Credible baseline setting and accurate and transparent Measurement, Reporting and Verification (MRV) of results are key conditions for the success of REDD (Reducing Emissions from Deforestation and forest Degradation). In this presentation, Britaldo Silveira Soares Filho from Universidade Federal de Minas Gerais discusses the challenges inherent in setting baselines and applying modelling for REDD. Britaldo Silveira Soares Filho gave this presentation on 8 March 2012 at a workshop organised by CIFOR, ‘Measurement, Reporting and Verification in Latin American REDD+ Projects’, held in Petropolis, Brazil. The workshop aimed to explore important advances, challenges, pitfalls, and innovations in REDD+ methods — thereby moving towards overcoming barriers to meeting MRV requirements at REDD+ project sites in two of the Amazon’s most important REDD+ candidate countries, Peru and Brazil. For further information about the workshop, please contact Shijo Joseph via s.joseph (at) cgiar.org

1. Delusional REDD baselines
il ho
F
Britaldo Silveira Soaress Filho
re
S oa
e ira
Si lv
al do
r it
B
MEASUREMENT, REPORTING AND VERIFICATION IN LATIN AMERICAN REDD+ PROJECTS
CIFOR WORKSHOP – MARCH 8-9, 2012 – PETRÓPOLIS, BRAZIL

2. Main concepts
• Additionality: Proof that reduction in emissionsil ho from
REDD is additional to reductions that would
F occur if
initiative were not in place. ar es
So
• Leakage: reduction in carbon emissions in one area
that results in increasedve
ira
l emissions in another
Si reduced emissions from
• Permanence: long-term
REDD. Depends on al do vulnerability to deforestation and/or
it
Br
degradation.
Concepts and methods borrowed from CDM projects

3. Measuring performance: the baseline approach
The CDM baseline
il ho
F
ar es
So
e ira
Si lv
al do
r it
Historical approach: introducing Forward-looking approach (ex ante):
B
an innovative process to an
installed industrial plant.
the building of a new factory with a
carbon neutral approach.
Depend strongly on the project and less on the context (Lesser risk of hot air).

4. REDD baselines
ho
30000 27772
km2
25000
Fil
20000
25396 19600
es
ar Baseline
So
19014 15405
18226
15000
14196 12911
ira
Reduction
10000 8935
Hitorical
lve7464
5000
Si
6238 6560
REDD 3805
do
0
t al
iapproach: Brazil´s
Br Change Plan.
Historical
National Climate
Forward-looking approach: Brazil overall GHG
target of 39% reduction by 2020 (emissions from
agriculture can increase from 480 to 627 MCO2)

5. What to do with countries or regions with low
deforestation rates and large extent of forests
Acre Madre de Dios Pando
il ho
F
ar es
So
e ira
Si lv
al do
r it
B
For more information: visit www.csr.ufmg.br/map

6. Forward-looking baseline
C stocks
12,000,000
il ho
F
10,000,000
ar es
So
8,000,000
ira
6,000,000
lv e with project
Si
4,000,000
Additionality = $$ credits
2,000,000
l do baseline
itaProject starts
-
time
Br

7. Forward-looking baseline
il ho
F
ar es
So
e ira
Si lv
al do
r it
B Project starts
Baseline is not solely based on the project itself, but on a reference
region. Therefore, there is a need to understand the effect of the project
on the deforestation regional trend.

8. Solution: Deforestation model??
BASIC STEPS OF MODELING il ho
F
ar es
1. understanding historical So trends
ira
ve
2. identifying drivers of deforestation
il
S
3. modelingofuture scenarios
ld
ita
Br A series of what if decisions !!!

9. Understanding historical trends
Modeling approaches for BAU baselines
il ho
160000 F
ha
140000
average
ar es
projected
Which one ?
120000 So (BAU)
simulated
100000
e ira Observed
• Historical fixed rates Si lv
80000
• Historical trend do 60000
it
• Business-as-usual al
simulation
r
B(future
40000
20000
scenarios)
0
time

10. Modeling deforestation trajectories
il ho
F
ar es
So
e ira
Si lv
al do
r it
B
2010
2007
Soares-Filho et al. 2010. PNAS.

