spatial patterns of lightning strikes in interior alaska ... · spatial patterns of lightning...

Spatial patterns of lightning strikes in interiorAlaska and their relations to elevation andvegetation

Dorte Dissing and David L. Verbyla

Abstract: The relationship between lightning strike density, vegetation, and elevation was investigated at three differentspatial scales: (i) interior Alaska (~630 000 km2), (ii) six longitudinal transects (~100 000 km2), and (iii) 17 individualphysiographic subregions (~50 000 km2) within Alaska. The data consisted of 14 years (1986–1999) of observations bythe Alaska Fire Service lightning strike detection network. The best explanation for the variation in lightning strikedensity was provided by a combination of the areal coverage of boreal forest and elevation. Each of these factors hasthe potential to influence the convective activity. Our study suggests that in a region that is climatically favorable forair-mass thunderstorms, surface properties may enhance local lightning storm development in the boreal forest. Light-ning strikes were found to occur frequently both in mountainous areas and at river flats, which is contrary to resultsfrom previous Alaskan studies.

Résumé : La relation entre la densité de la foudre, la végétation et l’altitude a été étudiée à trois échelles spatiales dis-tinctes: (i) la zone intérieure de l’Alaska (~630 000 km2), (ii) six transects longitudinaux (~100 000 km2) et (iii) 17sous régions physiographiques (~50 000 km2) à l’intérieur de l’Alaska. Les données proviennent des observations duréseau de détection de la foudre du Service des incendies de l’Alaska et couvrent une période de 14 ans (1986–1999).Une combinaison de la répartition spatiale de la forêt boréale et de l’altitude explique le mieux la variation dans ladensité de la foudre. Chacun de ces facteurs a la possibilité d’influencer l’activité convective. Notre étude porte àcroire que, dans une région où le climat est favorable aux masses d’air orageuses, les propriétés de surface peuvent fa-voriser le développement des orages localement dans la forêt boréale. Contrairement aux résultats d’études précédentesen Alaska, nous avons observé que la foudre frappe fréquemment tant dans les régions montagneuses que sur les battu-res de rivière.

[Traduit par la Rédaction] Dissing and Verbyla 782

Introduction

Thunderstorms and lightning at high latitudes are typi-cally rare compared with areas like the midwestern U.S.Great Plains, Florida, Alberta, or Saskatchewan (Orville andSilver 1997; Nash and Johnson 1996). Throughout interiorAlaska, however, lightning is not only relatively common,but plays a large role in the structure and diversity of thedominant ecosystem, the boreal forest (Kasischke andStocks 1999). For example, in the period 1990–1996, over2.7 × 106 ha burned in interior Alaska, 93% of which burnedas a result of lightning-initiated wildfires (Boles and Verbyla2000). Lightning, combined with remote, sparsely populatedregions (thus less need for fire suppression) and particularlyflammable fuels, cause lightning-initiated wildfires to burn9–10 times more area annually in Alaska than in any state ofthe contiguous United States (Court and Griffiths 1992).

There is insufficient evidence to indicate whether the pre-disposition to lightning activity is due to (i) the dry conti-nental climate of interior Alaska being particularlyconducive to thunderstorms, (ii) the surface–atmosphere ex-change characteristics of boreal forests providing effectivetriggering mechanisms for thunderstorm development, or(iii) a combination of the two. In Alberta and Saskatchewan,lightning-initiated fires were found to occur most frequentlyduring periods of local and mesoscale convection (Nash andJohnson 1996). Land surface – atmosphere interface hasbeen identified as a key area to reducing the uncertainties incontemporary regional and global climate models (Hardinget al. 2001) and major research has been done on the topic(Pielke and Vidale 1995; Serreze et al. 2001; Foley et al.1994; Bonan et al. 1995).

Climate is a dominant factor controlling the distribution ofvegetation. Since controls over climatic factors are exerted atmany spatial scales (globally, regionally, and locally), a sim-ilar scale dependence on vegetation distribution is observed(i.e., with latitude, continentality, and topography) at multi-ple spatial scales. In Alaska, for example, Viereck at al.(1992) noted the effect of continentality on the distributionof Sitka and white spruce. Sitka spruce occur typically inmaritime climates, whereas white spruce are largely re-stricted to continental climates (Viereck et al. 1992). Atsmaller scales, Van Cleve et al. (1991) relate the distribution

Can. J. For. Res. 33: 770–782 (2003) doi: 10.1139/X02-214 © 2003 NRC Canada

770

Received 6 November 2001. Accepted 3 December 2002.Published on the NRC Research Press Web site athttp://cjfr.nrc.ca on 31 March 2003.

D. Dissing1 and D.L. Verbyla. Department of ForestSciences, University of Alaska Fairbanks, Fairbanks,AK 99775-0100, U.S.A.

1Corresponding author (e-mail: [email protected]).

I:\cjfr\cjfr3305\X02-214.vpWednesday, March 26, 2003 3:00:54 PM

Color profile: DisabledComposite Default screen

of white and black spruce in interior Alaska to topographi-cally controlled microclimate.

Conversely, vegetation may also influence climate at vari-ous spatial scales. Surface albedo and roughness and the ef-ficiency with which incident solar radiation is converted tonet radiation are important factors influencing the mecha-nisms whereby vegetation can modify climate (Garrett 1982;Bonan et al. 1995). Foley et al. (1994) modeled Holoceneclimate warming at the global scale and found that an expan-sion of the boreal forest region had a greater effect on re-gional climate warming than did variations in solar inputassociated with Milankovitch cycles. Chambers (1998)found that removal of native vegetation at the mesoscale(10–200 km; Oke 1987) was sufficient to sustain circulationsat this scale. At the local scale (0.01–50 km; Oke 1987), the“oasis effect” happens when warm, dry air is blown over acool, moist oasis and gets latently cooled because of evapo-ration, thus modifying the local climate (Stull 1997).

There are two possible mechanisms in which edge effectscan influence convective triggering, depending upon the syn-optic wind conditions. One possibility is through mesoscalecirculations, another is due to surface roughness effects(Vidale et al. 1997).

Elements of surface patchiness can create horizontal gra-dients in surface and boundary layer heat fluxes, because ofdifferential solar radiation absorption, evaporation, transpira-tion, and aerodynamic transfer, which may in turn generatemesoscale circulations (Vidale et al. 1997). During periodsof low synoptic wind conditions, mesoscale circulations canresult in heating of the convective boundary layer and cool-ing aloft. When the synoptic flow becomes stronger, differ-ential surface roughness effects are the predominant factor inconvective triggering (Vidale et al. 1997). Both mesoscalecirculations and surface roughness effects depend on thescales on the surface inhomogeneity. For the BOREASstudy, Vidale et al. (1997) found that patches between 4 and280 km had the potential to influence the convective bound-ary layer development. They also found that the mesoscalecirculations may have significance even on days with sus-tained synoptic winds.

