Understanding the impacts of human activities on atmospheric
particles, clouds, storms and climate
Athanasios Nenes1Earth and Atmospheric Sciences, Georgia Institute of Technology
2Chemical and Biomolecular Engineering, Georgia Institute of Technology3ICE-HT, Foundation for Research and Technology-Hellas, Greece
Acknowledgments: NASA MAP/ACMAP, Phillips 66, NSF, EPA, DOE ONR, Georgia Tech team (Nenes, Kostantinidis groups)
Earth’s Energy BalanceSunlight
“visible radiation”
Heat
“infrared radiation”
When energy IN = energy OUT, climate is “in balance” (i.e., steady state)
235 Watts per square meter (Wm-2)
235 Watts per square meter (Wm-2)
Clouds: a significant contributor to Earth’s reflectivity of solar radiation
J.T. Houghton: “The science of climate change”
Facts:• Clouds account for ~50% of planetary reflectivity (albedo). • Small changes in clouds yield large changes in global energy balance. •A few % increase in global cloud cover can counteract warming from greenhouse gases.
Consequence:
Good representation of clouds in models is crucial for understanding climate change.
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) with ~ 0.1µm diameter.
Aerosols that can become droplets are called cloud condensation nuclei (CCN).
CCN that activatesinto a cloud drop
Aerosol particlethat does not activate
Cloud
Can humans affect clouds and the hydrological cycle?Yes! By changing global CCN concentrations (air pollution). Result: Clouds tend to be “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
Red: Clouds with low reflectivity.White: Clouds that reflect alot. Blue: Clear sky.
Observational evidence of indirect effect
Satellite observation of clouds near Australia.
Power plantLead smelter
Port
Oil refineries
Rosenfeld et al., Science
Wind direction
Satellite observation of clouds near Australia.
Red: Clouds with low reflectivity.White: Clouds that reflect alot. Blue: Clear sky.
Air pollution affects cloud reflectivityObservational evidence of indirect effect
Climate Models: the tools of understanding
Divide the Earth into small parts (“grid cells”). For each write equations describing:
• Conservation of Energy, Water, chemical constituents
• Aerosol population balances and their evolution
• Interactions of land/ocean with atmosphere … etc.
3°× 3° grid
climateprediction.net
Prescribe initial conditions (e.g., climatology).
Integrate the equations (numerically) over time.
Models Need Fast Physics: Simple expressions capturing important cloud physics
DynamicsUpdraft VelocityLarge Scale Thermodynamics
Particle characteristicsSize & ConcentrationChemical Composition
Cloud ProcessesCloud droplet formationIce crystal formationEffects of entrainment/mixingCollision/coalescence“Scaleup” of processes
Links/feedbacks need to be incorporated (at appropriate scales).VERY challenging problem (Stevens and Feingold, 2009)
aerosol
Activationnucleation
growth
Goal: Predict drop/ice number concentration for “characteristic” cloud types.
Calculating droplet number in a climate model
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 in each grid cell.
aerosol
activationdrop growth
S
Smax
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
Examine the equilibrium vapor pressure of a wet aerosol particle.Consider the effects of solute and droplet curvature
So… when does an aerosol particle act as a CCN ?
0.1 1 10
Wet aerosol diameter (µm)
Solute effect on vapor pressure
Curvature effect on vapor pressureKelvin+Raoult:
Equil. RH of wet particle
98
99
100
101
102
Rel
ativ
e H
umid
ity (%
)
The combined Kelvin and Raoult effects gives rise to the Köhler equation
(1922).
You can be in equilibrium even if
you are above saturation.
Wet Aerosol(Haze)
0.1 1 10
Wet aerosol diameter (µm)
98
99
100
101
102
Clo
ud, D
ropl
et E
quili
briu
mR
elat
ive
Hum
idity
(%)
“Critical” point
Cloud Droplet Particles act as CCN if the ambient relative humidity exceeds its “critical”
relative humidity
When does an aerosol particle act as a CCN ?Dynamical behavior of an aerosol in a variable relative humidity
environment.
Cloud humidity
When does an aerosol particle act as a CCN ?
