Roissy – 9 novembre 2010 REDD+ in Madagascar Carbon stocks assessment and baseline setting: project-level inputs to national level - Side-Event at COP 17 Durban - 2 dec 2011 Speakers : Romuald Vaudry: GoodPlanet Foundation Ghislain Vieilledent: CIRAD Clovis Grinand: GoodPlanet Foundation Jean-Roger Rakotoarijaona: National Office for the Environment Naomi Swickard: Verified Carbon Standard Charles Rakotondrainibe: Madagascar National Parks
87
Embed
Holistic Conservation Programme for Forests in Madagascar
REDD+ in Madagascar Side-Event at COP 17! Durban - 2 dec 2011 Foundation GoodPlanet - Action Carbone program Yann Arthus-Bertrand Romuald Vaudry: GoodPlanet Foundation Ghislain Vieilledent: CIRAD Clovis Grinand: GoodPlanet Foundation Jean-Roger Rakotoarijaona: National Office for the Environment Naomi Swickard: Verified Carbon Standard Charles Rakotondrainibe: Madagascar National Parkse
Welcome message from author
This document is posted to help you gain knowledge. Please leave a comment to let me know what you think about it! Share it to your friends and learn new things together.
Transcript
Roissy – 9 novembre 2010
REDD+ in Madagascar Carbon stocks assessment and baseline setting:
project-level inputs to national level!-!
Side-Event at COP 17! Durban - 2 dec 2011
Speakers:
Romuald Vaudry: GoodPlanet Foundation Ghislain Vieilledent: CIRAD Clovis Grinand: GoodPlanet Foundation Jean-Roger Rakotoarijaona: National Office for the Environment Naomi Swickard: Verified Carbon Standard Charles Rakotondrainibe: Madagascar National Parks
Roissy – 9 novembre 2010
!!!Keywords - 5M€ project exclusively funded by Air France - Implemented by GoodPlanet and WWF Madagascar
- Oct. 2008 - Feb. 2012 - Grant based project: no carbon credits issued by the end of the project
- To switch from conventional forestry measures (diameter, height) in carbon stocks"
- Africa: 30% of the tropical forests""- No generic allometric model for Africa"
- The choice of the allometric model is the most important source of error regarding AGB assessment (Keller 2001, Chave 2004)
"=> We made the decision to develop models for
Madagascar in partnership with the CIRAD and the Antananarivo University "
"
""
"
AGB AssessmentAllometric Models Development!
Materials: ""• 50 genera studied"
• 480 trees measured and entirely weighed, from stems to leaves"
• 2000 wood samples analysed in Cirad lab""""
AGB AssessmentAllometric Models Development!
Results:""• Models used in Madagascar (including DBH only) strongly overestimate the AGB of: "
" """Brown: "+ 35,5% in moist forests "Chave: "+ 36,9% in moist forests"
" " "+ 91,1% in spiny forests " "+ 63,3% in spiny forests""• Models now available for any project proponent in Madagascar"""
AGB AssessmentAllometric Models Development!
Results:""• Compared to our regional models, Chave’s models including DBH, Height and Wood
Density can also provide accurate tree biomass estimates.
AGB AssessmentAllometric Models Development!
Results:""• Compared to our regional models, Chave’s models including DBH, Height and Wood
Density can also provide accurate tree biomass estimates.
• When allometric models are not available (e.g in Africa) and since height is time-consuming to measure, a simple H-DBH allometry can feed the above model to accurately estimate biomass and carbon stock from plot inventories.
For more details: A universal approach to estimate biomass and carbon stock in tropical forests using generic allometric models - Vieilledent G et al (2011). Ecological Applications.
AGB AssessmentAllometric Models Development!
"
1. Land-Use analysis"
2. LiDAR technology"
3. Allometric models development"
4. Final results"
5. Key inputs for national level"
"
Above-Ground Biomass Assessment"
AGB AssessmentFinal results!
Very high correlation between Mean Canopy Height (m) and Above Ground Carbon density (Mg C ha -1) Uncertainty: 12% only
Mean Stocks"
Spiny forests"17 tC/ha"
Moist Forests"90 tC/ha"
Southern Mountains (East) and Plains (West)!
10% uncertainty"at the pixel level (1ha)"
AGB AssessmentFinal results!
Moist Forests"88 tC/ha"
Northern Mountains Region!
35% uncertainty"at the pixel level (1ha)"
For more details: Human and Environmental Controls over Aboveground Carbon Storage in Madagascar - Asner et al (2011). Carbon Balance Management (in review)
AGB AssessmentFinal results!
