landscape scale monitoring of forest treatments and disturbance s.e. sesnie 1,2, a.d. olsson 1, b.g....

LANDSCAPE SCALE MONITORING OF FOREST TREATMENTS AND DISTURBANCES.E. Sesnie1,2, A.D. Olsson1, B.G. Dickson1, A. Leonard3, V. Stein Foster3 and C. Ray1

Tornado

Thin

2006 2010

Lab of Landscape Ecology and Conservation Biology, NAU1, US Fish & Wildlife Service2, USFS Kaibab NF3

2

Collaborative Partnerships

NAU

USFS-FIA

Kaibab NF

USFS R3

FWS

Stakeholders

USFS-4FRI

3

Previous assessments

Region Landscape

Patch

Disturbance 4

Fire Urbanization Insects & Disease

Silviculture

Wind Drought

5

Kaibab NF Monitoring

Need for landscape-scale, up-to-date and repeatable forest data complementary to:

Monitoring planned and unplanned disturbance

Assessing changes wildlife habitat

Mitigating fire hazard and risk

Biomass & forest carbon assessment

Other planning objectives

6

Objectives

Model & map forest structure

Inexpensive

Validated

Consistent/repeatable

BA SDI HGT CCTPA QMDOthers…

7

Study Area – Kaibab NF

1.6 million acres (647,497 ha) Ongoing management & 4-FRI Three major forest types

8

Methods – reference data

USFS Forest Inventory and Analysis (FIA) plots

Measure10% of state’s forest per year

n = 648 plots (2001 – 2009)

Monitor forest status and trends

FVS compatible

~1 acre of forest land

9

Methods – remotely sensed dataLandsat TM (2006 & 2010)

6 spectral bands 2 dates (leaf on/off)

Spectral Derivatives NDVI NDVIc

Digital Elevation Model Elevation Terrain

39 Predictor VariablesLeaf-onTM1TM2TM3TM4TM5TM7BrightnessGreennessWetnessNDVI_onNDVIc_onNDVI4_onPCA1PCA2PCA3

Leaf-offTM1TM2TM3TM4TM5TM7BrightnessGreennessWetnessNDVI_offNDVIc_offNDVI4_offPCA1PCA2PCA3

DEMElevationSlopeAspect (trasp)RoughnessCTI

OthersNDVI_ratio

10

Methods – correction + modeling

Image correction: Uncorrected Landsat TM

Image normalization (MAD)

Atmospheric correction (FLAASH)

Terrain correction (C-correction)

Modeling: Random Forest (regression)

11

Literature & Tools

12

Results – outputs

2006 2010 Change (%)

13

Results – outputs

2006 2010

14

Results - Validation

Basal Area – Predicted vs. Actual

15

Results - Validation

Basal area Stand Density Index

All predictors Best subset Number of predictors

16

Results – Consistency

Compare 2006 and 2010 basal area

1km x 1km points (n = 4622)

Pearson correlation coefficients

Kruskal-Wallis One Way ANOVA on Ranks

17

Results – Consistency

BA 2006 (Uncorrected)

0 50 100 150 200

BA

201

0 (

Unc

orr

ect

ed)

0

50

100

150

200

BA 2006 (FLAASH)

0 50 100 150 200

BA

201

0 (

FLA

AS

H)

0

50

100

150

200

BA 2006 (MAD)

0 50 100 150 200

BA

201

0 (

MA

D)

0

50

100

150

200

BA 2006 (C)

0 50 100 150 200

BA

201

0 (

C)

0

50

100

150

200

BA 2006 (FLAASH/C)

0 50 100 150 200

BA

201

0 (

FLA

AS

H/C

)

0

50

100

150

200

BA 2006 (MAD/C)

0 50 100 150 200

BA

201

0 (

MA

D/C

)

0

50

100

150

200

r = 0.875 r = 0.89 r = 0.882

r = 0.862 r = 0.852r = 0.879

Between years for each correction technique (Basal area)

18

Results - Consistency

2006 basal area, H = 33.927 (P = <0.001)

2010 basal area, H = 26.090 (P = <0.001)

19

Monitoring applications

Treatment type

Fuel reduction Precommercial thin Group selection Broadcast burn

Cha

nge

(%)

0

10

20

30

40

50

Basal Area DensityCanopy Cover

Stand density index (SDI)

SDI change (%)

Treatment types 2006 - 2010

20

Conclusions

Terrain and atmospheric corrections important

We are likely to witness large and small disturbance events in the future

Forest structure data are consistent & sensitive to a variety of disturbance factors

The outputs support a range on of landscape scale analyses

21

Acknowledgments

Kaibab National Forest

USFS Forest Inventory and Analysis

program, Ogden UT

USFS Region 3 Office

US Fish & Wildlife Service