Humans have greatly affected the Earth9s land surfcea in recent centu- expected to further increase the risk for many of these hazards, such ries7311. In particular, the expansion of the built environment and the as higher wildfire frequency and intensity18,34,35. growth of settlements and their long-term resourcerequirements Here, we present a global map of the 2020 WUI at 10)m resolution have been dramatic across the globe12314. The growth of settlements using a globally consistent and validated approach based on remote can have remote effects via teleconnected processe1s5,16 but the most sensing-derived datasets of building area4 and wildland vegetation5. We immediate human3environmental conflicts arise wherebuildings are distinguished between two types of WUI: intermix WUI (where buildbuilt in or near wildland vegetation, an area known as the wildland3 ings and wildland vegetation intermingle) and interface WUI (where urban interface (WUI)1,17. The WUI is widespread across Australia, buildings are close to large wildland vegetation patches). We further disEurope and North America18322 and there is evidence for WUI in some tinguished between WUI dominated by forest, shrubland and wetland other countries23326. However, the worldwide distribution of the WUI versus that dominated by grassland. We then summarized population is unknown2. and biomass in the WUI for each country and biome, using the biome

The WUI is a desirable place to live for many people as a result of its definition of ref3.6. To identify areas of increased fire hazard in the proximity to natural amenities but it is also an area of manifold hazards WUI, we assessed wildfire occurrence using two remote sensing-based to both humans and natural ecosystems. Wildfires are a particular threat datasets4Moderate Resolution Imaging Spectroradiome ter (MODIS) to houses and lives, often caused by human ignition and facilitated by Active Fire data for 200332020 and Visible Infrared Imaging Radiomaltered fire regimes where settlements sprawl intofire-dependent eter Suite (VIIRS) Active Fire data for 201332020. Our identification of ecosystems. The availability of buildings themselves as fuel, along WUI types by dominant land cover allowed for a regionalized evaluawith swiftly moving fire, makes evacuations difficu19l,t27329. Indeed, the tion of fire hazard in the WUI. For example, whereas natural grasslands number of wildfires has increased in the WUI over hte past few decades2 exhibit frequent wildfires in some of the world9s WUI, wildfires are not owing to both housing growth and climate change. Other hazards to a concern where grasslands are highly managed pastures. In contrast, humans or their environment include the loss of biodiversity and car- both managed and wild forests provide fuel for wildfires. bon storage due to habitat loss and fragmentation,predation of wildlife We found that the total global WUI area in 2020 was 6.3)million)km2 by cats and dogs, light and noise pollution, the introduction of inva- or 4.7% of global land area, which is an order of magnitude larger than sive species, an increased risk for the spread of zoonotic diseases and the global urban area37 or twice the size of India. The global land share changes in hydrology3,30333. Quantifying any of these hazards requires a of intermix and interface WUI is 3.6% (4.8)millionk)m2) and 1.1% (1.5)milconsistent global assessment of the WUI. This is particularly important lion)km2), respectively. Two thirds of the overall WUI areaare dominated because the number of exposed buildings and peoplein the WUI is by forests, shrublands and wetlands, versus one third by grasslands. expected to grow as population grows and because climate change is Globally, 3.5)billion people live within the WUI (1.7)billion in intermix

São Paulo, Brazil 150° W

WUI area share (%) 0 10 20 30 40 50 100

Kingston, Jamaica 50 km 0 50 km Forest/shrubland/wetland-dominated Wildland–Urban Interface (WUI)

Intermix Interface and 1.8)billion in interface WUI) and two thirds ofthose live in WUI including East Africa, Brazil or Southeast Asia, widespread WUI has not dominated by forests, shrublands and wetlands. In total, nearly half been reported. Among the two most populated countries in the world, of all buildings and people on the globe are potentially affected by China has large WUI areas in southern and eastern regions, which are the human3environmental hazards that are concentrated in the WUI. previously undocumented in the literature. However, India has much However, only 4.1% of the total aboveground livingplant biomass occurs smaller WUI area in the southeast and the Himalayas, probably because within the WUI, with most of it in the intermix WUI. cropland density is high in other regions40, not providing enough wild

The WUI occurs on all continents. However, within continents, the land vegetation to create WUI. The area-adjusted overall accuracy of distribution of the WUI is highly uneven. Large WUI areas occur along our WUI map is between 79.6% (when distinguishing all WUI classes) the Pacific coast of North America; in eastern North America and the and 82.0% (when distinguishing WUI from non-WUI; Supplementary Caribbean; along the Brazilian coast; across Europe; in West, South and Data A).

