Impacts of Aerosols on Clouds and Climate Athanasios Nenes School of Earth & Atmospheric Sciences School of Chemical & Biomolecular Engineering Georgia Institute of Technology JPL Seminar, 6 May 2010 Acknowledgments : NASA, NSF, NOAA, ONR, CIRPAS Seinfeld/Flagan group (Caltech), Adams group (CMU), ARCTAS/ARCPAC Science teams (NASA/NOAA) How do (liquid water) clouds form? Clouds form in regions of the atmosphere where there is too much water vapor (it is supersaturated). This happens when air is cooled (primarily through expansion in updraft regions and radiative cooling). Cloud droplets nucleate on pre-existing particles found in the atmosphere (aerosols). This process is known as activation. Aerosols that can become droplets are called cloud condensation nuclei (CCN). CCN that activates into a cloud drop Aerosol particle that does not activate Cloud Humans can affect clouds and the hydrological cycle By changing global CCN concentrations (air pollution). Polluted clouds are whiter, change in precipitation efficiency. This yields a net cooling on climate and is called the indirect climatic effect of aerosols. Clean Environment CCN Lower Albedo (few CCN) Polluted Environment (more CCN) CCN Higher Albedo Increasing particles tends to cool climate (potentially alot). Quantitative assessments done with climate models. Rosenfeld et al., Science Satellite observation of clouds in the Black Sea. Red: Clouds with low reflectivity. White: Clouds that reflect alot. Blue: Clear sky. Observational evidence of indirect effect Power plant Lead smelter Port Oil refineries Rosenfeld et al., Science Wind direction Satellite observation of clouds in the Black Sea. Red: Clouds with low reflectivity. White: Clouds that reflect alot. Blue: Clear sky. Air pollution can affect cloud properties Observational evidence of indirect effect Aerosol Size Distribution and Chemical Composition Cloud Radiative Properties, changes in precipitation Cloud Droplet Number and Size This problem largely depends on the relationship between aerosol number concentration and cloud droplet number concentration. Empirical relationships are often used. Quantification of the Indirect Effect in Climate Models Spatial pattern of IF follows that of aerosol variations European and Asian pollution plumes Biomass burning North America pollution plumes Long-range transport Sotiropoulou et al., 2007 Global Annual Average Aerosol Indirect Radiative Forcing (W m -2 ) Problems with GCM assessments of aerosol indirect effect Cloud formation happens at smaller spatial scales than global climate models can resolve. Cloud formation happens at smaller spatial scales than global climate models can resolve. Aerosol-cloud interactions are complex. Aerosol-cloud interactions are complex. Climate models provide limited information about clouds and aerosols. Climate models provide limited information about clouds and aerosols. 3 3 grid climateprediction.net Describing cloud formation explicitly in global models is VERY expensive. These calculations need to be simplified (parameterized). Describing cloud formation explicitly in global models is VERY expensive. These calculations need to be simplified (parameterized). Approach for aerosol-N d : empirical Large variability. Why? Unaccounted: Meteorology Cloud microphysics Composition etc Many studies still utilize this type of approach. Large predictive uncertainty, without chances of improving. Droplet Concentration Aerosol sulfate concentration (proxy for pollution) (Boucher & Lohmann, 1995) Increasingly polluted Current Direction: Use simplified but physically based approaches for cloud processes Dynamics Updraft Velocity Large Scale Thermodynamics Particle characteristics SizeConcentration Chemical Composition Cloud Processes Cloud droplet formation Drizzle formation Rainwater formation Chemistry inside cloud droplets Links/feedbacks need to be incorporated (in some way). VERY challenging problem (Feingold and Stevens, 2009) aerosol activation drop growth collision/coalescence So when does an aerosol particle act as a CCN ? As the droplet size decreases, its equilibrium vapor pressure increases (Kelvin effect). Less molecules around in small drops to pull H 2 O in the droplet phase Wet diameter m Kelvin Relative Humidity (%) Start from a pure H 2 O drop When does an aerosol particle act as a CCN ? Take same drop and add some solute; e.g., (NH 4 ) 2 SO 4 Dissolved material decreases vapor pressure (Raoult effect). As particle grows, this effect is less important Put both effects together: You get the equilibrium vapor pressure of a wet