The issue of causal relationships between vegetation andclimate is still under debate. Bryson (1966) analyzed sum-mer air-mass distributions in relation to the boreal forest ex-tent, and found that the northern edge of the boreal forest isrelated to the mean summer position of the Arctic front.Pielke and Vidale (1995), using data from the BOREAS ex-periment, suggested that the northern boreal forest boundaryinfluences the summer location of the Arctic front across theNorth American continent. Conversely, other studies arguethat the position of the northern treeline represents a re-sponse to, rather than a forcing on, the summer position ofthe Arctic front (Serreze et al. 2001; Beringer et al. 2001).However, the authors conclude that the differences in energypartitioning between tundra and boreal forest could be sig-nificant enough to drive local-scale circulations of ecologicalimportance (Beringer et al. 2001).

Surfaces can act to trigger convection. The triggeringmechanisms are usually lifting, and could be due to oro-graphic effects, frontal uplift, low-level convergence, orheating from below (Schroeder and Buck 1970). Smallerconvective storms generally result from an unstable atmo-

sphere, which, when triggered, results in the movement ofboundary layer air to the level of free convection (Nash andJohnson 1996).

Two factors strongly influence surface properties and thuscontrols over triggering potential: vegetation and physiogra-phy. Convection can be affected by albedo, sensible heatflux, surface roughness, and landscape heterogeneity(Knowles 1993; Rabin et al. 1990), properties that differamong vegetation types. This is related to the differences be-tween the surface types and the resulting partitioning of en-ergy. In general, sensible heat fluxes are larger over borealforest than over tundra (Chapin et al. 2000; Lafleur andRouse 1995; Pielke and Vidale 1995; Eugster et al. 2000).

At the mesoscale, O’Neal (1996) found that deciduousforest in mountainous areas promoted more convectivefluxes and associated cloud cover than the surrounding flat,nonforested areas in midwestern North America. In SouthernCalifornia, local scale studies found that conifer forest andchaparral brush had higher lightning strike densities thanbroadleaf woodland, coastal sage shrub, and herbaceous veg-etation types (Wells and McKinsey 1993). However, the au-thors concluded that elevation (rather than vegetation) wasthe dominant factor influencing lightning distribution.

In mountainous areas, differential heating of mountainslopes results in upslope winds and thus low-level conver-gence, which is the major factor contributing to the sus-tained instabilities necessary for thunderstorm development(Biswas 1976). Slope winds are produced by the local pres-sure gradient caused by the difference in temperature be-tween air near the slope and air at the same elevation awayfrom the slope (Schroeder and Buck 1970).

The physiographic features can influence convective trig-gering through three factors with separate effects and mech-anisms that are included in the physiographic factor:(1) elevation, (2) aspect, and (3) topography. The most influ-ential of the three factors are topography and aspect. Topog-raphy influences convective activity through the mechanismof differential heating described above. Likewise, aspect, es-pecially in high latitudes, can cause significant differences insurface heating, leading to possible triggering. However,both topography and aspect vary with scale, and has less im-portance at the 1- or 100-km2 scale level; thus simple eleva-tion values are often used as a proxy for topography,assuming that higher elevation represents ridges and escarp-ments.

At the regional scale, topography has been found to havea dominant role in influencing lightning strike activity in theconterminous United States (Wells and McKinsey 1993;Lopez and Holle 1986). Court and Griffiths (1992) showed apositive correlation between the elevation of weather stationsand mean days per year of registered thunderstorms in thewestern United States. In Alaska, Reap (1991) found a gen-erally positive relationship between lightning strike densityand elevation below 800 m above sea level and a negativecorrelation above 800 m. At the local scale, Wells andMcKinsey (1993) found a positive correlation between ele-vation and the density of lightning strikes for San Diegocounty, California.

Based upon the demonstrated links between lightning andfire in the Alaskan boreal forest (Court and Griffiths 1992;Boles and Verbyla 2000), and the dominant role that fire

© 2003 NRC Canada

Dissing and Verbyla 771



plays in the spatial and temporal heterogeneity of the speciesthat constitute this ecosystem (Bourgeau-Chavez et al.2000), feedback systems among climate, vegetation, andwildfire may exist in this region. However, whether theseobservations should be mostly attributed to an atmospherethat is regionally favorable for convection of direct effectsof the landscape remains undetermined. Numerical modelingstudies suggest a positive feedback system between wildfireburn scars and mesoscale circulation patterns (Knowles1993). Through changes in surface properties and differ-ences in heat fluxes between the burned area and the un-burned areas (Chambers and Chapin 2003), large burnedareas may be capable of setting up “land-breeze” circula-tions patterns similar to “sea-breeze” systems. Knowles(1993) concluded that such positive feedback was poten-tially strong enough to result in increased convection that inturn produces lightning.

ObjectivesAs an independent means to verify the hypothesis that

landscape effects are responsible for the prevalent convectivedevelopment within interior Alaska, the objective of thisstudy was to analyze the relationship between lightningstrike density and both vegetation and elevation, at threespatial scales: (i) interior Alaska (630 000 km2), (ii) longitu-dinal transects (50 000 – 150 000 km2), and (iii) physio-graphic regions (6000 – 110 000 km2).

Study area

Interior Alaska is bound in the north by the Brooks Range(1000–2500 m above sea level) and in the south by theAlaska Range (1000–6000 m above sea level) (Fig. 1). Thetopographic corridor provided by the bounding mountainranges, and extensive Canadian continent to the east, mostlyrestricts the interior to a single water source for atmosphericmoisture, the Bering Sea. Maritime moisture is transportedfrom west to east across the region by large-scale advection(Reap 1991; Sullivan 1963). As a consequence of the west–east moisture and temperature gradients, and progressionfrom a maritime to a continental climate across the interior,the dominant vegetation cover changes from tundra, to bo-real forest with distance from the coast (Fig. 2). Boreal for-est is the dominant vegetation type in the study region,covering 63% of the total area.

The coastal areas experience a relatively mild climate,buffered by the stable maritime influence. The central andeastern regions of the interior experience large annual tem-perature extremes, with relatively hot summers (Hammondand Yarie 1996). In spite of the moisture gradient, there isgenerally sufficient moisture available throughout the inte-rior to fuel thunderstorms, either as a result of recent trans-port, or released from large stores in permafrost or deeporganic soils that trap snowmelt. In the western interior,where tundra is the dominant vegetation type, high surfacealbedo and near saturated soils result in minimal atmo-spheric heating and relatively stable air masses. These con-ditions inhibit extensive convective activity. The coastalareas have a milder climate with more stable air masses, lesssurface heating, and warmer air aloft, inhibiting extensiveconvective activity.

There were two primary considerations that led to thechoice of interior Alaska as a study region. Firstly, it is aconfined region where most thunderstorms develop withoutthe influence of synoptic-scale weather systems (Biswas andJayaweera 1976; Henry 1978). The infrequent intrusion ofsynoptic-scale weather systems is a crucial factor, becauselarge-scale forcing can confound the effects of vegetationand topography on convective activity. Secondly, almost90% of the annual Alaskan lightning strikes are recorded inthis area.