Northern Mountains Region! Costs
0,15 $/ha of forest This cost does not include allometric models development
Can easily drop around: 0,06-0,07 $/ha of forest Because: - Our sites are scattered in the country - 24 field plots of the 83 plots were
enough to calibrate the MCH/ACD model
- Basal area measurement can replace DBH/H (Mascaro et al.2011)
- LiDAR coverage: 5% of the project area is usually enough (vs 10% in our case)
± 300 000 $ for all the Malagasy moist forests (4,7 M ha)
AGB AssessmentFinal results!
Key inputs for (sub)national level • Malagasy colleagues trained to deal with Very High Resolution
imagery
• Critical information provided regarding forest definition (dry/spiny forests)
• Allometric models now available for any project proponent in Madagascar"
• LiDAR technology: a very interesting option to map carbon stocks of the remote moist forests of Madagascar (mainly located in mountains, on steep slopes). "
AGB Assessment!
Modeling deforestation and associated CO2 emissions
Introduction
Materials and methods Results Discussion
Outline
Introduction
Materials and methods Results Discussion
Outline
1- Modeling the deforestation process: l Deforestation rates 2000-2010 l Factors of deforestation
2- Forecasting deforestation 2010-2030 3- Estimating CO2 emissions associated to deforestation
Introduction Objectives
Introduction Area
4 study areas - 3 in moist forest (North, Middle, South) - 1 in spiny dry forest (South) Surface 7 800 000 ha
2 141 520 ha of forest in 2010
Introduction
Materials and methods l Land use change observation l Uncertainty l Deforestation model l Projected deforestation Results Discussion
Outline
Photo interpretation - Definition of forest: 0.1ha, 10% cover, 5m height - Landsat images (+GoogleEarth VHR images) - Training forest plots (1 for 500 ha of forest) - Land use change observations at 3 dates: FFF, FDD, FFD - Free Open Source (FOS) software: QGIS
Materials and Methods Land use change observation
Sept. 2001 April 2005 Nov. 2010
Land use change model
Materials and Methods Land use change observation
Photo interpretation Landsat images
model
FFF, FDD, FFD + cloud, water and shade
Pixel characteristics - NIR, VIS - NDVI - ...
FOS software: R with randomForest classification algorithm
Projected land use change on the whole image
Materials and Methods Land use chane observation
Landsat images
model
FFF, FDD, FFD + cloud, water and shade
Pixel characteristics - NIR, VIS - NDVI - EVI - ...
FOS software: R + GRASS GIS
Materials and Methods Land use change observation
Overland
Sept. 2001 April 2005 Nov. 2010
Past deforestation 2001-2005 and 2005-2010
Originality of the approach - Integrates expert knowledge (photo interpretation) - Automated statistical treatment of the images with an advanced algorithm (randomForest) - Direct classification of land use change (FFF, FDD) - Using Free Open Source software
Materials and Methods Land use change observation
Introduction
Materials and methods l Land use change observation l Uncertainty l Deforestation model l Projected deforestation Results Discussion
Outline
Uncertainty - Uncertainty associated to land use change observation - Assessed with cross-validation procedure → Land use accuracy: 85% → Land use change accuracy: 70% Consistency assessment - Comparison with other approaches: point sampling (pixel level)
Materials and Methods Uncertainty
Point sampling - Regular grid of 2 x 2 km - Pixel level - Photo interpretation on Landsat images (+ GoogleEarth VHR images)
Materials and Methods Uncertainty
Comparison
Materials and Methods Uncertainty
Introduction
Materials and methods l Land use change observation l Uncertainty l Deforestation model l Projected deforestation Results Discussion
Outline
Logistic regression model
Materials and Methods Deforestation model
Land-use change observations: Zit={1,0} 1: deforested and 0: forest i: pixel et t: date Zit ~Binomial(θit) logit(θit)=f(factors, β) θit: probability of deforestation for pixel i at date t β: model parameters to be estimated
Factors of deforestation - Landscape factor: forest fragmentation - Anthropogenic factor: population density - Policy factor: protected area network
Materials and Methods Deforestation model
Factors of deforestation - Landscape factor: forest fragmentation - Anthropogenic factor: population density - Policy factor: protected area network
Materials and Methods Deforestation model
Factors of deforestation - Landscape factor: forest fragmentation - Anthropogenic factor: population density - Policy factor: protected area network
Materials and Methods Deforestation model