aerosol particle. When does an aerosol particle act as a CCN ? Wet Aerosol (Haze) Wet aerosol diameter m Relative Humidity (%) Critical Wet Diameter, D c Critical S c Cloud Droplet When ambient saturation ratio exceeds S c, particles act as CCN. CCN = f (Size, Chemical Composition) When does an aerosol particle act as a CCN ? Dynamical behavior of an aerosol particle in a variable RH environment. Testing CCN activation theory: CCN Closure studies [CCN] measured [CCN] predicted 1:1 Compare measurements of CCN to predictions using Khler activation theory. Aerosol Size Distribution dN/dlogd p [cm -3 ] d p, nm Use theory to predict the particles that can act as CCN based on measured chemical composition and CCN instrument supersaturation. [CCN] predicted integrate CCN Closure Finokalia Finokalia Aerosol Measurement Campaign (FAME-07) Summer 2007 DMT CCN counter Supersaturation range: % TSI 3080 SMPS Size range: nm Low-vol impactor Ionic composition measured via IC WSOC/EC/OC also measured (Bougiatioti et al., ACP, 2009) 2% overprediction (on average). Introducing compreshensive composition into CCN calculation gives excellent CCN closure. Khler (CCN activation) theory really works. Finokalia Aerosol Measurement Campaign (FAME-07) CCN closure (Bougiatioti et al., ACP, 2009) CCN activation requires knowledge of cloud RH Approach : use the simple story of droplet formation Basic ideas: Solve conservation laws for energy and the water vapor condensing on aerosol particles in cloudy updrafts. aerosol activation drop growth S S max t Steps are: Air parcel cools Eventually exceeds dew point Water vapor is supersaturated Droplets start forming on existing CCN. Condensation of water on droplets becomes intense. S reaches a maximum No more droplets form A classical nucleation/growth problem Input: P,T, vertical wind, particle size distribution,composition. Output: Cloud properties (droplet number, size distribution). How: Solve one algebraic equation (instead of ODEs). Mechanistic Cloud Parameterizations efficiently solve drop/ice formation problem Characteristics: times faster than numerical cloud models. - Impressive agreement with numerical models. Examples for liquid clouds: Abdul-Razzak et al., (2000); Nenes and Seinfeld (2003), Fountoukis and Nenes (2005), Ming et al., (2007); Barahona and Nenes (2007) Examples for ice clouds: Liu and Penner (2005); Barahona and Nenes (2008, 2009ab) Krcher and Lohmann (2001) Apply them to real clouds. In-situ data obtained from airborne platforms is ideal for this purpose! Are these parameterizations good enough? CIRPAS Twin Otter NOAA P3 Observed Aerosol size distribution Observed Cloud updraft Velocity Observed Aerosol composition Parameterization Predicted Cloud droplet number Compare Observed Cloud Droplet Number Parameterization Evaluation CDNC closure Meskhidze et al.,JGR (2005) Parameterization agrees with observed CDNC Agreement to within a few % (on average) ! CIRPAS Twin Otter CRYSTAL-FACE (2002) Cumulus clouds Meskhidze et al.,JGR (2005) Parameterization agrees with observed CDNC Agreement to within a few % (on average) ! CIRPAS Twin Otter CRYSTAL-FACE (2002) Cumulus clouds Great agreement when you know the input. Aerosol Problem: Complexity An integrated soup of Inorganics, organics (1000s) Particles can have uniform composition with size or not Can vary vastly with space and time (esp. near sources) Organic species are a headache They can facilitate cloud formation by acting as surfactants and adding solute (hygroscopicity) They can facilitate cloud formation by acting as surfactants and adding solute (hygroscopicity) Oily films can form and delay cloud growth kinetics Oily films can form and delay cloud growth kinetics In-situ data to study the aerosol-CCN link: Usage of CCN activity measurements to constrain the above chemical effects on cloud droplet formation. Understanding & parameterizing CCN activity Organics and inorganics add solute and can be expressed in terms of a hygroscopicity parameter, k (Petters and Kreidenweis, 2007) org ~ and depends on oxidation state and precursor. If organics are composed of a soluble and an insoluble fraction, then org sol sol + insol insol ~ sol sol 0 See how the of water-soluble organics vary in samples org org + inorg inorg inorg ~ 0.6 for (NH 4 ) 2 SO 4 to 0.9 for NaCl Count CN Condensation Particle Counter, 3010 Count CCN Continuous Flow Streamwise Thermal Gradient Counter Aerosol Generation Shaker, Atomizer, or ambient sample Polydisperse Aerosol Monodisperse Aerosol Size Selection Scanning Mobility Particle Sizer Determining : size-resolved CCN measurements Particle Detection Monodisperse Aerosol Count CN Condensation Particle Counter, 3010 Count CCN Continuous Flow Streamwise Thermal Gradient Counter Aerosol Generation Burrell Wrist Action Shaker Polydisperse Aerosol Size Selection Scanning Mobility Particle Sizer Particle Detection Results: activation curves CCN/CN as a function of d Mobility Diameter (nm) Activated Fraction of particles (CCN/CN) Activated CCN Fraction CN = Determining : size-resolved CCN measurements Monodisperse Aerosol Count CN Condensation Particle Counter, 3010 Count CCN Continuous Flow Streamwise Thermal Gradient Counter Aerosol Generation Burrell Wrist Action Shaker Polydisperse Aerosol Size Selection Scanning Mobility Particle Sizer Particle Detection Mobility Diameter (nm) Activated Fraction of particles (CCN/CN) Determining : size-resolved CCN measurements d 50 : diameter for which 50 % of the particles activate into cloud droplets. d 50 : aerosol becomes more hygroscopic Parameterizing the CCN activity data using methods based on Khler-theory Determine d 50 dependence on supersaturation. Fit the measurements to a power law expression. Relate fitted coefficients to aerosol properties (e.g. hygroscopicity parameter ) by applying theory: Petters and Kreidenweis, ACP (2007); Padr et al., ACP (2007) (NH 4 ) 2 SO 4 but is a normalized average molecular weight of the solute in the aerosol and can be used to infer its nature. Constants from experimental data Average molecular weight over density (molar volume) This is Khler Theory Analysis (Padr et al., ACP, 2007) Inferring molar volume (molecular weight) For a particle containing only water-soluble organics: so Major findings on water-soluble organics Many aged soluble organics (SOA) from a wide variety of sources have a remarkably similar hygroscopicity. Aged organics in Mexico City aerosol from MILAGRO (Padr et al, in press). sol = 0.28 0.06, regardless of location and time ! Daytime samples contain organics ~200 amu on average and depress surface tension a few % Nighttime samples contain organics ~400 amu on average and depress surface tension % Molar volume/surfactant (KTA) analysis suggests that: Changes in surface tension partially compensates for shifts in average molar volume to give the constant sol Major findings on water-soluble organics Organic Aerosol from biogenic VOC emissions -pinene, monoterpene oxidation (Engelhart et al., ACP, 2009). sol ~ 0.28 Anthropogenic VOCs (Asa-Awuku et al.,ACP, 2010) terpinolene, cycloheptene, 1-methylcycloheptene ozonolysis -caryophyllene (Asa-Awuku et al., ACP, 2009). KTA suggests that: both systems contain organics ~200 amu on average and depress surface tension a few % sol ~ 0.26 sol ~ Data from SOA chamber experiments Major findings on water-soluble organics Aerosol samples from prescribed fire burning (Sullivan et al., 2006) sol ~ 0.33(Asa-Awuku et al., 2008) Even biomass-burning aerosol exhibits similar properties: Speciation across samples varies considerably, but their cumulative effects on CCN activity are about the same. Changes in surface tension partially compensates for shifts in average molar volume to give the constant sol What matters is the fraction of soluble organic Complexity sometimes simplifies things for us. org = (0.250.05) sol Date: April, Platform: NOAA WB-P3 Arctic haze / pollution; Flights over ice / leads Map of all Alaskan flights during the ARCPAC campaign. NOAA ARCPAC: Aerosol Radiation and Cloud Processes affecting Arctic Climate Flight tracks for the ARCTAS-B summer deployment. Jacob et al, ACP, 2009 NASA ARCTAS: Arctic Research of the Composition of the Troposphere from Aircraft and Satellites Test organic parameterization on Arctic CCN Date: April, June, July Platforms: NASA WB-P3, DC-8 Arctic haze / pollution; Boreal Forest Fires CCN In Fresh Biomass Burning Plumes: neglecting organics Tendency towards under-prediction. WSOC and AMS data suggest 30-60% of organics in fires are water-soluble. Large under-prediction for biomass burning! Closure Inputs: AMS bulk chemistry SS: % Surface Tension of Water Closure Error Histogram Lathem et al., manuscript in preparation Explicitly include organic impacts on CCN activity as rg = WSOC * 0.25 Most of the CCN prediction bias can be accounted for with a simple rule. The remaining scatter can be attributed (mostly) to size-dependent composition. Before (# 1) After (# 2) CCN In Fresh Biomass Burning Plumes: including organics Large under-prediction is now gone! Some take-home messages CCN theory is adequate for describing cloud droplet formation. Physically