Klaus Gierens Institut für Physik der Atmosphäre DLR Oberpfaffenhofen Modelling of Cirrus Clouds (MOD 10) (MOD 11)

Slides:



Advertisements
Similar presentations
Numerical Weather Prediction Parametrization of Diabatic Processes Cloud Parametrization 2: Cloud Cover Richard Forbes and Adrian Tompkins
Advertisements

Quantifying sub-grid cloud structure and representing it GCMs
Gas Laws and Thermal properties of matter
Precipitation I. RECAP Moisture in the air (different types of humidity). Condensation and evaporation in the air (dew point). Stability of the atmosphere:
Cirrus cloud evolution and radiative characteristics By Sardar AL-Jumur Supervisor Steven Dobbie.
The Problem of Parameterization in Numerical Models METEO 6030 Xuanli Li University of Utah Department of Meteorology Spring 2005.
Evaluation of ECHAM5 General Circulation Model using ISCCP simulator Swati Gehlot & Johannes Quaas Max-Planck-Institut für Meteorologie Hamburg, Germany.
Cloud Microphysics Dr. Corey Potvin, CIMMS/NSSL METR 5004 Lecture #1 Oct 1, 2013.
Diploma course special lecture series Cloud Parametrization 2: Cloud Cover
1 Clouds Consider a clean atmosphere with water vapor in it. Dry Atmosphere Water Vapor Given a long enough time, some water vapor molecules will run in.
Interfacial transport So far, we have considered size and motion of particles In above, did not consider formation of particles or transport of matter.
ENVI3410 : Lecture 8 Ken Carslaw
ABSORPTION Beer’s Law Optical thickness Examples BEER’S LAW Note: Beer’s law is also attributed to Lambert and Bouguer, although, unlike Beer, they did.
The Centre for Australian Weather and Climate Research A partnership between CSIRO and the Bureau of Meteorology The Effect of Turbulence on Cloud Microstructure,
CHAPTER 6 Statistical Analysis of Experimental Data
Initial 3D isotropic fractal field An initial fractal cloud-like field can be generated by essentially performing an inverse 3D Fourier Transform on the.
Lecture 11 Cloud Microphysics Wallace and Hobbs – Ch. 6
Water in the Atmosphere Water vapor in the air Saturation and nucleation of droplets Moist Adiabatic Lapse Rate Conditional Instability Cloud formation.
Uintah Basin WRF Testing Erik Neemann 20 Sep 2013.
Lapse Rates and Stability of the Atmosphere
Different heterogeneous routes of the formation of atmospheric ice Anatoli Bogdan Institute of Physical Chemistry, University of Innsbruck Austria and.
Influence of ice supersaturation, temperature and dynamics on cirrus occurrence near the tropopause N. Lamquin (1), C.J. Stubenrauch (1), P.-H. Wang (2)
Microphysics Parameterizations 1 Nov 2010 (“Sub” for next 2 lectures) Wendi Kaufeld.
Page 1© Crown copyright Distribution of water vapour in the turbulent atmosphere Atmospheric phase correction for ALMA Alison Stirling John Richer & Richard.
Rick Russotto Dept. of Atmospheric Sciences, Univ. of Washington With: Tom Ackerman Dale Durran ATTREX Science Team Meeting Boulder, CO October 21, 2014.
GEF2200 Stordal - based on Durkee 10/11/2015 Relative sizes of cloud droplets and raindrops; r is the radius in micrometers, n the number per liter of.
Today’s lecture objectives: –Nucleation of Water Vapor Condensation (W&H 4.2) What besides water vapor do we need to make a cloud? Aren’t all clouds alike?
Characterizing CCN Spectra to Investigate the Warm Rain Process by Subhashree Mishra.
1 Atmospheric Radiation – Lecture 9 PHY Lecture 10 Infrared radiation in a cloudy atmosphere: approximations.
Mass Transfer Coefficient
A canopy model of mean winds through urban areas O. COCEAL and S. E. BELCHER University of Reading, UK.
Chapter 8: Precipitation ATS 572. “Precipitation” Can be: 1.Rain 2.Snow 3.Hail 4.Etc. However, it MUST reach the ground. –Otherwise, it is called “virga”—hydrometeors.
On the distribution of relative humidity in cirrus clouds K. Gierens 1, P. Spichtinger 1, H.G.J. Smit 2, J. Ovarlez 3, and J.-F. Gayet 4 1 DLR Oberpfaffenhofen,
Marc Schröder, FUB Tutorial, De Bilt, 10.´04 Photon path length distributions and detailed microphysical parameterisations Marc Schröder Institut für Weltraumwissenschaften,