11. Modeling scenarios
Soares-Filho et al. 2006. Nature
il ho
F
ar es
So
e ira
Si lv
al do
r it
B End of Deforestation in the Brazilian Amazon
The
Nesptad, Soares-Filho et al. 2009. Science.
Trajectories can change drastically

12. Other aspects
• Delimiting the geographic reference o
ilh region. F
ar es
So
ira
ve
Project
il
S
al doReference region
r it
B

13. Other aspects
• Delimiting the geographic reference o
ilh region. F
ar es road
So
ira
ve
Project
il
S
al doReference region
r it
B

14. Other aspects
• What if there are other projects???ho
il F
ar es
Project
So
e ira
Si lv
Project
Project
al doReference region
r it
B
Madre de Dios, Peru, have 12 or more REDD projects

15. identifying drivers of deforestation
Probability map of deforestation
deforestation up to 1994
distance to main roads
ho
deforestation up to 1999
overlay
F il
cross-tabulation
buffer zone
ar es
deforested non-deforested total
----------------------------------------
distance > 10 km | 194187
distance < 10 km | 84228
1005296 | 1199483
199349 | 283577
So
ira
----------------------------------------
total | 278415 1204645 | 1483060
deforestation from 1994 to 1999
(gray tone)
lv e
Si
al do
They are in fact spatial determinants r it
B
Most models only incorporate spatial determinants!!!

16. modeling future scenarios
• Spatially-explicit model of deforestation to quantify
where deforestation is likely to occur. o
Land cover map of 1986 Land cover map of 1994
Filh
Simulated land cover map of 1994
O
-09 50` O
-09 50`
ar esO
-09 50`
project So project
ira
O
-10 00` O
-10 00`
O
-10 00`
lv e
O
-10 10` O
-10 10` Si O
-10 10`
al do
O
-10 20`
r it deforestation is not random but
occurs at locations that have a
O
-10 20`
O
-10 20`
B combination of bio-geophysical
and economic attributes that are
more attractive to the agents of
deforestation
O
-10 30` O
-10 30`
O
-10 30`
True or false?
O O
-55 00 ` -54 50` O
-55 00` O
-54 50`
O
-55 00` O
-54 50`

17. Stochastic nature of deforestation
Absolute frequency of transition probabilities of observed deforestation
3500
scene 23167
3000
2500
il
Performance ho x realism
number of cells
2000
1500
F
1000
ar esdecreases realism
Increasing spatial performance
So
500
0
0 50 100 150 200 250
ira
probability value
Half true, half false
lv e
Si
al do
r it
B
Dinamica EGO simulates landscape structure

18. Validation only regarding location, but
training not spatial structure
t1 t2 t3
il ho
Land cover map of 1986 Land cover map of 1994
F
Simulated land cover map of 1994
O
-09 50` O
-09 50`
ar esO
-09 50`
So
project project
Model ira
O
-10 00` O
-10 00`
O
-10 00`
ilve
validation
O
-10 10` O
-10 10` S O
-10 10`
al do
O
-10 20`
r it O
-10 20`
O
-10 20`
B
O
-10 30` O
-10 30`
O
-10 30`
O O
-55 00 ` -54 50` O
-55 00` O
-54 50`
O
-55 00` O
-54 50`

19. Model performance assessment
• Map comparison methods (how to lie with validation methods)
il ho
F
ar es
So
e ira
Label 1
Costanza Si lv2
Power et al.
3
Pontius Hagen
4
Van Vliet
et al.
5 6
Almeida et
7
Almeida et
do
map author (1989) (2001) (2002) (2003) (2011) al. (2008)1 al. (2008)2
degraded 5x5 0.98 0.83 0.65 0.70 0.66 0.83 1.00
degraded 9x9
degraded 15x15
it al 0.97
0.96
0.81
0.81
0.48
0.34
0.47
0.24
0.48
0.34
0.62
0.46
0.75
0.55
B r
degraded 31x31
degraded 51x51
0.95
0.95
0.80
0.80
0.18
0.12
-0.03
-0.13
0.18
0.12
0.26
0.19
0.32
0.23
random 0.90 0.84 0.002 1.00 1.00 0.04 0.05
line1 x line2 1.00 0.81 -0.02 0.59 -0.02 1.00 1.00
Soares-Filho et al. in review
And model comparisons are quite subjective (e.g. Vega et al. 2011)

20. Different models have different
assumptions, geographic scales, spatial resolutions,
scopes, and, therefore, purposes.
il ho
F
ar es
So
e ira
Si lv
al do Thank you
r it Reimer
B
Both simulations developed using Dinamica EGO, but none of them were designed to fix
baselines

21. This means trouble
il ho
F
ar es
So
e ira
Si lv
al do
r it
B

22. What about leakage?
In-out, out-out, diffuse
leakage
il ho
(Fearnside, 2009) F
ar es
So
Model wizard for rapid e ira Is modeling for dummies???
assessment ilv
S
ld o
Leakage
Enter value
Effectiveness = Success rate – Leakage
Success
------------
50% =
Effectiveness
------------
70% - 20%
ita
Br
Representation of a fictitious software
Leakage cannot be simulated, but can be evaluated.
See Soares-Filho et al. PNAS. 2010

23. Good job &Business
il ho
F
ar es
So
• But!!!!
ira (although vendors say so)
ve
No ready solution lfor REDD
Si
al do software wizard…
There is no magic solution, via
it
Br
MODEL must be built from the
ground to incorporate local
knowledge…

24. In sum
• Although various MRV methodologies for mapping carbon stocks and modeling
il ho
reductions as well as standards for international certification have been
developed, the criteria for establishing baselines for crediting purposes are
F
unsound; they do not take into account that forward-looking baselines are
es
questionable because deforestation trajectories can alter drastically in response to
ar
So
changes in a complex set of circumstances (Nunes et al. 2012; Soares-Filho et al.
IADB report, 2012).
•
ira
Furthermore, it is very difficult to isolate the local effect of a project from the
e
overall trajectory of deforestation as well as to assess leakage arising from the
Si lv
establishment of projects, and none of those projects have attempted to perform
such analyses (Soares-Filho et al. IADB report, 2012).
•
do
As more specific standards and indicators are established, more flawed they
al
become, because they are unfounded anyway (Personal Perception). And for
r it
financial crediting, every dollar counts. In addition, conflict of interest might lead
B
to gamming (Fearnside, 2012).

25. So, how can we apply modeling to
REDD? (let’s keep us in business)
il ho
F
ar
Knowledge es
Intervention So Modeling
e ira
Si lv
Information
al do
r it
B
Tool for policy design and planning

26. Tool for REDD planning and management. Application to Acre
Dinamica EGO software
il ho
F
ar es
So
e ira
Si lv
al do
r it
B Simulation as a tool for regional planning
www.csr.ufmg.br/map
Method adopted by SISA

27. Setting priority areas for REDD
• Index of threat (time dependent) (Soares-Filho et al. PNAS,
2010) o lh
Fi
ar es
So
e ira
Si lv
al do
r it
B
Priorização de áreas protegidas

28. Calculating costs of reduction
Amazon opportunity costs
il ho
F
ar es
So
e ira
Si lv
Carbon cost along the l do
U$
ita
Carbon Cost
Br
30 deforestation frontier
25
20
15
10
5
0
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30
year
year
(Nespdad, Soares-Filho et al. 2009; Soares-Filho et al. 2010)

29. Assessing effectiveness and leakage over time
Only monitoring deforestation is
not enough.
il ho
F
Soares-Filho et al. PNAS . 2010.
ar es
So Soares-Filho et al. Env. Con. 2012.
Nunes,
ira
Odds ratio adjusted
Land use zones 1997 2000 2005 Mean
Urban and periurban areas 6.15 2.17
lv e5.40 4.57
Mining concessions
Small-scale cattle ranching
1.26
1.42 Si
1.62
1.20
1.51
1.28
1.46
1.30
do
Small landowner lots 1.02 0.99 1.55 1.19
Brazil nut concessions
Logging concessions
ital 0.76
2.84
0.67
3.28
0.49
4.29
0.64
3.47
Native communities
Br 0.77 1.20 1.35 1.11
Reforestation concessions 0.19 0.39 0.93 0.51
Conservation concessions 3.85 2.78 2.06 2.89
Ecotourism concessions 0.08 0.08 0.76 0.31
Protected areas 4.85 8.27 5.55 6.22
Indigenous reserves 0.20 0.02 x 0.11
Unassigned land use 0.96 0.94 1.17 1.02

30. Which REDD projects in Madre de Dios can make a difference?
il ho
F
ar es
So Nunes, Soares-Filho et
al. Env. Con. 2012.
e ira
Si lv
al do
r it
Hajek et al. 2011
B

31. Channeling REDD+ investments
il ho
F
ar es
So
REDD funds
invested to
e ira upgrade the
production
Si lv chain of non-
timber forest
al do products
r it
B
Nunes, Soares-Filho et al. Economic benefits of forest conservation:
assessing the potential rents from Brazil nut concessions in Madre de Dios,
Peru, to channel REDD+ investments. Environmental Conservation, 2012.

32. Thank you/Obrigado/Gracias
Britaldo Silveira Soares Filho
il ho
(the culprit for developing deforestation models that are now being applied to REDD+)
britaldo@csr.ufmg.brs
F
re
S oa
For many more examples, please access
ira
ve
http://www.csr.ufmg.br/dinamica/publications/publications.php
il
S
al do
it
Br
Modeling in support of sound policy
Dinamica EGO freeware can be downloaded at www.csr.ufmg.br/dinamica