In this study, the data were examined at three differentspatial scales. The interior Alaska region represented the re-gional scale, the longitudinal transects represented the meso-scale, and the physiographic subregions, which sometimescorresponded to individual watersheds, represented the localscale (upper end). The east–west orientation of the topo-graphic bounds to the interior result in a good correlation oflongitude with climate (from maritime to continental) acrossthe study region. Since there was no physiographic classifi-cation of interior Alaska at the time of this study, we dividedinterior Alaska into 17 physiographic regions (Fig. 3), basedprimarily on physical landscape features such as relief andproximity to major rivers. The delineations were constructedfor this study and do not necessarily match existing man-agement units. The physiographic regions are delineated af-ter large-scale landscape forms, representing local large-scale topographical landscape changes (Yukon Flats toWhite Mountains, for example), to examine local semi-homogeneous subregions.

Alaska thunderstorm characteristicsBiswas and Jayaweera (1976) used NOAA Very High

Resolution Radiometer data to study the predominant pat-terns of thunderstorm meteorology in Alaska and found twodifferent types; air-mass and synoptic thunderstorms. An airmass is defined as any widespread body of air that is ap-proximately homogeneous in its horizontal and vertical ex-tent, particularly with reference to temperature and moisture(Huschke 1959). Air-mass thunderstorms consisted of iso-lated storms in confined areas and were mainly associatedwith sloping topography. Synoptic thunderstorms featuredwidespread and intense activity over large areas, triggeredby large-scale weather systems that were often tied to effectsof the jet stream (Biswas and Jayaweera 1976).

A large fraction of the annual number of lightning strikesrecorded in Alaska are due to a few days of intense synopticstorms. However, most days with recorded lightning are dueto air-mass storms, formed in the absence of significantlarge-scale circulation (Biswas and Jayaweera 1976; Henry1978). These low lightning strike frequency air-mass thun-derstorms are the starters of most of the lightning-causedfires in interior Alaska (Henry 1978). Under conditions con-ducive to air-mass thunderstorms, properties associated withtopography and vegetation such as sensible heat flux andalbedo, surface roughness and heterogeneity are more likelyto influence convection and subsequently lightning patterns.Vidale et al. (1997) looked at development of mesoscale cir-culation of the BOREAS region and found the circulation tobe strongest during periods of weak synoptic flow. However,the circulations were still present under stronger synopticflow patterns (Vidale et al. 1997).

© 2003 NRC Canada

772 Can. J. For. Res. Vol. 33, 2003



Data

Lightning detection and protection network dataThe Bureau of Land Management Alaska Fire Service

(AFS) operates an automated network of cloud-to-groundlightning sensors (Reap 1991). This lightning detection net-work was constructed as an aid for fire suppression, since in-terior Alaska has the highest potential for thunderstorms andthe highest number of lightning-ignited fires in the state. Thenetwork has been in operation since 1976 and consists ofnine stations in Alaska and three in the Yukon Territory(Fig. 1).

The position of a lightning strike is estimated by triangu-lation. Consequently, a strike is recorded only if it is de-tected by more than one of the sensors. Positional accuracyof estimated lightning strike locations varies with the num-ber of detectors sensing the strike, and with detector locationgeometry. For more details on how the lightning strike de-tection network functions, we refer to Hiscox et al. (1984).

The data recorded for each detected strike include an esti-mate of time, location, and the positional accuracy of thestrike. Before the start of the 1995 fire season, the networkunderwent corrections for systematic site errors. In the U.S.National Lightning Detection Network, data are reprocessedwithin a few days of real-time acquisition. These repro-cessed data are not subject to site errors (Cummins et al.1998). Lightning strike data within the interior Alaska re-gion processed after May 1995 are assumed to have a posi-tional accuracy of 2–4 km, at best (Global Atmospherics,Inc. 19952). The detection efficiency is assumed to be 60–80% in interior Alaska, decreasing rapidly towards thecoastal regions (Global Atmospherics, Inc. 19952). Reap(1991) using a subset of the same data (1987–1989) assumesthe location efficiency to be on the order of 5–10 km, andthe detection efficiency to be 70% or better.

Studies concerning detection efficiency and location accu-racy have not been done in Alaska. However, for the U.S.National Lightning Detection Network in eastern New York,

© 2003 NRC Canada


Fig. 1. Interior Alaska, the area south of the Brooks Range and North of Alaska Range, bound by the Alaska–Canada border in theEast (141°W), and the Bering Sea in the west. Nine lightning detection network sensors are shown. The meridians mark the boundariesfor the longitudinal transects.

2 Global Atmospherics, Inc. 1995. Written communication from Global Atmospherics to A.F.S. containing estimated lightning strike detec-tion efficiency and location accuracy of the ALDF and IMPACT sensors. Letters dated May 18, and September 7, 1995. Global Atmospher-ics, Inc., Tucson, Arizona.



Idone et al. (1998a, 1998b) did a performance evaluationcomparing the network lightning data with video-derived lo-cations of cloud-to-ground lightning and found a median lo-cation accuracy of 2–4 km. The detection efficiency wasfound to be 67–86%. A study of detection efficiency of theBritish Columbia network yielded results around 50–70%(Gilbert et al. 1987).

Unless otherwise stated, all analyses conducted as part ofthis study have been based on a 14-year data set, which cov-ers the 1986–1999 period and can be accessed at the Bo-nanza Creek Long Term Ecological Research Site Web site(http://www.lter.uaf.edu). The entire 1986–1999 lightningstrike data set consists of 272 079 lightning strikes. TheAlaska lightning detection network was, during the 2000 fireseason, upgraded to IMPACT sensors (AFS, personal com-munication). We have chosen to not include these data inthis study, because of the changed detection efficiencies andlocation accuracies.

The data set shown in this study uses 14-year averages orsums of individual 1-km grid cells, to which we have ap-plied a smoothing scheme in the areas around the stations,

where the site error corrections otherwise would have influ-enced our results.

Throughout this paper, a “lightning strike” refers to a re-corded cloud-to-ground lightning discharge. Nothing is in-ferred about whether a particular strike causes ignition.

GIS dataElevation at 1-km resolution was resampled from the

United States Geological Survey (USGS) 1 : 250 000 seriesdigital elevation models (Fig. 1; USGS 1990). We used avegetation grid produced by Markon et al. (1995) at 1 kmresolution. The vegetation classes were produced using anunsupervised classification of 1991 multitemporal AVHRRNormalized Difference Vegetation Index data (Markon et al.1995) and can be obtained from the Alaska Field Office ofthe EROS Data Center. We then aggregated the vegetationclasses to four major categories: boreal forest (63% of totalarea), tundra (21%), shrubs (12%), and other (4%) (Fig. 2).The “other” category consisted of snow, ice, water, and the1990 and 1991 wildfire burn scars.

© 2003 NRC Canada

774 Can. J. For. Res. Vol. 33, 2003

Fig. 2. Vegetation distribution in interior Alaska, based on vegetation classes produced by unsupervised classification of 1991 multi-temporal AVHRR Normalized Difference Vegetation Index data (Markon et al. 1995). Vegetation classes are simplified here into fourbroad categories.



