Outline phenomenon model aims methodical approach monsoon stability under uncertainty conclusions
1 / 13

Uncertainty Analyses of an - PowerPoint PPT Presentation

  • Updated On :

Outline Phenomenon, model, aims Methodical approach Monsoon stability under uncertainty Conclusions PIK - Potsdam Institute for Climate Impact Research, Germany http://www.pik-potsdam.de Michael Flechsig & Brigitte Knopf

I am the owner, or an agent authorized to act on behalf of the owner, of the copyrighted work described.
Download Presentation

PowerPoint Slideshow about 'Uncertainty Analyses of an ' - ostinmannual

An Image/Link below is provided (as is) to download presentation

Download Policy: Content on the Website is provided to you AS IS for your information and personal use and may not be sold / licensed / shared on other websites without getting consent from its author.While downloading, if for some reason you are not able to download a presentation, the publisher may have deleted the file from their server.

- - - - - - - - - - - - - - - - - - - - - - - - - - E N D - - - - - - - - - - - - - - - - - - - - - - - - - -
Presentation Transcript
Outline phenomenon model aims methodical approach monsoon stability under uncertainty conclusions l.jpg


Phenomenon, model, aims

Methodical approach

Monsoon stability under uncertainty


PIK - Potsdam Institute for Climate Impact Research, Germany http://www.pik-potsdam.de

Michael Flechsig & Brigitte Knopf

Uncertainty Analyses of an Indian Summer Monsoon Model: Methods and Results

The indian monsoon l.jpg

ITCZ N-Summer


ITCZ N-Winter

© Paul R. Baumann State University of New York

The Indian Monsoon

  • Semi-annual shift of the intertropical convergence zone ITCZ in conjunction with Temperature gradients in the atmosphere between land surface and ocean lead toIndian Monsoon: wet summers and relatively dry winters over the Indian sub-continent

  • Economic implications of the monsoon stability for India:

    • Agriculture accounts for 25% of the GDP

    • Agriculture employs 70% of the population

The model zickfeld et al grl 32 2005 l.jpg


Tibetan Plateau

stable states instable states

Indian Ocean

Indian Ocean

(20N , 75W)


2 Soil Layers




present value

The Model(Zickfeld et al., GRL 32, 2005)

  • One-dimensional (idealised) box model of the tropical atmosphere over India with about 60 parameters for qualitative studies

  • Prognostic state variables

    • Air temperature

    • Specific air humidity

    • Moisture in two soil layers

  • Drivers: boundary conditions for

    • Air temperature

    • Air humidity

    • Cloudiness

  • For the summer monsoon the model shows a saddle node bifurcationagainst parameters that govern the heat budget

    • Atmospheric CO2 concentration

    • Solar insolation

    • Albedo As of the land surface(AS for broad-leafed trees = 0.12, for desert = 0.30)

Aims and applied methods l.jpg
Aims and Applied Methods

  • Study the stability of the Indian summer monsoon under potential land use and climate change

  • Determine robustness of the bifurcation at SN1against the surface albedo AS under parameter uncertainty

  • Consider three parameter / initial value spaces (all without parameter As)

    • T38 the total space of all 38 uncertain parameters:determine most important parameters

    • S5 a 5-dimensional subspace of the most influential parameters: study parameter sensitivity

    • A5 a 5-dimensional subspace of anthropogenically influenceable parameters: get implications of potential climate change

  • Applied methods and used tools:

    • Combine a qualitative analysis (QA) of a model (“bifurcation analysis”)AUTO (Doedel, 1981)

    • with multi-run model sensitivity and uncertainty analysesSimEnv (Flechsig et al., 2005)

Multi run simenv approach l.jpg







Multi-Run SimEnv Approach

  • Consider Y = F(X) SN1 = QA ( model ( [ T38 | S5 | A5 ] ) )

    • X factor space: model parameters, initial values, boundary values, drivers

    • Y model output (multi-dimensional, large volume)

  • Apply deterministic and random sampling techniques in the multi-factor space Xto study model sensitivity and uncertainty of model output Y multi-run experiments

