Download presentation
Presentation is loading. Please wait.
Published bySuzanna Strickland Modified over 8 years ago
1
Module 2.7 Estimation of uncertainties REDD+ training materials by GOFC-GOLD, Wageningen University, World Bank FCPF 1 Module 2.7 Estimation of uncertainties Module developers: Giacomo Grassi, Joint Research Centre Suvi Monni, Benviroc Frédéric Achard, Joint Research Centre Andreas Langner, Joint Research Centre Erika Romijn, Wageningen University Martin Herold, Wageningen University Exercises: 1.Uncertainties in area and area change 2.Using IPCC equations to combine uncertainties 3.Using IPCC equations to assess trend uncertainties 4.The REDD+ matrix approach (see xls file and Country Example – This exercise is in common with module 3.3) 5.Preparations for Monte Carlo V1, April 2015 Source: IPCC GPG LULUCF Creative Commons License
2
Module 2.7 Estimation of uncertainties REDD+ training materials by GOFC-GOLD, Wageningen University, World Bank FCPF 2 Exercise 1: Uncertainties in area/area change (1) Example of accuracy measures for the forest class: ● Error of commission: (13+45)/293 = 19.80% ● Error of omission: (25+3)/263 = 10.65% ● User’s accuracy: 235/293 = 80.20% ● Producer’s accuracy: 235/263 = 89.35 Overall accuracy = (235+187+215+92+75)/986 = 81.54% Reference data Class. dataFAWUBTotal F235130450 293 A2518771820 257 W3021500 218 U0009235 127 B0001675 91 Total263200222171130986
3
Module 2.7 Estimation of uncertainties REDD+ training materials by GOFC-GOLD, Wageningen University, World Bank FCPF 3 Exercise 1.1 Classification accuracy assessment a) Calculate the error of commission, error of omission, user’s accuracy, producer’s accuracy for all land use classes b) Calculate the overall classification accuracy Use the Excel sheet to fill in the answers Exercise 1: Uncertainties in area/area change (2)
4
Module 2.7 Estimation of uncertainties REDD+ training materials by GOFC-GOLD, Wageningen University, World Bank FCPF 4 Error matrix of sample counts constructed from the accuracy assessment sample of a change map by Jeon et al., in press. (Extracted from Olofsson et al., 2013) 1. Expressing the error matrix in terms of area proportions: Reference data Class. data123TotalMap area (ha)WiWi 1970310022,3530.013 23279183001,122,5430.640 32197100610,2280.348 Total1022801185001,755,1231 Class 1: deforestation Class 2: no change in forest Class 3: no change in non-forest For deforestation [1,1]: 0.013*97/100 = 0.013 Exercise 1: Uncertainties in area/area change (3)
5
Module 2.7 Estimation of uncertainties REDD+ training materials by GOFC-GOLD, Wageningen University, World Bank FCPF 5 2.Calculation of the standard error: 3.Calculation of standard error of the area estimate: 4.Calculation of the 95% confidence interval: For deforestation: 1,755,123*0.00613 = 10758.9 ha For deforestation: 45580.54 +/- 21517.8 Exercise 1: Uncertainties in area/area change (4)
6
Module 2.7 Estimation of uncertainties REDD+ training materials by GOFC-GOLD, Wageningen University, World Bank FCPF 6 Exercise 1.2 Accuracy assessment of land change maps a) Convert the complete error matrix to estimated area proportions, using equation 1 b) Calculate the following estimates of area and accuracy for all classes: 1.Estimator of area per class 2.The standard error using equation 2 3.The standard error of the area estimate using equation 3 4.The 95% confidence interval using equations 4 c) Calculate the user’s accuracy, the producer’s accuracy and the overall accuracy for the change map, using the error matrix with stratified estimators Exercise 1: Uncertainties in area/area change (5)
7
Module 2.7 Estimation of uncertainties REDD+ training materials by GOFC-GOLD, Wageningen University, World Bank FCPF 7 Exercise 2: Combination of uncertainties The table below presents data for various types of conversions a. Calculate the total uncertainty for each category, in absolute amounts (Gg CO 2 ) and relative terms (%) b. Calculate the total net emissions and the corresponding uncertainty in absolute (Gg CO 2 ) and relative terms (%) a Estimate includes both AD and EF uncertainties. a When an estimate is the result of both emissions and removals, the net sum may be close to zero. This may lead to high values of % uncertainties, because estimated relative a small number. Category Net emissions (Gg CO 2 ) AD uncertain ty EF uncertainty Total uncertainty by category (Gg CO2) Total uncertainty by category (%) Forest converted to cropland 1324.1±16%±71% Forest converted to settlements 1522.7±64% a Cropland converted to grassland -84.7±23%±460% b TOTAL net emissions