The data detected by the AFS lightning strike detectionnetwork was converted to a GIS point coverage, based onthe location coordinates (latitude, longitude) assigned at thetime of detection. For each lightning strike in the 1986–1999time period, an elevation and a vegetation class were as-signed.

Materials and methods

Based on previous studies (i.e., Reap 1991; Wells andMcKinsey 1993; Lopez and Holle 1986) on the topic, we ex-pected elevation to be the most influential factor, thus ex-plaining the greatest variation in the lightning strike data.The hypothesis tested at all three scales was the following:How does elevation and percentage of real coverage of bo-real forest influence the lightning strike density?

At the three scales the data were summarized by elevationzone to obtain information like percentage of cover of boreal

forest and lightning strike density and to reduce the data setfrom more than 630 000 cells.

Regional scale (interior Alaska)At the regional scale, the data set contained four variables

for each data point (elevation zone, percentage of boreal for-est, lightning strike density, elevation zone × percentage ofboreal forest).

The data set was summarized by 100-m elevation zones(i.e., 0–100 m, 100–200 m, etc.) and the lightning strike den-sity and percentage of the area covered by boreal forest werecalculated for each elevation zone. The lightning strike den-sity data were based on the sum of all lightning strikes per1 km2 for the entire 14-year data set (1986–1999). The dataset was based on the 1-km2 grid cells to capture as much ofthe variation as possible, and the data set consisted, aftersummarization, of 19 data points.

For all statistical analyses, boreal forest was entered as theonly vegetation factor. This was done because boreal forest

© 2003 NRC Canada


Fig. 3. Physiographic regions within interior Alaska. The areas with the highest lightning strike densities are White Mountains, YukonUplands, Ray Mountains, and Kantishna River Flats in the eastern and central part of the interior. Th lowest densities are at NulatoHills, Innoko Flats, and Seward Peninsula in the western part. Note that boreal forest has a higher lightning strike density than tundraand shrubs for about 65% of the regions.



is the dominant vegetation type in interior Alaska. Com-bined, tundra and boreal forest cover 81–100% of the area atall elevations, and the relationship between the two variablesis inversely proportional. Thus similar analysis with severalvegetation types would have produced redundant results.

A 10 × 10 km grid of tundra and forest data was used forthe study of potential edge effects. This was done to mini-mize edge effects that a 1 × 1 km patch could have pro-duced. The mean lightning strike density for all (tundra andforest) the 100-km2 cells was calculated and stratified intolightning density classes of 0.1 strikes/km2.

Mesoscale (longitudinal transects)For the longitudinal transects a combined data set was

made, containing six sets of data, one for each transect. Eachdata set (one for each of the six transects) was summarizedby 100-m elevation zones (i.e., 0–100 m, 100–200 m, etc.)and the lightning strike density was calculated for every ele-vation zone.

The data set was based on the 1-km2 grid cells to captureas much of the variation as possible, and the data set con-sisted, after summarization, of 86 data points.

Local sale (physiographic regions)The physiographic regions were treated similarly to the

longitudinal transects data set, and were similarly based onthe 1-km2 grid cells. The data set consisted of 161 datapoints.

Statistical analysisThe data were tested for normality, using a χ2 normality

curve fit. The variables were found to not be significantlydifferent from normal with P values < 0.05.

To explore the effects of elevation and percentage of bo-real forest areal coverage on lightning strike density, we de-veloped a multiple linear regression model. To select themodel with the best fit we first tested the relation amongeach variable and lightning strike density separately, andthen applied models that included both variables as well asan interaction term. The multiple regression was done as aforwards stepwise regression, with a probability of 0.05 toenter the model and 0.1 to remove. The regressions were runfour times at each scale, examining the following models:(a) lightning strike density = a1 + b1 (elevation zone);(b) lightning strike density = a2 + b2 (percentage of boreal

forest);(c) lightning strike density = a3 + b3(percentage of boreal

forest) + c3 (elevation zone);(d) lightning strike density = a4 + b4 (percentage of boreal

forest) + c4 (elevation zone) + d4 (percentage of borealforest × elevation zone).

The model chosen as the best fit had the highest adjustedr2 value.

Proximity studyPrevious work has focused on the difference in surface

heat fluxes between boreal forest and tundra (Pielke andVidale 1995). We have chosen to focus the study of edge ef-fects similarly, because the difference in the lightning strikedensity was greater between boreal forest and tundra thanbetween boreal forest and shrubs. Additionally, these two

vegetation types are the two most abundant within the studyarea, providing for a larger sample size than any other vege-tation transitions. For each class the mean distance from atundra patch to the closest boreal forest patch, regardless ofsize, was computed. Similarly, mean distances from borealforest patches to tundra patches were computed.

Results

Interior regionThe relationship between lightning strike density and ele-

vation was generally positive up to a maximum elevation of1100–1200 m (Fig. 4). Above this elevation, the relationshipwas negative. The boreal forest covers 70–90% of the areabetween 200 and 1100 m in elevation. Above and belowthese elevations, the forest areal coverage is much less(Fig. 4). Most of the area below 200 m fell within thecoastal tundra and (or) shrubs, and most of the area above1100 m was above the altitudinal limit for forest.

Tundra lightning density was greater in tundra cells closeto boreal forest (Fig. 5). However, the distribution was suchthat only 1% of the grid cells fell in the 0- to 20-km dis-tance, 29% in the 20- to 30-km distance, and 70% in thegreater than 30-km distance to the nearest forest edge.Within the cells that are 20–30 km from the forest edge, thelightning strike density varied from 0.02 to 0.08 strikes/km2.The similar plot for the forest pixels, testing the proximity totundra pixels, showed similar trends with the curve shapebeing almost a mirror image of the curve in Fig. 5 (Fig. 6).Here the distribution showed that 53% of the pixels are inthe 0- to 10-km span, 47% in the 10- to 25-km span, andonly one pixel in the greater than 25-km span. The lightningstrike density within the 10- to 25-km distance to tundraedges varies from 0.02 to 0.17 strikes/km2.

The multiple regression at the regional scale showed thatthe best model, explaining the largest percentage of the vari-ance (r2 = 0.66) in the lightning strike density data wasmodel d, which was based on the interactive effect of eleva-tion and percentage of boreal forest (Table 1). The secondbest model (r2 = 0.57) was model c, involving the effects ofboth elevation and percentage of boreal forest, with borealforest entering the model first. Elevation did not enter themodel at all as a sole influence, whereas percentage of bo-real forest explains 19% of the variability in the lightningdata.

Longitudinal transectsThe boreal forest had the highest lightning strike density

across the climate gradient of continentality from the easterninterior to the first of western zones (Fig. 7). In the furthestwestern zone, shrub and tundra vegetation had lightningstrike densities that were similar to those of the forested ar-eas. However, the forest lightning strike densities in thistransect were smaller than those win the other five zones.The lightning strike densities of tundra were relatively con-sistent across the climate gradient (Fig. 7). In the shrub veg-etation the density remained at low in all but the twoclimatic transition zones.

