griscom_rel_e_kali
-
Upload
vdg777 -
Category
Technology
-
view
210 -
download
0
Transcript of griscom_rel_e_kali
Comparison of REL Methods for Districts of East Kalimantan, Indonesia.
Bronson Griscom, Sr. Scientist Forest CarbonJohn Kerkering, Conservation Analyst
REDDeX Conference, Cancun, July 14, 2010
Question: What is the most accurate method for predicting the amount of deforestation within districts of East Kalimantan?
Alternative methods for predicting deforestation (i.e. REL) at sub-national scale
Historical Rateof project area
Historical Rate(with adjustments)
Forward Looking
Simple Complex
“Planned” (e.g. legal license
to log/convert)
•Non-spatially explicit model (e.g. population-forest fraction)
•Spatially explicit modeling•Rate derived from “reference region.”
•Trend analysis.1
2 3
2000
1
2005
Predicted deforestation in each district = Historic rate in each district
Predicted deforestation in each district = Historic rate of reference region
ClusterAnal.
Deforestation Variables (District Mean Values)Crop Suitability IndexDeforestation Constraints IndexDistance from Converted AreasDistance from Major CitiesDistance from Navigable RiversDistance from SawmillsElevationPercent Histosol SoilsPercent Inceptisol SoilsPercent Oxisol SoilsPercent Remaining ForestPercent Ulfisol SoilsRoad DensitySlope
2
…where reference regions are determined by cluster analysis
Predicted deforestation in each district = Modeled future rate in each district
3
…using spatially explicit model at regional (province) scale.
2020
2015
Vulnerability
Projections
Prior Deforestation
2009
spatial plansoils
forest typesslope
dist. sawmillsdist. towns
dist. navigable riversdist. cities
topographydist. converted areas
dist. roads
2005
2000
“Driver” Variables
•neural network•no dynamic variables
LCM
3Here’s how…
Note: projections assume historic rate at province scale
Deforestation Variables Cramer's V
Distance from Converted Areas 0.67Elevation (DEM) 0.60Distance from Cities 0.46Distance from Navigable Rivers 0.42Distance from Roads 0.34Distance from Towns (ESRI) 0.23Distance from Sawmills 0.16Slope 0.16Population Density (GRUMP) 0.11Population Density (ICRAF) 0.07Distance from All Rivers 0.05
Forest Cover Types 0.76Soils 0.65Spatial Plan (National) 0.49Land Systems 0.46
Continuous
Categorical
Selection of Model “Drivers” 3
Deforestation Variables Cramer's V
Distance from Converted Areas 0.67Elevation (DEM) 0.60Distance from Cities 0.46Distance from Navigable Rivers 0.42Distance from Roads 0.34Distance from Towns (ESRI) 0.23Distance from Sawmills 0.16Slope 0.16Population Density (GRUMP) 0.11Population Density (ICRAF) 0.07Distance from All Rivers 0.05
Forest Cover Types 0.76Soils 0.65Spatial Plan (National) 0.49Land Systems 0.46
Continuous
Categorical
Selection of Model “Drivers” 3
Model Performance
AnalysisFigure of
MeritKappa for Location
This Analysis 0.3374 0.8330Harris et. al. 0.1869 0.7763
3
Predicted area deforested from
2006-2009
minus
Actual area deforested from
2006-2009
(as % of actual area deforested)
321
Comparison of Three Methods
Comparison of Three Methods
321
Comparison of Three Methods
321
Why do cluster reference regions seem to work?
2
Question: What is the most accurate method for predicting the amount of deforestation within districts of East Kalimantan?
•More complex doesn’t mean better.
•I suggest reference region method 2
Nested RELs National Scale:
Historic mean, with negotiated adjustments?
Sub-National Scale (e.g. State/Province): Modeled projection, to determine proportion of national emissions pie? Separate models for deforestation vs. degradation?
Project Scale: Mixed. Modeled projection for unplanned events? Book-keeping for planned events / strategies?