Master Course „Environmental Physics“ (MKEP4) http://www.iup.uni-heidelberg.de/institut/studium/lehre/MKEP4/ 19. Model Concepts, Box Models, Lumped-Parameter Models Summer Term 2011 Werner Aeschbach-Hertig Institut für Umweltphysik Lecture Program of MKEP4 Part 4: Methods and Models (6 sessions) 16. 17. 18 18. 19. 20. 21. Fundamentals of isotope methods Isotope methods in environmental physics Measurement methods of environmental physics Model concepts, box models, lumped-parameter models Complex models and numerical methods General circulation models and inversion techniques Part 5: Climate (5 sessions) 22. 23. 24 24. 25. 26. The carbon cycle Radiative transfer Aerosols and clouds Climate sensitivity, feedbacks, and predictions Climate variability and palaeoclimate reconstruction Last week: Written Exam (19.7.2011) 2 1 Contents of Today's Lecture Model concepts, box models, lumped-parameter models • Types of models, motivation for models • Environmental E i l system analysis: l i B Box models d l • Linear 1-box model • Linear 1-box model: Variable input • Lumped-parameter models, transit time distribution 3 Literature on Modeling / System Analysis Box Models, System Analysis • Imboden, D.M., and Koch, S., 2003: Systemanalyse. Springer. IUP 1876 • Schwarzenbach, R.P., Gschwend, P.M., Imboden, D.M., 2003: Environmental organic chemistry. Wiley. IUP 1865 • Gruber, N., 2010. Lecture "Systemanalyse" at ETH Zurich, see http://www.up.ethz.ch/education/system_analysis/index_DE Lumped Parameter Models, Groundwater Modeling • Mook, W.G. (ed.), 2001: UNESCO/IAEA Series on Environmental Isotopes in the Hydrological Cycle. Vol. 6. available at http://wwwnaweb.iaea.org/napc/ih/IHS_resources_publication_hydroCycle_en.html Climate Modeling • Stocker, T. 2009. Introduction to Climate Modelling. Lecture notes, University of Bern. See http://www.climate.unibe.ch/main/courses/ klimamodellierung_hs09/stocker09climmod.pdf 4 2 Environmental Systems • • • • Many parts of the environment can be understood as (complex) systems System: Boundary, internal relations, and external relations Complexity calls for description by models Only internal processes are modeled (def (def. of system/model boundary) external forcing (input) x no external feedback (output) IInfluence fl on environment (internal) 5 Types of Models • Physical or scale model – physical representation of the studied system (at smaller scale) • Conceptual model – simplified, abstract description of a system, defining its ts boundaries, bou da es, internal te a a and de external te a relations, e at o s, etc etc. • Mathematical model – a conceptual model translated into mathematical equations • Numerical (computer) model – numerical solution of mathematical model • Compare with a map – depiction of nature: 1. flat 2. miniaturized 3. simplified 4. explained – you can not use the same map for all purposes (e.g. scale), same is true for models: pick the appropriate one for the problem at hand 6 3 A Physical Model • physical model - wind tunnel model of flow in a city Univ. Hamburg In the following: Mathematical / numerical models … 7 Benefits of Models • • • • • • • • Models summarise our understanding of processes in an environmental system Model – data comparisons: test our current understanding off the th processes involved, i l d ttestt new ideas id Upscaling from local experiments bigger picture Models provide 4D info which experiments can never do Analysis of complex system behaviour Prediction (weather, climate, contaminant plumes,…) “what what ifif..” scenarios/numerical experiments: • switching off different processes • future/past scenarios • 2xCO2, 4xCO2, more/less aerosol, .. … 8 4 Limitations of Models • • • • • • • • Models are never a perfect description of reality Models can only represent current knowledge Not all processes can be included Resolution in time and space is limited (computing time) Approximations are necessary Initial and boundary conditions are crucial but often not available as needed – not enough data to use the model as exact depiction of the environment Sometimes: good results for the wrong reasons Models are only as good as modeler / user – “garbage in, garbage out” 9 Different Model Structures Dynamic models of environmental systems Describing temporal evolution of system variables yi(t) 1. Models without detailed spatial structure Box models ("0-dimensional") • • • • 1-box model (1 ore more variables, linear and non-linear) n-box models (n coupled 1-box models) Lumped parameter models (pseudo-structure: TTDs) Description by ordinary differential equations (ODEs) 2. Models with continuous spatial extension • • • • • 1-dimensional (e.g. vertical models of atmosphere, lakes) 2-dimensional (e.g. horizontal layers, vertical sections) 3-dimensional (full spatial extension of studied systems) Description by