl. Earth Obs. Geoinformation 14, 112–128 (2012).
Zhang, X., Pazner, M. & Duke, N. Lithologic and mineral information extraction for gold exploration using ASTER data in the south Chocolate Mountains (California). ISPRS J. Photogramm. Remote Sens. 62, 271–282 (2007).
Di Tommaso, I. & Rubinstein, N. Hydrothermal alteration mapping using ASTER data in the Infiernillo porphyry deposit, Argentina. Ore Geol. Rev. 32, 275–290 (2007).
Abdelkader, M. A. et al. Advanced exploration of rare metal mineralization through integrated remote sensing and geophysical analysis of structurally-controlled hydrothermal alterations. J. Geochem. Explor. 267, 107598 (2024).
Ninomiya, Y., Fu, B. & Cudahy, T. J. Detecting lithology with Advanced Spaceborne Thermal Emission and Reflection Radiometer (ASTER) multispectral thermal infrared “radiance-at-sensor” data. Remote Sens. Environ. 99, 127–139 (2005).
Yamaguchi, Y., Kahle, A. B., Tsu, H., Kawakami, T. & Pniel, M. Overview of Advanced Spaceborne Thermal Emission and Reflection Radiometer (ASTER). IEEE Trans. Geosci. Remote Sens. 36, 1062–1071 (1998).
Kalinowski, A. & Oliver, S. ASTER Mineral Index Processing Manual (Geoscience Australia, 2004).
Matsunaga, T. et al. HISUI status toward 2020 launch. IGARSS 2019 – IEEE International Geoscience and Remote Sensing Symposium 4495–4498, https://doi.org/10.1109/IGARSS.2019.8899179 (2019).
European Commission. Regulation (EU) 2023/1115 of the European Parliament and of the Council of 14 June 2023 on deforestation-free products. Official Journal of the European Union L 167, 1–52. https://environment.ec.europa.eu/topics/forests/deforestation/regulation-deforestation-free-products_en (2023).
Taskforce on Nature-related Financial Disclosures (TNFD). Guidance on the Identification and Assessment of Nature-related Issues: The Leap Approach v1.1 (TNFD, 2023).
Earth Insight Partners. Three Basins Threat Report: Fossil Fuel, Mining, and Industrial Expansion Threats to Forests and Communities (Earth Insight, 2023).
Kaunda, R. B. Potential environmental impacts of lithium mining. J. Energy Nat. Resour. Law 38, 237–244 (2020).
Kurmelovs, R. How Australia Became the World’s Greatest Lithium Supplier (BBC Future, 2022).
International Union for Conservation of Nature (IUCN). The IUCN Red List of Threatened Species (IUCN, 2011).
Ukpanah, I. Is lithium mining bad for the environment? GreenMatch Blog https://www.greenmatch.co.uk/blog/is-lithium-mining-bad-for-the-environment (2024).
Hansen, M. C. et al. High-resolution global maps of 21st-century forest cover change. Science 342, 850–853 (2013).
Tropek, R. et al. Comment on High-resolution global maps of 21st-century forest cover change. Science 344, 981 (2014).
European Commission. Regulation (EU) 2023/956 of the European Parliament and of the Council of 10 May 2023 establishing a carbon border adjustment mechanism. Official Journal of the European Union L 130, 52–104. https://eur-lex.europa.eu/eli/reg/2023/956/oj/eng (2023).
Volkswagen AG. Code of Conduct for Business Partners: Volkswagen Group Requirements Regarding Sustainability in its Relationships with Business Partners (Volkswagen AG, 2023).
International Energy Forum. How to Make Mining More Sustainable (International Energy Forum, 2024).
The World Bank. Climate-Smart Mining: Minerals for Climate Action (The World Bank, 2019).
International Council on Mining and Metals (ICMM). Mining Principles: Performance Expectations (ICMM, 2020).
Initiative for Responsible Mining Assurance (IRMA). IRMA Standard for Responsible Mining v2.0 (IRMA, 2025).
Martino, R., McCann, P., Ray, R. K. & van der Ende, O. Digging Deeper: Trends in Underground Hard-Rock Mining for Gold and Base Metals (McKinsey & Company, 2021).
Cheng, Y.-T., Hoang, N. T., Maupu, L. & Kanemoto, K. Classifying land use within 80,000 mining sites on a global scale. Sci. Data 13, 338 (2026).
Franks, D. M. et al. Tailings facility disclosures reveal stability risks. Sci. Rep. 11, 5353 (2021).
Werner, T. T. et al. Global-scale remote sensing of mine areas and analysis of factors explaining their extent. Glob. Environ. Change 60, 102007 (2020).
Breiman, L. Random forests. Mach. Learn. 45, 5–32 (2001).
Hernández, J., Sucar, L. E. & Morales, E. F. Multidimensional hierarchical classification. Expert Syst. Appl. 41, 7671–7677 (2014).
Clark, R. N. Spectroscopy of rocks and minerals, and principles of spectroscopy. In Manual of Remote Sensing (ed. Rencz, A.) (John Wiley and Sons, Inc., New York, 1999).
Hunt, G. R. Spectral signatures of particulate minerals in the visible and near infrared. Geophysics 42, 501–513 (1977).
Salehi, T. & Tangestani, M. H. Large-scale mapping of iron oxide and hydroxide minerals of Zefreh porphyry copper deposit, using Worldview-3 VNIR data in the Northeastern Isfahan, Iran. Int. J. Appl. Earth Obs. Geoinf. 73, 156–169 (2018).
Rowan, L. C. & Mars, J. C. Lithologic mapping in the Mountain Pass, California area using Advanced Spaceborne Thermal Emission and Reflection Radiometer (ASTER) data. Remote Sens. Environ. 84, 350–366 (2003).
Reuter, M. A. et al. The metrics of material and metal ecology: harmonizing the resource, technology and environmental cycles. Dev. Miner. Process. 16, 3–706 (2005).
Chawla, N. V., Bowyer, K. W., Hall, L. O. & Kegelmeyer, W. P. SMOTE: synthetic minority over-sampling technique. J. Artif. Intell. Res. 16, 321–357 (2002).
Lin, W.-C., Tsai, C.-F., Hu, Y.-H. & Jhang, J.-S. Clustering-based undersampling in class-imbalanced data. Inf. Sci. 409–410, 17–26 (2017).
Janitza, S. & Hornung, R. On the overestimation of random forest’s out-of-bag error. PLoS ONE 13, e0201904 (2018).
Zanaga, D. et al. ESA WorldCover 10 m 2021 v200 (European Space Agency, 2022).
Friedl, M. & Sulla-Menashe, D. MCD12Q1 MODIS/Terra+Aqua Land Cover Type Yearly L3 Global 500 m SIN Grid V006 (NASA Land Processes Distributed Active Archive Center, 2019).
Shinoda, Y., Kadoya, T., Takeuchi, Y. & Nansai, K. Decoupling the food supply from impacts on forest threatened species: rethinking suitable crop areas for food production. Resour. Conserv. Recycl. 226, 108696 (2026).
Cheng, Y.-T. yutongcheng/mining-rf-classification-example: example code and data for mining commodity classification (v1.0.0). Zenodo, https://doi.org/10.5281/zenodo.19648354 (2026).