1 8. One Function of Two Random Variables Given two random variables X and Y and a function g(x,y), we form a new random variable Z as Given the joint.
April Hansen et al. [1997] proposed that absorbing aerosol may reduce cloudiness by modifying the heating rate profiles of the atmosphere. Absorbing.
Random Variable The outcome of an experiment need not be a number, for example, the outcome when a coin is tossed can be 'heads' or 'tails'. However, we.
IACETH Institute for Atmospheric and Climate Science Cirrus Clouds triggered by Radiation Fabian Fusina ETH - Zurich 1 st EULAG Workshop - 9th October.
The Boltzmann Distribution allows Calculation of Molecular Speeds Mathematically the Boltzmann Distribution says that the probability of being in a particular.
Representation of Subgrid Cloud-Radiation Interaction and its Impact on Global Climate Simulations Xinzhong Liang (Illinois State Water Survey, UIUC )
Daniel Grosvenor, Thomas Choularton, Martin Gallagher (University of Manchester, UK); Thomas Lachlan Cope and John King (British Antarctic Survey). Daniel.
Simulation of boundary layer clouds with double-moment microphysics and microphysics-oriented subgrid-scale modeling Dorota Jarecka 1, W. W. Grabowski.
ECE-7000: Nonlinear Dynamical Systems 2. Linear tools and general considerations 2.1 Stationarity and sampling - In principle, the more a scientific measurement.
Stratiform Precipitation Fred Carr COMAP NWP Symposium Monday, 13 December 1999.
Horizontal Variability In Microphysical Properties of Mixed-Phase Arctic Clouds David Brown, Michael Poellot – University of North Dakota Clouds are strong.
Particle Size, Water Path, and Photon Tunneling in Water and Ice Clouds ARM STM Albuquerque Mar Sensitivity of the CAM to Small Ice Crystals.
COSMO General Meeting, WG3-Session, 7 Sep Cloud microphysics in the COSMO model: New parameterizations of ice nucleation and melting of snow.
Modeling. How Do we Address Aerosol-Cloud Interactions? The Scale Problem Process Models ~ 10s km Mesoscale Models Cloud resolving Models Regional Models.
Midlats: MOZAIC [40-60N, 0-75 W] 250 hPa layer Evaluation of upper tropospheric moisture in the GEOS5CCM and MERRA reanalyses and implications for contrail.
MODELING OF SUBGRID-SCALE MIXING IN LARGE-EDDY SIMULATION OF SHALLOW CONVECTION Dorota Jarecka 1 Wojciech W. Grabowski 2 Hanna Pawlowska 1 Sylwester Arabas.
Cloudnet Workshop April 2003 A.Tompkins 1 Changes to the ECMWF cloud scheme  Recent package of changes for 25r3:  COMPLETE REWRITE OF NUMERICS FOR CLOUD.
NUMERICAL SIMULATION CLOUDS AND PRECIPITATION CAUSED CATASTROPHIC FLOODS ALONG THE ELBE RIVER IN AUGUST 2002 Krakovskaia S.V., Palamarchuk L.V. and Shpyg.
Lecture 8: Measurement Errors 1. Objectives List some sources of measurement errors. Classify measurement errors into systematic and random errors. Study.
Update on progress with the implementation of a new two-moment microphysics scheme: Model description and single-column tests Hugh Morrison, Andrew Gettelman,
1)Consideration of fractional cloud coverage Ferrier microphysics scheme is designed for use in high- resolution mesoscale model and do not consider partial.
Radiative Equilibrium Equilibrium state of atmosphere and surface in the absence of non-radiative enthalpy fluxes Radiative heating drives actual state.
Full calculation of radiative equilibrium. Problems with radiative equilibrium solution Too hot at and near surface Too cold at and near tropopause Lapse.
ECMWF Radiation and Clouds: Towards McICA? Towards a McICA representation of cloud-radiation interactions in the ECMWF model Radiation: J.-J.
Clouds (Condensed PPT)
Diagnosing latent heating rates from model and in-situ microphysics data: Some (very) early results Chris Dearden University of Manchester DIAMET Project.
Review for Exam 2 Fall 2011 Topics on exam: Class Lectures:
Investigating Cloud Inhomogeneity using CRM simulations.
Clouds and Large Model Grid Boxes
H. Morrison, A. Gettelman (NCAR) , S. Ghan (PNL)
Condensational Growth
Melting of ice particles:
Relationships inferred from AIRS-CALIPSO synergy
Precipitation I.
Dorota Jarecka1 Wojciech W. Grabowski2 Hanna Pawlowska1
Particle formation and growth
Presentation transcript:

