Determining Goniferous Forest Gover andForest Fragmentation with NOAA'9

Advanced Very High ResolutionRadiometer Data

William l. Ripple

AbstractNoAA-g satellite data from the Advanced Very High Resolu-tion Radiometer (AVHtR) were used in conjunction withLandsat Multispectral Scanner (MSS) data to determine theproportion of closed canopy conifer forest cover in the Cas-cade Range of Oregon. A closed canopy conifer map, as de-termined from the MSS, was registered with AVHRI pixels.Regression was used to relate closed canopy conifer forestcover to AvHRR spectral data. A two-variable (band) rcgrcs-sion model accounted for morc variance in conifer coverthan the Normalized Difference Vegetation Index (Now). Thespectral signatures of various conifer successional stageswere qlso examined. A map of Oregon was produced show-ing the proportion of closed canopy conifer cover for each,wuan pixel. The Awran was responsive to both the percent-age of closed canopy conifer cover and the successionalstage in these temperate coniferous foresfs tn thr's experiment.

lntroductionOn 7 October 1989 the United States Congress approved leg-islation requiring that federal agencies preserve contiguousstands of old-growth forests in the Pacific Northwest ratherthan harvest in a patchwork pattern resulting in forest frag-mentation (Lehmkuhl and Ruggiero, 1991; Lord and Norton,1990). This legislation was intended to protect the habitat ofthe northern spotted owl which needs large stands of old-growth forest to survive extinction (Ripple et o1., 1991b).With this legislation in place, developing methods for mea-suring the extent of forest cover and fragmentation at theIandscape and regional level is important. Forest fragmenta-tion has been documented at the landscape scale using ageographic information system to describe patch size, shape,abundance, spacing, and forest matrix characteristics (Rippleet o/., 1991aJ. Spatial data on forest fragmentation at a re-gional scale is typically not available.-

The main objective of the research reported here was todescribe the relationship between the proportion of conifer-ous closed canopy forest cover and avgnn pixel radiancevalues. A second objective was to obtain spectral signaturesfrom various forest successional stages using AWIRn data.With a ground resolution of approximately 1.1 km, avnm

data can not detect individual disturbances such as clearcutsthat average between 10 and 20 ha on public lands' How-ever, it wai hypothesized that radiance as measured by theAVHRR would-have an i.nverse linear relationship with theproportion of coniferous closed canopy-forest cover,-

Apparently, few studies have been designed to determinecorrer of temperate coniferous forests with AVHRR data. Nelson(1989) attempted to use Global Area Coverage (GAc) AvHRRdata with a 4-tm resolution to estimate forest area for the en-tire United States. His results showed that GAC and MSs werenot highly corelated. Loveland et ol. (1991) included forests(conifJrous and deciduous) in the development of their land-cover database for the conterminous United States. Their re-mote sensing estimates of forest cover were compared to otherinventory approaches by Turner ef a/' (in press).

avnnn-satellite data have been used for regional esti-mates of forest cover in the tropical forests of South Americaand the eastern and southern hardwood forests of the UnitedStates. Nelson and Holben (19s6) and Woodwell et al. (7987)used At'ttRR data in conjunction with calibrations from thefiner resolution Landsat Multispectral Scanner (MSS) data todiscriminate cleared areas from primary forest in Brazil. Iver-son ef o1. (19891 and Zhu and Evans (1992) used regressioneouations based on estimates of forest cover from LandsatThematic Mapper (rrr.l) data to determine the relationship be-tween AVHRR-data and forest cover. Townshend and Tucker(1984) found that AwIRR data represented 70 percent of vssdata variation during a land-cover mapping study. The nor-malized difference vegetation index (wovI) [(near infrared -

visible)/(near infrared-+ visible)l explained 70 percent and79 percent of the variation in leaf area index in coniferousforest stands in western United States (Spanner et al.,1990a). The authors concluded that reflectance was related tothe proportion of surface cover types within AVHRR pixels aswell as changes in leaf area index.

Study AreaThe siudy area was in the western Cascade mountains of Or-egon andencompassed both public USDA Forest Service land

Environmental Remote Sensing Applications Laboratory (ER-SAL), Department of Forest Resources, Oregon State Univer-sity, Corvall is, OR 97331.