based formulations for description of cloud droplet formation in GCMs are comprehensive but are still computationally feasible. Size-resolved measurements of CCN activity are very useful for constraining the extent and sources of aerosol hygroscopicity on cloud droplet formation. The water-soluble fraction of oxidized organics is very hygroscopic, and is surprisingly constant. The cumulative effect of organics on CCN activity can likely be described by a simple relationship of the form: org = (0.250.05) sol org = (0.250.05) sol Cirrus (Ice) Cloud Parameterization Wet aerosol particles Homogeneous Freezing Mainly depends on RH i and T Heterogeneous Freezing (Immersion, deposition, contact, ) Also depends on the material and surface area + Insoluble Material (Ice Nuclei) Ice Crystal Population Multiple mechanisms for ice formation can be active. RH i (%) V Homogeneous Heterogeneous Liquid droplets + Insoluble material Ice Crystals Soluble and insoluble aerosol initial distribution Expansion cooling and ice supersaturation development Heterogeneous IN freezing begin forming ice Crystal growth, fresh IN continue to freeze and deplete vapor Homogeneous freezing of droplets Conceptual Model of Ice Formation in Cirrus Height (km) The Effect of IN on Ice Crystal Number Concentration Ice Crystal Concentration (cm -3 ) N c =N lim Homogeneous Homogeneous and Heterogeneous Heterogeneous Ice Nuclei Concentration (cm -3 ) Barahona and Nenes, ACP, 2009a. Parameterization Development Part 1: Pure Homogeneous Freezing freezing time Parcel Ice Saturation Ice Crystal SoSo growth Barahona and Nenes, JGR, 2008 Define freezing probability as a function of T and P Define freezing probability as a function of T and P Link the freezing probability to crystal size and concentration Link the freezing probability to crystal size and concentration Solve for crystal concentration at the maximum RH i Solve for crystal concentration at the maximum RH i Find a Solution of: (Crystal population balance) Parameterization Development Part 1: Pure Homogeneous Freezing Find a Solution of: (Crystal population balance) Parameterization Development Part 1: Pure Homogeneous Freezing Find a Solution of: Calculate freezing probability: (Crystal population balance) Parameterization Development Part 1: Pure Homogeneous Freezing Find a Solution of: Calculate freezing probability: Calculate size distribution: (Crystal population balance) Parameterization Development Part 1: Pure Homogeneous Freezing Find a Solution of: Calculate freezing probability: Calculate size distribution: Solve supersaturation balance: (Crystal population balance) Parameterization Development Part 1: Pure Homogeneous Freezing Find a Solution of: Calculate freezing probability: Calculate size distribution: Solve supersaturation balance: (Crystal population balance) Parameterization Development Part 1: Pure Homogeneous Freezing LOTS AND LOTS OF MATH (and scaling) The solution of the PDE system reduces to: A few lines of FORTRAN code. It is completely theoretical (i.e. rigorous and robust) and captures the (complex) physics of ice nucleation. We can also calculate the size distribution (not shown) Ice Crystal Number Concentration Homogeneously Frozen Fraction f (aerosol size and concentration, P and T ) Aerosol Number Concentration Barahona and Nenes, JGR, 2008 Parameterization Development Part 1: Pure Homogeneous Freezing IN tend to freeze early, grow and deplete water vapor IN tend to freeze early, grow and deplete water vapor Can inhibit homogeneous nucleation Can inhibit homogeneous nucleation IN also contribute to the total ice crystal concentration IN also contribute to the total ice crystal concentration Height RH i V Crystal Growth Homogeneous Heterogeneous (%) Liquid droplets + insoluble material Ice Crystals Homogeneous Heterogeneous Parameterization Development Part 2: Including Effects of IN Barahona and Nenes, JGR, 2008; ACP, 2009a Global water vapor balance Energy balance Ice water vapor condensation Ice crystal size distribution evolution = nucleation + growth Ice crystal growth even more math and scaling Parameterization Development Part 2: Including Effects of IN Barahona and Nenes, ACP, 2009a. Homogenously frozen fraction without IN (Part 1) Homogenously frozen droplet fraction with IN effects 2/3 2/3 lim, 1 N N ff IN c c Ice nuclei concentration that prevents homogeneous nucleation Analytical solution; explicitly depends on cloud formation conditions and ice nuclei characteristics Parameterization Development Part 2: Including Effects of IN Applies whether homogeneous