Post on 02-Jan-2016
transcript
Normalized Difference Fraction Index (NDFI): a new spectral index for enhanced detection of forest canopy damage caused by selective logging and forest fires
Carlos Souza Jr., ImazonDar Roberts, UCSB
Mark Cochrane, USDSamia Nunes, Imazon
Deforestation vs Forest Degradation
Creates a complex environment: Undisturbed forests Canopy gaps Exposed soils Dead vegetation
Caused by: Selective logging Forest fires Forest fragmentation
Souza Jr. and Roberts (2005) - IJRS
Ikonos Image – Paragominas, PA
Very Challenge with Landsat!
Selective logging
1998
1999
Old Selective logging
Selective logging and burning
2000
2001 Old selective logging andburning
Souza Jr. et al. (2003), Rev. Ciência Hoje
Objective
Develop a technique that combines spectral and spatial information to enhance the detection and mapping of canopy damage caused by selective logging and forest fires
New spectral index: Normalized Difference Spectral Index (NDFI)
Contextual classification algorithm (CCA) Spectral: NDFI Spatial: log landings
Forest Transect Classes
Forest class Field Description
Intact(n=4)
Mature and undisturbed forest
Non-mechanized logging(n=5)
Logged forest without the use of vehicles such as skidders and trucks, also known as traditional logging. Log landings, roads and skid trails are not built.
Managed logging(n=5)
Planned selective logging where a tree inventory is conducted, followed by road and log landing planning to reduce harvesting impacts.
Conventional Logging(n=2)
Conventional unplanned selective logging using skidders and trucks. Log landings, roads and skid trails are built.
Logged and burned(n=3)
Either non-mechanized or conventionally logged forests that have subsequently been damaged by intense surface fires.
Deg
rad
atio
n In
ten
sity
Souza Jr. et al. (2005), Earth Interactions
Forest Transects
Forest inventories Transects (10 m x 500 m = 0.5 ha)
Trees with DBH ≥ 10 cm Sub-plots (10 m x 10 m)
All trees Forest canopy cover Vine density % soil exposed % of dead vegetation
Aboveground biomass estimation Allometric equations (Gerwing, 2000)
Objective I
0
10
20
30
40
50
60
Intact Non-mechanizedLogging
ManagedLogging
ConventionalLogging
Logged andBurned
Fra
cti
on
Va
lue
(%
)
GV NPV Shade Soil
SMA Fraction Images
Forest Degradation Separability
Souza Jr. et al. (2005), Earth Interactions
Normalized Difference Fraction Index
SoilNPVGV
Soil)(NPVGVNDFI
Shade
Shade
Shade100
GVGVShade
-1 ≤ NDFI ≤1
NDFI low to moderate
NDFI near 1
High GVLow NPV and Soil
Low to moderate GVModerate to high NPV and Soil
Souza Jr. et al. (2005), RSE
SMA ResultsGVshade
NPV
Soil
NDFI
Fraction Value (%)
Fraction Value (%) Fraction Value (%)
NDFI
Souza Jr. et al. (2005), RSE
-20
-15
-10
-5
0
5
10
Non-mechanized
Logging
ManagedLogging
Logged Logged andBurned
Ch
an
ge
fro
m I
nta
ct
Fo
res
t (%
)
GV
NPV
Soil
Shade
NDFI % Change
NDFI vs. Fraction Images
Percent Change from Intact Forest
Souza Jr. et al. (2005), RSE
Class Separability
ClassClassIntactIntact
Non-Non-mechanized mechanized
LoggingLogging
Managed Managed LoggingLogging
Conventional Conventional LoggingLogging
Logged and Logged and BurnedBurned
Mean Stdev. Mean Stdev. Mean Stdev. Mean Stdev. Mean Stdev.
GVGV 40a 4 41a 5 41a 5 38b 9 25c 7
NPVNPV 6a 2 5a 2 6a 2 10bc 4 11bd 3
SoilSoil 2a 1 1a 1 3ab 1 4bc 3 7d 3
ShadeShade 51a 3 53a 5 51ab 4 49bc 3 56d 3
NDFINDFI 0.84a 0.08 0.87a 0.07 0.79b 0.07 0.58c 0.24 0.49d 0.22
Tukey test at P<0.01
Souza Jr. et al. (2005), RSE
Contextual Classification Algorithm - CCA
Step 1: Find log landingsStep 1: Find log landingsSoil > 10%Soil > 10%Find regions and calculate areaFind regions and calculate area1 ≤ Area ≤ 4 pixels1 ≤ Area ≤ 4 pixels
Step 2: Grow a canopy damage region around log Step 2: Grow a canopy damage region around log landingslandings Search for NDFI neighboring cell valuesSearch for NDFI neighboring cell values
If NDFI > 0.75 then If NDFI > 0.75 then Intact ForestIntact Forestif 0 if 0 ≤≤ NDFI ≤ 0.75 then NDFI ≤ 0.75 then Canopy DamageCanopy Damage
SoilFraction
LogLandings
CanopyDamage
CCA ResultsC
on
ven
tio
nal
Lo
gg
ing
Lo
gg
ed a
nd
B
urn
ed
NDFI Canopy Damage
Accuracy Assessment
ClassifiedPixels
Reference Pixels
Users Accuracy
(%)Non-forestNon-forest ForestForestCanopy Canopy DamageDamage Total
Non-Forest 454 0 17 471 96.4
Forest 6 625 117 748 83.6
Canopy Damage 40 0 616 656 93.993.9
Total 500 625 750 1875
Producers Accuracy (%) 91.0 100.0 82.0
Overall Accuracy = (1695/1875) = 90.4%90.4% Kappa Coefficient = 0.85
Generic Image Endmembers
Objective IV
Standardized Fractions and NDFIa) Paragominas,Pará State - 223/62
Soil
GV NDFI
NPV
Objective IV
b) Santarém, Pará State - 227/62
Soil
GVShade NDFI
NPV
Standardized Fractions and NDFI
Objective IV
c) Ji-paraná, Rondônia State - 231/67
Soil
GVshade NDFI
NPV
Standardized Fractions and NDFI
Objective IV
StandardizedNDFI (2001)
Rondônia
N
StandardizedNDFI (2001)
Summary
NDFI enhances the detection and mapping of degraded forests and performs better than any individual fraction.
CCA can unambiguously map canopy damage due to selective logging and forest fires.
These techniques can be applied over the Amazon region.