Parameter estimates Inference in a hierarchical Bayesian framework Iterations = 5001:9996 Thinning interval = 5 Number of chains = 1 Sample size per chain = 1000 1. Empirical mean and standard deviation for each variable, plus standard error of the mean: Mean SD Naive SE Time-series SE (Intercept) -2.149186 0.068773 2.175e-03 5.039e-03 as.factor(fragindex)2 -0.607695 0.100512 3.178e-03 7.411e-03 as.factor(fragindex)3 -2.053594 0.087759 2.775e-03 8.137e-03 as.factor(fragindex)4 -3.244258 0.313901 9.926e-03 3.165e-02 as.factor(fragindex)5 -6.777683 0.829412 2.623e-02 9.915e-02 as.factor(aire_prot)2 -0.807469 0.134398 4.250e-03 1.300e-02 dens 0.004556 0.001107 3.499e-05 7.449e-05
FOS software: R and C++ code
Materials and Methods Deforestation model
Introduction
Materials and methods l Land use change observation l Uncertainty l Deforestation model l Projected deforestation Results Discussion
Outline
Projecting explicative variables in the future
Population : Using demographic models Protected areas : Unchanged Forest fragmentation : Iterative method, forest fragmentation is re-computed at each time step
Materials and methods Projected deforestation
Computing probability of deforestation in the future logit(θit)=β0+β1j Fragmentationit+β2 DensPopit+β3k AireProtit θit: probabilité de déforestation du pixel i à la date t blue-green-yellow-orange-red θit=0 -----------------------> θit=1 We simulate the deforestation following the estimated probability of deforestation FOS software: R + GRASS GIS
Materials and methods Projected deforestation
Materials and methods Projected deforestation
Deforestation and associated C/CO2 emissions
Materials and methods Projected deforestation
Deforestation and associated C/CO2 emissions
Materials and methods Projected deforestation
Deforestation and associated C/CO2 emissions
Materials and methods Projected deforestation
Deforestation 2010-2030
Introduction Materials and methods Results Discussion
Outline
Carbon released on project areas
Results
Project area Forest surface 2010 (ha)
Deforestation rate 2000-2010
C from AGB 2010-2030 (t)
ANDAPA 197,763.13 0.44 1,332,641.00
Ivohibe 111,759.00 0.80 1,824,165.00
Fandriana 25,342.00 2.19 859,040.00
FD moist 51,433.00 0.84 529,995.00
FD dry 119,736.00 0.35 139,588.00
TOTAL 4,685,429.00
Introduction Materials and methods Results Discussion
Outline
Key inputs for the (sub)national level Work done on large areas (~25% of national forest) Methodological framework is now clearly defined Free Open-Source softwares with all scripts freely available Building capacity: students + engineers from national institutes ⇒ Work can be easily and rapidly transposed to other forest areas
Discussion
Soil Organic Carbon Assessment
Soil carbon accountancy at project scale
1. Introduction
2. Measuring soil carbon reference level
3. Spatial estimation of soil carbon stocks
4. Monitoring soil-C change
Why taking soil into account?
Values in billions of tC (PgC)
Soil is a major carbon pool
Why taking soil into account?
Values in billions of tC (PgC)
Higher stocks in tropical and boreal forests Soil is a major carbon pool
Why taking soil into account?
Values in billions of tC (PgC)
Dry subhumid, semiarid, arid and hyperarid lands
Higher stocks in tropical and boreal forests Soil is a major carbon pool
Soil carbon accountancy at project scale
1. Introduction
2. Measuring soil carbon reference level
3. Spatial estimation of soil carbon stocks
4. Monitoring soil-C change
Measuring soil carbon reference level
Objective : Soil carbon stock estimation at plot level Methodology : Soil collection and measurement from a 10x10 m plot, stratified by forest type, land use category and elevation Results were aggregated for 0-30 cm and 0-100 cm soil layer (IPCC requirements) Main activities : - Field survey - Formation of local survey team - Measurement of soil bulk density - Soil collection and database - Traditional lab measurement of C concentration - Infra red spectroscopy - Soil carbon stock estimation
Measuring soil carbon reference level
Some figures: - 121 sample plots - 302 kg of soil collected - 750 soil carbon measure using infra red spectroscopy
Infra red spectroscopy is used to produce accurate and cost-effective carbone measurement
Measuring soil carbon reference level C
sto
cks
(tC/h
a)
From 1,6 up to 3,8 times more carbon is stored on soil (1 m) compared to biomass
Humid Forest Spiny Forest
Soil layer
Soil carbon accountancy at project scale
1. Introduction
2. Measuring soil carbon reference level
3. Spatial estimation of soil carbon stocks
4. Monitoring soil-C change
Spatial estimation of soil carbon stocks
How much carbon is stored in the soil?
and here?
Spatial estimation of soil carbon stocks
Use of soil carbon inventory coupled with RS data and spatially explicit soil