Photogrammetric Engineering & Remote Sensing,Vol. 60, No. 5, May 1994, pp. 533-540'

oo99-1 1 1 2/94/6001-533$o3.oo/oO1994 American Society for Photogrammetry

and Remote Sensing

on the Willamette National Forest and private land (Figure1). The study area size was approximately 258,930 hectares,with elevations ranging from approximately 240 mIo 1700 mabove mean sea level. This area falls within portions of thewestern hemlock (Tsuga heterophylla) and Picific silver fir(Abies amabilis) vegetation zones with the major tree speciesconsisting of Douglas-fit (Pseudotsuga menziesii), westernhemlock, Pacific silver fir, noble fir (Abies procero), andwestern red cedar (Thuja plicato) (Franklin and Dyrness,1973). The topography is highly dissected by steep slopeswith slope gradients ranging between approximately 0 and60 degrees. The maritime climate has wet-mild winters andwarm-dry summers. The landscape consists of Iarge areas ofold-growth Douglas-fir/western hemlock forests over 400years old. Major disturbances include both fire and logging,resulting in a patchwork mosaic of herbaceous areas, decidu-ous shrubs and trees, and both natural and man-made closedcanopy conifer forests. Areas of big Ieaf maple (Acer macro-phyllum) and red aider [r{1nus rugosa) are also found in thestudy area.

MethodsPart of a Landsat MSS scene acquired on 31 August 19BB wasrectified using 7.5-minute orthophoto quadrangles. A nearestneighbor interpolation method was used to resample to a 50-by 50-m pixel size from the original 57-by 79-m pixel size.Using an unsupervised classified scheme, the delineated datawere grouped into "closed canopy conifer forests" and"other areas" as part of a Landsat forest change project(Spies et o1., in press). The "closed canopy conifer forestcover" was old-growth, mature, and other conifer standsgreater than approximately 30 to 40 years of age, and within-stand conifer canopy cover exceeded 60 percent (Brown,1985). The term "closed canopy conifer forest cover" wil l bereferred to as simply "closed conifer cover" in this paper.The "other areas" included clearcuts, young pre-canopy clo-sure conifer plantations, shrubs, bare soil and rock, naturalmeadows, and water, Approximately 100 spectral classeswere generated and were aggregated into the binary classifi-cation described above.

An accuracy assessment was conducted using a system-atic sampling of 135 points across the study area. High-alti-tude, color-infrared photographs (scale 1:60,000) from July1.988 were used to check each of the 135 points. The cover ateach point was identified on the aerial photography andcompared to the category at the corresponding location onthe classified image.

The AVHRR data were acquired in a local area coverageformat from NOAA-9 orbit 18558 on the afternoon of 19 July1988. A sub-scene of these data was rectif ied to UTM coordi-nates and registered to the MSS data set. AVHRR pixels wereresampled to 1.,000 by 1,000 m using a nearest-neighbor algo-rithm. The mean-square-error was 0.6 AwIRR pixels for theregistration of the AVHRR data to the MSS data. Eighty-nineA'VTIRR pixels were selected systematically from the AVHRRset, and a window of 20- by 20-MSS pixels was selected tocorrespond to each AVHRR pixel. The visible (0,58 to 0.68pm) and near infrared (NIR, 0.72b to 1.100 pm) band valueswere recorded for the Bg AVHRR pixels, and the percentclosed conifer cover of each corresponding 20 by 20 MSSpixel window was computed. The percent closed coniferiorrer *as defined as th-e proportion of an area or pixel con-taining closed conifer stands. Correlations were computed forthe closed conifer cover,'visible, and NIR variables albng withthe wm/visible band ratio and the NDVI. In addition. the

534

Bil:g'

OREGON

Figure 1. Map showing thestudy area which was locatedon the west slopes of theCascade Mountains in west-ern Oregon.

|i

AVHRR band values were converted to albedo using pre-Iaunch calibration equations for the computation of a cor-rected band ratio and corrected NDVr (NOAA, 1991).Computing albedo from the AVHRR band values helps to al-low for a direct comparison of Novr values with other studiesusing AVHRR data (Spanner et al.,1990b), A stepwise multi-ple regression was used to develop a model for predictingclosed conifer cover from the A''TIRR data. The model wasvalidated by comparing predicted proportions of closed coni-fer cover for the entire study area to the proportion of closedconifer cover for the study area as determined from the MSSdata set.

