Assessing forest cover from the air and space

1
Assessing forest cover from the air and space

• Brian Turner

2
Group Work

• How well equipped is your country to use remotely
  sensed data?
• How will remotely sensed data fit into your REDD
  plans?

3
How well equipped is your country to use remotely
sensed data?

• Human resource
• FA GIS/RS unit 8people
• GIS specialist 3 people but need more training,
  especially skill in data interpretation and the
  use of other algorithms in data processing
• Data Currently has satellite data for 2002-2006
  (Landsat ETM) Need more time series satellite
  image data with higher resolution.
• Financial support for Improving GIS unit within
  the Forestry Administration to monitor forest
  cover change on regular basis.

4
How will remotely sensed data fit into
Cambodia REDD plans?

• Remote sensing data might be fit well as we
  already have a GIS unit that has done forest
  cover change analysis. But the current unit need
  to be improved for enabling the system to do
  forest monitoring in regular basis

5
Laos

• Use of RS data
• Use is made of Spot 1, 2, 3, 4, ALOS
• Several institutes are involved
• Forest Department has RS/GIS facility
• Capacity must be upgraded, training, software
• RS REDD
• RS for baseline survey, forest classification
• Pilot studies for carbon stock determination in
  shifting cultivation areas

6
Nepal

• Use of RS
• Institutional Department of Forest Survey, well
  equipped, not enough resources, technical experts
  in RS and GIS but need more training
• Monitoring systems, but not compatible with REDD
• Financially not a priority, expensive to afford
• REDD RS
• Identification of priority areas
• Identification of deforestation and degradation

7
Vietnam

• Use of RS
• Forest inventory started long ago, 1950s, with
  aerial photography in northern Vietnam
• In 1990s forest assessment and monitoring, using
  RS Spot 1
• 1996-2000 use of Landsat ETM
• 2004 Landsat ETM
• Ongoing program with plans for 2005-2010
• Also use of QB, Aster, PALSAR, etc
• Software Erdas Imagine, ENVI, ILWIS
• RS for REDD
• Spot-5 for baseline survey in 10 out of 45
  provinces area, volume
• RS for biomass in the future, needs more research
• Capacity building in latest technology, also in
  univerities
• Knowledge on biomass assessment required

8
Why assess forest cover?

• Easiest forest attribute to measure remotely from
  above (air or space)
• Measure of deforestation (forest cover goes from
  100 to 0) and degradation (forest cover goes
  from 100 to between, say, 20 and 80)
• But is it human-induced or natural decrease in
  health?
• Difficulties
• Comparability over time
• Scale (resolution) needs to be right
• Image problems clouds, topographic shadowing
• Relating to ground estimates

9
Forest cover, degradation deforestation
FC60
Degradation
Deforestation
FCgt100
FCgt0
10
Aerial photography

• Positives
• 110 000 to 180 000 OK for interpreting forest
  cover
• Past (historic) images often available
• Easily handled manually
• Familiar view (especially if natural colour)
  stereo
• Can be collected on demand (avoiding cloud cover)
• Negatives
• Geometric distortions make transfer to maps
  difficult
• Usually, limited to visible band of spectrum
• Difficult to handle large numbers and to automate
• Manual interpretation open to bias
• Quite expensive unless multi-purpose

11
Satellite imagery

• Positives
• Minor geometric distortions easily corrected
• (Relatively) Easily transferred to maps
• Measurement of forest cover easily automated -
  unbiased
• Not limited to visible region of spectrum
• Wide range of sensors and resolutions now
  available
• Now (except very high resolution) relatively
  cheap
• Historical data back to 1972
• Negatives
• Rarely can collect on demand (so cant dodge
  clouds, etc)
• Satellites have limited life no guarantees of
  carry-on
• Each satellite/sensor has unique
  characteristics/challenges

12
Remote sensing systems for REDD

• What kinds of remotely sensed data are needed for
  REDD projects?
• Reasonably high resolution (1 ha or better)
• Spectral bands that can detect vegetation changes
  and differences (e.g., visible, NIR, radar)
• Previous historical data needed
• At least annual, cloud-free complete coverage
• Assured continuity of data-types
• No one system meets all these criteria but
  Landsat and SPOT come closest

