wildland–urban interface maps vary with purpose and...

Wildland–Urban Interface MapsVary with Purpose and Context

Susan I. Stewart, Bo Wilmer, Roger B. Hammer,Gregory H. Aplet, Todd J. Hawbaker, Carol Miller, andVolker C. Radeloff

Maps of the wildland– urban interface (WUI) are both policy tools and powerful visual images. Althoughthe growing number of WUI maps serve similar purposes, this article indicates that WUI maps derivedfrom the same data sets can differ in important ways related to their original intended application. Wediscuss the use of ancillary data in modifying census data to improve WUI maps and offer a cautionarynote about this practice. A comparison of two WUI mapping approaches suggests that no single mapis “best” because users’ needs vary. The analysts who create maps are responsible for ensuring thatusers understand their purpose, data, and methods; map users are responsible for paying attention tothese features and using each map accordingly. These considerations should apply to any analysis butare especially important to analyses of the WUI on which policy decisions will be made.

Keywords: wildland– urban interface, fire policy, GIS, Healthy Forest Restoration Act, National FirePlan

T he dramatic losses and costs associ-ated with recent wildland fires inthe United States have captured

widespread media coverage and subsequentpublic attention. Recent in-depth newspa-per stories have singled out housing growthin high fire risk areas as a major problem(Heath 2007, Johnson 2007, Boxall 2008),drawing new attention to what resourcemanagers, scientists, and policymakers havelong known: that housing growth in thewildland–urban interface (WUI) is a seriousconcern in the United States. Both the Na-tional Fire Plan (NFP) and the Healthy For-

est Restoration Act (HFRA) have providedincentives for focusing wildland fire riskmitigation in this zone. In this context,WUI maps are both pragmatic policy toolsand powerful visual images with broad ap-peal, and both roles contribute to their re-cent proliferation (Radeloff et al. 2005,Wilmer and Aplet 2005, Theobald andRomme 2007).

Most WUI maps that cover a broad ex-tent (such as the whole United States) arebased on the definitions used in either theNFP or the HFRA and use population orhousing data from the US Census and veg-

etation data from satellite image classifica-tions such as the National Land Cover Data-set (NLCD). However, although they oftenrely on the same general definitions anddata, differences in the analytical methodsused to produce them can result in very dif-ferent WUI classifications. We show thatWUI maps derived from the same standarddata sets can differ in important ways thatreflect the purpose or application for whichthe map originally was developed. We alsodiscuss the practice of using ancillary dataand dasymetric mapping to modify censushousing data for use in WUI mapping andother resource management tools and rec-ommend caution with this practice.

Comparing Methods and MapsThe two WUI mapping methods com-

pared here were generated independentlyand were published in 2005, by Wilmer andAplet (2005) (WA) and by Radeloff, Ham-mer, and Stewart (2005) (RHS). Both meth-ods use the same data: the US Census data tomeasure housing (US Census Bureau 2002)and the National Land Cover Data to char-acterize vegetation (Vogelmann et al. 2001).Each method involved two analyses, one to

Received December 7, 2007; accepted October 14, 2008.

Susan I. Stewart ([email protected]) is research social scientist, US Forest Service, Northern Research Station, 1033 University Place, Suite 360, Evanston, IL60201. Bo Wilmer ([email protected]) is landscape scientist, The Wilderness Society, 950 W. Bannock St., Suite 850, Boise, ID 83702. Roger B. Hammer ([email protected]) is assistant professor, Department of Sociology, Oregon State University, 307 Fairbanks Hall, Corvallis, OR 97333. Gregory H. Aplet([email protected]) is senior forest scientist, The Wilderness Society, 1660 Wynkoop St, Suite 850, Denver, CO 80202. Todd J. Hawbaker ([email protected]) is research ecologist, US Geological Survey, PO Box 25046, MS 507, Denver, CO 80225. Carol Miller ([email protected]) is research ecologist,Wilderness Fire, Aldo Leopold Wilderness Research Institute, 790 E Beckwith Ave., Missoula, MT 59801. Volker C. Radeloff ([email protected]) is associateprofessor, University of Wisconsin–Madison, Madison, Wisconsin. This work was supported by the US Forest Service Northern Research Station and Rocky MountainResearch Station via National Fire Plan funding and by Oregon State University, the University of Wisconsin-Madison, and The Wilderness Society.

