The document summarizes discussions from a Global Forest Watch partnership meeting about advancing forest monitoring technologies. It covered several topics:
1) Differentiating forest types and disturbances using remote sensing to provide more detailed context for tree cover loss. This included classifications for managed vs. natural forests, primary vs. degraded forests, and stand-replacement vs. non-stand-replacement disturbances.
2) Increasing the spatial resolution of forest monitoring using data from Planet Labs that can image the entire land surface of the Earth nearly every day. This high frequency of observations allows for more accurate tracking of infrastructure development and natural disasters.
3) Increasing the temporal frequency of monitoring using multi-sensor approaches that combine Landsat, Sentinel
2. #GFWPartners17Presenters: Mikaela Weisse, Doug Muchoney, Sasha Tyukavina, Joe Mascaro and Johannes Reiche
PARALLEL DISCUSSIONS 3:
THE NEXT FRONTIER OF FOREST MONITORING
3. The Next Frontier of
Forest Monitoring
GFW Partnership Meeting
February 9, 2017
Mikaela Weisse, Research Analyst
4. GFW’s Vision
“Global Forest Watch uses cutting
edge technology and science to
provide the timeliest and most precise
information about the status of forest
landscapes worldwide.” and most
useful!
5. Forest Monitoring Systems
PRODES DETER
Terra-I
Global Forest Change/
Hansen
FORMA
GLAD alerts
Update frequency
Coverage
Country-levelGlobal
Annual Weekly
?
6.
7. Limitations
• Too coarse to find on the ground
• Cloud cover limits frequency
• Accuracy is unknown in my area of interest
• Doesn’t define forest, only tree cover
• Treats all tree cover as equal
• Treats all tree cover loss as equal
• Can’t detect forest degradation
8. Where do we go from here?
Tech providers:
• What is the “cutting edge” in
forest monitoring?
• What is feasible?
Users:
• What advances in forest
monitoring would be most
beneficial in your work?
9. Agenda
• Lightning updates from partners: new frontiers of forest monitoring
• Defining forest vs tree cover – Doug Muchoney, FAO
• Differentiating types of forest and forest disturbance - Sasha Tyukavina, UMD
• Increasing spatial resolution – Joe Mascaro, Planet
• Increasing temporal frequency - Johannes Reiche, Wageningen University
• User priorities: sticky note exercise
• Karimah Hudda, Mondelez
• Morgan Erickson-Davis, Mongabay
• Discussion
• Wrap-up
10. Defining forest vs tree cover
Global Forest
Resources Assessment
(FRA)
Douglas Muchoney – Chief, FAO, Forest
Policy and Resources Division
11. Global Forest Resources Assessment (FRA)
Definitions
• FAO-FRA:
Forest: Land spanning more than 0.5 hectares with trees higher than 5 meters and a canopy cover of more
than 10 percent, or trees able to reach these thresholds in situ. It does not include land that is
predominantly under agricultural or urban land use.
Source: FRA 2015 Terms and Definitions
• University of Maryland (UMD)/World Resources Institute (WRI):
Tree cover: Trees are defined as woody vegetation taller than 5m in height and are expressed as a percentage per
output grid cell as ‘2000 Percent Tree Cover’. ‘Tree Cover Loss’ is defined as a stand-replacement disturbance, or a
change from a treed to non-treed state, during the period 2000–2014. ‘Tree Cover Gain’ is defined as the inverse of loss,
or a non-treed to treed change entirely within the period 2000–2012.
Source: University of Maryland
Forest land vs tree cover (1/2)
12. Global Forest Resources Assessment (FRA)
Forest land vs tree cover (2/2)
• The FRA reports forest land use area, and the data reported are official
statistics submitted to FAO by the countries. Data collection is based on
a combination of methods including national forest inventories, remote
sensing, aggregated local-level reporting and expert opinion. The last
FRA reports data from 1990 to 2015 and allows calculation of net forest
area change between different reporting years.
• UMD/WRI data report remote sensing-based estimates on annual tree
cover loss, during the period 2000–2014, which can be disaggregated
according to tree cover density classes. It also reports tree cover gain
within the period 2000–2012, but does not recommend aggregating
loss and gain to calculate net change.
13. Global Forest Resources Assessment (FRA)
Forest land use approach:
each of these stages is
considered forest
Tree cover approach:
only young secondary or
older is considered as tree
cover
Forest use vs tree cover
14. Global Forest Resources Assessment (FRA)
The forest or the trees?
IS THIS FOREST (YES =) ?
Tree cover...
FRA
Other sources,
Remote Sensing
only
...in agricultural production systems (oil palm
plantations, coffee plantations, etc.) X
...on land that is predominantly under agricultural
or urban land use X
...temporarily removed as part of a forest
management scheme X
...temporarily lost through natural disturbances
X
Newly established forest
X
15. Global Forest Resources Assessment (FRA)
• The Global Forest Resources Assessment (FRA)
has reported on status and trends on global forest
resources for national and international decision-
makers and the public at large since 1948.
