MIT OpenCourseWare http://ocw.mit.edu 15.023J / 12.848J / ESD.128J Global Climate Change: Economics, Science, and Policy Spring 2008 For information about citing these materials or our Terms of Use, visit: http://ocw.mit.edu/terms. Click to edit Master title style Methods of Uncertainty Analysis • Click to edit Master text styles • • • • Second level Third level Fourth level Fifth level 15.023 Lecture 9 April 2008 Travis Franck Eunjee Lee (material from previous lectures by Mort Webster, Ian Sue Wing, Marcus Sarofim, and Travis Franck) 1 “Persons to forecast the Click topretending edit Master title style future shall be considered disorderly •under Click to edit Master text subdivision 3, styles section 901 of •the Second level code and liable to a fine criminal • Third level of $250 and/or six months in • Fourth level prison.” • Fifth level Section 889, New York State Code of Criminal Procedure 2 ClickLecture to edit Master Highlights title style • • • •• • • How Clickto touse editDecision Master text Trees styles Second Understanding level mean vs. median, standard deviations, Third level etc. How tolevel read PDFs (and make one from a Fourth histogram) Fifth level The role of uncertainty in climate change science 3 Elements of a Decision Tree Click to edit Master title style Uncertainty Node: Shows different possible outcomes foredit an uncertain event, and the respective • Click to Master text styles probabilities for each outcome • Second level p • Third level • Fourth level 1-p • Fifth Node: level possible actions at a choice point Decision act BAU 4 Example: Biking in the Rain Figure by MIT OpenCourseWare. Decision tree without Memory Click to edit Master title style Period 1 Decision Outcome • • • • • Period 2 Period 3 Click to H edit Master text H styles R R Second level L L Third Hlevel H R FourthRlevel L L Fifth level H R L H R L No connection between periods 6 Decision Tree with Memory Period 1 Period 2 Period 3 Click to edit Master title style Decision Outcome H • • • • • Click to edit Master text styles Second level H R Third level L Fourth level H H R Fifth levelR L H R L R L H R L L Depends on the path 7 Emissions Scenarios Click to edit Master title style • • • • • Growth AEEI Shorthand for Click to edit Master text styles Emissions H H Second level R R Third levelL L Growth H Fourth level R Fifth level L AEEI H R L H R L H R L 8 Climate Scenarios Click to edit Master title style Growth AEEI Kv S • Click to edit Master text styles H L L H • Second level R R R • LThird level H H L • Fourth level Collapse into RR, HL, LH • Fifth level In Homework Economic Uncertainty Faer L H Climate Uncertainty Probability distribution 9 Handling scientific uncertainty in AR4 Click to edit Master title style • In general, uncertainty ranges for results given in this for Policymakers are 90% uncertainty intervals • Summary Click to edit Master text styles unless stated otherwise, that is, there is an estimated 5% likelihood level that the value could be above the range given in • Second square brackets and 5% likelihood that the value could be below that range. Best estimates are given where available. • Third level Assessed uncertainty intervals are not always symmetric about the level corresponding best estimate. Note that a number • Fourth of uncertainty ranges in the Working Group I TAR corresponded to 2 standard deviations (95%), often using • Fifth level expert judgment. (IPCC, 2007: Summary for Policymakers. In: Climate Change 2007: The Physical Science Basis. Contribution of Working Group I to the Fourth Assessment Report of the Intergovernmental Panel on Climate Change) 10 Why ClickWorry to editabout Master Uncertainty? title style • To identify important and assumptions Click to edit Masterfactors text styles underlying disagreements • Second level • To know where more information is needed • Third level • To attach a range to model forecasts •• To Fourth level attitudes toward risk understand •• To Fifth levelfor learning over time account (Attaching uncertainty to predictions might be as natural as attaching cost/benefit to climate targets) 11 