78 Journal of Forestry • March 2009

AB

ST

RA

CT

fire

determine where housing and vegetationcoincide, designated as intermix WUI, andanother to determine where houses and veg-etation are in close proximity, designated in-terface WUI. To determine where housingand vegetation coincide, areas were identi-fied where both housing criteria (housingdensity and/or pattern) and vegetative landcover criteria (type and density) were met.To determine where houses and vegetationwere in close proximity, a geographic infor-mation system (GIS) buffer analysis speci-fied some distance for “close proximity” andindicated the areas within that distance.However, the order in which the criteriawere applied and the units of analysis usedfor including and excluding potential WUIareas differed, yielding markedly differentmaps (Figure 1, a–f).

Using California as an example, we il-lustrate these differences. Both of the analy-ses start by intersecting census block bound-aries with public lands boundaries andassigning housing units in census blocks thatare partially in public ownership to the pri-vately owned portion of the census block.The landownership data available for Cali-

fornia are accurate and complete (CaliforniaSpatial Information Library [CalSIL] 1997),facilitating this dasymetric modificationwithout the loss of accuracy [1]. Workingfrom this same modified housing densitydata, the WUI analysis in both methodsstarted by identifying census blocks wherehousing density exceeded one housing unitper 40 ac (Figure 1, a and d). Then, theanalyses diverged. The RHS method se-lected those census blocks that also havemore than 50% wildland vegetation anddesignated these as intermix WUI (Figure1b) and then used a buffering process toidentify interface WUI, where housing wasin close proximity (within 1.5 mi) of a large(more than 1,325 ac) contiguous area ofwildland vegetation (Figure 1c). The WAanalysis used a buffering process to add allareas in the vicinity (within 0.5 mi) of hous-ing (Figure 1e), and, finally, removed pixels(30-m cells) in which the land cover was notconsidered wildland fuel, such as urban, ag-ricultural, or barren (Figure 1f).

A geographic comparison quantifies theproportion and extent of the areas classifiedas WUI by each, by both, and by neither

method (Figures 2 and 3; Table 1). Thoseareas identified as WUI by both methods(yellow in Figures 2 and 3; upper left cell ofTable 1) contain more than one house per40 ac and wildland vegetation. Areas identi-fied only by RHS (orange, upper right cell)contain housing adjacent to wildland vege-tation, but do not identify or measure theextent of that adjacent vegetation. Areasidentified only by WA (purple, lower leftcell) consist of wildland vegetation within0.5 mi of houses but do not identify or mea-sure the extent of that housing. Statewide(Figure 3), the two methods identified over5 million ac in common, but the WAmethod classified twice as much WUI area asthe RHS method (Table 1).

Differences between the two WUImaps stem from variations in the analysismethods that reflect the purpose and defini-tion underlying the maps. The first differ-ence is in buffering for proximity analysis,which stems from the way in which proxim-ity is treated in the NFP and HFRA WUIdefinitions. The WA method buffers 0.5 miaround all blocks that meet the housing cri-teria, in keeping with the HFRA’s specifica-

d e f

ba c

Figure 1. Major steps in the delineation of the wildland–urban interface (WUI) in the Los Angeles area using the approach by (a–c) Radeloff,Hammer, and Stewart (2005) and (d–f) Wilmer and Aplet 2005. (a) Housing density >1 housing unit per 40 acres. (b) Removal of blocks<50% vegetated. (c) Addition of area within 1.5 mi of blocks >75% vegetation. (d) Housing density >1 housing unit per 40 acres. (e)Addition of area within 0.5 mi of WUI blocks. (f) Removal of pixels with non-wildland fuels. (c and f) Representation of the final outcomeof each approach.

Journal of Forestry • March 2009 79

tion that a 0.5-mi zone around communitiesis part of the WUI (US Congress 2003). TheRHS method is based on the older NFPWUI definition, which does not specify abuffer for each community but does includeinterface WUI where homes are near wild-land vegetation (US Department of the In-terior [USDI] and USDA 2000, 2001). Thewider buffer (1.5 mi) used by RHS identifiesinterface areas but extends only from areaswith dense wildland vegetation and retainsonly those blocks or portions of blocks withadequate housing density. Proximity is cap-tured differently in the two definitions: theHFRA intent was to identify communitiesand establish a buffer around them to betreated as mitigation zones in the creation ofCommunity Wildfire Protection Plans,whereas the NFP interface definition wasbased on the notion that fire brands carrywildland fire into communities, creating anarea of potential risk. Thus, one majorsource of difference in the two maps is theHFRA-based definition used by WA and theNFP-based definition used by RHS.