A: parameter related to surface tension (controls the Kelvin
effect)
B: parameter related to the moles of solute
in the particle (controls the Raoult
effect)
2/3~ −dryc ds 2/1~ −
solubleεcs
Thermodynamic theory links the critical RH to particle size and composition
2/13
274
=
BAsc
0.1 1 10
1.00
1.01
1.02
Wet aerosol diameter (µm)
Rel
ativ
e H
umid
ity (%
)
98
99
100
101
102
aerosol
activationdrop growth
S
Smax
t
1. Calculate smax (approach-dependent) 2. Nd is equal to the CCN with sc < smax
Algorithm for calculating Nd
(Mechanistic parameterization)
“Putting it together” for droplet number in models
Input: P,T, vertical wind, particle size distribution,composition.Output: Cloud properties (droplet number, size distribution).
Mechanistic Parameterizations: Twomey (1959); Abdul-Razzak et al., (1998); Nenes and Seinfeld, (2003); Fountoukis and Nenes, (2005); Kumar et al. (2009), Morales and Nenes(2014), and others.
Comprehensive review & intercomparison:Ghan, et al., JAMES (2011); Morales and Nenes (2014)
:
Aerosol Problem: ComplexityAn integrated “soup” of
Inorganics, organics (1000’s)Particles can have uniform composition with size…… or notCan 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) 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.
Continuous-Flow Streamwise Thermal Gradient Chamber
wet wallwet
wall
Outlet: [Droplets] = [CCN]
Inlet: Aerosol
Roberts and Nenes (2005), US Patent 7,656,510Lance et al., (2006), Lathem and Nenes (2011),
Raatikainen et al. (2012)
Metal cylinder with wetted walls
Streamwise Temperature Gradient
Water diffuses faster than heat
Supersaturation, S, generated at the centerline = f (Flowrate, Pressure, and Temp. Gradient)
Operated as a spectrometerusing Scanning Flow CCN Analysis(Moore and Nenes, 2010)
Development phases of cloud chambersc
ale
= 1
m
1st versionApril 2002
2nd versionJanuary 2003
Commercial ver. July 2004
Mini-instrumentAugust 2015
Roberts and Nenes, AS&T (2005); Lance et al., AS&T (2006)
Locations sampled over the past 8 years…
Measured:CCN, Aerosol concentrations and size distributions, and aerosol chemistry
Cloud hydrometeor distributions (liquid/ice) and dynamics.
We have sampled:The arctic, urban pollution, biomass burning, marine aerosol, hurricanes, Oil spills, the tropics….
Test the ability to predict CCN concentrations against ambient measurements.
Testing CCN activation theory: CCN “Closure” studies
[CCN]measured
[CC
N] pr
edic
ted
1:1
Compare measurements of CCN to predictions using theory and a simple description of molar volume for organics
Aerosol Size Distribution
dN/dlogdp
[cm-3]
dp , 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
2% overprediction(on average).
CCN prediction theory reallyworks.
The simple treatment of organic hygroscopicitytreatment reallyworks too.
(Bougiatioti et al., ACP, 2009; 2012)
Example: Finokalia Aerosol Measurement Campaigns (Crete, Greece)
Film-forming compounds (e.g., Feingold & Chuang,2002)Slow solute dissolution kinetics (e.g., Asa-Awuku & Nenes, 2007)Glassy states (e.g., Virtanen, 2010)
They can slow down the condensation of water onto growing droplets, because they provide an additional kinetic barrier in addition to gas-phase mass transfer.
These effects are parameterized in terms of changes in the water uptake coefficient, α
Water uptake kinetics: a “big” (important) unknown
water molecule
Slow growthα << 1
water molecule
Rapid growthα ∼ 1
Papers published over 40 years… suggest a from 10-5 to 1.
Water uptake kinetics: a “big” (important) unknown
Varying α changes the condensation rate of water during nucleation of droplets hence smax (and Nd)
NCAR CAM5 Global annual average Nd
Current day Nd (cm-3)
α Uncertainty in the value of α can have a huge impact on predicted droplet droplet number …
Water uptake kinetics: the “big” (important) unknown
Variability of Nd from changes in α can overwhelm the Anthropogenic (Preind-Current day) aerosol indirect effect.