partial differential equations (PDEs) Numerical solution on discrete grid ( link to box-models) 10 5 0-D: “Box Model” Independent variable: time Example: Concentration c(t) of a chemical in a wellmixed air volume Graedel and Crutzen 11 1-D: Vertically Resolved Profile Independent variables: time, altitude Example: Concentration profile c(z,t) of a chemical in a column of air Graedel and Crutzen 12 6 2-D: Horizontal Groundwater Flow Independent variables: time, E and N coordinate Example: Horizontal flow field v(x,y,t) of groundwater with transport: concentration c(x,y,t) Wollschläger et al., Aquat. Sci. Vol. 69, 2007 13 3D: General Circulation Models Independent variables: time, altitude, E and N Example: Global distribution of system variables y(x,y,z,t) in the atmosphere and ocean http://www.bom.gov.au/info/GreenhouseEffectAndClimateChange.pdf 14 7 Why use Box Models? Does it make sense to use simple box models, when 3-D models are obviously more complete descriptions? Yes! Simple models have advantages: • Th They provide id iinsights i ht iinto t th the characteristic h t i ti b behaviour h i off a system t • They can often be solved analytically • They are adequate if few data are available Some problems that can be addressed with box models • • • • • General: Input and reaction of substances in simple compartments Chemicals in lakes (contaminants, nutrients, oxygen etc.) Nitrate in groundwater (residence time, concentration in wells) Accumulation of anthropogenic trace gases in the atmosphere Carbon cycle: CO2 in atmosphere, ocean, biosphere 15 3-Box Model of the Carbon Cycle Fossil fuel C is released into the atmosphere (box 1). The atmosphere Th t h exchanges with the ocean (box 2) and the biosphere (box 3). System analytical question: The CO2 level in air shall be stabilised at 8% above the present. How do the emissions have to change? Gruber, 2010 16 8 One Box or Mixed Reactor Model Simple model for a substance in an environmental system: Fully mixed reservoir with input, output, and 1st order reaction Mass balance: Q,cin V M, c Q,c mixed V: System volume [L3] Q: Flow rate [L3T-1] kf = Q/V: Flushing rate [T-1] M: Mass of substance [M] c = M/V: Concentration [ML-3] cin: Input concentration [ML-3] kr: Reaction rate constant [T-1] dM Qcin Qc krM dt dc k f cin k f kr c k f cin k totc dt 1st order linear inhomogeneous ODE General solution for constant coefficients (cin const ) and c(0) = c0: i , kf, kr = const.) c t c0 c ektot t c with the stationary concentration: c kf kf cin cin k tot k f kr 17 One Box or Mixed Reactor Model Generalisation: Input and output do not have to be tied to volume fluxes of a fluid. Mass in/output per time Jin, Jout. Mass balance: dM Jin Jout kM J kM dt J Jin Jout dc J kc dt V 1st order linear inhomogeneous ODE General solution for constant coefficients (J (J, k = const.) const ) and c(0) = c0: V: System V S t volume l [L3] M: Mass of substance [M] c t c0 c ekt c -3 c = M/V: Concentration [ML ] with the stationary concentration: Jin: Input of substance [MT-1] Jout: Output of substance [MT-1] J j J c with j ML3T1 k: Total 1st order removal rate [T-1] kV k V 18 9 One Box Model: Temporal Evolution Discussion of solution for constant input c t c0 c ektot t c c0ektot t c 1 ektot t Concentrattion dilution/decay of c0 ingrowth of c∞ Ingrowth g to C∞ Washing out of C0 Gruber, 2010 19 One Box Model: Equilibration Time How quickly does the system reach its steady state? Depends on how small the deviation shall be. c c c c0 c ln k tot E.g.: 1/ 2 ektot ln2 k tot 1/ e 1 k tot 0.05 3 k tot e-folding time, system time constant Steady state concentration Initial disturbance Equilibration time Initial concentration Gruber, 2010 20 10 One Box Model with Variable Input Generalisation: External input (forcing) can depend on time Mass balance: f dc k f cin t k totc dt f 1st order linear inhomogeneous ODE with time-variable inhomogeneous term General solution for constant rate coefficients (kf, kr = const.) and c(0) = c0: t c t c 0 e k tot t k f c in t e V: System volume [L3] kf = Q/V: Flushing rate [T-1] c(t): Concentration [ML-3] cin(t): Input concentration [ML-3] kr: reaction rate constant [T-1] k tot t t dt 0 decay of c0 ingrowth to c∞(t) Time-dependent "stationary state": c t kf cin t k tot Equilibrium conc. at momentary input 21 One Box Model: Temporal Evolution (2) Discussion of solution for variable input "Ingrowth term" t k c t e f in k tot t t dt 0 past inputs weighted according to age t - t' Superposition of decay curves from a series of "input p events". Gruber, 2010 22 11 Variable Input: Exponential Forcing Exponentially increasing input c in t c in0 et t c t c 0 e k tot t k f c in0 et e k tot t t dt 0 c t c 0 e k tot t k f c in0 et e k tot t k tot for ≠ -ktot fast growing forcing: ≈ ktot slow (adiabatic) forcing: << ktot C∞(t) t System follows the