The second major distinction betweenWA and RHS is in the way areas withoutwildland vegetation were treated. The RHSmethod excludes entire census blocks wherefewer than 50% of the 30-m NLCD pixelswithin the block are wildland vegetation.The WA method removes only the individ-ual 30-m NLCD pixels that lack wildlandvegetation cover types; and they do so afterbuffering, so that nonvegetated areas are re-moved from the WUI community bufferzones as well. The pixel-by-pixel retention ofwildland vegetation in the WA map showsprecisely where wildland vegetation ispresent near housing.

Maps Vary Depending on TheirPurpose

The differences between these analyseshighlight the underlying differences in themotivations for mapping the WUI. The goalof the WA approach was to identify vege-tated areas near housing to prioritize treat-ment of wildland fuels. In contrast, the goal

of the RHS approach was to identify hous-ing near forests and other wildland vegeta-tion and its consequences for wildland fire.Although both methods start by identifyingareas with housing and adjust based on thevegetative characteristics within or near theblocks, the WA method is primarily focusedon vegetation, whereas RHS is primarily fo-cused on housing. The maps and statisticsthat result from the two methods conveysomewhat different information that is con-sistent with the motivation for each ap-proach. Both specify where the WUI is lo-cated, but the WA method providesadditional detailed information about wild-land vegetation in and near communities,information that is useful for managers fo-cused on reducing hazardous fuels. TheRHS method quantifies the housing units inor near wildland vegetation, additional de-tails that allow policymakers to assess thescope of human settlement in areas affectedby wildland fire management.

There are advantages to each ap-proach, and both are relevant to current

RHS Only OverlapWA Only

RHS

WA

+ =

Forest Service

Figure 2. The combined area identified as wildland–urban interface in the Los Angeles area by both approaches. The Radeloff, Hammer,and Stewart (2005) (RHS) method focuses on identifying housing near wildlands; the Wilmer and Aplet 2005 (WA) method focuses onidentifying treatable wildland fuels near housing.


debates on wildfire policies. The RHS dataprovide specific, high-confidence countsof housing units. Census blocks are bro-ken only where the 1.5-mi buffer bisectsthem, and no block is intersected morethan once. Hence, the extent of interpola-tion needed to provide a count of housingunits is minimal. Counting the housingunits in the WUI is less feasible using theWA method because areas as small as 30m2 may be eliminated from census blocks

and it is not possible to determine howmany of the housing units in a censusblock were contained in the eliminated ar-eas versus the remaining areas. However,the WA map clearly shows the locationand extent of wildland vegetation acrossthe landscape, including the vegetation inmore densely settled areas, identifyingwhat the NFP Federal Register notice clas-sified as “occluded” WUI areas, islands ofwildland vegetation surrounded by devel-opment (USDI and USDA 2001). Inlandscapes such as southern California,these occluded areas may be at risk undersevere wildfire conditions.

Taken together, what these maps il-lustrate most clearly is the elusiveness of asingle or “actual” WUI zone; differentideas and maps of the WUI may be moreor less relevant to the user, depending ontheir needs. There is growing recognitionthat wildland fire is a “wicked” problem,where simple solutions are elusive and anyeffort to define a problem reveals a new set

of problems (Carroll et al. 2007, Hammeret al. in press). As such, map users shouldlet the situation dictate which facts (and amap is nothing more than a representationof facts) are best suited to addressing theirparticular problem. Whether one shouldfocus on the housing or on the vegetationwill depend on whether one is treatingwildland fuels, preparing a communitywildfire protection plan, or identifying ar-eas where housing growth affects forestmanagement.

More Transparency Is NeededWe can not expect map users to choose

the right map unless analysts make theirpurpose, data, and analysis methods trans-parent. Like many GIS products, WUImaps often lack the full documentationneeded for transparency. For example, whilethe general analytical approach is usually ex-plained in the metadata, manipulation of in-put data is also common and is not alwaysfully explained.

Dasymetric mapping (Mennis 2003) ofhousing data is a prime example. Most WUImaps use housing data from the US Censusto measure human presence, because struc-ture protection is important in wildland fire-fighting and because housing counts includeall residences, while population counts ex-clude seasonal residents. Unfortunately, thespatial extent of a census unit (e.g., block,block group, and tract) varies with housingdensity, tending toward larger units wherehousing is more sparse. The result can be alarge census block with a small cluster ofhomes in one area but large uninhabitedspaces in the rest and an average density toolow to meet the WUI criteria. To workaround this problem, ancillary data can beused in dasymetric mapping to modify theboundaries of census blocks. Public landboundaries are the ancillary data most com-monly used for this purpose. Where publiclands are included in a census block, the re-sult is often a large, sparsely settled block.However, public lands generally do not con-tain houses, so by intersecting public landboundaries and census block boundaries, anoriginal census block can be split into twomodified blocks, one being the part of theblock that is in public ownership and can beassumed to have no housing units, and theother—the area outside the public landboundary—which is assumed to be the cor-rect location of all of the block’s housingunits.