… unless a is between 0.1-1.0 OR remains constant over time.This has never been looked at to date (in a global sense)
wet wallwet
wall
Outlet: [Droplets] = [CCN]
Inlet: Aerosol
Roberts and Nenes (2005), US Patent 7,656,510Lance et al., (2006), Lathem and Nenes (2011),
Raatikainen et al. (2012, 2013)
Standard CCN measurement (>100 instruments in operation).
Metal cylinder with wetted walls
Streamwise Temp. Gradient Water diffuses faster than heat
Supersaturation, S, generated at the centerline = f (Flowrate, Pressure, and Temp. Gradient)
They offer THE opportunity to infer kinetics of water uptake for global aerosol
Water uptake kinetics from CCN observations
Num
ber
Droplet Size (µm)
Initial AerosolDistribution
(polydisperse)
Kinetic modeling: tool to interpret CCN measurements
Num
ber
Droplet Size (µm)
Initial AerosolDistribution
(polydisperse)Aerosol composition,
CCN Instrument OperatingConditions
Iterate to find supersaturation,water vapor uptake coefficient, α
that produces observeddroplet distribution
Lathem and Nenes (2011)Raatikainen et al., (2012)
Kinetic modeling: tool to interpret CCN measurements
Activation Kinetics of SOA sampled during the Deep Water Horizon Gulf Oil Spill
Opportunity to study fresh,
hydrocarbon-rich (i.e.,“oily”) aerosol above
a clean background
Moore et al. (2012)
• Model captures with remarkable fidelity the observed size variations.• Predictive understanding of droplet size ensures that we fully understand the
CCN instrument AND the kinetic analysis• Oily SOA does not exhibit kinetic limitations (α range: 0.1-1.0)• Kinetic analysis with instrument model crucial for this conclusion.
Moore et al. (2012)
June 8, 2010
Activation Kinetics of SOA sampled during the Deep Water Horizon Gulf Oil Spill
Predictions (fully coupled model)Observations
Extend kinetic analysis to global data
• Major global airmass types sampled over 10 field campaigns.• No activation kinetics delays observed in any data to date• Any value of α between 0.1 and 1.0 fits the kinetic activation data• α uncertainty is not a big problem (so far) for climate modeling.
Raatikainen et al., PNAS (2013)
Karydis et al., ACP (2012)
Understanding now the “big picture”
hygroscopicity of soluble particles
updraft velocity
Water uptake kinetics
water adsorption for insoluble particles
aerosol number
aerosol mass
How important is each parameter? Much difference across parameterizations? … and why?
Ice Nuclei Concentrations
Karydis et al., ACP (2012)
hygroscopicity of soluble particles
updraft velocity
Water uptake kinetics
water adsorption for insoluble particles
aerosol number
aerosol massIce Nuclei
Concentrations
Ultimate Goal: SENSITIVITIES
for any parameter that affects crystal/droplet number
Over space and time….
∂Nd
∂p
How important is each parameter? Much difference across parameterizations? … and why?
Understanding now the “big picture”
Traditionally: finite differences • Multiple simulations required per input• Truncation or approximation errors
possible• One sensitivity calculation for one grid
cell
We use automatic differentiation.• One simulation needed for all inputs• Analytical precision • Flexibility and portability
𝜕𝜕𝜕𝜕𝜕𝜕𝜕𝜕𝜕𝜕𝜕𝜕𝜕𝜕
≅ ∆𝜕𝜕𝜕𝜕∆𝜕𝜕𝜕𝜕𝜕𝜕
= 𝜕𝜕𝜕𝜕 𝜕𝜕𝜕𝜕𝜕𝜕+∆𝜕𝜕𝜕𝜕𝜕𝜕 −𝜕𝜕𝜕𝜕 𝜕𝜕𝜕𝜕𝜕𝜕−∆𝜕𝜕𝜕𝜕𝜕𝜕2∗∆𝜕𝜕𝜕𝜕𝜕𝜕
For all grid points and every timestep!
Behind droplet/crystal number sensitivity
size distribution
hygroscopicity of soluble particles
updraft velocity
uptake coefficient
AFHH and BFHH of insoluble particles
aerosol number
Number of DropletsNd = f{Fs(s), smax}
Estimate maximum supersaturation, smax
Soluble ParticlesCalculate supersaturation integral by Fountoukis and
Nenes (2005)
Insoluble ParticlesCalculate supersaturation integral by Kumar et al.
