REDD Methodological Module
“Location and quantification of the threat of unplanned baseline
deforestation”
Version 1.0 – April 2009
I. SCOPE, APPLICABILITY, DATA REQUIREMENT AND OUTPUT PARAMETERS
Scope
This module provides methods for quantifying and locating the threat of unplanned
deforestation in the baseline case.
Applicability conditions
This module is applicable for quantifying and locating the risk of unplanned conversion
of forest land to non-forest land in the baseline case.
The forest landscape configuration can be either mosaic or frontier.
In case or the mosaic configuration, location analysis is not required if the carbon stock
is homogeneous in more than 80% of the project area. The module “Methods for
stratifying the project area” (X -STR) shall be used to determine whether this criterion
is satisfied.
Data requirements
Spatial data on historical deforestation within the reference region1 and data on spatial
driver variables that either increase or reduce the risk of deforestation must be
available to apply this module.
This module calls upon the following other VCS-approved Modules and Tools:
BL-UR
“Estimation of the baseline rate of unplanned deforestation” – Version 1.0
X -STR
“Methods for stratifying the project area”– Version 1.0
Output parameters
This module provides methods to determine the following parameter:
Parameter
1
SI Unit
Description
See the most recent version of the module “Estimation of the baseline rate of unplanned
deforestation” (BL-UR) for criteria to determine the boundary of the reference region.
1
%
RDef,loc,t
Risk of deforestation at the location loc within the reference
region at year t
II. PROCEDURE
The procedure is based on the assumption that deforestation is not random but a
phenomenon that occurs at locations that have a combination of bio-geophysical and
economic attributes that is particularly attractive to the agents of deforestation. For
example, a forest located on fertile soil, flat land, and near roads and markets for
agricultural commodities is likely to be at greater risk of deforestation than a forest
located on poor soil, steep slope, and far from roads and markets. Locations at higher
risk are assumed to be deforested first2.
This concept can be described empirically by analyzing the spatial correlation between
historical deforestation and geo-referenced “proxy driver” variables. In the previous
example, soil fertility, slope, distance to roads and distance to markets are the likely
spatial proxy driver variables (SDV) or “predisposing factors”. These variables can be
represented in a spatial data layer (or “driver map”) and overlaid on a map showing
historical deforestation using a Geographical Information System (GIS). From the
combined spatial dataset, information is extracted and analyzed statistically to
produce a map that shows the level of deforestation risk at each spatial location (=
“pixel” or “grid cell”). The risk at a given spatial location may change at the time when
one or more of the spatial driver variables included in the model will change, e.g. when
population density increases within a certain area or when infrastructure develops.
The basic steps needed to perform the analysis described above are:
STEP 0.
Selection of the procedure or model
STEP 1.
Preparation of proxy driver maps
STEP 2.
Preparation of risk maps for deforestation
STEP 3.
Selection of the most accurate deforestation risk map using an acceptable
validation metric
STEP 4.
Mapping of the locations of future deforestation
STEP 0. Selection of the procedure
The REDD project activity may be located:
1.
2
in a region for which no regional deforestation baseline has been determined yet
(Scenario 1); or
Several models and software have been proposed to analyze where deforestation is most likely to
happen in a future period. This methodology is inspired by the “GEOMOD” model; because, landuse and land-cover change modeling is an active field, all models that implement at least the steps
described in this module can be used.
2
2.
in a region for which a regional deforestation baseline has already been assessed
by a third party (Scenario 2).
If Scenario 1 applies:
Steps 1 to 4 must be applied.
•
If Scenario 2 applies:
•
If the third party that determined the regional deforestation baseline is
approved or sanctioned by the national or regional government, the existing
baseline must be used, unless it is not applicable according to the criteria
listed below.
•
If the third party is not the national or regional government, project
participants can decide not to use the existing regional deforestation baseline
if they consider that it does not reflect the baseline circumstances expected to
occur in the project area during the crediting period. In this case Steps 1 to 4
of this module will apply.
An existing regional deforestation baseline is applicable under the following
conditions:
a)
The regional deforestation baseline has been projected for a reference
region that includes the entire project area of the proposed REDD project
activity.
b)
If the area for which the existing baseline rate has been projected is larger
than the project area, the projected baseline must include the location of
the expected baseline deforestation, so that areas subject to baseline
deforestation can transparently be located within the project area. If no
location analysis exists, Steps 1 and 4 of this module must be applied.
c)
If the area of the region is equal to the project area, and:
d)
3
1.
A location analysis exists: use the existing location analysis to identify
the areas subject to baseline deforestation within the project area.
2.
No location analysis exists, and:
•
The landscape configuration is mosaic: assume that all locations
have about the same risk to be deforested3.