8
Module 2.7 Estimation of uncertainties REDD+ training materials by GOFC-GOLD, Wageningen University, World Bank FCPF 8 Exercise 3: Trend uncertainty in forest land (1) The data presented in the table on the next slide is reported for Forest land (i.e. Forest land remaining forest land plus Land converted to forest land) in the year 1 and year 2 Errors in ADs are typically not correlated over time, while errors of EFs may be correlated or not, depending on the methods used for estimation a. Calculate the trend in net emissions from forest land between years 1 and 2 b. Calculate the trend uncertainty assuming errors of EF not correlated c. Calculate the trend uncertainty assuming errors of EF fully correlated
9
Module 2.7 Estimation of uncertainties REDD+ training materials by GOFC-GOLD, Wageningen University, World Bank FCPF 9 Exercise 3: Trend uncertainty in forest land (2) Category Net emissions (Gg CO 2 -eq) Uncertainty (%) Trend uncertainty (%) year 1year 2ADEF assuming no EF correlation assuming full EF correlatio n Forest land remaining forest land -23440.0-36152.4 10%35% Land converted to forest land 488.4164.1 20% 150% Total Forest land TREND
10
Module 2.7 Estimation of uncertainties REDD+ training materials by GOFC-GOLD, Wageningen University, World Bank FCPF 10 Exercise 4: The REDD+ matrix approach (1) to Natural Forest Land Boundary Forest LandOther Land from Natural Forest LandForest conservationForest DegradationDeforestation 1 Boundary Forest Land Enhancement of C stocks Sustainable Management of Forests Deforestation 2 Other Land- Enhancement of C stocks (afforestation / reforestation) REDD+ activities matrix in expanded IPCC categories
11
Module 2.7 Estimation of uncertainties REDD+ training materials by GOFC-GOLD, Wageningen University, World Bank FCPF 11 Exercise 4: The REDD+ matrix approach (2) A country with wet tropical forests uses the IPCC tier 1 factor (plus additional assumptions) for above ground C stock. Only the uncertainties of EF are considered here. A virtual REDD+ period 2015-2020 is simulated, and "emission reductions" are estimated in comparison to an hypothetical area- based reference level (RL) adopted in 2015. Calculate: a) the combined total uncertainty (referred to the EF) of the "reduced emissions" as compared to Reference Level b) the conservative accounting of "reduced emissions" as compared to Reference Level (see ppt “country example” and file xls “exercise 3” for more details)
12
Module 2.7 Estimation of uncertainties REDD+ training materials by GOFC-GOLD, Wageningen University, World Bank FCPF 12 Exercise 5: Preparation for Monte Carlo (1) The following equations are used to estimate relative change in Biomass Expansion Factor (BEF) between two years: Relative change in BEF = (BEF 2 -BEF 1 )/ BEF 1 (1) BEF = Exp{3.213 - 0.506*Ln(BV)} (2) where BV is biomass of inventoried volume in t/ha, calculated as the product of Volume of Biomass per ha (VOB/ha: m 3 /ha) and wood density (t/m 3 ) The parameter values and their uncertainties as 95 % confidence intervals are presented in the table below VOB values are obtained empirically for the two years. Wood density and empirical parameters (3.213 and 0.506) are from literature, and are the same for the two years Exercise: Prepare data needed for Monte Carlo simulation to estimate uncertainties relative change in BEF (equation 1)
13
Module 2.7 Estimation of uncertainties REDD+ training materials by GOFC-GOLD, Wageningen University, World Bank FCPF 13 Exercise 5: Preparation for Monte Carlo (2) Solution: For purposes of Monte Carlo simulation, the actual parameter values used in calculation are taken as mean values of the distributions Symmetric uncertainties are assumed to be normally distributed and positively skewed distributions lognormally distributed, using the upper bound of the distribution as 97.5%-tile The values which are the same for both years are assumed to be fully correlated whereas the VOB values are assumed not to be correlated between years. The data needed to prepare for Monte Carlo simulation are summarized in the table below.
14
Module 2.7 Estimation of uncertainties REDD+ training materials by GOFC-GOLD, Wageningen University, World Bank FCPF 14 Recommended modules as follow up Module 2.8 to learn more about the role of evolving technologies for monitoring of forest area changes and changes in forest carbon stocks Modules 3. to proceed with REDD+ assessment and reporting
Similar presentations
© 2024 SlidePlayer.com. Inc.
All rights reserved.