Figure 3. The combined area identified as wildland–urban interface across California. RHSis the approach by Radeloff, Hammer and Stewart 2005; WA is the approach by Wilmerand Aplet 2005).

Table 1. A Comparison of the Radeloff,Hammer, and Stewart 2005 (RHS) andWilmer and Aplet 2005 (WA) wildland–urban interface (WUI) classifications forthe state of California.

WA classification

WUI Non-WUI

Acres % Acres %

RHS ClassificationWUI 5,211.044 5 1,843.011 2Non-WUI 9,504.548 10 83,255.375 83


In both WUI mapping methodsshown here, this dasymetric method wasused to modify the housing data before theanalysis. The same technique has also beenused by Wilmer and Aplet (2005) in theiranalysis of WUI areas in Idaho, Califor-nia, and Colorado; by Hammer et al.(2007) in a fire risk analysis of Washing-ton, Oregon, and California; and byTheobald and Romme (2007) in creationof a national WUI map.

The quality of dasymetric mappingresults is highly dependent on the qualityof the ancillary data used, and currentlandownership data suffers from unevenquality. Omissions and inaccuracies arecommon. Although the California dataused in this analysis are accurate and com-plete, the Protected Areas Database (PAD;DellaSala et al. 2001) for the entire UnitedStates is more variable. It is compiled fromwhatever landownership information canbe attained in each state, so data qualityand errors vary by state depending on thedata available and the extent of participa-tion and assistance provided by state offi-cials. Complicated ownership patterns insome areas of the United States introduceadditional problems with data quality andconsistency. In the West, where designa-tion as public land predates widespreadsettlement of the region, there are rela-tively few private inholdings within thenational forests. The situation is differentin the East, where forests may have exten-sive inholdings that are also attractivebuilding sites and small inholdings are notrepresented in the public land boundaries.Under these circumstances—which arenot uncommon—assuming houses occuronly outside public land boundaries is notcorrect, and dasymetric mapping intro-duces errors into the housing data. Fur-thermore, the East also has extensive non-federal public forestland. The majority ofstate public land is represented in thePAD, but the large expanses of county for-ests in some eastern states (e.g., Wiscon-sin) still have not been included. Hence,the quality of landownership data variesfrom East to West, as it does from state tostate. These current problems argue forcaution in the use of landownership data,particularly for national-level mappingwhere the regional variations in data qual-ity complicate interpretation of the re-sults. When the landownership data itselfare supplemented or modified before use

in dasymetric mapping, this too should betransparent to any data user, so that thequality of the methods and the end prod-uct can be understood and judged.

Data set improvement is an ongoingprocess, and some shortcomings we notewith the PAD are likely to be remedied insubsequent versions of the data. Beyondthese particular issues, though, the generalpoint is that dasymetric mapping, likeoverlay analyses, converting GIS datafrom vector to raster (“gridding”), andother common GIS operations all have thepotential to introduce error. But errors arehidden from the map users by map appear-ance that does not vary with data qualityand even from other analysts when the op-erations and adjustments used are notfully documented in the metadata. Workmust continue to develop better data,minimize errors introduced through anal-ysis, and better communicate and docu-ment the characteristics of the data onwhich a map is built.

Conclusions and ImplicationsWUI maps have proliferated in recent

years, and different WUI patterns on thesemaps have caused confusion among landmanagers. The underlying cause for thesedifferences, though, are the different pur-poses for which the maps were developed;such as focusing on vegetation versus fo-cusing on housing or mapping for localplanning purposes versus national policy-making.

The solution is not to declare a singlemap best or, conversely, to tell managers notto use a given map because it is wrong, butrather to consider the purpose for whicheach map was developed and critically eval-uate the quality of the data and analysis onwhich it is based. Most people who use WUImaps, including scientists, land managers,planners, and wildland fire policymakers,could benefit from reviewing different WUImaps. These users often have a sophisticatedunderstanding of the WUI and recognize itas a complex place. Building on that under-standing, we suggest that because it is com-plex, we can represent the WUI in differentways, depending on what information isneeded for the task at hand. As research addsto our knowledge about how people, ecosys-tems, and fire interact in the WUI, the vari-ety of maps needed to convey this informa-tion will also expand.