Klaus Gierens Institut für Physik der Atmosphäre DLR Oberpfaffenhofen Modelling of Cirrus Clouds (MOD 10) (MOD 11)

Overview MOD11: Numerical modelling of important microphysical processes in cirrus clouds MOD12: Stochastic cloud modelling

Klaus Gierens Institut für Physik der Atmosphäre DLR Oberpfaffenhofen Numerical modelling of important microphysical processes in cirrus clouds MOD 11

Overview Problems special to cirrus modelling Model types Bulk microphysics models Processes and their representation in my bulk model Some modelling examples

Problems special to cirrus modelling - Radiation 1. Cirrus clouds may heat or cool the Earth-Atmosphere System depending on micro-/macrophysical properties temperature (altitude) generation mechanism (incl. synoptic situation, geogr. location) 2. Complex ice crystal shapes (inter alia T- and S i - dependent) render calculation of radiative transfer a tough problem

Problems special to cirrus modelling – Ice formation 3. Various modes of ice crystal formation  homogeneous freezing of aqueous solution droplets  heterogeneous modes:  deposition freezing  immersion freezing  condensation freezing  contact nucleation  and still other modes from Vali, 2004

Problems special to cirrus modelling - Supersaturation 4. Cirrus clouds have only a loose relation to ice saturation, viz.  they do not form at saturation  once formed, they are not very strongly attracted by the equilibrium state Consequently: there is plenty of ice supersaturated, yet clear air in the UT (sometimes marked by persistent contrails) Cirrus clouds are embedded in supersaturated air masses RHi pdfs within cirrus have long tails into the supersaturated regime Spichtinger et al., 2004 cloudy air INCA data

Cirrus and Climate Change — an unsolved problem Freezing/nucleation thresholds are high above saturation  extremal states in the RHi field Extremal states react much more sensitive to changes of background conditions than do averages. Hence it is difficult to estimate, how the probability will change in a changing climate that in the RHi field the nucleation thresholds will be surpassed. Example: mean S i increases from 10 to 11% but probability to surpass 40% decreases by about 1/3 wrt to the earlier pdf

Model types Models are a compromise between -numerical effort oCPU costs, ocomputing and turnaround time, omemory and storage requirements -and scientific ambition. Models with clouds usually combine -sophisticated dynamics with simple microphysics (NWP, GCM) simple microphysics: bulk microphysics -simple dynamics with a elaborate microphysics; mostly box models with size resolved microphysics: bin microphysics -trajectory calculations with single particle microphysics (recent development)

Peculiar model types Models with both elaborated microphysics and detailed dynamics (e.g. Grabowski’s superparameterisation) are extremely expensive (in terms of computing power). Box models with bulk microphysics are almost never used. But they are very quick and one can learn a lot playing around with such a model (see Gierens, ACP, 2003).

Bulk microphysics models Bulk microphysics: balance equations only for few total concentrations that characterise a cloud. Typically -Mass concentration (1st moment of mass distribution) -Number density (0th moment) -Traditionally, many bulk models only transport the 1st moment (e.g. the classical Kessler scheme) -Now, more dual-moment schemes (0th and 1st moments) Bulk schemes are usually used in NWP and GCM models and in many mesoscale models. Bulk schemes are much faster than bin microphysics schemes, at the expense of giving up information on size distribution (and probably also realism). The ECMWF model uses cloud coverage as a prognostic variable in addition to vapour and liquid/ice water concentrations.

Mathematical modelling of clouds, bulk version Needs assumption on probability density function type for the masses (or sizes) of the various hydrometeor and aerosol classes considered in the model. Marshall-Palmer (i.e. exponential) gamma log-normal uni-modal, bi-modal, multi-modal Note: only the type of the pdf is chosen initially. The parameters generally change with time during evolution of the model cloud.

Common mass pdfs and their moments

avoid too many parameters! Number of parameters that fix the pdf should not exceed the number of prognostic variables by much. Parameters should be functions of the prognostic variables. The functional dependence should be understandable. It is difficult to determine a priori, how higher moments (skewness, curtosis, etc.) will evolve with the evolution of a cloud. Higher moments are difficult to determine from data -sensitive to outliers.