To determine spectral variation associated with succes-sional changes, spectral signatures were extracted from thisAVHRR data set for a set of homoeeneous landscapes consist-ing of four different successionafstages. These included avery large, disturbed clearcut area with an herbaceous/shrubcanopy cover of approximately 90 percent (21 ArvT{RR pixels,3 by 7); managed, closed-canopy Douglas fir forests approxi-mately 40 years old with approximately 1.5 percent of thearea in bigleaf maple and red alder (16 nVHRR pixels; aby a);natural mature Douglas fir forests approximately 90 to 130years old (21 AVHRR pixels, 3by 7); and old-growth Douglasfir and western hemlock forests 400 to 600 years old (20A\.HRR pixels, by 5). The mean AVHRR band values andstandard deviations for each of these forest successionalstages were plotted for the visible, un, and NDVI response.

ResultsThe vtss classification showed 57.8 percent of the study areain closed conifer cover with a total accuracy of gj. percent. Atotal of 74 oul of B3 samples (89 percent) feil correlt ly intothe closed conifer class, and a total of 49 out of 52 samoles(94 percent) fell correctly into the "other land" category-. Therelationship between AVHRR pixel band values and closedconifer cover was inverse for both bands as expected fFigure2J. The v is ib le ( r : - 0 .64, 12 = 0.41, P<0.0001) and NrR (r: -0.63, r2 : 0 .4O, P<0.0001.) bands showed much h iehercorrelations with closed conifer cover than the albedo-cJr-rected band rat io ( r : -0 .1,4, 12 = 0.02, P:0.190) and thealbedo-corrected NDVI ( r : -0 .15, 12 = 0.02, P=0.168J. Theuncorrected band rat io ( r : - 0 .46, 12 : 0 .27, P<0.0001)and the uncorrected NVDI (r = -0.46, 12 = 0.21,, P<0.0001)had significantly higher correlations with closed conifercover than the albedo-corrected transformations (Figure 3).

Scatter diagrams between closed conifer cover and red orinfrared AVHRR bands showed inverse linear relationshioswith random dispersal of points. The step-wise multiple

PE&RS

1 - -

f r o+

) o o

y = 6 1 . 4 - 0 0 7 Xr = -0.64t2 = o ,4 lp < 0.0001

40 60Closed Conifer Cover (7o)

- 144

NrA

f ror

z84

y = 1 2 9 . 9 - 0 . 3 0 Xr = -0.63r2 = 0.40p < 0.0001

- - ' - -€ ' ' -

Closed Conifer Cover (7o)

Figure 2. Scatter diagram and simple linear regres-sion of (a) AVHRR visible band values and percentclosed conifer cover along with simple linear regres-sion of (b) nvHnn near infrared band values and per-cent closed conifer cover.

regression resulted in the model: percent closed conifercover : 335.53 - 3.42 (visible) - 0.73 (NIn)with an ad-justed coefficient of determination of 0.46 (P<0.001). Whenthe above model was applied to the entire study area of 2405AVHRR pixels, the mean percentage of closed conifer coverwas 54.7 percent which compared favorably with the oriSinalMSS estimate of 57.B percent. Spatial patterns in the A'rrHRRdata from the model output were similar to the MSS patterns(Figure 4).

Figure 5 shows AVHRR band values related to forestsuccessional stage. Band values decreased with forest succes-sion for both the visible and utR bands. It should be notedthat these decreases were after the establishment of herbsand shrubs on the landscape. The visible band showed theereatest difference between mean band values for herb/shrubind young closed canopy conifers. The NIR showed the great-est difference between mean band values between youngconifer and mature forests as well as between mature andold-growth band values. The NDvt values were Iowest for the

PE&RS

o.4

zzoZ 0.3

20 40 60 80

Closed Conifer Cover (7o)

1 0 0

o.7

z v 'o

o u ' )

4.