Multiple regression analysis indicated that the best modelwas model d, where percentage of forest enters the modelfirst and explains 21% of the variability in the data of the

© 2003 NRC Canada

776 Can. J. For. Res. Vol. 33, 2003



overall 57%, the inclusion of elevation zone and elevation ×percentage of boreal forest improves the model significantly.Both variables are positively related to lighting strike den-sity, whereas their combined product is negatively related(Table 2). The second best model was model c, which ex-plains 45% of the variation in the data. Like at the regionalscale level, elevation did not enter the model at all as a soleinfluence.

Physiographic regionsFor each of the 17 physiographic subregions the mean

lightning strike density for the 1986–1899 time period wascomputed (Table 3). The three physiographic subregionswith the highest lightning strike densities were mountain re-gions, followed by five regions of river flats. The lowestlightning density occurred at Nulato Hills. In 65% of thesubregions, the highest average lightning strike density wasrecorded in the boreal forest. In general, the highest light-ning strike densities occurred in the subregions with the

higher (mean) elevation. A definitely east–west trend canalso be traced, identifying the high lightning density subre-gions as the eastern regions.

The multiple regression analysis indicated that, again, thebest model was model c, where percentage of forest entersthe model first and explains 18% of the variability in thedata of the overall 27%, the inclusion of elevation zone isimproving the model significantly. Again, both variables arepositively related to lighting strike density (Table 4). Thesecond best model was model b, using percentage of borealforest as the sole influential factor, explaining 18% of thelightning data variability. At this scale, elevation did enterthe model as a sole influential factor but could only accountfor 1.2% of the lightning variation.

Discussion

The highest lightning strike density occurred in the borealforest vegetation at all three scales. The trend was consistentfor all of interior Alaska, and persisted along the climatolog-ical gradient of continental (eastern interior) to maritime(western interior) climate (Fig. 7). More than half of thephysiographic regions had the highest lightning strike densi-ties within boreal forest (Table 3). The two regions with thelowest lightning strike density in boreal forest were YukonDelta and Innoko Flats. Both of these regions are in thewestern part of the interior, and their low strike densitiesmay likely reflect the maritime climatic influence.

For all the examined scales, boreal forest and elevationcombined were found to be positively related to lightningstrike density (Tables 1, 2, and 4). The statistical resultsstrongly indicate that it is the interaction or combination be-tween elevation and boreal forest that accounts for a largefraction of the variability seen in the lightning strike data.

The interactive influence of elevation and boreal forestcoverage at the regional level accounted for approximately66% of the variability in the lightning strike data. At the twosmaller scales, the combined effects of elevation and borealforest coverage explained the greater part of the variation(mesoscale: r2 = 0.57; local scale r2 = 0.27). This suggests

© 2003 NRC Canada


Fig. 4. Mean (1986–1999) lightning strike density (per km2) di-vided into elevation intervals (100 m) for interior Alaska (darktriangles). Horizontal lines represent ±1 standard deviation. Theopen squares represent the percent of the total area within eachelevation zone covered by boreal forest.

Fig. 5. Mean (1986–1999) tundra lightning strike density (perkm2) as a function of the mean distance from the 10 × 10 kmtundra patches to the nearest forest patch of 10 × 10 km orgreater size. Numbers of grid blocks per lightning strike densityclass are included next to the markers.

Fig. 6. Mean (1986–1999) forest lightning strike density (perkm2) as a function of the mean distance from the 10 × 10 kmforest patches to the nearest tundra patch of 10 × 10 km orgreater size. Numbers of grid blocks per lightning strike densityclass are included next to the markers.



that longitude (i.e., the degree of continentality) is a moreimportant factor explaining the lightning strike distributionin interior Alaska than the physiographic subregions, andthat the longitudinal transects provide a built-in climatic(continentality) control factor.

However, this also presents a surprising result in showingthat for the scales shown and the data used in this study, it isthe interaction, or combined effects, of boreal forest cover-age and elevation that influences the lightning strike densi-ties, not elevation as a factor by itself. On the contrary, thisfactor seems to have the least influence. An important thingto consider here are potential biases in the lightning detec-tion network, as mentioned by Gilbert et al. (1987) fromtheir study in B.C., where the detection efficiencies were be-low average in mountainous regions. Also, this lightning de-tection network is focused on cloud-to-ground lightning,leaving the possibility open that thunderstorms can still de-velop over higher terrain, but our lightning strike data willnot portray this adequately. However, our main focus is thecloud-to-ground lightning strikes, because of their role aspotential fire starters.

The persistent trend of a positive relationship betweenlightning strike density, boreal forest, and elevation repre-sents a difference from most previous results in the field,which have concentrated on the positive relationship be-tween elevation and lightning strike density (Wells andMcKinsey 1993; Lopez and Holle 1986; Reap 1991). For allof interior Alaska (Fig. 4), the decrease in boreal forest areal

coverage at upper elevation tree line of 1000–1200 m corre-sponded with a decrease in lightning strike density. Thiscould reflect the differences in energy partitioning betweenthe boreal forest below, and the tundra above the tree line,similar to the arguments from Pielke and Vidale (1995). Asecond small lightning strike density peak occurs at 1600–1700 m elevation. These high lightning strike densities at the1200–1700 m elevation occurred only in the eastern interior,and have huge standard deviations associated with them,making it hard to draw any conclusions about them.

Some discrepancies exist between the data from our anal-ysis and the study of Reap (1991). The positive relationshipbetween elevation and lightning strike density up to 1100–1200 m that we found for all of interior Alaska followed thegeneral trend observed by Reap (1991), although Reapfound the maximum lightning strike density to occur at800 m. This difference in the elevation at which the maxi-mum lightning strike density occurs is likely attributable tothe differences in grid-cell sizes between our studies (ourstudy: 1-km grid, Reap’s study: 48 km). The larger grid sizescompute mean elevation over a larger area, which wouldproduce lower values.

Furthermore, our study focused on the interior of thestate, while Reap’s study area included the whole state ofAlaska, thus incorporating the low elevation, low lightningstrike density areas along the north and west coast andsoutheastern Alaska. Additionally, the lightning data onwhich Reap’s study was based (1987–1989, also AFS data)included two of the three lowest lightning strike years in the14-year (1986–1999) data series that were used for thisstudy.

The highest lightning strike physiographic regions werefound to be in mountainous areas of eastern Alaska (Ta-ble 3). However, the Kantishna River, Tetlin, Koyukuk, Yu-kon, and Tanana Flats show relatively high lightning strikedensities, higher than several mountainous regions(Kuskokwim and Purcell Mountains, Nulato Hills, Table 3).This somewhat contradicts the conclusions of Reap (1991)and Biswas and Jayaweera (1976), who found that no riverflats, with the exception of the Tanana Flats, showed anysignificant thunderstorm development or lightning strike fre-quencies. Both of these studies were based on relatively fewyears of data (3 and 1, respectively), and at least 2 of theseyears were very low lightning strike years. Possibly this dis-crepancy is explainable if low-lightning years also can be as-sumed to have fewer air-mass storms, and thus the influenceof elevation could be of greater importance than that of bo-real forest, therefore resulting in a higher lightning strikedensity over higher terrain.