Processes to be included in a (pure) cirrus model Nucleation of the ice phase from -aerosol oliquid (homogeneous) osolid (heterogeneous, various modes) -water droplets Crystal growth and evaporation Crystal sedimentation Crystal aggregation Aerosol dynamics and chemistry (parts of it implicit in nucleation) Radiation (may feed back on growth/evap rates) Processes marked in red are currently included in the Spichtinger/Gierens bulk cirrus physics of EuLag.

a typical set of equations Equations used in the two-moment bulk cirrus scheme by Spichtinger and Gierens. Note also the two forms of ice!

Homogeneous nucleation of aqueous solution droplets Parameterisation after Koop et al. Critical supersaturation Nucleation rate J given as polynomial of aw  awi. In equilibrium the water activity equals the relative humidity wrt liquid water. Non-equilibrium occurs in strong updraughts. Integration over droplet size distribution: -Actual droplet volume derived by inversion of Köhler equation. -For a log-normally distributed dry aerosol mass Gauss- Hermite integration works fine (Gierens and Ström, JAS, 1998).

Köhler equation For a given ambient relative humidity the equilibrium size of a solution droplet is given by the Köhler equation. Simplest form: S = A/r  B/r 3 A/r is the Kelvin term, B/r 3 is the Raoult term

Heterogeneous nucleation simplest assumption possible: -a certain number of solid aerosol particles (typically 1 to 50 per cubic centimetre) -freeze to ice at a certain supersaturation (typically 130%). On evaporation of het. ice, these aerosols are set free and can form new ice afterwards.

Deposition growth and evaporation Parameterisation after Koenig (JAS, 1971): -dm/dt = a m b with temperature, pressure, and supersaturation dependent coefficients a,b. Corrections for kinetic growth regime (small ice crystals)

Integration over mass distribution Integration over mass distribution: -Diffusion regime: -dIWC/dt = a µ b -Kinetic regime: -dIWC/dt = a µ b+  / m 0  a > 0 implies growth of the ice mass concentration: dIWC/dt > 0 -the ice number density is the constant. a<0 implies crystal evaporation (dIWC/dt < 0). -the ice number concentration decreases then, but with a higher relative rate than the ice mass: -(N t-1  N t )/ N t-1 = [(IWC t-1  IWC t )/ IWC t-1 ]  with  =1.1 -(Harrington et al., 1985)

Two-moment sedimentation scheme Flux densities for ice mass and number concentrations Empirical relation between crystal mass and terminal velocity

Two-moment sedimentation scheme, cont’d Allows to express mass and number related terminal velocities as: Since large crystals fall faster than small ones, one must have -v t,m > v t,n -in other words: µ  +1 µ 0 > µ  µ 1. This inequality is always fulfilled (Gierens and Spichtinger, SPL, subm.)

Simulation of different sedimentation Shape of ice crystals: columns Initialising of a thin cirrus cloud at t=0s ( IWC = 10 mg / m 3, N = 100 / dm 3 ) in the altitude range km Simulation time: t = 3600 s

v iwc  v nc v iwc = v nc The two-moment sedimentation scheme nicely obyes the principle that large crystals fall faster than smaller ones. This is not so in the one- moment scheme.

Two vs. One-moment sedimentation, other effects Effect on vertical distribution of ice water mass and number concentration. Effect on SW and LW extinction per model layer. Larger optical thickness in the 2-moment scheme.

Sensitivity studies: homogeneous vs. heterogeneous nucleation Results from the DFG project „Dünner Zirrus“ (thin cirrus). Setup for idealised 2D simulations Model domain: horizontal resolution dx = 100 m, horizontal extension: 6.3 km vertical resolution dz = 50 m, vertical extension: 6 km, i.e km time step dt = 1s, simulation time 6 h = s constant vertical motion for whole model domain (i.e. adiabatic cooling) w = 3 / 4.5 / 6 cm/s Set of number densities of ice nuclei: N i = 1 / 3 / 5 / 7 / 10 / 30 / 50 L -1 Set of thresholds for heterogeneous nucleation: RHi het = 110 / 130 / 140 % Additional temperature fluctuations:  T = 0.1 / 0.05 / 0.01 / / K

Start profiles

Varying ice nuclei number density N i In the following mean values over all 64 columns are shown: x-axis: time in minutes z-axis: altitude in metres colour bar: relative humidity with respect to ice Isolines of equal ice crystal number densities purple: ice crystals formed by homogeneous nucleation black: ice crystals formed by heterogeneous nucleation