East Africa, including Nigeria and Uganda; in Southeast Asia, including The characteristics of the WUI vary considerably among continents. India, China, Indonesia and Japan; and in Australia (Fig. 1a). In some The WUI covers only 3% of South America but 15% ofEurope. Europe and of these places, such as in California, Mediterranean Europe or South Asia have especially high shares of interface WUI area, whereas intermix Africa, the WUI has been well studied because manybuildings and peo- WUI dominates in North America. South America is the only world ple are affected by wildfires ther3e8,39. In many other places, however, region where grassland WUI area dominates, whereas Asia has the least. b 0 e

In South America, 33% of the population live in grassland-dominated high share of biomass in the WUI (60%) because mostWUI occurs near WUI but only 7% in Asia. In Oceania, 56% of the total population lives its coastal regions where biomass is concentrated. In Japan, more than in WUI dominated by forest, shrublands and wetlands, compared to half of all wildfires occurred within the WUI and most of the people livonly 24% in Asia. Europe has the largest share of its biomass, 10.5%, ing in the WUI live in the interface because settlements are generally within the WUI (Fig. 2a3c). Seventy per cent of the global WUI area is well demarcated and abut wildland vegetation. France and Poland have in very low or low density rural areas and only 8% in urban clusters and an especially high share of WUI area and populationin the grassland centres (Extended Data Table 1, classes according to ref.41). However, WUI. In Indonesia, the Philippines, Brazil and Ecuador, WUI area share this pattern differs strongly by world region: in North America, 84% of is small but high proportions of people live in those small WUI areas the WUI is rural (5% in urban areas) but in Asia only 53% (14% in urban and are affected by wildfires (0.7 to 13.1)milliond,epending on the areas). Lastly, the WUI occurs in countries across all income classes country). Those different WUI patterns reflect thdeiversity of reasons (Extended Data Table 2). for both WUI development and wildfire occurrence and highlgi ht that

We selected 12 hotspot countries on the basis of WUI area share different management responses are required to mitgi ate the human3 and wildfire occurrence for closer examination: Uganda, Lebanon, environmental conflicts that are concentrated in the WUI3. Sri Lanka, Japan, France, Poland, Jamaica, El Salvador, Indonesia, the Among biomes, the WUI is highly concentrated in a few (Fig. 2g3i Philippines, Brazil and Ecuador (Fig. 1b3g, Extended Data Fig. 1 and and Extended Data Fig. 2). The temperate broadleafand mixed forFig. 2d3f). Uganda, Sri Lanka, Jamaica and El Salvador have an excep- est biome covers only 9% of global land area, yet contains 35% of total tionally high share of population in the WUI (>80%) and many people WUI area. Similarly, subtropical and tropical moist broadleaf forests there were affected by fires since 2003 (for examep,l8.7)million in represent only 15% of the global land but contain 26% of WUI area. In Uganda and 1.4)million in El Salvador). Lebanon hasan exceptionally contrast, deserts and xeric shrublands cover 20% of the land area but Africa Asia ENuorortpheAmericaOSceoaunthiaAmerica

World region Forest/shrubland/wetland-dominated WUI Grassland-dominated WUI

Biome detection from 2003 to 2020. We analysed active fire point data in a 1)km grid and we considered all grid cells with at least one fire as being wildfire areas and all people living in a grid cell where a fire occurred 200332020 as affecetd by wildfire. Relative patterns are confirmed using VIIRS Active Fire data from 2013 to 2020 (Supplementary Data B3E). contain only 3% of the global WUI area. We also observed large dif- relatively few people are affected (Fig3.c ). Some biomes, for example, ferences in population patterns: in both boreal foersts/taiga and in Mediterranean forests, woodlands and shrublands, are small in area subtropical and tropical coniferous forests over 80% of the population but are hotspots of recurring severe wildfire and destruction42. In other lives in the WUI, whereas in the deserts and xeric shrublands, only 10% biomes, the large share of population living in the WUI (for example, does. The distribution of WUI area across biomes isimportant because 80% in boreal forests/taiga) and the fact that most people affected by WUI-related hazards, such as wildfires and their efects on people prob- wildfire during 200332020 lived in the WUI suggeststhat changing wildably differ among biomes. Wildfire hazard is higher either where the fire regimes could quickly increase the likelihoodof wildfire exposure WUI is widespread and where many people are affected by wildfire (such in the future. Particularly, there is a high probability that temperate as in subtropical and tropical moist broadleaf forests) or where WUI broadleaf and mixed forests and subtropical and tropical moist broadarea itself is small but both people and wildfiresare concentrated there leaf forests, the biomes with the largest WUI areaand home to 130)mil(such as boreal forests/taiga and mangroves that are highly vulnerable lion people previously affected by wildfire in the WUI, will experience to fire). Wildfire hazard is also driven by the available biomass. In both increased fire hazard towards the middle of the twenty-first century the broadleaf/mixed forests and in subtropical andtropical coniferous if rising trends in wildfire fire frequency and intensity continu6,e44346. forests, a substantial share of their biomass is in the WUI (about 13% in both biomes) as a result of their high overall share of land in the WUI (18% and 14%, respectively). By contrast, in the temperate grassland, Discussion savanna and shrubland biome, the overall WUI area share is low but a The WUI is where people live within or near wildland vegetation. We large portion of the biomass occurs in vegetation-rich coastal areas, found that the WUI covers nearly 5% of the globalalnd area, even which is also where the WUI is concentrated. though the WUI has not been a class in any previous global land cover