© 2003 NRC Canada

778 Can. J. For. Res. Vol. 33, 2003

Dependant variable* Pr2

(adj.)Beta(1)

Beta(2)

Beta(3) Intercept

SEestimate

SEintercept n F

(a) E The variable did not enter the model at all(b) %BF 0.002 0.19 0.485 40.84 23.62 10.97 19 5.23(c) E (1) and %BF (2) 0.0021 0.57 0.839 1.05 –411.17 17.19 112.38 19 13.09(d) E (1), %BF (2),

and E × %BF (3)<0.0001 0.66 0.824 35 15.29 5.79 19 30.07

*E, elevation; %BF, percentage of boreal forest. The letters refer to the model tested, and the numbers after the variablesrefer to the beta value numbers.

Table 1. Results from forwards stepwise multiple regression, interior region.

Fig. 7. Mean (1986–1999) lightning strike density (per km2) as afunction of the longitudinal transects, divided into the three veg-etation classes: tundra, shrub, and boreal forest. The right side ofthe figure is the eastern part of the interior Alaska, the left sideis the western part. The climatic zone is marked above the chartpanel.



One question that arises from our results is the following:Can the boreal forest itself enhance thunderstorm develop-ment or does it simply occur in a climatic region that sus-tains convective activity? The most likely explanation is acombination of both: in a region already favorable for con-vective activity, the differential heating due to landscape het-erogeneity and therefore variation in surface energypartitioning over the boreal forest provide the triggeringmechanism to enhance thunderstorm development. Lynch etal. (2001) suggest from their studies of correlations among

vegetation, topography, and the Arctic frontal zone, that al-though vegetation contrasts were found to be insufficient forinducing frontal activity, they could and would contribute totopographically generated preferred frontal zones. Thesefindings support the idea of convective triggering being pro-vided by a combined effect of the vegetation and topogra-phy, although our statistical results suggest that vegetation(boreal forest) is the most important.

Our proximity study showed that most of the higher light-ning strike densities, both within pixels classified as tundra

© 2003 NRC Canada


Dependent variable* Pr2

(adj.)Beta(1)

Beta(2)

Beta(3) Intercept

SEestimate

SEintercept n F

(a) E The variable did not enter the model at all(b) %BF <0.0001 0.21 0.466 31.20 31.59 6.50 86 23.27(c) E (1) and %BF (2) <0.0113 0.45 0.639 0.856 –28.74 26.29 11.10 86 35.94(d) E (1), %BF (2),

and E × %BF (3)<0.0001 0.57 0.669 0.971 –0.16 –51.46 21.20 0.11 2 879 853 1 287 967

*E, elevation; %BF, percentage of boreal forest. The letters refer to the model tested, and the numbers after the variables refer to the beta valuenumbers.

Table 2. Results from the forwards stepwise multiple regression analysis of the longitudinal transects.

Region PhysiographyMeanelevation

TotalLSD Tundra Shrub Forest

White Mountains Mountains 647 8.0 8.8 8.6 8.8Ray Mountains Mountains 365 5.0 5.5 4.4 5.6Yukon Uplands Mountains 443 5.0 3.9 7.3 5.4Kantishna River Flats Flats 277 4.4 6.0 3.3 4.8Tetlin Flats Flats 666 3.8 0.6 0.3 4.6Koyukuk Flats Flats 105 3.5 2.8 5.2 3.7Yukon Flats Flats 155 3.1 0.3 0.1 8.0Tanana Flats Flats 325 2.9 0.1 3.4 4.8Kuskokwim Mountains Mountains 293 2.8 1.3 2.1 5.1Nowitna Flats Flats 155 2.7 0.5 4.0 5.5Selawik Flats Flats 123 1.6 0.8 0.0 1.4Yukon Delta Flats 48 1.1 1.2 1.2 0.9Stony Flats Flats 136 1.0 1.3 1.0 1.5Purcell Mountains Mountains 298 0.9 1.3 0.0 2.0Innoko Flats Flats 100 0.6 0.6 0.6 0.5Seward Peninsula Mountains and

(or) flats208 0.5 0.4 0.6 0.7

Nulato Hills Mountains 277 0.3 0.1 0.5 0.4

Table 3. Main physiographic property, lightning strike density (LSD) (strikes/km2), totalfor the subregion, and separately for the three vegetation types, and mean elevation (m) forthe 17 physiographic subregions.

Dependent variable* Pr2

(adj.)Beta(1)

Beta(2)

Beta(3) Intercept

SEestimate

SEintercept n F

(a) E <0.0001 0.012 0.135 38.00 30.60 4.29 161 2.94(b) %BF <0.0001 0.18 0.428 22.00 27.91 4.30 161 35.63(c) E (1) and %BF (2) 1 × 10–6 0.27 0.334 0.549 0.67 26.91 6.15 161 30.7(d) E (1), %BF (2),

and E × %BF (3)<0.0001 0.12 0.846 0.799 –0.84 –34.69 28.62 0.11 4 893 601 230 840

*E, elevation; %BF, percentage of boreal forest. The letters refer to the model tested, and the numbers after the variables refer to the beta valuenumbers.

Table 4. Results from the forwards stepwise multiple regression analysis of the physiographic regions.



© 2003 NRC Canada

780 Can. J. For. Res. Vol. 33, 2003

and as boreal forest, fall within the 10- to 15-km range ofthe edges of the vegetation patches. However, a large varia-tion in lightning strike density exists within this zone(Figs. 5 and 6). It is beyond the scope of this paper to exam-ine the meteorological conditions associated with these re-sults, although this information would provide someimportant understanding of the processes taking place. Spec-ulations based on the studies of Vidale et al. (1997),Knowles (1993), and Chambers (1998) would suggest thatthe orientation of a mesoscale circulation across a borealforest – tundra boundary as the one examined here would befrom the cool surface air towards the warmer surface air.The effects of these circulations would be strongest aroundthe edges of a developed circulation cell, which is dependenton the size of the forest patches, but likely within 10 timesthe convective boundary layer depth (~2 km), so 5–20 km,approximately. This coincides with the distance withinwhich we see the higher lightning strike densities on Figs. 5and 6. However, this phenomenon is observed on both sidesof the vegetation boundaries, whereas, according to abovereasoning, the effects would be expectable on the forest sideof the edge. Wind patterns could influence the location ofthese effects.

Knowles’ (1993) modeling study suggested that under fa-vorable meteorological conditions wildfire burn scars werecapable of enhancing convective activity of sufficientstrength to produce lightning. Combining the conclusions ofKnowles (1993) and Pielke and Vidale (1995), perhaps theboreal forest, through relatively high sensible heat fluxes,can enhance convective activity if an already unstable atmo-spheric environment is present. Beringer et al. (2001) sug-gest that although surface heating differences due to energypartitioning between the boreal forest and tundra are un-likely determinants of the summer position of the Arcticfront, they still may be sufficient to support local-scale cir-culations.