N i = 1L -1, w = 4.5 cm/s, RHi het = 130 % Time (min) Altitude (m)

N i = 3L -1, w = 4.5 cm/s, RHi het = 130 % Altitude (m) Time (min)

N i = 5L -1, w = 4.5 cm/s, RHi het = 130 % Altitude (m) Time (min)

N i = 7L -1, w = 4.5 cm/s, RHi het = 130 % Altitude (m) Time (min)

N i = 10L -1, w = 4.5 cm/s, RHi het = 130 % Altitude (m) Time (min)

N i = 30L -1, w = 4.5 cm/s, RHi het = 130 % Altitude (m) Time (min)

N i = 50L -1, w = 4.5 cm/s, RHi het = 130 % Altitude (m) Time (min)

Results 1 If one of these competing nucleation mechanisms (heterogeneous/homogeneous) can produce many ice crystals, relative humidity can be reduced effectively. Two different regimes: -few heterogeneous ice nuclei: homogeneous nucleation is effective -many heterogeneous ice nuclei: heterogeneous nucleation is effective between these two regimes the cloud is very sensitive to the number of ice nuclei; often there is persistent ice supersaturation within the simulated clouds, reaching rather high values. transition between the two regimes depends on the relation between three time scales: growth - sedimentation - cooling

varying threshold humidity for heterogeneous nucleation In the regimes where one formation mechanism is dominant only marginal changes are due to different thresholds In the range where no process is dominant a change in the threshold affects the properties of the clouds quite seriously For low thresholds a “secondary cloud formation” is observed: -Ice crystals sediment and evaporate in the sub saturated layers below the cloud  Moistening of the sub saturated layer  Collection of aerosols in this layer  Due to cooling cloud formation by heterogeneous nucleation

N i = 5L -1, w = 4.5 cm/s, RHi het = 110 % Altitude (m) Time (min) Secondary cloud formation

N i = 5L -1, w = 4.5 cm/s, RHi het = 130 % Altitude (m) Time (min)

N i = 5L -1, w = 4.5 cm/s, RHi het = 140 % Altitude (m) Time (min)

Varying ice nuclei number density N i with additional temperature fluctuations In the following mean values over all 64 columns are shown: x-axis: time in minutes z-axis: altitude in metres colour bar: relative humidity with respect to ice Isolines of equal ice crystal number densities purple: ice crystals formed by homogeneous nucleation black: ice crystals formed by heterogeneous nucleation Temperature fluctuations: Gaussian,  T = 0.1 K

N i = 1L -1, w = 4.5 cm/s, RHi het = 130 %,  T =0K

N i = 1L -1, w = 4.5 cm/s, RHi het = 130 %,  T =0.1K

N i = 7L -1, w = 4.5 cm/s, RHi het = 130 %,  T =0K

N i = 7L -1, w = 4.5 cm/s, RHi het = 130 %,  T =0.1K

N i = 50L -1, w = 4.5 cm/s, RHi het = 130 %,  T =0K

N i = 50L -1, w = 4.5 cm/s, RHi het = 130 %,  T =0.1K

Varying IN number density N i with temperature fluctuations In the regimes where one formation mechanism is dominant only marginal changes are due to temperature fluctuations In the range where no process is dominant temperature fluctuations affect the properties of the clouds quite seriously The effect is in two directions: Temperature fluctuations can oenforce the reduction of relative humidity oslow down the reduction of relative humidity

Klaus Gierens Institut für Physik der Atmosphäre DLR Oberpfaffenhofen Stochastic cloud modelling MOD 12

Stochastic cloud modelling (statistical schemes) Main problem here: Parameterisation of cloud fraction (i.e. fractional cloud cover). Problem for large scale models, not for cloud resolving models. In a CRM a grid box is either cloudy or cloud free (binary or 0-1 scheme). Some old GCMs also use this binary assumption of total or zero cloud cover. The 0-1 schemes neglect sub-grid variability. This leads to errors in all computations, where quantities depend nonlinearly on liquid or ice water path or concentration. Statistical cloud schemes would allow to consistently treat sub-grid variability in cloud microphysical processes and in radiation.

A cloud resolving model runs reasonably well with a 0-1 scheme

In a large scale model the results of a 0-1 scheme are unsatisfying

A cloud fraction looks somewhat better, although problems of cloud overlap assumptions arise, in particular for radiative transfer. Most (all?) models do not assume a variable cloud fraction in the vertical within one grid layer.