freezing contributes crystals (i.e., if N IN N lim ). Keeps the heterogeneous nucleation spectrum in its most general form: Can account for several freezing modes and aerosol species Can use theoretical and/or empirical IN data Barahona and Nenes, ACP, 2009b. Heterogeneously Frozen fraction Parameterization Development Part 2: Including Effects of IN Final Answer: Crystal concentration Cirrus Parameterization evaluation with numerical model Evaluation over a broad range of updraft, temperatures, and aerosol (soluble and insoluble) concentrations Average error below 5 10 %, for combined homogeneous and heterogeneous freezing Barahona and Nenes, ACP, 2009b. Problem with Implementation of all this in Climate Models Cloud formation: < 1 km GCM grid cell size: 100 km Mesoscale grid cell size: 16 km Cloud formation: < 1 km GCM grid cell size: 100 km Mesoscale grid cell size: 16 km Clouds form at much smaller scales than models can resolve. Implementing all this parameterized physics is not at all trivial! Problem with Implementation of all this in Climate Models Use the probability density function (PDF) of vertical velocity, p(w), in the grid cell Treating subgrid cloud processes Probability that vertical velocity lies between w and w+dw : Typical Assumption: Gaussian PDF w (m s -1 ) p(w) Fitting parameters: Average updraft velocity Diagnosing the PDF from GCM Assume (boundary layer, stratiform) Average updraft velocity: Turbulent kinetic energy (gravity wave energy, etc.) resolved in model can be used to diagnose and e.g., TKE ~ (in boundary layers) Using PDF for grid-average N d Fraction of droplets from vertical velocities between w and w+dw PDF integrations have been used in large scale models [e.g., Ghan et al., 1997] but is slow because of numerical integration. Add up the droplets emerging from each updraft in the PDF. Upon integration, we get for grid-average N d : Characteristic Velocity: The fast and consistent approach Use the characteristic velocity, w *, that gives the PDF-averaged property: If we knew w *, then we get PDF-averaged values by calling a parameterization once. Much faster than a numerical integration, but of comparable accuracy. Morales and Nenes, in review Characteristic Velocity Each property/process has its own Each property/process has its own w * Droplet number Effective Radius Autoconversion rate different between each other and from Because N d, r eff and A scale differently with N d ( w ) Morales and Nenes, in review Deriving Characteristic Velocity The most simple (elegant) form possible Twomey [1959] parameterization a simple power law. k is an aerosol-dependent parameter. Substituting into gives or 89% of the average updraft Morales and Nenes, in review gives Applying Characteristic Velocities in a GCM simulation Kharoutdinov & Kogan (2000) Diagnose from simulation (dynamics) PDF-Average N d PDF-Average r eff Call N d parameterization PDF-Average A Biases associated with NOT using characteristic Velocity ~8% overestimation in grid-average N d (not bad) Use instead of appropriate w * (current practice) ~10-15% underestimation in grid-average r eff 1-2 m bias (important) Factor of 5 underestimation in grid-average A Kharoutdinov & Kogan (2000) (very important) Using characteristic velocity can take care of an important source of tuning done in GCMs Morales and Nenes, in review Some take home messages PDF-averaging is a good (first-order) way to address the scale-gap when applying parameterizations in models. Characteristic velocity is a fast and consistent way of calculating PDF (grid-) average cloud parameters and processes in models. Each process (cloud quantity) has a characteristic velocity, but it always scales nicely to the PDF characteristics Using characteristic velocity can account for important biases that GCMs currently treat through empirical tuning. Summary Physically-based representations of droplet and ice formation in GCMs is becoming sophisticated but still computationally feasible. CCN theory is adequate for describing cloud droplet formation. Simplified treatments of aerosol composition get most of the CCN number right. Size-resolved measurements of CCN activity are very useful for constraining aerosol complexities (especially organic effects) on cloud droplet formation. Correct coupling of the parameterizations with dynamics and consideration of its subgrid variability is a key issue. Uncertainty of climate model predictions of the indirect effect will still be significant (and probably grow), but we are getting much better at knowing how much it really is. THANK YOU !!