Wildfires are of increasing concern across the globe, as their fre- or land-use maps. We quantified the global extent of the WUI at high quency, intensity and season-length have increased because of climate spatial resolution, characterized it by dominant land cover and related change, more human ignitions and rising fuel loads42. Wildfires are par- it to wildfires of the last two decades. Our analysis yielded three pri nticularly problematic in the WUI and cause substantial losses of houses cipal insights. First, we found that the WUI is a global phenomenon. and lives there43. Indeed, more than two thirds of all people affecetd by Although previous work showed that the WUI is widespread in Mediwildfires during 200332020 (those experiencing a fire within 1)km of terranean Europe, the United States and Australia2, our results show their homes) live in the WUI. This is partly because population density large WUI areas in all continents, including previously undocumented in the WUI is higher than in non-urban non-WUI areas but nevertheless hotspots in eastern Asia, East Africa and parts of South America. Secsubstantial because only a small share of all global wildfire occurrences ond, the WUI is highly diverse in terms of population density, biomass was directly in the WUI (Fig. 3a). Effects of WUI wildfires on the popu-la quantity and dominant vegetation type. Third, the WUI is where wildtion differ among world regions and countries. In North America, 85% fires affect the most people. Globally, two thirds of all the people that of the population affected by wildfire lives in thWe UI but in Africa only experienced wildfires live in the WUI. Among our WUI hotspots with 55% (nearly 150)million) does. In all world regions, except Europe and frequent wildfires, many lack assessments of wildfire regimes, sett-le South America, most people were affected by wildfires that occurred ment patterns and wildfire risk, as is true for many WUI hotspots where in WUI dominated by forests, shrublands and wetlands (Fig.3 b). This wildfires are likely to become more prevalent as aresult of changing suggests that, despite the WUI9s small overall area, and, despite the climatic conditions. comparatively rare occurrence of wildfires, buildings and people in Local and regional patterns of the WUI are highly variable. We found the WUI may face an elevated wildfire hazard across the world. major differences in the proportion of area that iWsUI in different

The role of wildfires in the WUI differs among bioems but wildfire countries, how many people live there, how much biomass occurs in the distribution suggests increasing effects of wildfeirs on people in the WUI and what the dominant land cover is. Although some WUI areas are future. In some biomes, for example, in the tundraor in deserts and xeric well-known due to a history of disastrous wildfires(for example, Medishrublands, wildfires in the WUI are not a widespread phenomenon and terranean areas of California and Europe and in Southeastern Australia), we found many more places where the WUI is widespread. Some of these also have frequent wildfires, whereas other hazards and conflicts may Online content dominate in other WUI areas. Furthermore, the patterns of the WUI vary Any methods, additional references, Nature Portfoloi reporting summagreatly, from large continuous WUI areas in East Africa, stretching over ries, source data, extended data, supplementary information, acknowlhundreds of kilometres, to small and patchy WUI inthe heterogeneous edgements, peer review information; details of author contributions landscapes of Mediterranean Europe (Fig.1 d,e). Irrespective of whether and competing interests; and statements of data and code availability a WUI area is large or small, it is likely to be a current or future hotspot are available at https://doi.org/10.1038/s41586-023-06320-0. of human3environmental conflicts affecting many people and many different types of ecosystems. As fire-prone areaesxpand globally, our 1. Radeloff, V. C. et al. The wildland3urban interface in the United States. Ecol. Appl. 15, WUI data can help guide proactive actions to prepare for future wildfire 7993805 (2005). in the WUI and tailor such preparations according to the dominant 2. aBnednteov-aGlounatçiaolnvemse,Ath.o&dVoileoigraie,sA..SWciil.dTfoirteasl Einnvthireonw.i7ld0l7a,n1d335u5r9b2an(2i0n2te0r)f.ace: key concepts vegetation types and associated fire regimes. 3. Bar-Massada, A., Radeloff, V. C. & Stewart, S. I. Biotic and abiotic effects of human