RH-controlled parameterisation of cloud cover In some GCM schemes cloud cover is parameterised as a function of relative humidity, e.g. the so-called Sundqvist scheme of ECHAM. Simple statistical scheme: Clouds already form at U c >100%, i.e. at sub-saturated conditions. Interpretation: Fluctuations of RH in the grid box  Supersaturation somewhere  clouds form in a fraction of the box. Some schemes use also vertical wind speed to parameterise C.

Statistical schemes Working principle Consider a phase space (T,RH). In a certain part of the phase space clouds can form, in the remaining part not. Examples: Water clouds: RH>100% Ice clouds: RHi > RHi crit (T) The model predicts at every time step and for each grid box a mean state.

Statistical schemes, cont’d If we know the probability density function of fluctuations of the phase point around the grid-box mean value, we can compute, how probable it is that a fluctuation reaches into the supercritical regime. I call this probability the Overlap Integral .  can be interpreted as the actual cloud coverage C. For numerical reasons it might be better to compute d  / dt and from that dC/dt.

Examples for homogeneous nucleation and contrail formation Red line: critical supersaturation for homogeneous nucleation (Koop theory). Green dots: fluctuations of temperature and relative humidity around the grid mean state (-50°C, 140%).  = (number of dots above the red line) / (total number of dots)

Phase diagram for formation of persistent contrails (for two pressure levels).

Probability density functions for fluctuations Problem: how are fluctuations of the phase state variables distributed. The distribution may depend on -location, -time (e.g. season), -in particular on the spatial scale ospatial resolution of the model ospatial resolution of data (correlation lengths)

Probability density functions for fluctuations, cont’d Generally, pdfs are chosen in an ad hoc way, -data on fluctuations almost non-existent. -pdf selection according to criteria outside of physics, -more inside of mathematics and numerics. -symmetrical pdfs often used, BUT -symmetric pdfs cannot be the true nature of the fluctuations since temperature and relative humidity (or other humidity variables) cannot be negative. Apart from measurements, distributions of fluctuations are also sometimes obtained from cloud resolving model runs e.g. Adrian Tompkins). It is clear that many runs are needed to get a good statistical ensemble.

measured statistics of instantaneous fluctuations MOZAIC data (one year) Gierens et al., Ann. Geophys., 1997

Analytical formulation of the fluctuations The measured fluctuations (on a T42 grid scale, i.e. 250×250 km 2 ) follow closely a Cauchy distribution (Lorentz line shape):   (  ) = (  /  ) / (  2 +  2 ) Cauchy distribution: no moments! (not even a mean value). Widely extended tails. BUT: convolution of two Cauchy distributions yield another Cauchy distribution   1 (  )    2 (  ) =   1+  2 (  )

joint probability density of (T,RH)-fluctuations linearise RH(T) sum of two random variables  convolution evaluate convolution integral

joint probability density of (T,RH)-fluctuations, cont’d Insert the two Cauchy distributions (  ). Result is a product of the original Cauchy distribution for  T with a “rotated” Cauchy distribution for  RH + A  T

Theoretical joint pdf of (T,RH) fluctuations joint pdf of (T,RH) fluctuations constructed from MOZAIC data

Overlap integral for contrail formation, analytically The calculation of the overlap integral effectively smears out the boundaries in the phase space.

overlap integral for Koop parameterisation, numerically

practical considerations – random numbers with certain pdf Random number generators usually produce uniform distribution of variable R on [0,1). Cauchy distribution: distribution of tan , with  uniformly distributed within [  /2,  /2]. hence set random  = tan (  R -  /2) For other distributions: inversion of cumulative distribution function F (integral of the pdf) R F 0 1 x random number x is: x = F -1 (R) where R is a random number in [0,1) produced by a generator.

practical considerations - d  /dt total derivative -d  /dt = (  /  T) (dT/dt) + (  /  RH) (dRH/dt) How to compute the partial derivatives of  wrt the phase variables? Analytical expression (at least with 2D-Cauchy distribution) are VERY complex, unfeasible… Numerical approximation. (  /  T)  [  (T)   (T+dT)] / dT yields noisy results and needs a lot of computing time for computing the random numbers. Better idea: see next slide!

practical considerations - d  /dt, cont’d Temperature derivative: shift the red line in ±T-direction by ±dT, count the number of points between the black lines, divide by total number of points and divide by 2 dT. RH derivative: shift the red line in ±RH-direction by ±dRH, count the number of points between the black lines, divide by total number of points and divide by 2 dRH.