?o 40 60 80

Closed C-onifer Cover (%)

1 0 0

0.7y = 0.a6 +^0.89 xr = 0.88, ra = 0.77P < 0.00001

c

zr 0.6oI

6E o.s

o.0fr.0.3 0.4

Raw NDVI

Figure 3. Scatter diagrams and simplelinear regression of (a) raw NDVI and per-cent closed conifer cover, (b) albedo-cor-rected NDVI and percent closed conifercover, and (c) albedo-corrected NDVI andraw NDVI.

v = 0.36 - 0.0006 Xt= .O .qA , r2=O.Z tP = 0.0001

" 8 " "

k*$t, ,B *" = '

o 3 s " r t: _ oTFe-- t r .

, t " - r t " F

E

t " " t r " t tE G E

oEF

oE

aE E o

,+-; '_ _ o l t . ' E FO E E E -

O -

E gy = 0.57 - 0.00O2 X

r=-0.15,2=0.02P = 0.168

herb/shrub class, highest for the young conifer class, and in-termediate for mature and old-growth forests.

DiscussionThe results from this experiment indicate that it may be pos-sible to use A'r'HRR data to obtain landscape and regional es-timates of closed conifer cover in the temperate coniferousforests. Bands 1 and 2 were related to both the proportion of

535

Figure 4. The image on the left shows closed conifer coverin black and other lands in white. lt was produced fromclassified Landsat MSs data (50-m pixels). The right imageis a gray scale of AVHRR data (1,000-m pixels) showing thepercentage of closed conifer cover within RvHnn pixels (lowpercentages are light and high percentages are dark).

closed conifer cover within AvHRR pixels and the seral. stageof the vegetation. Both visible and t'uR band values decreaiedlinearly with the proportion of closed conifer cover within apixel. This result is in contrast to the typically found directrelationship between MR and vegetation amount for manyvegetation types. Researchers have found that this direct re-lationship between MR with measures of conifer amountdoes not always exist, especially for canopies that are notclosed. When the background or understory is brighter thanthe conifer canopy, the relationship can be inverse (Ripple efa1., 1991c), and when the background is darker or the canopyis closed, it can be direct (spanner et a1.,1990a). When thebackground has about the same brightness as the conifers oris highly variable, the relationship to mn may be flat orweak. It has also been hypothesized that conifer canopyshadowing contributes to an inverse Mn relationship as thecanopy structure becomes more complex and shaded in olderstands (Ripple ef o/., 1991c). The reason for the inverse rela-tionships between the band values and closed conifer coverin this study was attributed to the deciduous shrubs andtrees along with herbaceous vegetation being more highly re-flective than the conifer canopy in both the visible and tttRbands (Fiorella and Ripple, 1993). Areas within AvHRRrn pixels

herba-Dands tlroreua and Kipple, 1993J. nre€not covered by a conifer canopy consisnot covered by a conifer canopy consisted mostly of herbeceous or deciduous woody vegetation with very little bareceous or deciduous woody vegetation with very little baresoil. Therefore, band values decreased as the proportion ofthe darker conifer cover increased.

The Nnvt relationship with closed conifer cover was alsoinverse, but the correlation was not as high as with the visi-ble and utn bands. The two-variable (band) regression modelaccounted for more variance than either the Novt or the indi-vidual bands. These results are similar to those found byIverson ef o/. (1989). The variance not accounted for in thisregression model was attributed to slight misregistration be-tween the AvHRR and uss data sets and variabilitv in the

536

broad successional stages. The mean-square error of 0.6AVHRR pixels, equivalent to approximately 600 metres on theground, may have been the cause for some of the scatteraround the regression lines. Overall, the model worked well,considering that each AVHRR pixel represented 400 timesmore land area than each Landsat MSS pixel,

It should be noted that the albedo-corrected Novr andthe uncorrected NDVI were not perfectly correlated (r : 0,88,12 : 0.77, Figure 3c). The albedo-corrected Novr had a lowercorrelation to the percentage of closed conifer cover than theuncorrected NnvL The differences between the uncorrectedand corrected ND'r'I values were greater than the rounding er-rors inherent in the transform from the original awnn bandvalues to the albedo-corrected NDW. Most recent research onthe calibration of AvgRn has been concerned with oost-launch performance of the sensor, and specificallyihe degra-

-E

co- q 7 0c vG l

- r oB g* 6 6 0

o

t

140

E E i r s o6 = F

E S : 120: P o

2rd i ! l o

100

o.7ao

E

to 0.6zooEv 0.s

I

Herb/ Young OtdShrub Conifer Mature Gro\./th

Herb/ Young OtdShrub Conifer Mature GroMh

Herb/ Young OldShrub Coniter Malure Growth

Figure 5. Band value means and standard de-viations using the visible, near infrared, andthe normalized differences vegetation index(ttovr) from AVHRR data for four successionalstages found on the west slope of the Cas-cades of Oregon.