  • Simple model interface to SimEnv for factors X and model output Y

    “Include for each factor and for each model output field one SimEnv function call into the model source code”

    • at programming language level: C/C++ Fortran Python

    • at modelling language level: MatLab Mathematica GAMS

    • at shell script level

Simenv experiment types l.jpg





SimEnv Experiment Types

  • SimEnv provides generic multi-run simulation experiment typesthat differ in their sampling strategies

  • To generate a sample in the factor space under study a selected experiment type has to be equipped with numerical information

o = default value

x = 1 single run

x = 2nd sample

Monsoon model uncertainty analyses l.jpg
Monsoon Model Uncertainty Analyses

Model interface:


  • Global sensitivity analysis in T38

  • Behavioural analysis in S5

  • Monte Carlo analysis in T38 and A5

Morris design 1991 model free l.jpg

k=2 factors p=5 levelsNTraject=4 trajectoriestrajectory


nonlinear effect on model output 


sensitivity w.r.t. model output 

Morris’ Design (1991)model free

  • Modified by Campolongo et al. (2005)

  • Grid factor space x = (x1 ,…, xk) with p levels for each factor and constant grid widths Δi (i=1,…,k)

  • Define a local elementary effect di of xifrom two grid points in xthat differ only in one factor xi by Δi bydi := Y(x+eiΔi) - Y(x)

  • Select randomly NTraject trajectories of length k (from k+1 points) where exactly one elementary effect dij (j=1,…,NTraject) can be derived from two consecutive points

  • Consider distributions Fiabs = { |dij| } and compute μiabs = mean of Fiabs Fi = { dij } and compute σi = standard deviation of Fi


  • high μiabs :factor xi has an important overall influence on model output Y

  • high σi:factor xi is involved in interactions with other factors w.r.t. Yoreffect of factor xi on Y is nonlinear

Global sensitivity analysis l.jpg

σ -nonlinear effects with respect to SN1

μabs - sensitivity with respect to SN1

Global Sensitivity Analysis

  • Morris’ design for all 38 parameters T38

    • p = 7-level grid for the variation rangesof the 38 parameters

    • NTraject = 1,000 trajectories

    • Resulting in 39,000 single model runs

  • 93.1% of all runs show a bifurcation

  • Some outstanding parameters and one cluster

S5most influential parameters

A5anthropogenically influenceable parameters

Behavioural analysis l.jpg

Maximum value Asat the bifurcation pointover the 5*5 single runs of the two dimensions that are not shown

rank 1

rank 5

Behavioural Analysis

  • Deterministic screening exercise for the 5 most sensitive parameters S5

    • for deep insight into the model

    • 5 equidistant values per parameterin its variation range result in 55 single model runs

  • All runs show a bifurcation

  • Most sensitive parameters show largest variation

Monte carlo analyses l.jpg



value without uncertainty

present value

Monte Carlo Analyses

  • For all 38 parameters T38 and the 5 anthropogenic parameters A5

    • Uniform marginal distributions on their variation ranges

    • Latin hypercube sampling

    • 20,000 single model runs

  • 94.4% of all runs in T38,all runs in A5show a bifurcation

  • According to the model it is not likelythat the system reaches the bifurcation point under influence of human activity

  • Variation of As for T38 at the bifurcation point SN1 is the same as variation of As for current vegetation

Conclusions l.jpg


  • Combination of a bifurcation analysis with multi-parameter uncertainty studies enabled qualitative considerations for the whole parameter space

  • SimEnv as a multi-run simulation environment with the focus on model sensitivity and uncertainty studies

    Model results:

  • Bifurcation for surface albedo in the model is robust under parameter uncertainty though the value of the bifurcation point varies

  • The present state of the system is far away from the variability range of the bifurcation point

  • More detailed studies are necessaryExample:

    • System: air pollutants  aerosols  optical thickness of stratum clouds

    • Model: parameter τst bifurcation point for surface albedo

Thank you for your attention l.jpg
Thank you for your Attention

SimEnv on the Internet: