A stepwise approach to reference levels Louis Verchot, Arild Angelsen, Martin Herold, Arief Wijaya
Nov 13, 2014
A stepwise approach to reference levels
Louis Verchot, Arild Angelsen, Martin Herold, Arief Wijaya
A stepwise approach to FREL/FRLs
Criteria for comparing country circumstances and strategies
THINKING beyond the canopy
Deforestation/degradation drivers for each continent
39%
35%
13%
11%
2%
57%36%
2%4%
1%
41%
36%
7%
10%
7%
26%
62%
4% 8%
70%
9%
17%
4%
67%
19%
6%7%
Forest degradation driver
Deforestation driver
AMERICA AFRICA ASIA
Deforestation
Degradation
RLs using regression models– Simple, easy to understand and test new variables
– But, data demanding
– Predicting deforestation in a period: Pt – Pt+1, based on deforestation in the previous period Pt-1 – Pt and a set of other factors (observed at time t).
– Using structure (coefficients) from the estimated regression equation to predict deforestation in period Pt+1 – Pt+2, based on observed values at time t+1
5
2000 200920052004 2010
Estimated/Predicted deforestation Historical deforestation
Predictive model, based on structure from regression model
Regression model
Step 1 case for 4 countries using FAO FRA data
1985 1990 1995 2000 2005 2010 2015 2020 20250
500
1,000
1,500
2,000
2,500
3,000
3,500
Cameroon
Year
Fore
st
C s
tock (
Mt)
1985 1990 1995 2000 2005 2010 2015 2020 20250
2,0004,0006,0008,00010,00012,00014,00016,00018,000
Indonesia
Year
Fore
st
C s
tock (
Mt)
1985 1990 1995 2000 2005 2010 2015 2020 20250
300
600
900
1,200
1,500 Vietnam
Year
Fore
st
C s
tock (
Mt)
1985 1990 1995 2000 2005 2010 2015 2020 20250
10,000
20,000
30,000
40,000
50,000
60,000
70,000
80,000
Brazil
Year
Fore
st
C s
tock (
Mt)
Step 2: Brazil
Predict deforestation rates for legal Amazon2005- 2009
7
Category Regression coefficient
Deforestation rate (2000-2004) 0.395Trend variable -0.136 -0.145Deforestation dummy -0.373 -0.773Forest stock 2.18 4.756Forest stock squared -1.8 -3.826Log per capita GDP -0.034 -0.13Agric GDP (%GDP) 0.28 0.28Population density 0.081 -0.81Road denisty 0.039 0.076
R2 0.831 0.789N 3595 3595
Step 2: Vietnam
Predict deforestation rates 2005- 2009
8
Category Regression coefficient
Deforestation rate (2000-2004) 1.464Trend variable -0.006 0.003Deforestation dummy -0.011 -0.031Forest stock 0.067 0.260Forest stock squared -0.189 -0.463Population density -1.177 1.036Road denisty 0.004 -0.001
R2 0.515 0.052N 301 301
THINKING beyond the canopy
Preliminary conclusions Historical def. is key to predict future deforestation
– Coefficients below one simple extrapolation can be misleading
Some evidence of forest transition (FT) hypothesis– Robustness of FT depends on the measure of forest stock
FT supported when forest stock is measured relative to total land area, otherwise mixed results emerge
Other national circumstances have contradictory effects
Contradictory relationships may be linked to data quality and interrelations of econ. & institutions differ
9
THINKING beyond the canopy
In-depth case study: Indonesia definitions matter
FAO forest definition – minimum 10% crown cover, minimum 0.5 ha and minimum height 5 m
Indonesia national forest definition – vegetation cover dominated by intertwined tree crowns with canopy cover of more than 60%
Indonesia – vegetation cover dominated by trees, with canopy cover between 25 and 60% is defined as bush
Natural forest definition – no plantations
THINKING beyond the canopy
Forest definitions affect estimates of deforestation
THINKING beyond the canopy
Assessment of national REL/RL for Indonesia
SourceCumulative Emission from
LUCF 2000 -2009(in Gg CO2e)*
Methods
FAOStat 3,140,033 FRA country report(EF = 138 ton C/ha)
MoE - Second National Communication to UNFCCC 7,443,064 IPCC Guidelines 2006
Winrock International (Harris, 2012) 3,468,150 Carbon Book keeping model
(RS + Field)
MOF (official) 1,760,000 Approach 1 + NFI(Tier 1 or 2)
MOF + Saatchi (CIFOR) 1,811,396 Approach 1 + Global EF(Tier 1 or 2)
* does not include peat emissions and peat fire
Comparison of national deforestation estimates
Validation of deforestation maps of Indonesia
Source: Wijaya, et.al, (In prep)
Validation of deforestation maps
1990-2000
2000-2001
2001-2002
2002-2003
2003-2004
2004-2005
2005-2006
2006-2007
2007-2008
2008-2009
2009-20100
100
200
300
400
500
600
700
800
900
1000
Indonesia MOFOR Indonesia Hansen
Indonesia JRC Indonesia Mean
Annu
al D
efor
esta
tion
(x 1
000
ha)
Previous deforestation rates are good predictors of future rates
National Bali Java Kalimantan Maluku
& Papua Sulawesi Sumatera
Log his def.
0.942 0.781 1.270 1.059 1.187 0.563 1.032
R2 0.574 0.517 0.187 0.869 0.848 0.589 0.524
Num. of obs
372 32 114 43 25 47 111
Using deforestation rates in 2003 to 2006 to predict deforestation in 2006 to 2009
Including socioeconomic factors improves the regressions
National Bali Java KalimantanMaluku &
Papua Sulawesi SumateraLog his def. 0.289 0.507 0.532 0.277 0.662 0.299 0.116Forest stock 10.121 -2.019 27.345 23.192 1.166 14.658 18.523
Forest stock sq -8.829 2.342 -43.279 -19.797 6.328 -8.653 -19.510
Log District GDP per capita
1.432 0.456 -0.255 -0.038 0.381 -1.136 1.688
Agric. GDP 0.033 0.015 -0.027 0.032 0.002 0.004 0.069
Log Pop. den. -0.357 0.291 0.145 0.089 -0.738 -0.404 -0.853
Road density -2.816 -4.355 0.000 0.494 5.134 6.912 1.089
R-square 0.777 0.665 0.549 0.980 0.965 0.707 0.858Num. of obs 371 32 114 43 25 47 110
Observations so far…
Forest definition matters
Selection of minimum mapping unit is important to determine the smallest units of deforested areas
Different satellite image classification methods may result in different estimate
There are several useful approaches to integrating drivers of deforestation and forest degradation into assessments of RELs
Thank you