PE&RS

( ) l i [ ( i ( ) N , \ V l ] l i l i l N l A ( i t

T{ ' r

rf,-1,-&[ ' l4,{

ot i t -GoN CoNt t t -n t o l tFS I cov t - l l

f*

I l

0 . , 'l " j t , : l " I

Plate 1, Raw AVHRR image of Oregon (top) with the visible and NIR bands andclosed conifer cover patterns in Oregon (bottom) with four broad closed conifercover classes (25 percent cover intervals). Film recording couftesy of ERDAS pro-duction seruices, Atlanta. Georgia,

9 6 n

l

ni

3 + o

I

- r o

- l

0 20I

40 60

Closed conifer cover (7o) AVHRR

Figure 6. Comparison of estimates of closed conifer coverfrom AVHRR satellite data and from U-2 aerial photographyfor 32 7.S-minute quadrangles in Oregon.

dation in the instrument gain values over time because noon-board calibration is available (Che and Price, 1992; Kauf-man and Holben, 1993 j. Because of this degeneration overtime, it is especially important to use post-launch calibrationprocedures in studies comparing multiple AVHRR images.Off-nadir viewing, atmospheric variability, and instrumentprecision also creates errors in NDVI observations (Goward efo/,, 1991). Little information is available on the effects trans-forming raw AVHRR values into calibrated values and then inturn converting those calibrated values into Novt space. Thediscrepancy between raw NDVI and Albedo-corrected NDVIvalues apparently result from two different locations incartesian space (tltR, red coordinates) being nonJinearlytransformed into polar (NDVI) space. Similar discrepanciescould also exist in other situations where satellite band val-ues are corrected before transforming them into a ratio likethe Nnu.

The relationship between visible band values and forestsuccessional stage was inverse and somewhat asy.rnptoticafter canopy closure. The largest mean differences betweenthe seral stages of young and mature, and mature and old-growth were found using the NIR band. NIR band values be-tween the serai stages of herb/shrub and closed canopyyoung conifer decreased because of decreased deciduousvegetation exposed to the sensor. This trend continued be-tween the closed canopy young conifer and the mature coni-fer seral stage as the increase in canopy shadowing anddecrease of the bright deciduous component apparently dom-inated off-setting increases in NIR due to increases in leafarea index (Ripple et al.,7gg7ct Spanner ef a1., 1990a). MRvalues for o)d-growth were slightly Iower than mature forest.This was probably due to the increase in shadowing andcanopy gaps dominati.ng the signal resulting in a lower radi-ance for old-growth (Fiorella and Ripple, 1993). This was not

s38

Iikely due to the higher leaf area index in the old-growth, be-cause light becomes asymptotic above a LAI of 6 (Spanner eto/., 1990b1 and old-growth LAIs are normally higher than 6 inthis area (Franklin ef o/., 1981).

NI/DI increased from the herb/shrub stage to the youngconifer stage and then decreased through the mature to theold-growth stage indicates a potential problem with usingNDVI for successional stage mapping. Within these temperateconiferous forests, it appears that both pre-and post-canopyclosure NDVI values could be confused, Spanner ef a1. (19S9)had similar results when using the similar rv band 4/3 ratioin these forests. Box ef o/. (1g8g) also concluded that theredid not seem to be any reliable relationship, across differentvegetation structures, between biomass and NDvt when usingAVHRR data. Multiple regression of individual bands may bethe best approach for successional stage mapping withAVHRR.

