ESTIMATING WOODY BROWSE ESTIMATING WOODY BROWSE ABUNDANCE IN REGENERATING ABUNDANCE IN REGENERATING
CLEARCUTS USING AERIAL IMAGERYCLEARCUTS USING AERIAL IMAGERY
Shawn M. Crimmins, Alison R. Mynsberge, Shawn M. Crimmins, Alison R. Mynsberge, Timothy A. WarnerTimothy A. Warner
INTRODUCTIONINTRODUCTION
•Timber harvestTimber harvest•Common in eastern hardwood forestsCommon in eastern hardwood forests•Can create wildlife habitatCan create wildlife habitat•Abundance of available forageAbundance of available forage
•Assessment of forage abundanceAssessment of forage abundance•Time consuming, labor intensive field workTime consuming, labor intensive field work•Method to quantify browse remotely?Method to quantify browse remotely?
INTRODUCTIONINTRODUCTION
•Remote sensing in forestryRemote sensing in forestry•Forest species classificationForest species classification•Forest healthForest health•Forest regenerationForest regeneration
•Remote sensing in wildlifeRemote sensing in wildlife•Habitat classificationHabitat classification•Habitat quality?Habitat quality?
OBJECTIVEOBJECTIVE
•Use readily available aerial imagery to Use readily available aerial imagery to estimate the abundance of woody browse in estimate the abundance of woody browse in regenerating clearcutsregenerating clearcuts
STUDY AREASTUDY AREA
•Penn-Virginia (formerly MeadWestvaco) Penn-Virginia (formerly MeadWestvaco) Wildlife and Ecosystem Research ForestWildlife and Ecosystem Research Forest
•3413 hectares3413 hectares•Randolph County, WVRandolph County, WV•Northern hardwood forest communityNorthern hardwood forest community•Actively harvestedActively harvested•Dominated by clearcuttingDominated by clearcutting
METHODSMETHODS
•12 regenerating clearcuts12 regenerating clearcuts•Surveyed as part of large-scale deer researchSurveyed as part of large-scale deer research•6 age classes (0 – 5 years post harvest)6 age classes (0 – 5 years post harvest)•2 cuts in each class2 cuts in each class•30 plots in each cut (15 edge, 15 interior), 0.5m30 plots in each cut (15 edge, 15 interior), 0.5m22
•Identified and counted all woody stems Identified and counted all woody stems << 1.5 m 1.5 m•Maximum browse height of deerMaximum browse height of deer•Variable of interestVariable of interest
METHODSMETHODS
METHODSMETHODS
•Aerial imageryAerial imagery•National Agriculture Imagery Program (USDA)National Agriculture Imagery Program (USDA)•2 meter resolution2 meter resolution•Visible spectrum onlyVisible spectrum only•Taken on September 24 (37 days after last field Taken on September 24 (37 days after last field survey)survey)•Chosen due to availabilityChosen due to availability
•Publicly availablePublicly available•Representative of data available to forest/wildlife Representative of data available to forest/wildlife managersmanagers
METHODSMETHODS
•Generated 12 image metricsGenerated 12 image metrics•Mean and varianceMean and variance
•Red, Green, BlueRed, Green, Blue•Ratios calculated, but abandoned laterRatios calculated, but abandoned later
•Hand digitized clearcuts in ArcGIS 9.1Hand digitized clearcuts in ArcGIS 9.1
•Multiple Linear Regression (PROC REG)Multiple Linear Regression (PROC REG)•Forward variable selectionForward variable selection•αα = 0.10 for entry into model = 0.10 for entry into model
RESULTSRESULTS
•Censored one cut from analysis due to Censored one cut from analysis due to excessive shadows covering edge plotsexcessive shadows covering edge plots
•Ratio terms removedRatio terms removed•Insufficient degrees of freedomInsufficient degrees of freedom
•12 predictor variables12 predictor variables•11 observations11 observations
RESULTSRESULTS
Variable Estimate p
Intercept 261.42376 0.0052
Red-mean -1.82409 0.0134
•Forward selection-first step (Forward selection-first step (rr22 = 0.5109) = 0.5109)
•Forward selection-second step (Forward selection-second step (rr22 = 0.7684) = 0.7684)Variable Estimate p
Intercept 424.79332 0.0005
Red-mean -2.38020 0.0010
Green-variance -0.12022 0.0175
RESULTSRESULTS
Variable Estimate p
Intercept 345.49928 0.0007
Red-mean -1.93586 0.0013
Blue-variance 0.13138 0.0226
Green-variance -0.19743 0.0015
•Forward selection-third step (Forward selection-third step (rr22 = 0.8953) = 0.8953)
RESULTSRESULTS
Variable Estimate p
Intercept 240.05731 0.0166
Red-mean -2.23253 0.0006
Blue-variance 0.16328 0.0072
Green-mean 0.81805 0.0910
Green-variance -0.18525 0.0014
•Forward selection-final step (Forward selection-final step (rr22 = 0.9375) = 0.9375)
Variable Partial rr22 F value
Red-mean 0.5109 9.40
Blue-variance 0.2575 8.90
Green-mean 0.1269 8.48
Green-variance 0.0422 4.04
•Variable selection summaryVariable selection summary
RESULTSRESULTS
•Final model (Final model (pp = 0.0009) = 0.0009)
)(16328.0)(23253.205731.240 2bluered
)(18525.0)(81805.0 2greengreen
•Global model (Global model (pp = 0.0062, = 0.0062, rr22 = 0.9671) = 0.9671)
RESULTSRESULTS
•Accurate across abundancesAccurate across abundances
0
10
20
30
40
50
60
70
0 10 20 30 40 50 60 70
Estimated browse
Ob
serv
ed b
row
se
DISCUSSIONDISCUSSION
•High (High (rr22 > 0.9) predictive accuracy > 0.9) predictive accuracy•Minimal complexityMinimal complexity
•4 predictor variables4 predictor variables•Visible spectrumVisible spectrum
•Normally distributed error termNormally distributed error term•Slight deviance in tailsSlight deviance in tails•Low abundance sites (logging slash)Low abundance sites (logging slash)•High abundance sites (canopy growth)High abundance sites (canopy growth)
DISCUSSIONDISCUSSION
•Visible spectrum onlyVisible spectrum only•Available in almost all imageryAvailable in almost all imagery•Still allowed high predictive accuracyStill allowed high predictive accuracy•Lack of IR bands (required for NDVI)Lack of IR bands (required for NDVI)
•Growth > 1.5 metersGrowth > 1.5 meters
•Technique available to most managersTechnique available to most managers•No knowledge of remote sensing/GIS requiredNo knowledge of remote sensing/GIS required
FUTURE DIRECTIONSFUTURE DIRECTIONS
•Temporal replicationTemporal replication•Only 1 yearOnly 1 year•Track individual cuts across growing seasonsTrack individual cuts across growing seasons
•Spatial replicationSpatial replication•Local weather/climate affecting growth?Local weather/climate affecting growth?•Different forest types?Different forest types?
ACKNOWLEDGMENTSACKNOWLEDGMENTS
•Matt Shumar, Chris RunnerMatt Shumar, Chris Runner•MeadWestvaco CorporationMeadWestvaco Corporation•West Virginia Division of Natural ResourcesWest Virginia Division of Natural Resources•WVUWVU
•Division of Forestry and Natural ResourcesDivision of Forestry and Natural Resources•Cooperative Fisheries & Wildlife Research UnitCooperative Fisheries & Wildlife Research Unit•Department of Geology and GeographyDepartment of Geology and Geography
•WV ViewWV View
QUESTIONS?QUESTIONS?