Image Processing Steps

1
Biodiversity Implications of Forest Disturbance
and Related Landscape Dynamics in the Brazilian
Amazon
Mark A. Cochrane1,2, David P. Roy1, Carlos Souza
Jr.2, Jos Barlow3, Eugenio Arima4, Izaya Numata1,
Christopher P. Barber1, Luiz Mestre1, Rafael
Andrade1, and Sanath Kumar1
1 Geographic Information Science Center of
Excellence, South Dakota State University,
Brookings, SD USA 2 IMAZON, Instituto do Homem e
Meio Ambiente da Amazônia, Belém, PA Brazil 3
Lancaster University, Lancaster United Kingdom 4
Hobart and William Smith Colleges, Geneva NY USA
2
What is the project?

• The fundamental hypothesis underlying this
  project is that the biodiversity levels of
  Amazonian forests are strongly related to two
  competing factors forest disturbance and time
  since last disturbance

3
The Brazilian Amazon
Amazon humid tropical forest biomes 6.4 km2
Portion in Brazil 4 km2
Source WWF
4
(No Transcript)
5
Human Access to Forest
85 of deforestation within 50 km of main roads
73,000 km of official roads in region
240,000 km of unofficial roads
Expansion rates gt 40 km / 10,000 km2 / year
Source IBGE, IMAZON
6
(No Transcript)
7
(No Transcript)
8
(No Transcript)
9
(No Transcript)
10
(No Transcript)
11
(No Transcript)
12
(No Transcript)
13

• Forest wildfires Interact with ongoing threats
  to the Amazon

14
How do we propose to test this?

• Our approach is straightforward
• 1) Determine recent forest disturbance history
  across the Brazilian Amazon (2000-2009)
• 2) Conduct extensive field studies of indicator
  taxa, stratified by disturbance history, to
  determine biodiversity responses
• 3) Model the determinants of fire ignition and
  fire spread
• 4) Predict the current and future levels of
  biodiversity similarity in disturbed forests
  spatially across the Brazilian Amazon.

15
Phase 1. Imagery Acquisition and Processing
16
Image Processing Steps
(2) Build Spectral Library
(1) PRE-PROCESSING
Image Registration Radiance Conversion
Estimate Visibility and water vapor
Correct Haze?
Reflectance Space
Atmospheric Correction (ACORN)
No
Yes
Apply Carlottos Technique
Pixel Purity Index - (PPI)
(4) Enhance and Detect Canopy Damage
(3) SMA
40 million pixels
NDFI
CCA
Visualization
SVDC
NDFI 0.75
Scatter matrix Spectral curves
Extract Patios
Generic Image Endmembers
Canopy Damage
Soil 10 1 pixel Area 4 pixels
GV NPV Soil Shade 1
Souza Jr. et al. (2005), RSE
17
Haze Correction
Contaminated Image
Corrected Image
Ji-Parana, 231/67 R3, G2, B1
18
Normalized Difference Fraction Index
Souza Jr. et al. (2005), RSE
19
NDFI
226/68 - 2001 (Sinop - MT)
20
NDFI
226/68 - 2000 (Sinop - MT)
21
NDFI
226/68 - 2001 (Sinop - MT)
22
NDFI
226/68 - 2003 (Sinop - MT)
23
Mapping Forest Damage History
24
Characterizing Forest Fragmentation
25
Age map

• Derived from Landsat time series
• Used for calculation of fragmentation features

Time series Land cover map
Age map
S.G.Forest
26
Persistence of Forest Edge (Ariquemes)
Remaining edge
27
Class 1
28
Phase 2. Biodiversity Field Studies

• The spatial database of forest disturbance is
  used to stratify and interpret our field studies
  investigating the response of 4 major indicator
  taxa (birds, dung beetles, trees and ants) as a
  function of disturbance history and time since
  last disturbance.