(2009)
πγρw
2aVIK 0,spart( )+ IK spart ,smax( )+ IFHH 0,smax( )
− 1= 0
Smax determined
iterate
Compute CCN spectrum ( Fs(s))
Adjoint Sensitivity Analysis: how it works
Karydis et al., ACP, 2012
Algorithm (calltree) for calculating droplet number
Track calltree & propagate a differential perturbation in Nd
(differential calculus)
updraft velocity
uptake coefficient
AFHH and BFHH of insoluble particles
aerosol number
Number of DropletsNd = f{Fs(s), smax}
Estimate maximum supersaturation, smax
Soluble ParticlesCalculate supersaturation integral by Fountoukis and
Nenes (2005)
Insoluble ParticlesCalculate supersaturation integral by Kumar et al.
(2009)
πγρw
2aVIK 0,spart( )+ IK spart ,smax( )+ IFHH 0,smax( )
− 1= 0
Smax determined
iterate
Compute CCN spectrum ( Fs(s))
Adjoint Sensitivity Analysis: how it works
Karydis et al., ACP, 2012
size distribution
hygroscopicity of soluble particles
updraft velocity
uptake coefficient
AFHH and BFHH of insoluble particles
aerosol number
Number of DropletsNd = f{Fs(s), smax}
Estimate maximum supersaturation, smax
Soluble ParticlesCalculate supersaturation integral by Fountoukis and
Nenes (2005)
Insoluble ParticlesCalculate supersaturation integral by Kumar et al.
(2009)
πγρw
2aVIK 0,spart( )+ IK spart ,smax( )+ IFHH 0,smax( )
− 1= 0
Smax determined
iterate
Compute CCN spectrum ( Fs(s))
Adjoint Sensitivity Analysis: how it works
Karydis et al., ACP, 2012
size distribution
hygroscopicity of soluble particles
Track calltree & propagate a differential perturbation in Nd
(differential calculus)
dN∂
Number of DropletsNd = f{Fs(s), smax}
Estimate maximum supersaturation, smax
Soluble ParticlesCalculate supersaturation integral by Fountoukis and
Nenes (2005)
Insoluble ParticlesCalculate supersaturation integral by Kumar et al.
(2009)
πγρw
2aVIK 0,spart( )+ IK spart ,smax( )+ IFHH 0,smax( )
− 1= 0
Smax determined
iterate
Compute CCN spectrum ( Fs(s))
Adjoint Sensitivity Analysis: how it works
Karydis et al., ACP, 2012
Track calltree & propagate a differential perturbation in Nd
(differential calculus)
dN∂
p∂ for allparameters
size distribution
hygroscopicity of soluble particles
updraft velocity
uptake coefficient
AFHH and BFHH of insoluble particles
aerosol number
Number of DropletsNd = f{Fs(s), smax}
Estimate maximum supersaturation, smax
Soluble ParticlesCalculate supersaturation integral by Fountoukis and
Nenes (2005)
Insoluble ParticlesCalculate supersaturation integral by Kumar et al.
(2009)
πγρw
2aVIK 0,spart( )+ IK spart ,smax( )+ IFHH 0,smax( )
− 1= 0
Smax determined
iterate
Compute CCN spectrum ( Fs(s))
Adjoint Sensitivity Analysis: how it works
Karydis et al., ACP, 2012dN∂
p∂ for allparameters
size distribution
hygroscopicity of soluble particles
updraft velocity
uptake coefficient
AFHH and BFHH of insoluble particles
aerosol number
∂Nd
∂pNd
One Nd calculation gives:
for all p’s and !