13
Landsat

• Common characteristics
• Altitude about 700 km
• Repeat cycle 18-16 days vertical views
• Sun-synchronous (about 930am), near polar orbit
• Global coverage, now mostly available at nominal
  cost
• Landsat MSS (multispectral scanner) (NASA, USA)
• First 1972-85 Landsats 1, 2, 3 4
• Pixel size about 60m X 80m
• 3 visible (blue, green, red), 1 NIR bands
• Landsat TM (Thematic Mapper)
• 1982- Landsats 3, 4 5
• Pixel size 30 m
• 7 bands (3 visible, 3 IR, 1 TIR)
• Landsat ETM (Enhanced TM plus)
• 1999- Landsat 7 rest similar to TM but 15m
  panchromatic added

14
(No Transcript)
15
SPOT (Satellites pour lObservation de la Terre)
(France)

• Common characteristics
• Altitude about 800 km
• Repeat cycle 26 days but pointable (oblique)
  views possible
• SPOT 1, 2 3
• 1986-
• Resolution 20 m MSS 10 m Panchromatic
• 3 spectral bands (2 visible, 1 NIR)
• SPOT 4
• 1998-
• Extra 20m band in MIR
• SPOT 5
• 2002-
• 4 MSS bands with 10m resolution panchromatic
  with 2.5/5m resolution
• Stereo-pairs (N-S) possible

16
SPOT 5 data bands 3 (NIR) 4 (MIR)
17
Other possible RS systems

• MODIS (NASA)
• Launched 1999 one-off
• Frequent coverage (1-2 days), so easier to find
  cloud free
• Much larger pixels than L/S (250 m 1 km)
• 36 bands over visible to MIR (better
  discrimination)
• ALOS PALSAR (Japan)
• Launched 2006 (10 years after JERS-1)
• 10 m resolution
• L-band active radar
• Excellent delineation of topography, cuts through
  cloud smoke
• Indian and Chinese Landsat-look-alikes
• IRS-1A-D, first in 1988
• Chinese launches in 2006/7

18
Classification of multispectral data

• Supervised classification
• Locate a number of typical sites on the ground,
  e.g. mature natural forest, recently disturbed
  forest, secondary forest, plantation, rice field
• Determine average spectral pattern (signature)
  for each
• Examine every pixel in scene and classify it
  according to the closest signature and map the
  scene accordingly
• Unsupervised classification
• Examine every pixel and put them into classes
  according to how close they are to each other
• Determine the class means and limits and map the
  scene
• Use ground truth information to give the
  classes a label.

19
(No Transcript)
20
More complex classification

• Classify groups of pixels rather than individual
  ones, e.g., stands rather than trees
• Classify based (partly) on texture, i.e.,
  relationship of pixel to its neighbours
• Classify based on time-series (multi-date)
  information, e.g. if a pixel shows the following
  (annual) temporal pattern
• WWWWRWWW -? mature forest
• WWBBRRRWR -? regenerating forest
• WWBBAAABAA -? cleared forest
• WWRRWWRW -? degraded forest
• where Wforest, Bundergrowth, Rregrowth,
  Afarming

21
A REDD monitoring system based on Landsat

• Basic characteristics
• Baseline dataset
• Annual updates
• Classification of each pixel according to whether
  woody vegetation or not
• Multi-date algorithm to determine whether forest
  or not
• More detailed analysis to determine whether
  degradation has occurred and degree
• Models to determine change in biomass ? C.

22

• Australian National Carbon Accounting System-
• Forest subsystem

23
REDD monitoring system more details

• Desirably should be part of a national carbon
  accounting system for consistency
• Only way to detect (within-country) leakages?
• What remote sensing data are most suitable?
• Overall, Landsat TM (or something like it) seems
  best appropriate resolution spectrally
  spatially, temporal frequency OK, continuity
  probable, historical data available,
  idiosyncracies known
• But radar (like PALSAR) may be necessary if
  clouds are a problem and to detect (degree of)
  degradation (experiments on this currently in
  Tasmania Kalimantan)

24
PALSAR data over the Amazon(2006)
(http//www.eorc.jaxa.jp/ALOS/img_up/l_jers_palsar
_amazon_e.htm)