29
Both birds and dung beetles are good indicators
of community change in most other faunal groups
Barlow et al. 2007 PNAS
30
Both birds and dung beetles hold large numbers of
species that are restricted to primary forest
Many wide ranging species
Mostly territorial or habitat specialists
Barlow et al. 2007 PNAS
31
Both birds and dung beetles are highly cost
effective to sample you get good information on
habitat integrity for a low cost
Gardner, Barlow et al. 2008 Ecology Letters
32
Selecting effective biodiversity indicators

• Wide variability in cost of surveying different
  taxa
• Some taxa (e.g. birds and dung beetles) are of
  high performance for evaluating forest management
  systems in the Brazilian Amazon because they
• Are sensitive indicators of changes in forest
  integrity
• Can be surveyed cost effectively
• Ants share similar ecological attributes as birds
  and dung-beetles
• They are cheap to sample
• Species rich, with many different functional
  groups
• Most do not move large distances from their
  colonies (if you find them in a habitat, they
  come from that habitat).

33
Methods Dung Beetles

• Baited pitfall traps (human faeces)
• 5 traps per transect, run for 4 days.
• Follows methods discussed at global Scarabnet
  meetings, and used effectively in many Amazonian
  studies
• Complemented by un-baited flight intercept traps

34
Fire-mediated dieback and compositional cascade
Barlow and Peres 2008
35
POINT COUNTS
Observation and recording 10 point counts per
day per site 0630 h - 0900 h 10 min. bird
observation and recording, spaced at least 150m
each other, intending to avoid double
bird-counting (Parker, 1991). Along each
mist-net transects and other trails.
36
MIST NETS
Less individuals sampled compared to point
counts BUT It is independent of observer
accuracy Capture mainly understory birds Samples
species that are not singing Possibility to mark
(banding) and measure
37
MEASURING We will measure
Wings Tarsus
Bill Weight
38
Expected Results
A large-scale comparison of effects of fire on
Amazonian bird communities. A long-term
comparison of effects of fire on Amazonian bird
communities. One of the best overviews comparing
Amazonian bird communities in different Amazonian
States (500 sp).
39
Phase 3 Characterizing Fires (Starting June 2008)

• Integration of Landsat based analyses of forest
  disturbance with MODIS-derived fire products will
  enable us to accurately separate fires into their
  three main types
• (1) deforestation fires, where slash is burned,
  creating relatively hot fires that burn for
  several hours
• 2) maintenance fires, which rapidly burn as
  narrow fire lines through grass and early second
  growth
• 3) forest fires, escaped fires in standing
  forests which vary from extremely low intensity
  in previously undisturbed forests to high
  intensity in previously burned or logged forests

40
Phase 4 Spatio-socioeconomic modeling (starting
June 2008)

• We will use spatial regressions of economic
  (farmgate prices for soybean and beef),
  physical-geographic (precipitation, soil types,
  vegetation types, distance to previous
  deforestation, and land protection status (e.g.
  indigenous lands, conservation units)) and land
  cover (disturbance history) factors to model
  probability surfaces of fire ignition and fire
  spread.
• MODIS fire detections will be used to validate
  the ignition event model for 2000-2009 and the
  composite burned area product (Phase 3) will be
  used to validate the fire spread model over the
  same time period.
• Once validated, the models will be run using
  likely economic and rainfall scenarios to create
  spatio-temporal predictions of disturbance
  frequency and expected biodiversity impacts for
  the 2010-2019 time period.

41
Reprise of Project Objectives

1. Develop a basin-wide spatial database of all
   forest disturbance (selective logging,
   fragmentation, fire, deforestation) from
   2000-2009, based on NDFI analyses of annual
   Landsat imagery.
2. Derive regional estimation functions of expected
   biodiversity similarity based on disturbance
   history (disturbance metric) and
   time-since-last-disturbance (resilience metric)
   derived from stratified field data collected for
   four separate taxa (woody plants, birds, dung
   beetles and ants).
3. Develop a basin-wide spatial and temporal
   datasets of all fires by type (1) deforestation
   fires 2) maintenance fires 3) forest fires,
   using MODIS and Landsat data.
4. Model economic, physical-geographic and land
   cover factors affecting fire ignition and spread
   from 2000-2009 to create probability surfaces of
   fire ignition and fire spread.
5. Create a basin-wide map of probable biodiversity
   alterations in current standing forests across
   the Brazilian Amazon and predictions of future
   changes in these conditions over the next 10
   years (2010-2019) based on likely economic and
   climate scenarios. (Starting late 2009)

42
(No Transcript)