Wildfire damage to buildings is a global problem and we show that settlements in the wildland3urban interface. BioScience 64, 4293437 (2014). most people who experienced wildfire live in the WUI. This applies 4. PMeuslatirteesmi,pMo.r&alP(1o9l7it5is3,2P0.G30H)S(EBuurilotp-uepanSuCrofamcemGisrsidio,nD,eJoriivnetdRFersoemarSchenCteinnetle2r,a2n0d2L2a)n.dsat, even in areas where wildfires are common but population density is 5. Zanaga, D. et al. ESA WorldCover 10 m 2020 v100. Zenodo https://zenodo.org/record/ low, such as in boreal forests or where wildfires are rare and few people 5571936 (2021). live, such as in deserts. However, the subtropical and tropical moist 6. iEnllwisi,lTd.fMire.,rBisokwdmueanto,Dc.liMm.aJt.eS-.d,rJiavienn,Pd.,eFcllainnensiginanfu,Mel.mD.o&istWurileli.aGmlosobna,l GC.hJa.nGgleobBaiolli.n2c8re,ase forests, subtropical and tropical grasslands and shrublands and tem- 1544315 59 (2021 ). perate broadleaf and mixed forests are the biomes where the most 7. Crutzen, P. J. Geology of mankind. Nature 415, 23 (2002). people live in the WUI and experienced wildfires. In these biomes, a 8. tShteefgferena,Wtf.o,rCcreustozefnn,aPt.uJr.e&.AMMcBNIeOil3l,6J.,R6.1T4h3e62A1n(t2h0ro0p7)o.cene: are humans now overwhelming future increase in the exposure of people to wildfrei is probably due 9. Song, X.-P. et al. Global land change from 1982 to 2016. Nature 560, 6393643 (2018). to (1) population and WUI growth45; and (2) an increasing frequency of 10. Winkler, K., Fuchs, R., Rounsevell, M. & Herold, M. Global land use changes are four times extreme weather events due to climate change, such as longer and more 11. Pgoretaapteorvt,hPa.netp arle.vGioloubslayl emsatipmsaotfecdr.oNpalta.nCdoemxtmeunnt.a1n2d, c25h0an1 g(2e0s2h1o).w accelerated cropland severe drought, causing lower fuel moisture and a higher frequency expansion in the twenty-first century. Nat. Food 3, 19328 (2022). and severity of wildfire4s6. How climate change affects wildfires will 12. Melchiorri, M. et al. Unveiling 25 years of planetary urbanization with remote sensing: differ by vegetation type though. Grasslands, forxeample, can respond 13. Eplehrascpheacmtiv,eEs.,fBroemn-Uthrei, gL.l,oGbraolzhouvmskain,Js.,eBttaler-mOenn,Yt.laMy.e&r. MReilmo,oRte.GSleonbsa.l1h0u, m76a8n-(m20a1d8e).mass rapidly even to incremental climate change47 and herbaceous fuels can exceeds all living biomass. Nature 588, 4423444 (2020). increase rapidly after periods of high precipitation. Given that 15% of 14. Wiedenhofer, D. et al. Prospects for a saturation of humanity’s resource use? An analysis the global population lives in the grassland-dominated WUI, more oCfhmanagteerhiatltpstso:/c/kdsoai.nodrgf/l1o0w.1s0i1n6n/ji.ngeloweonrvlcdhrae.g20io2n1s.1f0ro24m1019(02002t1o).2035. Global Environ. intensive grassfires could become a big challenge for both wildfire 15. Seto, K. et al. Urban land teleconnections and sustainability. Proc. Natl Acad. Sci. USA preparedness and response44. 109, 768737692 (2012).

