A Comparison Of Five Sampling Techniques To Estimate Surface Fuel .

Comparison of five surface fuel sampling methodsInt. J. Wildland Fire365(a)(b)Fig. 1. Range of fuel loads examined with five sampling techniques in the present study. (a) Site C1 with mostlyfine fuels, herbs, and grass; and (b) site M5 with abundant logs.five sampling techniques that were tested. Sampling occurredin the following order on each macroplot: (1) planar transects,(2) photoload microplots, (3) photoload macroplot, (4) photoseries, (5) fixed-area plot, and (6) reference clipped sample.Although it is difficult to design a field-sampling plot structurethat provides an objective and fair comparison of the samplingmethods without any site or procedural bias, the protocols thatfollow were a compromise that accommodated all the types of

366Int. J. Wildland FireP. G. Sikkink and R. E. KeanePlotPlanar intersect lines (50 m)Photoload FWD plots( microplot/reference)1 1mNSubplot(10 10 m)Microplots(1 1 m)50 m50 mFig. 2. Sample layout of the macroplot divided into subplots withmicroplots placed in the north-east corner of each subplot.sampling for the downed woody debris and herbaceous materialinvestigated in the current study.Sampling techniques for the reference sampleCreating the perfect reference sample design that captured actualloadings by the six components for each sample site was logistically impractical because we did not have the resources to clip,collect, and weigh all the herb, shrub, and woody fuels within the2500-m2 plot and we could not handle the large volume of heavyand unwieldy log material in our laboratory. Therefore, we subsampled four woody size classes and ground cover componentsusing nested microplots (Fig. 2). In the north-east corner of eachsubplot, we established a 1 1-m microplot using a plot framemade out of 1.9 cm of plastic PVC pipe (Fig. 2). Within thetwenty-five 1 1-m microplots, we collected all of the FWDand clipped and collected all of the aboveground living and deadshrub and herbaceous material. Because this method of samplingwas destructive, it was done only after data collection for all othersampling methods was completed. Only material that fell withinmicroplot boundaries was collected. If it extended beyond theboundary, it was cut off and the in-plot portion collected. Wesorted shrub, herb, and FWD by size class into labelled paperbags in the field, brought all samples back to the laboratory tobe oven-dried for 3 days at 90 C, and finally weighed each tothe nearest milligram. The average of the 25 microplot samplesby size class constituted the loading estimates for FWD, shrub,and herbaceous material in each plot.For the 1000-h fuels, or CWD, we measured the small-enddiameter, large-end diameter, and length of each piece of debrisgreater than 7.62 cm that fell within each of the 25 subplots to geta 100% inventory of all logs on each site. If a log extended beyonda subplot boundary, the ends were measured at the boundary tocalculate only the in-plot portion. We also assigned a decay class(i.e. classes 1 to 5) to each log using FIREMON guidelines (Luteset al. 2006).NFixed-area plots (1 50 m)Fig. 3. Sample design for fixed area, planar-intersect, and photoloadmethods within each site. Fixed-area strip plots were established along thenorthern subplot edge using a width of 1 m. Planar intersect transects were2 m apart in the north–south and east–west directions. Photoload fine woodydebris (FWD) was assessed in the same microplots that were used to collectthe reference fuel loads.Sampling techniques for other tested methodsFive strip plots (1 50 m) were established at each site to assessthe fixed-area plot technique (see Fig. 3 for placement). TheFWD within each strip was sampled by measuring the lengthof each fuel particle (cm) by 1-h, 10-h, and 100-h diameter sizeclasses within each subplot. The total lengths of debris for allsubplots were then summed by size-class to get totals for theentire macroplot. The five strips effectively sampled 10% of thearea at each site. Some portions of strip plots had such heavyfuel accumulations (i.e. the fine woody fuels were greater than100 particles m 1 ) that it was impractical to measure every fuelparticle in each diameter size class; therefore, we counted allfuel particles and multiplied by an estimated average length foreach particle in each subplot strip to get total length to use in theloading calculations. Lengths for the 100-h fuels and dimensionsof all CWD (small- and large-end diameters and length) werealways measured, never estimated, within the strip.The sampling design for the planar-intersect technique used52 line transects that were each 50 m long to estimate fuel loadings within the base plot. The beginnings and ends of each ofthese transects were located systematically at 2 m intervals alongthe outside edges of the plot (see Fig. 3). We tried to minimisebias from systematic sampling by taking the 1-h, 10-h, and 100-hsamples in different 10-m sections along each of the 52 lines. Thesampling plane was 2 m high with the bottom located at the baseof the litter layer. For the 1-h and 10-h fuels, we assessed loadingby counting the number of intersections crossing the samplingplane in 5-m sections along each of the 52 transects. For the100-h fuels, we counted the number of intersections along 10 m

Comparison of five surface fuel sampling methodsof plane length and then summed all subplot values to get plottotals. For the 1000-h fuels, we recorded the diameter and rotclass of each log at the point where it intercepted the samplingplane for the entire transect length (50 m). To keep samplingprotocols consistent among sites, all planar-intersect samplingwas guided by procedures detailed in FIREMON (Lutes et al.2006). Shrub and herbaceous sampling were not applicable toplanar-intersect techniques.The development and evaluation of the photoload samplingmethod are discussed in detail in Keane and Dickinson (2007a).We invited 29 participants to visually estimate loadings ofour six fuel classes using the photoload technique. Estimateswere made within the same 1 1-m microplot that was usedfor the reference fuel sampling. Each participant was askedto match the fine-fuel, shrub and herb loading conditions thathe or she observed within each of the 25 microplots to conditions portrayed in a set of downward-looking photographs offuelbeds showing graduated picture sequences of increasing load(Fig. 4a). For the 1000-h fuels, each participant was asked toestimate load at the subplot (10 10 m2 ) scale, instead of themicroplot, because the subplot best matched the scale of thephotoload log pictures (Fig. 4b). We recorded the total time ittook each participant to complete their photoload estimate of allsix fuel components on all 25 microplots at each site. This timewas used as a measure of efficiency for the technique. The participants varied in fuel sampling experience from those with littleor no prior experience measuring fuel loads to those with extensive experience in all phases of fuel sampling. Each was givena 2-h training session in applying the photoload technique andin using the photoload pictures to estimate both fine and coarsefuels. There was not an established crew that worked all five sitestogether nor were the numbers of participants measuring loadsconstant from site to site.In addition to applying the photoload technique on themicroplots, as described above, the same participants were askedto estimate fuel loading within the entire macroplot (50 50 m)using two related visual-assessment methods. First, each participant used the photoload sequences to estimate one loadingvalue for each woody size class using a general walk-through ofthe macroplot, which will be referred to in the current paper asphotoload macroplot estimates. Next, they were asked to estimate loadings using the Fischer photo series (Fischer 1981a),which was specifically created for estimating downed woodydebris in western Montana forests. Participants walked the entiremacroplot and tried to determine which of the oblique photos most closely matched the observed downed woody debrisconditions. Loadings were assigned to each fuel componentusing summaries presented by Fischer (1981a) for each selectedphoto. The photo series technique did not assess shrub and herbloadings.Calculating loadings for comparative analysisReference plotsWe standardised the reference loads for 1-h, 10-h, 100-h, liveand dead herb, and live and dead shrub from each microplot tothe base plot by summing the weights from the laboratory analysis of samples from each microplot by size class and dividingthe total weights by total microplot area. For the 1000-h debrisInt. J. Wildland Fire367class, the weight of each log greater than 7.62 cm in diameterwas computed from its volume and its density. Volumes werecalculated as follows:V l[(as al ) (as al )]3where as is the cross-sectional area (m2 ) of the small end ofthe log, al is the cross-sectional area of the large end (m2 ),and l is the length (m) (Keane and Dickinson 2007a). We used400 kg m 3 for the density of sound logs in decay classes 1, 2,and 3, and 300 kg m 3 for the density of rotten logs in decayclasses 4 and 5 because Brown (1974) suggested these densitiesfor coniferous forests based on experimental work in wood specific gravity. The 1000-h loadings were also standardised to thebase plot by summing the individual log weights and dividing thetotal weight by the total plot area (2500 m2 ). These log loadingsand the total loadings for fine woody debris, shrubs, and herbsfrom the laboratory analyses were combined to represent our‘actual’ loadings for each site. The values became our referencedataset to evaluate the performance of all other tested methods onthe plot.Choosing an appropriate wood density value was an important decision for calculating the reference loading values andthe load values for the other methods tested in the present study.Many of the traditional methods for measuring load assumed thatthe density of fuel (kg m 3 ) was constant across all size classesand species but different across various classes of decay (Brown1974). However, research has shown that there are significantdifferences in fuel wood density between different species, rotclasses, and size classes (van Wagtendonk et al. 1996). Eventhough we measured site-specific wood densities for each of oursites (Keane and Dickinson 2007a), we did not have the properequipment (i.e. a Kraus Jolly specific gravity balance) to getreliable density estimates for the FWD components. Therefore,we decided to use Brown’s (1974) density values in all load calculations, which allowed us to focus on results that were dueto differences among methods, rather than due to differences indensity values for each technique.Other sampling methodsWeights of the 1-h, 10-h and 100-h woody fuel particles foreach 50 1-m strip plot were calculated using the same volume–density procedure described above. The weights were summedacross all twenty-five 10-m sections and then divided by totalstrip plot area (i.e. (50 1-m strip) 5 strips 250 m2 ) to getloading values (kg m 2 ) for the entire macroplot.The diameter ofthe fine woody fuel particles was the midpoint of each size class.The large- and small-end diameters were considered equal. Thelength was the total measured length of debris in each size class.We followed the procedures detailed in Brown (1971, 1974)to calculate downed woody fuel loadings for the planar-intersectmethod. For most FWD calculations, we chose the averagequadratic mean diameter (non-slash) values based on the dominant overstorey tree at the site (see table 2 of Brown 1974). ForFWD on site K4, we used the composite value for mixed-speciesoverstorey. We also used Brown’s (1974) density values for eachsize class.

368Int. J. Wildland Fire0.01 kg m 2 (0.05 tons acre 1)0.03 kg m 2 (0.13 tons acre 1)0.05 kg m 2 (0.23 tons acre 1)0.07 kg m 2 (0.32 tons acre 1)0.10 kg m 2 (0.45 tons acre 1)0.30 kg m 2 (1.35 tons acre 1)0.50 kg m 2 (2.25 tons acre 1)0.70 kg m 2 (3.15 tons acre 1)1.00 kg m 2 (4.50 tons acre 1)2.00 kg m 2 (9.00 tons acre 1)3.00 kg m 2 (13.50 tons acre 1)1m1m1m(a)P. G. Sikkink and R. E. KeaneFuel type:1h1mDiameter reference1 h, 0–7.62 10 2 m100 h, 0.30–0.91 m3 10cm10 h, 7.62 10 2–0.30 m8641000 h, 0.91 m 1m1mFig. 4. Examples of photoload sequences for (a) microplot estimation of 1-h fuels and (b) subplot estimations of 1000-h fuels (reprinted from Keane andDickinson 2007b).

Comparison of five surface fuel sampling methods(b)Fuel type: 1000 hSpecies: Pseudotsuga menziesii (Douglas-fir)imitation Diameter: 15.20 cm (6.00 in)0.10 kg m 2 (0.45 tons acre 1)Total log length: 1.22 m (4 ft)0.40 kg m 2 (1.80 tons acre 1)Total log length: 5.49 m (18 ft)0.80 kg m 2 (3.60 tons acre 1)Total log length: 10.67 m (35 ft)1.00 kg m 2 (4.50 tons acre 1)Total log length: 13.41 m (44 ft)1.50 kg m 2 (6.75 tons acre 1)Total log length: 20.12 m (66 ft)Fig. 4.(Continued)Loading values for all of the photo-based techniques were calculated in similar ways. Load estimates made at each microplotwere averaged for the photoload method using estimates madeby all participants at each site. Site totals were obtained by summing the average values for each microplot by fuel class. Rangewas also computed to show the variation in loading estimatesat each site. Estimates by all participants at each site were alsoaveraged to obtain loading values for each photoload macroplot.For the photo series method, loadings were assigned to eachInt. J. Wildland Fire369component based on each participant’s photo choice and thenaveraged by site.Statistical comparisonsStatistical comparisons in the present study needed to accountfor two major issues: (1) the different sampling scales used foreach method; and (2) the non-normal distribution of collecteddata for most fuel classes. To address the differences in samplingscales used for each method, the measured loadings from thereference sample and estimated loadings from the five samplingtechniques were all standardised to the macroplot level by siteas described in the previous section. To address the non-normaldistribution of debris, we used two different procedures. Thefirst procedure was used only to test (i) how expertise affectedestimates in the photo-based methods and (ii) how a samplemethod affected estimates in individual fuel classes (e.g. 1-hfuels) without separating the fuels by site. We tested each fuelclass for normal distribution and homogeneity of variance usingQ-Q normal plots and Levene’s tests (Levene 1960). Data weretransformed to natural log values in all fuel classes except 10-hfuels to comply with parametric assumptions of the analysis ofvariance (ANOVA) and Tukey’s B tests. Log-transformations ofthe 10-h fuel loadings only increased the lack of homogeneity,so we used the raw data for these comparisons. We used a secondprocedure for tests on how the quantity of fuel load at each siteaffected the estimates obtained using each sampling method. Thedifference between a site’s reference sample and a method’s estimate were computed from actual load data for each of the six fueltypes and for total FWD and CWD. The 1-h, 10-h and 100-h fuelswere grouped as FWD. The 1000-h fuels constituted the CWD.We calculated a mean error and standard deviation for the FWD,CWD, shrubs, and herbs from the difference values. We calculated between- and within-methods standard deviations for theoverall project using a one-way analysis of variance (ANOVA)with site as the analysis factor. All statistical comparisons in thecurrent study were considered significant if P was 0.05.Determining which method might be appropriate for differentsampling applications was most easily evaluated using a rating scale. Ratings were assigned to each method based on howclosely its loading values matched the reference sample. Ratingsranging from 1 to 5 were given for total FWD, the 1000-h fuelclass, and the total loading on site. Shrub and herb loadings werenot included individually in the rankings because their loadingswere sampled in only three of the five methods. Loadings that differed least from the reference sample were given a ‘1’, sites thatwere second-best were given a ‘2’, and sites that differed mostwere assigned ‘5’. To determine overall ‘best’performance of themethods for each fuel class, a rank total was obtained by summing individual rankings over all sites. Low rank totals indicatethat the method was consistently close to the results obtained forthe reference sample for the respective fuel class.The time needed to complete each technique was directlycompared among methods. The two most time-consuming methods, namely the fixed-area and planar intersect, were alsoevaluated using bootstrapped samples to determine if puttingin fewer lines might improve efficiency without sacrificing sampling error. Bootstrapping is a statistical way to increase samplesize that randomly selects data values from the original dataset

370Int. J. Wildland FireLoading (kg m 2)4P. G. Sikkink and R. E. Keane1h10 h100 h1000 hHerbShrub31000 h21000 h11000 h1000 hSite C1Site C21000 h0Fig. 5.Site S3Site K4Site M5Total fuel loading by site (reference samples only).and creates a new set of observations of specified size. Weused standard, with-replacement bootstrapping techniques inS-Plus (Insightful Corporation 2003) to create 2000 bootstrapobservations for a range of sample sizes at each site. For theplanar-intersect method, sample size ranged from 2 to 52 lines(i.e. distances of 20 to 2600 m using a horizontal plane). For thefixed-area plots, sample sizes ranged from 1 to 25, so sampleareas increased in 10-m2 increments from 10 to 250 m2 . For eachfuel class, we calculated the mean loading for the 2000 bootstrapobservations at each sample size, and tested how the variance ofthe means changed with increased sample size (see Jalonen et al.1998 for details). We considered the recommended sample sizefor each 50 50-m macroplot to be the point where differencein variance was minimal compared with the effort needed toadd additional samples (i.e. usually the inflection point foundin the graphs). Although we did a bootstrap analysis for eachfuel size class, only the results of the 1-h and 1000-h will bepresented here.Resu

Comparison of five surface fuel sampling methods Int. J. Wildland Fire 365 (a)(b)Fig. 1. Range of fuel loads examined with five sampling techniques in the present study. (a) Site C1 with mostlyfine fuels, herbs, and grass; and (b) site M5 with abundant logs.five sampling techniques that were tested.