Application of ResultsThe results presented above have applications in both mea-suring the total extent of coniferous forests and the level offorest fragmentation for Iarger regions. For example, theregression model was also applied to the entire state of Ore-gon to determine the spatial distribution of conifer forestsand levels of forest fragmentation, The resulting state mapshowed closed conifer cover ranging from 0 to 100 percent in1 percent increments just as in Figure 4 above for the smallarea. Because it was not feasible to show a map of all 100classes for the entire state, a generalized map was producedshowing the estimates of the proportion (in 2S percent steps)of closed conifer cover for each A\,TIRR pixel in the state(Plate 1). Within western Oregon, the map showed the high-est levels of forest fragmentation in the Oregon Coast Rangeand the lowest fragmentation in the southern Cascade Moun-tains, especially in and around the Umpqua National Forest.The map can also be used to analyze landscape linkages forbiodiversity planning and ecosystem management. Foi exam-ple, it showed the various levels of fragmentation on forestcorridors linking the Cascade range with both the CoastRange and the Klamath Mountains.

The A\alRR-based Oregon map showed 10,891,000 ha ofconifer forests in Oregon. This is 4 percent higher than esti-mated by the U.S. Forest Service under the Resources Plan-ning Act (nen), f,he RpA estimated 10,463,82T ha of coniferand 890,750 ha of hardwoods for a total of 1,1,384,527 ha offorest land in Oregon. The land-cover database for the con-terminous United States shows total forest land in Oregon at11,990,000 ha [Loveland et a].,7997; Turner et al., in press).

This state-wide data set of percentage of closed conifercover was verified through a comparison with U-2 color-in-frared aerial photography flown by NASA on 19 July 19S8.The flight covered parts of the central Oregon Coast range,the north Cascade range, the central Cascade range, and theBlue Mountain range in northeastern Oregon. Eight 7.S-min-ute quadrangles were randomly selected in each of these fourgeographic areas to compare with the A\TIRR set. The propor-tion of closed conifer cover in each of these 32 quadrangleswas estimated using a dot grid on the u-2 photography.These same 32 quadrangles were windowed out of theAVHRR closed conifer cover data set.

The proportion of closed conifer cover as estimated fromAVHRR was highly correlated (r, : 0,81) with estimates fromthe u-z photography (Figure 6). The regression between thesetwo sets showed a linear relationship with the U-2 photosproviding slightly higher estimates of closed conifei cover

PE&RS

than the AVHRR set (intercept : 9.04).This difference wasprobably due to the ability to detect very small coniferpatches on the U-z photography.

The relationship between A\,TIRR pixel values with theproportion of closed conifer cover indicate that these sensordata may be well suited as an integrator for large area forestfragmentation studies in this Pacific Northwest region. Forexample, these AVHRR results were used in designing pro-posed federal legislation for locating biological forest re-serves to protect the northern spotted owl and otherimportant wildlife species (Forest Ecosystem ManagementTeam, 1993).

ConclusionsThis research has shown that band values in the AVHRRchannels 1 and 2 are related to the continuum of forest land-scape conditions. The regression model was successful be-cause areas that were highly fragmented by clearcutting hadhigh band values in both the visible and near infrared bands.Conversely, areas dominated by late-successional forests andlow fragmentation consistently had the lowest band values.

Additional research should be conducted to confirm thespectral/successional stage relationships found here. Thiswork could include research using pixel mixture models(Smith ef o1., 1990) to develop a better understanding of howthe proportions of successional stages and forest cover deter-mine the spectral response in large AVHRR pixels. Cautionshould be used when making calibration corrections beforecalculating the NDVI and in attempting to use the NDVI whenconsidering data sets that include both natural and managedstands with a range of structural characteristics and succes-sional stages (i.e., herbs, deciduous shrubs, conifers, etc.). Itappears that AVFIRR has potential for estimating forest frag-mentation in temperate coniferous forests and for mappingthe spatial distribution of forest resources at continental andglobal scales.

AcknowledgmentsThis project was funded in part by NASA grant number NAGW- 1460. The author would Iike to thank G.A. Bradshaw, LouisIverson, RJay Murray, Mike Spanner, Dave Turner, and DenisWhite for their helpful comments on an earlier version of themanuscript. Maria Fiorella and Jon Greninger assisted in var-ious aspects of the mapping.

ReferencesBox, E.O., B.N. Holben, and V. Kaib, 1989. Accuracy of AVHRR Veg-

etation Index as a Predictor of Biomass, Primary Productivityand Net CO, FIux, Vegetation, B0:71-89.