Funding

K.K. discloses support for the research of this work from the Environment Research and Technology Development Fund (JPMEERF20234004).

Author information

Authors and Affiliations

Contributions

Y.T.C. contributed to conceptualisation, methodology, data curation, investigation, formal analysis, software development, validation, visualisation, and writing of the original draft, as well as review and editing of the manuscript. N.T.H. contributed to conceptualisation, methodology, validation, resources, and review and editing of the manuscript. Y.S., K.I., M.M. and T.K. contributed to validation, resources, and review and editing of the manuscript. K.K. contributed to conceptualisation, methodology, supervision, funding acquisition, validation, resources, and review and editing of the manuscript.

Corresponding author

Ethics declarations

Competing interests

The authors declare no competing interests.

Peer review

Peer review information

Nature Communications thanks Tim Werner, Zhujun Gu and the other anonymous reviewer(s) for their contribution to the peer review of this work. A peer review file is available.

Additional information

Publisher’s note Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.

Supplementary information

Rights and permissions

Open Access This article is licensed under a Creative Commons Attribution-NonCommercial-NoDerivatives 4.0 International License, which permits any non-commercial use, sharing, distribution and reproduction in any medium or format, as long as you give appropriate credit to the original author(s) and the source, provide a link to the Creative Commons licence, and indicate if you modified the licensed material. You do not have permission under this licence to share adapted material derived from this article or parts of it. The images or other third party material in this article are included in the article’s Creative Commons licence, unless indicated otherwise in a credit line to the material. If material is not included in the article’s Creative Commons licence and your intended use is not permitted by statutory regulation or exceeds the permitted use, you will need to obtain permission directly from the copyright holder. To view a copy of this licence, visit http://creativecommons.org/licenses/by-nc-nd/4.0/.

About this article

Cite this article

Cheng, YT., Hoang, N.T., Shinoda, Y. et al. Mapping global resource driven nature loss in the mining sector from 2001 to 2022. Nat Commun 17, 6936 (2026). https://doi.org/10.1038/s41467-026-73792-9

DOI: https://doi.org/10.1038/s41467-026-73792-9