Each mapping approach must be clearabout the problem being addressed, themethods used, and the effect of data qual-ity on the results. These considerationsshould apply to any analysis, but they areespecially important to analyses such asthe location and extent of the WUI, onwhich policy decisions will be made.

Endnote[1] Because public land boundary data are not

accurate and complete for the whole conter-minous United States, Radeloff (Radeloff etal. 2005) and others did not use this methodin their 2005 analysis; but in Hammer et al.2007, the multistate study area was one thatdid permit use of this method.

Literature CitedBOXALL, B. 2008. In harm’s way. Los Angeles

Times, July 31, Sect. 1.CALIFORNIA SPATIAL INFORMATION LIBRARY (CAL-

SIL). 1997. Government ownership. Availableonline at www.gis.ca.gov/meta.epl?oid�293;last accessed Aug. 15, 2008.

CARROLL, M.S., K.A. BLATNER, P.J. COHN, AND

T. MORGAN. 2007. Managing fire danger inthe forests of the US Inland Northwest: A clas-sic “wicked problem” in public land policy. J.For. 105(5):239–244.

DELLASALA, D.A., N.L. STAUS, J.R. STRITTHOLT,A. HACKMAN, AND A. JACOBELLI. 2001. An up-dated protected areas database for the UnitedStates and Canada. Natur. Areas J. 21(2):124–135.

HAMMER, R.B., V.C. RADELOFF, J.S. FRIED, AND

S.I. STEWART. 2007. Wildland-urban interfacegrowth during the 1990s in California, Ore-gon and Washington. Int. J. Wildland Fire 16:255–265.

HAMMER, R.B., S. I. STEWART, AND V.C. RADE-LOFF.0000 Demographic trends, the wildland–urban interface, and wildfire management.Soc. Nat. Resour. (in press).

HEATH, B. 2007. Wildfire areas get influx of res-idents. USA Today, May 11, Sect.

JOHNSON, K. 2007. Rethinking fire policy in thetinderbox zone. New York Times, October 28,Sect.

MENNIS, J. 2003. Generating surface models ofpopulation using dasymetric mapping. Prof.Geogr. 55(1):31–42.

RADELOFF, V.C., R.B. HAMMER, S.I. STEWART,J.S. FRIED, S.S. HOLCOMB, AND J.F. MCKEE-FRY. 2005. The wildland-urban interface in theUnited States. Ecol. Applic. 15:799–805.

THEOBALD, D.M., AND W.H. ROMME. 2007. Ex-pansion of the US wildland–urban interface.Landsc. Urb. Plan. 83(4):340–354.

US CENSUS BUREAU. 2002. 2000 Census of pop-ulation and housing: Summary file 3A. USCensus Bureau, Washington, DC.

UNITED STATES CONGRESS. 2003. Healthy Forests


Restoration Act of 2003. Available online at www.//frwebgate.access.gpo.gov/cgi-bin/getdoc.cgi?dbname�108_cong_bills&docid�f:h1904eh.txt.pdf; last accessed Sept. 2, 2008.

US DEPARTMENT OF INTERIOR (USDI) AND

USDA. 2000. The national fire plan: Managingthe impact of wildfires on communities and theenvironment. Available online at www.forestsandrangelands.gov/NFP/overview.shtml; lastaccessed Sept. 2, 2008.

US DEPARTMENT OF INTERIOR (USDI) AND

USDA. 2001. Urban wildland interface com-munities within vicinity of federal lands thatare at high risk from wildfire. Fed. Regis. 66(3):751–777. Available online at www.access.gpo.gov/su_docs/fedreg/a010104c.html; lastaccessed Sept. 2, 2008.

WILMER, B., AND G. APLET. 2005. Targeting thecommunity fire planning zone: Mapping matters.The Wilderness Society, Washington, DC.

Available online at www.wilderness.org/Library/Documents/upload/TargetingCFPZ.pdf; lastaccessed Sept. 2, 2008.

VOGELMANN, J.E., S.M. HOWARD, L. YANG, C.R.LARSON, B.K. WYLIE, AND N. VAN. DRIEL.2001. Completion of the 1990s national landcover data set for the conterminous UnitedStates from Landsat thematic mapper data andancillary data sources. Photogramm. Eng. Rem.Sens. 67:650–652.