Brown, E.R. (editor), 1985. Manogement of Wildlife and Fish Habi-tats in Forests of Western Oregon and Washington, No. R6-FLWL-192-1985, Pacif ic Northwest Region, USDA Forest Ser-r r i n o P n " t l a n r l O r o o n n

Che, N., and J.C. Price, 1992. Survey of Radiometric Calibration Re-sults and Methods for Visible and Near Infrared Channels ofNOAA-7, -9, and -11 A\tIRRs, Remote Sensing of Environment,41,:79-27.

Fiorella, M., and W.J. Ripple, 1993. Determining Successional Stateof Temperate Coniferous Forests with Landsat Satellite Data,Photogrammetric Engineering & Remote Sensrng, 59(2)t239-246.

Forest Ecosystem Management Assessment Team, 1993. Forest Eco-system Management: An Ecological, Economic, and Social As-sessmenf, USDA Forest Service, Portland, Oregon.

PE&RS

Franklin, J.F., and C.T. D1'rness, 7973. Natural Vegetation of Oregonand Washington, Oregon State University Press, 452 p.

Franklin, J.F., K. Cromack, Jr., W. Denison, A. McKee, C. Maser, l.Sedell, F. Swanson, and G. Juday, 1981. Ecological Characteris-tics of Old-Grovdh Douglas-Fir Forests, General Technical Re-port PNW-118, USDA Forest Service, Pacific Northwest Forestand Range Experiment Station, Portland, Oregon, 417 p.

Goward, S.N., B. Markham, D.G. Dye, W. Dulaney, and J. Yang,1991. Normalized difference vegetation index measurementsfrom the advanced very high resolution radiometer, RemoteSensing of Environment, 35:257 -27 7.

Iverson, L.R., E.A. Cook, and R.L. Graham, 1989. A technique for ex-trapolating and validating forest cover across large regions cali-brating AVHRR data with TM data, International /ournal ofRemote Sensing, 70(77):7805-7812.

Kaufman, Y,J., and B.N. Holben, 1993. Calibrat ion of the AVHRRVisible and Near-IR Bands by Atmospheric Scattering, OceanGlint, and Desert Reflection, lnternational Journal of RemoteSensr'ng, 74(1):27-52.

Lehmkuhl, J.F., and L.F. Ruggiero, 1991. Forest fragmentation in thepacific northrvest and its potential effects on wildlife, Wildlifeand Vegetation of Unmanaged Douglas'fir forests, General Tech-nical Report PNI/V-GTR-285, USDA Forest Service, Pacific North'west Research Station, Port land, Oregon, pp. 35-46.

Loveland, T.R., I .W. Merchant, D.O. Ohlen, and J.F. Brorvn, 1991.Development of a land-cover characteristics database for theconterminous U.5., Photogrammatic Engineering & RemoteSensrng, 57(1 1) :1453-1463.

Lord, J.M., and D.A. Norton, 1990. Scale and the spatial concept offragmentation, Conservation Biologr, 4(2):197 -2O2.

Nelson, R., 1989. Regression and ratio estimators to integrate AVHRRand MSS data, Jlemote Sensing of Environment, 301270-216.

Nelson, R., and B. Holben, 1986. Identifying deforestation in Brazilusing multiresolution satellite data, /nternaffonal Journal of Re-mote Sensing, 7(3):429-448.

NOAA, !997. Polar Orbiter Data Users Guide, National Oceanic andAtrnospheric Administration, National Environmental Satellite,Data, and Information Service, National Climatic Data Center,Satellite Data Services Division, Washington, D.C.

Ripple, W.J., G.A. Bradshaw, and T.A. Spies, 1991a. Measuring for-est landscape patterns in the Cascade Range of Oregon, USA,Biolo gical Conservation, 57 i7 3- 88.

Ripple, W,J., D.H. Johnson, K.T. Hershey, and E.C. Meslow, 1991b.Old-Growth and Mature Forests Near Spotted Owl Nests inWestern Oregon, lournal of Wildlife Management, 55(2):31.6-318.

Ripple, W.J., S. Wang, D.L. Isaacson, and D.P. Paine, 1991c. A pre-liminary comparison of Thematic Mapper and SPOT-1 HRVmultispectral data for estimating coniferous forest volume, lnter-national Journal of Remote Sensing, 72(9):7977-7977.