16. Liu, J. et al. Framing sustainability in a telecoupled world. Ecol. Soc. https://doi.org/10.5751/

Our WUI maps provide an accurate and high-resolution global per- ES-05873-180226 (2013). spective on a land-use type that is home to nearly half of the global 17. Cohen, J. D. Preventing disaster4home ignitabil ity in the wildland3urban interface. J. For. population. The overall accuracy of our WUI maps was consistently 18. 9G8a,n1t5e3a2u1m(2e0,A0.0,B).arbero, R., Jappiot, M. & Maillé, E. Understanding future changes to fires high across regions and classes with only slight differences caused by in southern Europe and their impacts on the wildland3urban interface. J. Saf. Sci. Resil. 2, uncertainties in the underlying building and land cover datasets (for 20329 (2021). example, 79% for Oceania versus 86% in Africa). Our WUI maps enable 19. Radeloff, V. C. et al. Rapid growth of the US wildland3urban interface raises wildfire risk. Proc. Natl Acad. Sci. USA 115, 331433319 (2018). researchers to identify WUI hotspots where climatechange, population 20. Modugno, S., Balzter, H., Cole, B. & Borrelli, P. Mapping regional patterns of large growth, land-use change and increasing wildfire andother hazards forest fires in Wildland3Urban Interface areas in Europe. J. Environ. Manag. 172, 1123126 are likely to cause the most pressing problems. Our maps are valu- 21. (K2a0i m16,)D.., Radeloff, V. C., Szwagrzyk, M., Dobosz, M. & Ostafin, K. Long-term changes of able because they offer a consistent global assessment at a resolution the wildland3urban interface in the Polish Carpathians. ISPRS Int. J. Geo-Inf. 7, 137 that is sufficiently fine to inform local and regnioal management, in (2018). addition to showing how global fire regimes are caused by and affec-t 22. iLni,CSa.,liDfoaron,iVa.,uKsiunmgarer,mMo.,teNgseunyseinn,gP.d&atBaa.nSecri.jeRee,pT..1M2a,5p7p8in9g(2th0e22w).ildland3urban interface ing humans48. Future research will need to assess wildfire riskin the 23. Argañaraz, J. P. et al. Assessing wildfire exposure in the Wildland3Urban Interface area of WUI in detail because that risk and the associated social vulnerability the mountains of central Argentina. J. Environ. Manag. 196, 4993510 (2017). are affected by a multitude of factors, includinga nld management 24. aSraerariscaonledap,Po.peut laalt.ioRneceexnptowsuilrdefiirnetshine wCeilndtlaranldCuhriblea:ndientteecrtfaincge.liSncksi. bToettawleEennvibrounrn.7e0d6, practices, ecological and economic value, community preparedness, 135894 (2020). natural disturbance regimes, regional precipitation, temperature and 25. Vilà-Vilardell, L. et al. Climate change effects on wildfire hazards in the wildland3 vegetation patterns and wildfire management and prevention49. For 26. uCrhbraisnt-,iSn.t,eSrfcahcwe4arBzl,uNe.p&inSelifuozraess,tRs.oWfBilhdulatnand.uFrboar.nEicnotel.rMfaacneaogf.t4h6e1C,1it1y79o2f7C(a2p0e2T0o).wn example, our maps treat both wild steppes and managed pastures as 199032019.Geogr. Res. 60, 3953413 (2022). grasslands, yet the former are highly susceptible to wildfire whereas the 27. Bar-Massada, A., Radeloff, V. C., Stewart, S. I. & Hawbaker, T. J. Wildfire risk in the latter are not. Similarly, our maps do not distinguish between natural wMialdnlaagn.d235u8r,b1a9n9i0n3te19rf9a9ce(2:0a0si9m).ulation study in northwestern Wisconsin. For. Ecol. forests and plantations, yet forest type can affecftire dynamics and 28. Kramer, H. A., Mockrin, M. H., Alexandre, P. M. & Radeloff, V. C. High wildfire damage in wildfire likelihood. In WUI areas, fire risk can etiher increase as a result interface communities in California. Int. J. Wildl. Fire 28, 641 (2019). of higher fuel loads and more human ignitions or decrease as a result 29. iMnitehtekieUw.Si.c(z1,9N9.2e3t2 a0l1.5In).tFhireeli3n,e5o0f(f2ir0e2:0co).nsequences of human-ignited wildfires to homes of fire suppression and fuel treatments, especiallywhen buildings 30. Gavier-Pizarro, G. I., Radeloff, V. C., Stewart, S. I., Huebner, C. D. & Keuler, N. S. Housing is and people are threatened45. The global WUI is, and will be, an area of positively associated with invasive exotic plant species richness in New England, USA. both human3wildlife conflicts and coexistence. Itsi, thus, a key area to 31. LEacrosle.An,pAp.l.E2.,0M, 1a9c1D3o3n19al2d5, (A2.0J1.0&).Plantinga, A. J. Lyme disease risk influences human discover how to shape resilient, sustainable and livable settlements, in settlement in the wildland3urban interface: evidence from a longitudinal analysis of addition to minimizing human3environmental conflicts50. Although counties in the northeastern United States. Am. J. Trop. Med. Hyg. 91, 7473755 (2014). fine-scale research is required to understand local drivers of WUI pa-t 32. aSnedtod,Kir.eCc.t,iGmüpnaecrtaslpo,nBb.&ioHdiuvteyrrsai,tyL.aRn.dGcloabrbaolfnorpeocoalsst.sPorfoucr.bNaantleAxpcaadns.iSocni.toUS20A31009, terns, our globally consistent assessment highlights that WUI occurs 16083316088 (2012). on all continents, reveals its broad-scale patterns and provides a basis 33. Jenerette, G. D. et al. An expanded framework for wildland3urban interfaces and their for future research on global WUI dynamics and thesocioeconomic 34. Smcahnoaegnenmagenetl,. TF.roetn at.l.EAcdola.pEtntvoirmono.r2e0w,i5ld16fi3re5i2n3w(2e0st2e2r)n. North American forests as climate and biophysical processes that make the WUI unique. changes. Proc. Natl Acad. Sci. USA 114, 458234590 (2017).

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

Disclaimer Any use of trade, firm, or product names is for descriptive purposes only and does not imply endorsement by the U.S. Government.

Open Access This article is licensed under a Creative Commons Attribution 4.0 International License, which permits use, sharing, adaptation, 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 changes were made. 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/4.0/. © The Author(s) 2023 Methods data approach. Most notably, we could not apply the building density threshold of 6.17)km22.