• First report: 1948
• Responsible of the FRA Programme: FAO Forestry Department
• Frequency of the most recent reports: Every 5 years
• Website: http://www.fao.org/forest-resources-assessment/
16. Global Forest Resources Assessment (FRA)
FRA goes beyond forest area estimates …
7 Thematic Elements of Sustainable
Forest Management of FRA
1) Extent of forest resources
2) Biological diversity
3) Forest health and vitality
4) Protective functions of forest resources
5) Productive functions of forest resources
6) Socio-economic functions
7) Institutional and legal framework
• 110+ variables
• 234 countries
• FRA 2015 time series
covers the last 25 years
17. Global Forest Resources Assessment (FRA)
Data sources and partners
Country Reports
- Governments through National Correspondents
- International and regional organizations and processes through the
Collaborative Forest Resources Questionnaire (CFRQ) initiative:
- Central African Forest, Commission (COMIFAC/OFAC)
- FAO Forestry (FRA)
- FOREST EUROPE
- International Tropical Timber Organization (ITTO)
- Montréal Process
- United Nations Economic Commission for Europe (UNECE)
Remote Sensing
FAO with:
- Joint Research Centre of European Commission
- Regional partners
- National focal points and specialists from the countries
21. Global Forest Resources Assessment (FRA) 21
Towards FRA 2020 – reporting, review, analysis and dissemination
INTERACTIVE
DATA ENTRY
ON-THE-FLY
LOGICAL CHECKS
NATIONALDATA
REVIEW AND
COMMUNICATION
REPORTING
HARDCOPY
ONLINE
MAPS
DB LINKS
REPORTS
ON
REQUEST
NATIONAL DATA
REPOSITORY
22. Global Forest Resources Assessment (FRA) 22
NATIONAL DATA
RS FIRES
RS TREE COVER
PROTECTED
AREAS
ECOZONES
ADMINNFI MAPS
FREE TOOLS
ANALYSISACCESS
UNFCCC
SDG
REGIONAL REPORTING
PROCESSES
FAOSTATINCREASED CONSISTENCY
REDUCED REPORTING BURDEN
IMPROVED QUALITY
RELEVANCE
Towards FRA 2020 – reporting, review, analysis and dissemination
24. Alexandra (Sasha) Tyukavina, Matthew Hansen, Peter Potapov,
Svetlana Turubanova, Alexander Krylov, Marc Steininger, Belinda Margono
The 4th annual Global Forest Watch Partnership Meeting
Washington D.C., February 8-9 2017
Differentiating types of forest and forest disturbance
25. Background
Tree cover and loss products don’t differentiate between the
types of forest and forest disturbance;
Value-added analysis is needed to adequately interpret tree
cover loss products for carbon accounting and other purposes;
There are no standard classifications of forest and disturbance
types detectable via remote sensing.
26. Forests: by management type
Natural (unmanaged) Human-managed
Natural forest loss in Mato Grosso, Plantation clearing, Parana, Brazil
• Primary forests
• Mature secondary forests
• Natural woodlands
• Forest plantations
• Agroforestry systems
• Secondary regrowth in shifting agriculture
27. Why this matters:
45% of gross 2000-2012 tropical forest loss and 42% of gross aboveground
carbon (AGC) loss comes from human-managed forests (Tyukavina et al.
2015)
Forests: by management type
28. Primary intact Primary degraded
Undisturbed and unfragmented mature natural
forests retaining natural structure and
composition
Fragmented or subjected to forest utilization
(e.g. selective logging) that have led to partial canopy
loss and altered forest structure and compositionIntact Forest Landscapes (Potapov et al. 2017),
Hinterland forests (Tyukavina et al. 2016)
Selective logging in primary forest, Brazil
2006
Forests: by degree of degradation
29. Why this matters:
Fragmentation edge effects and selective logging increase humid tropical
forest susceptibility to fire (Cochrane 2003; Cochrane & Laurance 2008)
Forests: by degree of degradation
Fire-degraded primary forest fragment in São Félix do
Araguaia municipality in the state of Mato Grosso, Brazil
Photo from INPE Fototeca
http://www.obt.inpe.br/fototeca/fototeca.html
30. Why this matters:
98% of primary forest loss in Indonesia occurs within primary degraded
forests (Margono et al. 2015) -> logging often precedes conversion
Forests: by degree of degradation
32. Why this matters:
In 2013 in Brazilian Legal Amazon clearing of primary forests accounted only for 47%
of forest disturbance area (Tyukavina et al., in review), compared with 70% in 2003.
Forest disturbance types
33. Post-deforestation land-use
Forest
Palm estates
Tree plantations
Small-holder land use
Mining
Road
Settlement
Other
Forest land
Grassland
Cropland
Wetlands
Settlements
Other
IPCC Land Use classes
Custom classifications
(national- or regional-scale)
Example: Indonesia
35. Conclusions
Differentiating types of forest and forest disturbance allows to
place tree cover loss stats into a more specific context;
Non-primary forest loss is an increasingly large proportion of the
tree cover and carbon dynamics;
There is an ongoing effort to map forest types (e.g. primary vs.
non-primary forests) and forest disturbance types (e.g. fire vs.
non-fire forest loss);
Sampling can be used to supplement wall-to-wall tree cover and
loss mapping to identify types of forest, forest disturbance and
post-deforestation land-uses, as well as to assess accuracy of
forest and disturbance type maps.
40. The Planet
Explorer
Noor solar facility in
Morocco:
Planet imagery is
tracking its
construction nearly
each day, with
recent looks Dec 15,
14, 13, 10, 9, 8…
50. Some of Our
Ambassadors:
Andreas Kääb and Bas Altena –
University of Oslo
Greg Asner –
Carnegie Airborne Observatory
Ulyana Horodyskyj –
Science in the Wild
Matt McCabe –
King Abdullah University of Saudi Arabia
Doug Edmonds and Sam Roy –
Indiana University
Matt Finer –
Amazon Conservation Associtaion
59. Dimensions of Accuracy
1. Amount of forest cover
change
2. Amount of forest carbon
change
3. Type of change (forest
loss, deforestation,
degradation, disturbance)
4. Cause of change (mining,
logging, selective logging,
agriculture, etc.)
Deforestatio
n
Disturbance
Degradation
Forest loss
66. Illegal gold mining in Peru: Planet imagery resulted in news coverage and later intervention to destroy illegal mining equipment.
In the Peruvian
Amazon
rainforest, an
illegal gold
mine
encroaches
into the
Tambopata
National
Reserve.
Matt Finer
67.
68. Johannes Reiche
Wageningen University & Research, The Netherlands
Credits to: Martin Herold, Jan Verbesselt, Eliakim Hamunyela, Dirk Hoekman
Next Frontier of Forest Monitoring
“Increasing temporal frequency”
70. Landsat observations for Peru, 2014 (Hansen et al, 2016)
Potential observations Cloud free observations
71. Temporal frequency
Annual Monthly
How to increase the temporal frequency?
30 m
Landsat (GLAD)
Weekly
Near real-time capacity
dense cloud cover sparse
72. 3
Optical and RADAR satellites
BOLD Free data access (G) Global acquisition strategy
Landsat
+ 40 years of data & open archive
+ Ready-to-use data
Limited RADAR capacity in the past
- Commercial data distribution
- Fragmented archives
74. The “game changer” Sentinel-1
BOLD Free data access (G) Global acquisition strategy
Sentinel-1
+ First time dense RADAR time series for the tropics
+ Open data access
75. Sentinel-1 global acquisition strategy (from 10/2016)
https://sentinel.esa.int/web/sentinel/missions/sentinel-1/observation-scenario
76. 2015-09-07 0 10 km
Sentinel-1 data for Santa Cruz, Bolivia
2014-10-18
0 5 km
2015-09-07
0 5 km
Logging
roads
Logging
patterns
79. Temporal frequency
Annual Monthly
Multi-sensor approaches
30 m
Landsat + S2
Landsat (GLAD)
Weekly
Near real-time capacity
+
S1
Level of operationalisation
dense cloud cover sparse
+
24 revisit 12
(days)
Landsat + S1
80. Santa Cruz, Bolivia
Tropical dry forests (10.000 km²)
Industrial logging
Data
Landsat
Sentinel-1
ALOS-2 PALSAR-2
Methods
Probabilistic approach for time series
combination and NRT change detection
(Reiche et al., 2015)
Multi-sensor near real-time deforestation monitoring
(Reiche et al., under review)
83. Temporal accuracy
(Mean time lag of detected changes)
Landsat = 60 days
Sentinel-1 = 27 days
Multi-sensor = 19 days
Detected deforestation
(10/2015 - 09/2016)
84. Temporal frequency
Annual Monthly
Opportunities and key challenges
30 m
Landsat + S2
Landsat (GLAD)
Weekly
Near real-time capacity
Landsat + S1
+
S1
Level of operationalisation
24 revisit 12
(days)
dense cloud cover sparse
+ Near real-time
(24 | 12 days guaranteed obs. )
+ Deforestation, Degradation (?)
- RADAR pre-processing
- Image co-registration
85. Increasing free RADAR data stream
BOLD Free data access (G) Global acquisition strategy