forcing Gruber, 2010 t System lags behind the forcing 23 Fast Exponential Forcing System with fast exponentially growing forcing: What happens, if input is stabilised? (e.g.: keep CO2 emissions constant) Answer: Option 4 After tstab we have the case with constant input exponential ingrowth to equilibrium Stabilising emissions does not stabilise system concentration immediately Gruber, 2010 24 12 Variable Input: Periodic Forcing Periodically oscillating input c in t c in0 c1in sin t t k t t c t c 0 e k tot t k f c in0 c1in sin t e tot dt 0 kfc k c0 k f c1in k c1 c 0 f in ek tot t sin t 2f in 2 e k tot t k tot k tot k tot 2 k 2tot 0 in c t arctan k tot terms remaining for ktott >> 1 c ≈ ktot << ktot >> ktot C(t) C∞(t) t 25 Gruber, 2010 1-Box Model as Lumped Parameter Model Re-Interpretation of the box model with variable input: Not a mixed reactor but a system with transit time distribution Q General solution for system starting at t = 0 and e evolving ol ing to time tt: cin((t)) t V cout(t) c t c 0 e k tot t k f c in t * e dt * k tot t t * 0 t*: forward running time between 0 and t t' = t - t*: backward running "age" before t t V: System volume [L3] Q: Flow rate [L3T-1] cin(t): Input concentration [ML-3] cout(t): Output concentration [ML-3] kf = Q/V: Flushing rate [T-1] = 1/kf: Mean transit time [T] c t c 0 e k tot t k f c ini t t ek tot tdt 0 Put system start to t = -∞ or maximum "age" t' to ∞ and use = 1/kf: 1 c t c in t t e kr t e t dt 0 26 13 Lumped Parameter Models: Interpretation Discussion of generalised lumped parameter model equation c t c in t t e t g t dt 0 input of age t't before present weight of input at t't , tranist time distribution (TTD) decay for radioactive tracer g(t'): Transfer or response function, TTD Special case: Exponential model (= mixed reactor) Input 1 g t e t Environ. System = Black Box Output 27 Lumped Parameter Models: Mathematics • Time axis: t = (present) time, t' = time span before t, age • Input in the past: c in t t • Transit time distribution (TTD) g t with g t dt 1 0 Depends on mean transit time + other model parameters • Mean age or transit time: t g t dt 0 • Output (convolution of input and transfer function + decay): c out t c in t t e t g t dt 0 28 14 Piston-flow Model (PM) • Transfer function: g t t – Parameter : Age • Output: c out t c in t t e t t dt c in t e 0 • Picks out the input of one specific age • Describes closed system, no mixing • Piston-flow age = apparent tracer age 29 c out t c in t e 30 15 Exponential Model (EM) 1 t g t e – Parameter : Mean age • Transfer function: • Output: 1 t c out t c in t t e t e dt 0 • Weight of input events decreases exponentially with age – recent input (t' << ): highest contribution to output – very old input (t (t' >> ): very small contribution to output • Describes completely mixed system (mixed reactor) • Model age = mean residence time in mixed system 31 32 16 Dispersion Model (DM) g t • Transfer function: – Parameter : Mean age e 3 4t t 2 4 t – Parameter : Dispersion • Output: c out t c in t t e 0 t e 3 4t t 2 4 t dt • Weight of input events: ~ Gauss-type distribution – highest g contribution to output p for certain input p age g t' – decreasing contribution to output for younger/older input • Describes throughflow system with limited mixing (dispersion) • Model age = mean transit time 33 34 17 Transit Time Distributions tran nsfer function g(t') 0.4 Exponential Model (EM) 0.35 0.3 =3 = 10 = 30 0.25 0.2 0.15 0.1 0.05 0 0 10 20 30 40 50 t' [yr] 0.4 0.4 Dispersion Model (DM), = 1 transfer functio on g(t') =3 = 10 = 30 0.25 Dispersion Model (DM), = 0.1 0.35 0.3 0.2 0.15 0.1 0.3 =3 = 10 = 30 0.25 0.2 0.15 0.1 0.05 0.05 0 0 0 10 20 30 40 0 50 10 20 30 40 t' [yr] t' [yr] 50 35 Example: Tritium Output of Expon. Model 1000 Tritium [TU] transfer functio on g(t') 0.35 100 input decay exp modd = 3 exp mod = 7 exp mod = 15 10 1952 1960 1968 1976 1984 1992 2000 Year 36 18 Example: Tritium Time Series in Rivers Mixing of fast and slow runoff components in a river from Mook, 2001 37 Tritium Time Series from a Spring 80 70 =3 =6 Input Samples = 10 Tritium [TU] 60 50 40 decay 30 20 10 0 1980 1984 1988 1992 1996 Year 38 19 Summary • Models are simplified representations of complex systems. Different types of models: Physical, conceptual, mathematical, numerical models • Numerical models can be 0-D (box models), 1-D, 2-D, 3-D • Box models can help to analyse system characteristics • 1-Box or mixed reactor model: 1st order linear inhomogeneous ODE Exponential approach to stationary concentration (equilibrium) System time constant = 1/ktot determines reaction to forcing • Lumped-parameter models/ Transit-time distributions (TTD) Input – output description of (tracers in) environmental systems Internal dynamics described by TTD with few "lumped" parameters TTDs can be derived from conservative tracers 39 20