Defining the wildland–urban interface We reclassified the land cover data into wildland versus non-wildland. Although the WUI is defined in different ways forifdferent regions Wildland vegetation included tree cover/forest, shrubland, grassland, and applications19324,51354, we used the conceptual WUI definition of herbaceous wetland, mangroves and moss and lichen. Non-wildland the US Federal Register53, first operationalized by ref1., which is the vegetation included cropland, built-up area, bare soil and sparse vegmost widely used WUI definition in the United States and many other etation, snow and ice and water. For the interface WUI, we performed countries1,19321,54. This approach defines the intermix WUI as areas with two reclassifications, where grassland was either included as wildland more than 6.17 buildings per km2 (or one building per 40)acres) and a vegetation or not. Accordingly, we mapped two sets of large vegetation wildland vegetation area share of greater than or equal to 50%. The patches. We used the reclassified datasets to compute the wildland interface WUI is defined as areas with more than 6.17 buildings per vegetation share within a circular kernel for intermix mapping (radius km2 but less than 50% wildland vegetation that lie in proximity (less of 500)m following precedent)54. We also identified patches of more than 2.4)km) of a large patch (at least 5)km2) of wildland vegetation than 5)km2 where wildland vegetation share is greater than 75% for (with a share of more than 75%). The minimum patch size excludes interface mapping. Pixels within 2,400)m (followingprecedent54) of small urban parks from wildland vegetation1. The minimum distance large vegetation patches were included as interface WUI. of 2.4)km (1.5)miles) is included in the US Federal Register definition53 On the basis of initial tests, we set building density to zero where and represents the distance embers can fly during a wildfire55. slope is more than 25° (based on a gap-filled SRTM63/ASTER64 digital

We further extended this WUI classification approach by stratifying elevation model) or where intra-annual water occurrence in the Global WUI areas on the basis of the dominant land cover, into intermix or Water Surface dataset65 is more than 20%, that is, where water was preinterface WUI dominated by forest, shrubland and wetland versus WUI sent during at least 20% of the year to reduce commission errors on dominated by grassland (Extended Data Fig. 3). We added that strati- steep bare rock and in temporary river beds where false detections fication because grasslands are among the most diverse and dynamic of buildings were more common (Supplementary Information). Also, land cover types across the globe, with a large range of management we only considered pixels with an estimated building density of more practices, from wild steppe to managed pasture, almost resembling than 20%, thereby removing areas with very low building density for agricultural use56,57. As a result, grasslands in a given place may or may which data accuracy can be limited. We defined all pixels with an aggrenot cause wildfire risk in the WUI, which is why previous national-level gated building density greater than 0.5% in their surrounding (500)m WUI maps purposefully either included58 or excluded21 grasslands. radius) as candidate WUI pixels. Compared to the commonly used Our separation of grassland-dominated WUI supports subsequent definition of 6.17 buildings per km2, this threshold is usually slightly map interpretation. higher (depending on local building sizes). We chose this threshold to avoid WUI commission errors in low building density areas, which means that our WUI estimates are conservative. Similarly, we defined pixels with an average building density greater than 15% in a 500)m radius as having an urban character. These areas, for example, densely vegetated and high-density suburban environments, could not be classified as intermix WUI because urban vegetation often differs from wildland vegetation in terms of species identity, management practices and habitat restrictions and stronger fire control systems are in place that prevent fires. WUI mapping was subsequently performed as illustrated in Extended Data Fig. 3.

mix WUI, we determined the dominant land cover type within a pixel based on the area share of wildland vegetation. We distinguished pixels dominated by forests, shrubland and wetlands from those dominated by grassland.

We identified candidate hotspot countries as the top ten countries in their respective world region with the highest WUI area share, that had more than 20% of their wildfire area within the WUI and were more than 10,000)km2 in size. Among these, we selected the two countries with the most people affected by wildfire in the WUI. If their borders were within 200)km, we replaced the second-ranked country with the third-ranked (until border distance greater than 200)km).

cover and buildings to map the WUI. Both datasets are derived from

We used the European Space Agency WorldCover dataset to capture land cover information5. It is representative for 2020 (v.100) and provides land surface cover information distinguishing 11 classes globally with 10)m resolution. The overall accuracy of this dataset is about 75% (ref. 59). The information was derived from Sentinel-1 and Sentinel-2 satellite imagery using an ensemble of gradient-boosting decision trees with expert rule-based postprocessing to map many land cover classes at the same time (Supplementary Information)5.

We used the Global Human Settlement GHS-BUILT-S4R2022A data set (hereafter, GHS-BUILT-S) as a reference for building location and density4. GHS-BUILT-S is representative for 2018 and provides pixelwise estimates of built-up surface area (from 0% to 100% in steps of 1%) globally at 10)m resolution. The dataset contains all building types (with residential, commercial, industrial, agricultural, service or other purposes). The information was derived from Sentinel-2 satellite imagery using a symbolic machine learning approach designed to accurately capture built-up surface area (Supplementary Information).

(Africa, Antarctica, Asia, Europe, North America, Oceania and South

age and efficient processing, meanwhile avoiding spatial grid oversampling and raster distortion. We used tiles over land for all EQUI7 world regions excluding Antarctica (Extended Data Fig. 4 and Supplementary

cell, with 100)m resolution. This dataset is basedon the building density dataset we used to map the WUI but excludes non-residential buildings. It was created by disaggregating census data to grid cells using building density as weight. We computed area-weighted summaries of population data.

bon density for 2010 (ref. 67), with 300)m resolution. We converted biomass carbon density (MgC)ha21) to mass (kg) and applied a factor of two to convert carbon equivalent mass to dry matter biomass68. We computed area-weighted summaries of biomass data.

For wildfire data, we analysed the MODIS Collection 6.1 Active Fire dataset (MCD14ML)69 and extracted grid cell-based fire frequency data from 2003 to 2020 (the years with complete data records for both Aqua and Terra). We selected only fires categorized as vegetation fires and excluded those representing active volcanoes and static land sources such as gas flares. Furthermore, we distinguished wildfires from agricultural or structural fires by only including fires for which the share of wildland vegetation in that MODIS pixel was more than 50% according to the WorldCover dataset. We reduced fire frequency data to fire presence by setting many fire occurrences within one grid cell to one.

est consistent spatially explicit global time series information of fire. However, MODIS active fire data have some limitations, for example caused by the wide sensor swath of 2,230)km, which can result in pixel area differences between nadir and the swath edges of a factor of 10, thereby potentially underestimating fire area at the swath edges70. This is a particular issue as active fires are detected by thermal anomalies that, if classified as fire, are represented by a single point in the centre location of the pixel. Furthermore, its nominal pixel resolution of

low tree cover areas. This is why we also analysed data from the VIIRS

data are comparable to the MODIS product but overcome some of its challenges as a result of higher resolution and narrower sensor swath71.

a stratified random sample72. Validation sites were stratified on the basis of the mapped area shares of our five classes: forest/shrub/ wetland-dominated and grassland-dominated intermix and interface

According to ref. 72, the number of required validation sites is based on the mapped area proportion Wi of each validated class i, the target user9s accuracyUi and the target standard error S for the estimated overall accuracy (equation (1)).

n = ∑i5=0 (Wi × Ui × (1 − Ui) )   S    2 (1)

The number of sites n was drawn for each world region. The sites were then equally allocated to the five classes. A class-area proportion of the distribution would have complicated data handling, as non-WUI was expected to be by far the dominant class. The total number of validation sites was 1,504 per world region, that is, 300 or 301 per world region and class, based on a target user9s accuracy of 0.75 and a target standard error of 0.01. The sites were randomly drawn within the respective strata (Extended Data Fig. 5).

WUI versus non-WUI classes was 82.1% (Extended Data Fig. 6). The area-adjusted overall accuracy when all five classes were separately assessed was 79.6%. Class-wise user9s and producers9accuracies ranged considerably and so did overall accuracies without area-adjustment (Supplementary Data A). Area-adjusted accuracy is largely affected by the high area share and high user9s accuracy of the non-WUI class, whereas all WUI classes have a minor area share across the globe. The overall accuracy varied only slightly among the world regions (between five and ten percentage points), with no clear recurring patterns. The overall accuracy between different interpreters differed by similar margins. Only 3% of all validation points were labelled as 8uncertain9 during the validation process. The quality of the WUI map is largely a function of the quality of the underlying land cover data. For example, despite the overall high accuracy of the ESA WorldCover product, the user9s accuracy of shrublands, one of the key land cover types for WUI mapping, is only 39% (ref. 59). The quality of our WUI map also depends on the quality of building data. However, we found that because WUI requires only to be above the minimum building density threshold, even fairly widespread omission errors in areas with scattered buildings typically do not lead to missed WUI. On the other hand, in areas where small, isolated buildings are missed, the mapped WUI area was not greatly affected either because such isolated buildings do not form

census-based and building location-based WUI maps across the United

tion of 0.80) (Extended Data Fig. 7a). In densely populated northeastern states (for example, Connecticut, Massachusetts, New Jersey and Rhode Island), we found considerably more WUI area than census-based and building location-based approaches. In most other states, our WUI area estimate is very close to or slightly higher than estimates from the census-based approach and slightly lower than from the building location-based approach. We also compared our map results with data from two previous studies across 36 European countries20,73 and found high agreement in total WUI area with ref. 27 (r)=)0.94 across all countries) and medium agreement with ref. 20 (r)=)0.55) (Extended

those two studies. Compared to ref. 73, we mapped more WUI because our distance threshold for interface mapping is larger (2.4)km versus 0.6)km) and ref.20 defined the WUI as the overlap of the buffers around built-up land cover (200)m buffer) and vegetation (400)m buffer), which resulted in considerably less WUI.

We developed our WUI maps on the basis of a well-established definition of the WUI that was originally developed in the United States and successfully applied in other world regions (for example, Argentina23,74 and Poland21) However, WUI maps depend on the mapping criteria, especially the radius that is considered when computing mean building density for a given area and the distance to a large vegetation patch that determines the interface WUI. Previous sensitivity analyses confirmed the general suitability of the parameters that we selected, that is, a 500)m radius for density calculations and a 2,400)mdistance to a large vegetation patch. In the United States, radii smaller than 500)m make the resulting WUI maps highly sensitive to commission or omission errors in the underlying building dataset, whereas larger radii resulted in minimal changes in WUI area54. In Europe, overall WUI area is 25% lower when limiting interface WUI to areas within 600)m of a large vegetation patch compared to 2,400)m but WUI area estimates based on either distance were highly correlated (R²)=)0.94; ref.73). Because there are no published WUI-mapping thresholds for most parts of the globe, we decided to apply the most established approach across the globe but acknowledge the value of further regionalized research that accounts for local particularities.

points in any world region. The slight difference is probably due to the ability of VIIRS to capture smaller fires and potentially more fires in areas located at the MODIS swath edges (see Supplementary Data B3E for more detailed information and comparisons by biome, region, country and subnational administrative units).

org/record/7941460). The data are also accessible at https://geoserver. silvis.forest.wisc.edu/geodata/globalwui. We share the data for visualization purposes in an interactive data view at https://silvis.forest. wisc.edu/data/globalwui. Source data are provided with this paper.

Additional information Supplementary information The online version contains supplementary material available at https://doi.org/10.1038/s41586-023-06320-0.

Correspondence and requests for materials should be addressed to Franz Schug.

Peer review information Nature thanks Kirsten Thonicke and the other, anonymous, reviewer(s) for their contribution to the peer review of this work.

Reprints and permissions information is available at http://www.nature.com/reprints. El Salvador (b), Ecuador (c), Poland (d), Lebanon (e), Japan (f), and the Philippines (g). Interactive global map at https://silvis.forest.wisc.edu/data/ globalwui/. Extended Data Fig. 2 | WUI area share by biome. Per cent area share of the total of the four mapped WUI classes per biome. Black lines: Biome boundaries as defined by Olson et al. (2001)36. Map projection: Robinson. Grid coordinates: WGS 84.

Extended Data Fig. 3 | The Wildland–Urban Interface mapping workflow.

Classifying the Wildland3Urban Interface based on a, moving window mean building density and b, wildland vegetation in a circular kernel (r = radius = 500)m, black outline) andc, distance to the closest large wildland vegetation patch (d = distance = 2,400)m). G = grassland, F/S/W = forest/shrubland/ wetland. d, Combining building density and wildland vegetation to map the Wildland3Urban Interface. Extended Data Fig. 4 | World regions and tiling scheme, EQUI7 reference grid. Tile counts: Africa 3 3814 (purple), Asia 3 4547 (orange), Europe 3 1374 (green), North America 3 3155 (blue), Oceania 3 2026 (yellow), South America 3 2055 (red). Adapted from Bauer-Marschallinger et al. (2014)62. Map projection: Robinson. Grid coordinates: WGS 84.

Extended Data Fig. 5 | Global Wildland–Urban Interface, Validation Sites. Global distribution of all validation sites, 1,504 per world region for a total of 9,024, world regions defined according to Extended Data Fig. 4. Map projection: Robinson. Grid coordinates: WGS 84. Extended Data Fig. 6 | Overall Accuracy of global Wildland–Urban Interface (WUI) mapping. Iterative overall and area-adjusted accuracy (%) globally (top) and by world region (bottom). Columns represent different class aggregations.

All classes (left): all mapped classes individually (non-WUI, forest/shrubland/ wetland-dominated intermix WUI, grassland-dominated intermix WUI, forest/ shrubland/wetland-dominated interface WUI, grassland -dominated interface WUI). Intermix/Interface/Non-WUI (centre left): intermix and interface classes aggregated respectively. WUI vs. Non-WUI A (centre right): all WUI classes aggregated. WUI vs. Non-WUI B (right): aggregates forest/shrubland/ wetland-dominated WUI as WUI, and grassland-dominated WUI as non-WUI.

Extended Data Fig. 7 | Total WUI area comparison with previous studies states in the conterminous United States and for selected European countries. a, Comparison of aggregated global mapping results for 48 conterminous US states to Census block-level and building centroid point-based WUI mapping (Radeloff et al. 201819, Carlson et al. 202257) . The data in Radeloff et al. are representative for 2010. The data in Carlson et al. are representative for 201532018. b, Comparison for 36 European countries: The data in Modugno et al. are representative for 200620. The data in Bar-Massada et al. are representative for 20193202072. Neither European study considers grassland areas as wildland vegetation. Differences in WUI area share are due to smaller distance thresholds for WUI Interface mapping (600)m distance to closest large vegetation patch in Bar-Massada et al.73) and lower mapping resolution with a simplified buffering approach for building detection (only buffer overlaps of built-up land cover and vegetation are considered WUI in Modugno et al.20).

Article Extended Data Table 1 | The global Wildland–Urban Interface and the degree of urbanization WUI area in different degree of urbanization classes. Classes according to Dijkstra et al. (2021)41).

Extended Data Table 2 | The global Wildland–Urban Interface and income