ofMaster Uncertainty ClickTypes to edit title style • • • • • • •• Parametric Uncertainty Click to edit Master text styles – Uncertainty in the value of a quantity Second level Model or Structural Uncertainty Third level – Uncertainty in the form of a model Fourth level vs. Quadratic relationship – e.g. Linear Fifth level Surprise/Indeterminacy – Don’t know what we don’t know 12 Describing an Uncertainty Quantity Click to edit Master title style ∞ μ x = ∫ xf x ( x)dx = E[ x] • Mean −∞ • Click to edit Master text styles •• •• • •• ∞ 2 Variance σ = ( x − μ x ) fx ( x)dx = Var[ x] Second level ∫−∞ Third level σ x = Var (x) Standard Deviation Fourth level Covariance cov( X 1 , X 2 ) = E [X 1 ∗ X 2 ] (if E[x]=0) Fifth level 2 x 13 Describing an Uncertain Quantity Click to edit Master title styleII •• Mode: value (peak) Click toMost editlikely Master text styles • Median: Value of x such that • •• • • • Second level – Prob (x<x0) = Prob (x>x0) = 0.50 Third level Fractile: The p fractile is the value x0 such that – Prob (x<x 0) = p Fourth level Probability Density Function (pdf) Fifth level – The integral under a portion of the function is the probability that the event will fall into that range. • Cumulative Density Function (cdf) 14 3 Probability Density Functions Click to edit Master title style Median Mean 1 Standard Deviation (67% bounds) •2.5Click to edit Master text styles • 2Second level •1.5Third level • 1Fourth level • Fifth level 2 Standard Deviations (95% bounds) 0.5 0 -0.1 0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 15 Click Cumulative to editDensity Master title Function style 1.2 • • • • • 1 0.8 0.6 0.4 0.2 Click to edit Master text styles Second level Third level Fourth level Fifth level 1 0.9 0.8 0.7 0.6 0.5 0.4 0.3 0.2 0.1 0 0 0 0.5 1 1.5 2 2.5 3 3.5 4 4.5 5 0 0.5 1 1.5 2 2.5 3 3.5 4 4.5 16 5 Input distributions Click to edit Master title style 8 2.5 0.6 5 4 3 2 1 0 • • • • • 0 Click to edit Master text styles Second level Third level Fourth level Fifth level 0.4 0.3 0.2 2 Probability Density 6 0.5 Probability Density Probability Density (Values x 10-2) 7 1.5 1 0.5 0.1 0 20 40 Ocean Diffusivity (Kv) (cm 2/s) 0 2 4 Climate Sensitivity (°C) 6 0 0 0.2 0.4 0.6 0.8 Faer (Aerosol Forcing) (W/m 2) Methods of Input Distribution Acquisition: Expert Elicitation, Historical Analysis, Modeling Ranges 17 1 Correlation between Distributions Click to edit Master title style Click to edit Master text styles 7 Second6 level 5 Third 4level Fourth3 level 2 Fifth level CS • • • • • 8 1 0 0 1 2 3 Sqrt Kv 18 Measuring Click to edit Uncertainty Master title - Global style • • • • •• Range Click toSensitivity edit Master text styles – Varying each x from low to high (tornado plots) Second level Joint Analysis ThirdParametric level – Vary level lows and highs for multiple variables Fourth Montelevel Carlo Simulation Fifth – Use pdfs of inputs to produce pdfs of outputs 19 Click Monte to edit Carlo Master Simulation title style • • • • • Crudeto Click Monte edit Master Carlo text styles Second Stratified level Sampling Methods – Latinlevel Hypercube Sampling Third Importance Fourth levelSampling Methods Response Fifth levelSurface Methods – Probabilistic Collocation Method 20 Click Monte to edit Carlo Master Simulation title style • •• Example 1: Throwing Click to edit Master textDice styles Examplelevel 2: Second – Y = X1+ X2 • Third level – X1 = N(50,20) • Fourth – X2 = level N(40,25) • Fifth level Analysis: Y = N(90,32) – Analytic – Compare to Monte Carlo samples of size n=10,100,1000,10000 21 Monte Carlo, n=10,100 Figure by MIT OpenCourseWare. Monte Carlo, n=1000,10000 Figure by MIT OpenCourseWare. 23 Click Webster to edit (2001) MasterGraph title style • • • • • Click to edit Master text styles Second level Third level Fourth level Fifth level 24 Click ThetoGreenhouse edit MasterGamble title style • • • • • Click to edit Master text styles Second level Third level Fourth level Fifth level Business As Usual Policy 25 Click to edit Master title style • • • • • Click to edit Master text styles Second level Questions? Third level Fourth level Fifth level 26