Smith, M.O., S.L. Ustin, l .B. Adarns, and A.R. Gil lespie, 1990. Vege-tation in deserts: L A regional measure of abundance from mul-tispectral images, Remote Sensing of Environment, 3717-26.

Spanner, M,A., C.A. Hlauka, and L.L. Pierce, 1989. Analysis of forestdisturbance using TM and AVHRR Data, Proceedings of lnterna'tional Geoscience ond Remote Sensing Symposium, Vancouver,- ^ - ^ i ^ r ^ r . . r . . - - 1 3 9 7 - 1 3 9 0 .u a r r d u d , r u . , u r y , P P . r

Spanner, M.A., L.L. Pierce, D.L. Peterson, and S.W. RunninS, 1990a.Remote sensing of temperate coniferous forest leaf area index:The influence of canopy closure, understory vegetation andbackground reflectance, International Journal of Remote Sens-i n g , 1 1 ( 1 ) : 9 5 - 1 1 1 .

Spanner, M.A., L.L. Pierce, S.W. Running, and D.L. Peterson, 1990b.The seasonality of AVHRR data of temperate coniferous forests:Relationship with leaf area index, Remote Sensing of Environ-ment,33:172.

Spies, T.A., W.J. Ripple, and G.A. Bradshaw, in press. Dynamics andpattern of a managed coniferous forest landscape, Ecological Ap'plications.

539

Townshend, J.R.G., and C.J. Tucker, 1984. Objective assessment ofAdvanced Very High Resolution Radiometer data for land covermapping, lnternational Journal of Remote Sensing, 5(2):a97-50a'

Turner, D.P., G. Koerper, H. Gucinski, and C. Peterson, 1993. Moni-toring global change: Comparison of forest cover estimates usingremote sensing and inventory approaches, Environmental Moni-toring and Assessmenf , 26 :295-305.

Woodwell , G.M., R.A. Houghton, T.A, Stone, R.F. Nelson, and W.Kovalick, 1987. Deforestation in the tropics: New measurementsin the Amazon Basin using Landsat and NOAA AVHRR im-agery, I. Geophys, nes., 92(02):2757 -2763.

Zhu,2., and D.L. Evans, 1992. Mapping midsouth forest distr ibu-tions, /ourno,l of Forestry, 9O(72):27 -30.

(Received 19 February 1993; revised and accepted 10 August 1993)

tg94 ASPRS/ACSMTECHNICAT PAPERS

Proceedings of the 1994 ASPRS/ACSM Annual Meetingheld in Reno, Nevada, April 1994.

Even if you couldn't attend the 1994 ASPRS/ACSM Annual Convention in Reno, this 2 volume set can help youdiscover what's new in the industry. karn the latest research and theory in GIS, Remote Sensing, Photogram-metry, Surveying, and Cartography.

vol.. r - ASPRSStock # 4934-1. pp.762.

r GIS Mapping and Highwaysr Photogrammetry Standards and Instrumentationr Remote Sensing - Precision/Prescription Farming

with GIS/GPSr Remote Sensing - Water and WetlandsI Photogrammetry - Softcopyr Remote Sensing - Technologyr Remote Sensing - Environmentr Land and Real Estater Photogrammetry - Close RangeI Remote Sensing - AgricultureI Photogrammetry - Geodesy/Sensorsr Photogrammetry - GPSr GIS - Collection and Mappingr Remote Sensing - The Flood of 1993I Photogrammetry - Triangulation and Orthophoto-

graphy

YOL.2 - ACSMStock # 49t4-2. ort.3ll3.

r DahBase/GPS Issuesr Survey Management Issuesr Surveying Computationsr Surveying EducationI Digiial MappingI Global Change, EOS and Nale Issuesr Battelle Research in Remote Sensingr Battelle Research in GISr Advanced lmage Processingr GIS Issuesr Surveying and Geodesy Issuesr Water Resource lssuesr Advanced Applicationsr Surveying History and Related Issuesr North American Landscape Characterization

(NALC): LANDSAT Pathfinderr Advanced Application

1994. Twovolumes. 1,164pp. $30each(soficover);/4SPRSMembers$20each. Stock#4934-1,4934-2.

To Order, See The ASPRS Store.

540 PE&RS