Examples of Nd and its sensitivity to input parmeters in the Community Aerosol Model 5.1
[ cm-3 ]
Morales and Nenes, ACP (2014)
Nd
Present Day Emissions (2000)mean: 117 cm-3
Ultrafine particles Coarse particlesFine particles
Nd variability in the CAM 5.1:
Morales and Nenes, ACP (2014)
• How to determine the contribution from each parameter ( χ j ) to the total droplet number variability for each grid cell?• We may be interested in seasonal, monthly or annual average variability…
The adjoint sensitivities allow us to efficiently attribute the contribution from each parameter over space and
time
Aerosol Number
Kappa
Mode Diameter
Ultrafine particles Coarse particles
Updraft Velocity Contribution (%) to Nd variability
Fine particles
Aerosol Number
Kappa
Mode Diameter
Updraft Velocity
2 variables captures most of the Ndvariability (>90%)
% of variability on Nd
Ultrafine particles Coarse particlesFine particles
Lifetime and albedo effect as originally proposed (and implemented in global climate models).
Stevens and Feingold (2009)
Challenge: aerosol-cloud feedbacksDirect aerosol-cloud links for describing drop and ice formation is now included in climate models.
Stevens and Feingold (2009)
Cloud feedbacks partially buffer “perturbations”.
Feedbacks span from cloud-to-global scales
Feedbacks not well understood nor represented in models.
Particle feedbacks on storm intensity
Clouds developing under low aerosol loads:• Precipitation develops early on (before freezing level)• Much of the water “falls” near its source• Relatively little water reaches freezing level• Cold pool from downdrafts generates secondary clouds• Convection localized and generally not intense
Rosenfeld (2008)
Clouds developing under high aerosol loads:• Aerosols reduce drizzle - precipitation delayed• More water reaches freezing level – additional latent
heat “energizes” the storm• Cold pools generates stronger secondary clouds• Convection is intensified
Rosenfeld (2008)
Particle feedbacks on storm intensity
“Invigoration of clouds and the intensification of rain rates is a preferred response to an increase in aerosol concentration.”
Koren et al., Nature Geosci. (2012).
Aerosol-Precipitation interactions
Can particles affect Hurricanes?Violent tropical cyclonic storm systems.
Predictions of storm intensity are particularly challenging, not improving over time.
Are we missing something important in models?
Some more slides on the topic:
In-situ observations of aerosol-cloud interactions that we collected from
hurricanes.
Properties & Responses• Measured CCN / IN activity of bacteria.• Relate observations to surface properties – and see if it
can be linked to bacterial membrane structure.
CCN/IN activity
Contact Angle
In-situ sampling of microbesCIRPAS Twin Otter
NASA DC8
Sampling system
Finokalia, Crete
Collect samples from: • Aircraft for high/low altitudes
and many locations.• Ground-based sites for
seasonal variations.
Bacteria-Cloud interactions:Approach and Goals
Characterization of microbes
Amplification Sequencing
DNA extraction
Assembly Gene search
Molecular & bioinformatics tools used for: • Community composition (16S rDNA)• Characterizing the metagenome of samples.• Look for IN (inaZ) or CCN-relevant genes.• Detect stress response genes.
Current Focus: CCN ActivityCurrent state of knowledge: • Range of critical supersaturations: 0.1-1.0%• Few studies published to date (<10).• Low diversity of bacteria characterized.• Metabolic state (vegetative/spore, live/dead) unknown.Importance: • Role of bacteria as Giant CCN (warm rain initiation)• Cloud scavenging very important for long range
transport of microbes and biogeography.• Mechanistic understanding of how water condenses on
bacteria that act as immersion mode IN.
Some more slides on the topic:
In-situ observations of bioaerosolsand their lifecycle in the marine
boundary layer. Methods development can also be
shown.
Take-home messagesPhysically-based representations of droplet and ice formation in climate models is becoming sophisticated… but still computationally feasible.
Measurements of ambient CCN were critical for bypassing complexities (related to composition and kinetic constants) on cloud droplet formation.
Instrumentation that we helped develop really answered a lot of the outstanding questions around droplet formation.
Direct sensitivity (adjoint) methods are very useful for understanding the important parameters for droplet formation, and what causes its variations in models.
Take-home messagesThere is a lot of work to be done to understand the feedbacks of particles on storm formation and the climate system.
Neglecting these feedbacks can explain the inability to predict e.g., hurricane intensity.
There is also the realization that new particle types (e.g. bacteria and other bioparticles) are present – and with surprising impacts.
Understanding their potential impacts and source function is an open area of research… so keep tuned!
For more information and PDF reprints, please go to
http://nenes.eas.gatech.edu
THANK YOU !!