If boreal forest is capable of enhancing thunderstorm de-velopment then a feedback loop that creates more firesmight exist. An increased frequency of wildfires, despite acooler, moister climate in the boreal forest ecosystem fol-lowing the appearance of black spruce (Hu et al. 1993;Lynch et al. 2003) could reflect an increased availability offlammable fuels. The increased fire frequency could also bedue to the energy partitioning in a black spruce forest, re-sulting in higher sensible heat fluxes than over other veg-etation types, thus providing triggering mechanisms forthunderstorm development (Chambers and Chapin 2003).

Such a feedback is less likely to develop in the atmo-spherically more stable (e.g., maritime) environment, be-cause stable conditions would suppress the circulationcenters induced by the heating differences between the wild-fire burn scars and the surrounding unburned surface, thusresulting in fewer thunderstorms than the unstable environ-ment (Knowles 1993). This explains our finding of a weakercorrelation between boreal forest and lightning strike densityin the maritime western longitudinal transects. The maritimeclimate is also cooler, providing much less heating differ-ences to induce circulations.

ConclusionsWe examined the interrelationships among vegetation, ele-

vation, and lightning strike density for a 14-year time seriesat three spatial scales.

Regional scale (interior Alaska)(1) Lightning strike density was consistently higher in bo-

real forest than in tundra or shrub vegetation.(2) The interaction between increasing boreal forest area

and elevation is correlated with increasing lightningstrike density, providing an explanation for 66% of thevariation in the lightning strike densities at this scale.

(3) Most of the 10 × 10 km pixels with the highest lightningstrike densities fall within 10–25 km on either side ofthe boundary between boreal forest and tundra.

Mesoscale (longitudinal transects)(1) Consistent with the observed trend for the regional

scale, the mesoscale showed higher lightning strike den-sities within the boreal forest, followed by shrubs andtundra. The effect is more pronounced in the continentalthan in the maritime climate.

(2) The combination of boreal forest areal coverage, eleva-tion, and a built-in continentality factor provides an ex-planation for ~45% of the variation in the lightningstrike density. The boreal forest areal coverage carriesthe highest weight in the model.

Local scale (physiographic subregions)(1) More than half of the subregions showed the highest

lightning strike densities within boreal forest.(2) Elevation and percentage of boreal forest do not provide

as good an explanation for the variability in lightningstrike data as at the mesoscale (~27%). However, thebest model was still the combined model of elevationand boreal forest areal coverage.

Overall interpretation and suggestions for future workBoreal forest, because of its high sensible heat fluxes,

may enhance convective activity of air-mass thunderstorms.Another explanation is that the boreal forest biome exists ina climatic region that sustains the convective activity. Ourdata suggest that both explanations are important. Our statis-tical analysis indicates that it is the interaction or combina-tion of elevation, boreal forest, and a climatic (continental)factor, that influence the variation in lightning strike density.This has strong implications for future climate change sce-narios, where lots of speculation about extension of the bo-real forest and tree line have been made.

The combined effect of boreal forest and elevation on in-creased lightning strike activity was found in the largerscales (regional and mesoscale), but was less prevalent at thesmallest scale (local). The zones at the local scale that wereexamined were delineated based on physiographic featuressuch as proximity to major rivers, or mountainous areas.Possibly not all of these delineations were appropriate forshowing the above effects, thus we only saw trends similarto the larger scales at some zones, whereas the statisticalanalysis, incorporating all the zones, did not show as stronga pattern. Likewise, the proximity study did not portray aconsistent pattern at the patch scale (regardless of size).

Several factors have not been included in this study: patchsizes (and shapes), effects of wildfire burn scars, topography,and aspect and meteorology. All of these could be likely



candidates for explaining some of the results from thisstudy.

AcknowledgmentsThis research was funded by the National Science Foun-

dation Bonanza Creek Long Term Ecological Research Pro-ject. We thank Scott Chambers, Terry Chapin, Merav Ben-David, John Fox, Carolyn Kremers, Peter Olsson, T. ScottRupp, David Valentine, and the anonymous reviewers forhelping us improve the manuscript. We also thank ThorWeatherby, Alaska Fire Service, for providing informationand data from the lightning strike detection network.

ReferencesBiswas, A.K. 1976. Climatology of thunderstorms in Alaska. M.S.

thesis, University of Alaska Fairbanks, Alaska.Biswas, A.K., and Jayaweera, K.O.L.F. 1976. NOAA-3 satellite

observations of thunderstorms in Alaska. Monthly Weather Rev.104: 292–297.

Beringer, J., Tapper, N.J., McHugh, I., Chapin, F.S., III, Lynch,A.H., Serreze, M.C., and Slater, A. 2001. Impact of Arctictreeline on synoptic climate. Geophys. Res. Lett. 28: 4247–4250.

Boles, S.H., and Verbyla, D.L. 2000. Comparison of threeAVHRR-based fire detection algorithms for interior Alaska. Re-mote Sens. Environ. 72: 1–16.

Bonan, G.B., Chapin, F.S., and Thompson, S.L. 1995. Boreal forestand tundra ecosystems as components of the climate system.Climatic Change, 29: 145–167.

Bourgeau-Chavez, L.L., Alexander, M.E., Stocks, B.J., andKasischke, E.S. 2000. Distribution of forest ecosystems and therole of fire in the North American boreal region. In Fire, climatechange, and carbon cycling in the boreal forest. Edited by E.S.Kasischke and B.J. Stocks. Springer Verlag, New York. Ecol.Stud. 138. pp. 111–131.

Bryson, R.A. 1966. Air masses, streamlines, and the boreal forest.Geogr. Bull. 8: 228–269.

Chambers, S.D. 1998. Short- and long-term effects of clearing na-tive vegetation for agricultural purposes. Ph.D. thesis, FlindersInstitute for Atmospheric and Marine Sciences, The FlindersUniversity of South Australia, Adelaide, Australia.

Chambers, S.D., and Chapin, F.S., III. 2003. Fire effects on sur-face-atmosphere energy exchange in Alaskan black spruce eco-systems: implications for feedbacks to regional climate. J.Geophys. Res. 108D(1): 8145.

Chapin, F.S., III, Eugster, W., McFadden, J.P., Lynch, A.H., andWalker, D.A. 2000. Summer differences among Arctic ecosys-tems in regional climate forcing. J. Climate, 13: 2002–2010.

Court, A., and Griffiths, J.F. 1992. Thunderstorm climatology. InThunderstorm morphology and dynamics. Edited by E. Kessler.University of Oklahoma Press, Norman, Okla. pp. 9–40.

Cummins, K.L., Murphy, M.J., Bardo, E.A., Hiscox, W.L., Pyle,R.B., and Pifer, A.E. 1998. A combined TOA/MDF technologyupgrade of the U.S. National Lightning Detection Network. J.Geophys. Res. 103D: 9035–9044.