•
The landscape configuration is frontier: apply Step 1 to 4 of this
module
The existing regional deforestation baseline is applicable to the entire
period of time during which the project baseline must not be revisited (< 10
years), after which the deforestation baseline needs to be reassessed for its
continued applicability. If it has been determined for a number of years
fewer than the crediting period, Steps 1 to 4 of this module must be used
This assumption does not imply that the region cannot be divided in strata with different
deforestation rates per stratum.
3
for the years of the crediting period for which the existing regional
deforestation baseline is not applicable.
e)
Methods used to project the baseline deforestation rate are transparently
documented so that assumptions and data used to do the projections can
be verified. This provision does not apply in case of deforestation baselines
established by the national or regional government having adopted a REDD
scheme recognized by the UNFCCC or VCS.
f)
The existing regional deforestation baseline must either be:
•
independently validated by a VCS accredited verifier, or is registered
under a VCS acknowledged system, or has been established by the
national or regional government having adopted a REDD scheme
recognized by the UNFCCC or VCS, in which case it can be used;
or
•
it has been determined by an independent team and has been peerreviewed, in which case it can be used. If the previous two requirements
are not satisfied, VCS verifiers shall do an independent validation of the
existing regional deforestation baseline rate.
STEP 1. Preparation of proxy driver maps
Identify the spatial variables that most likely explain the pattern of deforestation in the
reference region, such as:
•
Landscape factors, e.g. vegetation type, soil fertility, slope, elevation, distance
to navigable rivers and water bodies, etc. (as relevant).
•
Human infrastructure, e.g. distance to roads, railroads, sawmills, settlements,
etc. (as relevant); and
•
Actual land tenure and management, e.g. private land, public land, protected
land, logging concession, etc. (as relevant).
Obtain spatial data for each variable identified and create digital maps representing
the Spatial Features of each variable. Some models, such as Geomod, will require
producing, for each of the digital maps, Distance Maps from the mapped features (e.g.
distance to roads) or maps representing continuous variables (e.g. slope classes) and
categorical variables (e.g. soil quality classes). For simplicity, let’s call these maps
“Factor Maps”. Other models do not require Factor Maps for each driver variable, and
analyze all the driver variables and deforestation patterns together to produce a risk
map.
Where some of the spatial proxy driver variables are expected to change, collect
information on the expected changes from credible and verifiable sources of
information and prepare different Factor Maps for the same spatial driver variable, to
represent the changes that will occur in different future periods.
In case of planned infrastructure (e.g. roads, industrial facilities, settlements) provide
documented evidence that the planned infrastructure will actually be constructed and
4
the time table of the construction. In case of new roads or road improvements,
provide credible and verifiable information on the planned construction of different
segments (e.g. how many kilometers will be constructed, where and when). Evidence
includes: approved plans and budgets for the construction, signed construction
contracts or at least an open bidding process with approved budgets and finance. If
such evidence is not available use one of the two following options:
•
Exclude the planned infrastructure from the driver variables considered in the
analysis; or
•
Adjust the baseline post facto, based on monitoring of actual infrastructure
development during each monitoring period.
To create the Factor Maps, use one of the following two approaches:
•
Heuristic approach: Define “value functions” representing the likelihood of
deforestation as a function of distance from point features (e.g. saw mills) or
linear features (e.g. roads), or as a function of polygon features representing
classes (e.g. of soil type, population density) based on local expert opinion or
other sources of information. Specify and briefly explain each value function in
the PD.
A useful approach to estimate value functions is to sample spatially
uncorrelated points in the Distance Maps and their corresponding location in
the maps representing historical deforestation and to use regression
techniques4 to define the probability of deforestation as a function of
“distance”.
•
Empirical approach: Categorize each Distance Map in a number of predefined
distance classes (e.g. class 1 = distance between 0 and 50 m; class 2 = distance
between 50 and 100 m, etc.). In a table describe the rule used to build the
classes and the deforestation likelihood assigned to each distance class5. The
deforestation likelihood is estimated as the percentage of pixels that were
deforested during the period of analysis (i.e. the historical reference period).
The empirical approach should be preferred over the heuristic approach. Use the
heuristic approach only where there is insufficient information about the spatial
location of historical deforestation or where the empirical approach does not produce
accurate results when validated against a historical period.
In the finalized Factor Maps, the value of a pixel must represent the deforestation risk
or, as an approximation, the percentage of area that was deforested during the period
of analysis in the distance class to which the pixel belongs.
4
e.g. logistic regression.