Eugster, W., Rouse, W.R., Pielke, R.A., McFadden, J.P., Baldocchi,D.D., Kittel, T.G.F., Chapin, F.S., III, Liston, G.L., Vidale, P.L.,Vaganov, E., and Chambers, S.D. 2000. Land–atmosphere energyexchange in Arctic tundra and boreal forest: available data andfeedbacks to climate. Global Change Biol. 6(Suppl. 1): 84–115.

Foley, J.A., Kutzbach, J.E., Coe, M.T., and Levis, S. 1994. Feed-backs between climate and boreal forests during the Holoceneepoch. Nature (London), 371: 52–54.

Gilbert, D.E., Johnson, B.R., and Zala, C. 1987. A reliability studyof the lightning locating network in British Columbia. Can. J.For. Res. 17: 1060–1065.

Hammond, T., and Yarie, J. 1996. Spatial prediction of climaticstate factor regions in Alaska. Écoscience, 3: 490–501.

Henry, D.M. 1978. Fire occurrence using 500mb map correlation.NOAA technical memorandum, NWS AR-21. NWS, Anchor-age, Alaska.

Hiscox, W.L., Krider, E.P., Pifer, A.E., and Uman, M.A. 1984. Asystematic method for identifying and correcting “site errors” ina network of magnetic directions finders. International Aero-space and Ground Conference on Lightning and Static Electric-ity, 26–28 June 1984, Orlando, Fla. National Inter-AgencyCoordination Group, Orlando, Fla.

Hu, F.S., Brubaker, L.B., and Anderson, P.E. 1993. A 12 000 yearrecord of vegetation change and soil development from WienLake, central Alaska. Can. J. Bot. 71: 1133–1142.

Huschke, R.E. 1959. Glossary of meteorology. American Meteoro-logical Society, Boston, Mass.

Idone, V.P., Davis, D.A., Moore, P.K., Wang, Y., Henderson, R.W.,Ries, M., and Jamason, P.F. 1998a. Performance evaluation of theU.S. National Lightning Detection Network in eastern New York.1. Detection efficiency. J. Geophys. Res. 103D: 9045–9055.

Idone, V.P., Davis, D.A., Moore, P.K., Wang, Y., Henderson, R.W.,Ries, M., and Jamason, P.F. 1998b. Performance evaluation ofthe U.S. National Lightning Detection Network in eastern NewYork. 2. Location accuracy. J. Geophys. Res. 103D: 9057–9069.

Kasischke, F.S., and Stocks, B.J. (Editors). 1999. Fire, climatechange, and carbon cycling in the boreal forest. Springer Verlag,New York.

Knowles, Captain, J.B. 1993. The influence of forest fire inducedalbedo differences on the generation of mesoscale circulations.M.S. thesis, Department of Atmospheric Science, ColoradoState University, Fort Collins, Colo.

Lafleur, P.M., and Rouse, W.R. 1995. Energy partitioning attreeline forest and tundra sites and its sensitivity to climatechange. Atmos.-Ocean, 33: 121–133.

Lopez, R.E., and Holle, R.L. 1986. Diurnal and spatial variabilityof lightning activity in Northeastern Colorado and CentralFlorida during the summer. Monthly Weather Rev. 114: 1288–1312.

Lynch, A.H., Slater, A.G., and Serreze, M. 2001. The Alaskanfrontal zone: forcing by orography, coastal contrast and the bo-real forest. J. Climate, 14: 4351–4362.

Lynch, J.A., Clark, J.S., Bigelow, N.H., Edwards, M.E., and Finney,B.P. 2003. Geographic and temporal variation in fire history inboreal ecosystems of Alaska. J. Geophys. Res. 108D(1): 8152.

Markon, C.J., Fleming, M.D., and Binnian, E.F. 1995. Characteris-tics of vegetation phenology over the Alaskan landscape usingAVHRR time series data. Polar Rec. 31: 179–191.

Nash, C.H., and Johnson, E.A. 1996. Synoptic climatology oflightning-caused forest fires in subalpine and boreal forests.Can. J. For. Res. 26: 1859–1874.

Oke, T.R. 1987. Boundary layer climates. 2nd ed. Routledge Jour-nals, London.

O’Neal, M. 1996. Interactions between land cover and convectivecloud cover over Midwestern North America detected fromGOES satellite data. Int. J. Remote Sens. 17: 1149–1181.

Orville, R.E., and Silver, A.C. 1997. Lightning ground flash den-sity in the contiguous United States: 1992–95. Monthly WeatherRev. 125: 631–638.

Pielke, R.A., and Vidale, P.L. 1995. The Boreal forest and the Po-lar front. J. Geophys. Res. 100D: 25 755 – 25 758.

© 2003 NRC Canada




© 2003 NRC Canada

782 Can. J. For. Res. Vol. 33, 2003

Rabin, R.M., Stadler, S., Wetzel, P.J., Stensrud, D.J., and Gregory,M. 1990. Observed effects of landscape variability on convec-tive clouds. Bull. Am. Meteorol. Assoc. 71: 272–280.

Reap, R.M. 1991. Climatological Characteristics and objective pre-diction of thunderstorms in Alaska. Weather Forecasting, 6:309–319.

Schroeder, M.J., and Buck, C.C. 1970. Fire weather. AgriculturalHandbook 360, U.S. Department of Agriculture, Forest Service,Washington, D.C.

Serreze, M.C., Lynch, A.H., and Clark, M.P. 2001. The ArcticFrontol Zone as seen in the NCEP-NCAR Reanalysis. J. Cli-mate, 14: 1550–1567.

Stull, R.B. 1997. An introduction to boundary layer meteorology.Kluwer Academic Publishers, Dordrecht, Netherlands.

Sullivan, W.G. 1963. Low-level convergence and thunderstorms inAlaska. Monthly Weather Rev. 91: 89–92.

Van Cleve, K., Chapin, F.S., III, Dyrness, C.T., and Vierick, L.A.1991. Element cycling in taiga forests: state-factor control. Bio-science, 41: 78–88.

Vidale, P.L., Pielke, R.A., Sr., Steyaert, L.T., and Barr, A. 1997:Case study modeling of turbulence and mesoscale fluxes overthe BOREAS region. J. Geophys. Res. 102D: 29 167 – 29 188.

Viereck, L.A., Dyrness, C.T., Batten, A.R., and Wenzlick, K.J.1992. The Alaskan vegetation classification. USDA For. Serv.Gen. Tech. Rep. PNW-GTR-286.

Wells, M.L., and McKinsey, D.E. 1993. The spatial and temporaldistribution of lightning strikes in San Diego County, California.GIS/LIS Proceedings, 2–4 Nov. 1993, Minneapolis, Minn.American Society for Photogrammetry and Remote Sensing,Bethesda, Md.



spatial patterns of lightning strikes in interior alaska ... · spatial patterns of lightning...

Documents