5
When building classes of continuous variables it is important to build classes that are meaningful in
terms of deforestation risk. This implies the parameterization of a “value function” based on specific
measurements. For instance, the criterion “distance to roads” might not have a linear response to
assess the deforestation risk: a forest located at 50 km from the nearest road may be subject to the
same deforestation risk of a forest located at 100 km, while at 0.5 km the risk may be twice as much
as at 1.0 km. Data to model the value function and build meaningful classes can be obtained by
analyzing the distribution of sample points taken from historically deforested areas.
5
STEP 2. Preparation of deforestation risk maps
A Risk Map shows, for each pixel location l, the risk (or “suitability”) of deforestation as
a numerical scale (e.g. from 0 = minimum risk to some upper limit representing the
maximum risk).
Models use different techniques to produce Risk Maps. The Geomod model produces
Risk Maps by calculating different weighted average combinations of the Factor Maps
prepared with the previous step. Choose different combinations of Factor Maps and
weights, taking into account expert opinion and the analysis performed in the previous
steps.
RDef ,loc,t
 N
  N

=  ∑WSDV * PSDV ,loc,t  ÷  ∑WSDV 
 SDV =1
  SDV =1

(1)
Where:
RDef,loc,t
Risk of deforestation at the location loc (pixel or grid cell) at year t; %
yr-1
SDV
A particular factor; number
N
Total number of factors; number
WSDV
Weight of the driver image SDV; %
PSDV,loc,t
Value of the grid cell of factor map SDV at location l and time t;
number
The weights (WSDV) of each Factor Map can be determined heuristically through expert
consultations or empirically using statistical analysis. For instance, Geomod-2 uses
non-linear multiple-regression to weight each Factor Map.
Other published models can also be used to produce Risk Maps.
STEP 3. Selection of the most accurate deforestation risk map
A model validation is needed to determine which of the deforestation risk maps is the
most accurate. A good practice to validate a model (such as a risk map) is “calibration
and validation”.
Model calibration and validation:
Two options are available to perform this task:
a)
calibration and validation using two historical sub-periods; and
b)
calibration and validation using tiles. To build tiles, divide the reference
region in n equal-area subsets.
Option (b) will be used when two historical sub-periods are not available for applying
option (a).
6
a)
Where two or more historical sub-periods are available, data from the most
recent period can be used as the “validation” data set, and those from the
previous periods as the “calibration” data set.
Using only the data from the calibration period, prepare for each Risk Map a
Prediction Map of the deforestation in the validation period. Overlay the
predicted deforestation with locations that were actually deforested during
the validation period. Select the Prediction Map with the best fit6 and identify
the Risk Map that was used to produce it. Prepare the final Risk Map using
the data from the calibration and the validation period.
b)
Where only one historical sub-period is representative of what is likely to
happen in the future, divide the reference region into tiles and randomly
select half of the tiles for the calibration data set and the other half for the
validation set. Perform the analysis explained above.
Briefly report in the PD the procedures used to select the most suitable Risk Map.
STEP 4. Mapping of the locations of future deforestation
Future deforestation is assumed to happen first at the pixel locations with the highest
deforestation risk value.
To determine the locations of future deforestation do the following:
•
Mask out all current “non-forest land” from the selected Deforestation Risk
Map7.
•
In the transformed Deforestation Risk Map select the pixels with the highest
value whose total area is equal to the area expected to be deforested in
project year one (or first monitoring period). The result is the Map of Baseline
Deforestation for Year 1 (or first monitoring period, respectively).
•
Repeat the above pixel selection procedure for each successive project year (or
monitoring period) to produce a Map of Baseline Deforestation for each future
project year (or monitoring period). Do this at least for the upcoming crediting
period and, optionally, for the entire project term.
•
Add all yearly (or periodical) baseline deforestation maps in one single map
showing the expected Baseline Deforestation for the Crediting Period and,
optionally, Project Term.
6
The map with the best fit will be the map that best reproduced actual deforestation in the validation
period. Parameters such as % of area of correct prediction, % of area of omitted prediction (area that
was actually deforested but not predicted), and % of area of commission (area that was predicted as
deforested but that was actually not deforested) can be used to identify the map with the best fit.
7
The GEOMOD model refers to these maps as “Potential for Land Use Change” (PLUC).
7
III. Data and parameters used and generated in this module
Data/parameter
Unit
Used in
equations
N
PSDV,loc,t
number
number
1
1
RDef,loc,t
%
1
SDV
WSDV
number
%
1
1
Descripiton
Source of
data
Measurement
procedure (if any)
Comments
Total number of factors
Value of the grid cell of factor map SDV at
location l and time t
Risk of deforestation at the location l (pixel or
grid cell)
A particular factor
Weight of the driver image SDV
8