Betsy AndrewsLauren SchmeisserJohn Ogren

Acknowledgements to these model providers: GOCART (Chin et al.), MPIHAM (Stier et al.), SPRINTARS (Takemura et al.), TM5 (Krol et al.)

Preliminary comparison of in-situ measured and AeroCom model-simulated aerosol optical properties

Evaluate model predictions of aerosol optical properties using long-term, in-situ

measurements

Improve the predictive capability of global climate models

• Models often cannot reproduce surface aerosol trends or annual cycles (e.g., Shindell et al., 2008), but the models still are used for predicting atmospheric behavior/climate

• Several studies have used AERONET retrievals of absorbing aerosol to suggest that models significantly underestimate black carbon aerosol

OBJECTIVE

MOTIVATION

Why long-term, in-situ, surface aerosol optical data?

NOAA Long-term Surface Monitoring

Aircraft Campaigns

AERONET Satellite

Length of dataset

Long-term Short-term Long-term Long-term

Temporal resolution

Continuous (1min)

Variable Intermittent Intermittent

Geographical Coverage

Sparse Very Sparse Medium Sparse

Global

Vertical Resolution

Surface only Vertically resolved

Column Column (mostly)

Aerosol optical properties

Complete RFE suite; @ low RH

Various Complete RFE suite (at high loading); @ ambient RH

Various

There are advantages and disadvantages for each data set.

• In-situ data Measurements Locations Aerosol parameters Data availability

• Preliminary comparisons Aerosol Climatology Aerosol Characteristics and Behavior

• Where do we go from here?

Talk Outline

In-situ Aerosol Optical Properties

Aerosol light scattering• 3λ nephelometer (TSI or Ecotech)• Total & hemispheric backscattering

Aerosol light absorption• Instruments: MAAP, PSAP, CLAP• Single and multi-wavelength

Data Collection• Low RH (<40% RH)• 1 min freq• 1 & 10 um size cuts

Data Processing• Edited and corrected• Averaged (H, D, M, Y), • Absorption and scattering reported

at STP

MLO aerosol rack

Optical Parameters Available for Comparison

IN-SITU MODEL OUTPUT

Absorption

Scattering (Extinction-Absorption)

Extinction

Single scattering albedo

Scattering Ångström exponent

Absorption Ångström exponent

Phase function parameterization

Fine mode fraction

Current low RH surface data from models doesn’t capture breadth of parameters available from in-situ measurements:• Climatically important phase function parameterization (e.g. asymmetry parameter)• Source characterization (Ångström exponents, fine mode fraction)

NOAA Stations

BRW

MLO

SMO

SPO

MAO

CPR

ALTSUMRSL

WSA GRW

ETLEGB

PVCAPP

BND

SGPSPL

WHI

PYETHD

NIM

CPT

ARN

FKBKPS

BEOPGH

WLG AMYGSN

LLNHFE

• Currently 25 operational stations• Baseline station measurements go back to 1970s, most sites came online in 2000s• Most NOAA stations have QC’d data in NILU/EBAS database

In-situ Measurements – 2006

• Sites with both scattering AND absorption measurements in 2006• Data may not be in EBAS database

In-situ Measurements – All Years

• Any sites with record of scattering AND absorption measurements• Data may not be in EBAS database• Still fewer sites than AERONET• Gaps in S. America, Africa, Middle East, Russia, Australia

• Number of stations almost doubled from 2006 to 2009• 2006 is not ideal for model/in-situ measurement comparison

Plot shows stations with simultaneous scattering and absorption data*

*Data may not be in EBAS database

Number of Monitoring Stations is Growing!

Maybe a slide with basic model inventory…?

GOCART MPIHAM SPRINTARS TM5

Dry ext&abs2006

HourlyDailyMonthly

DailyMonthly

DailyMonthly

DailyMonthly

Extended time

2000-2007Daily

2006-2008Daily

2000-2008Daily

2000-2009Daily

Humidity (specific) Monthly

DailyMonthly Monthly Monthly

Wasn’t sure what else to add…Need to check if monthly data is available for extended time

Can compare models/measurements from several perspectives…

Preliminary Comparisons

Tells us how well the model is simulating aerosol sources, transport processes,

removal mechanisms, etc.

Tells us how well the model is simulating aerosol aging processes, chemistry,

sources, etc.

CLIMATOLOGY

CHARACTERISTICS &

BEHAVIOR

???What other diagnostics should we consider to analyze the models?

Betsy 2 cents: climatology tells how well models are doing at given locations, but I don’t think climatology is so helpful for picking apart issues with sources/processes…could be right for wrong reasons…In contrast, the behavior stuff (esp. systematic variability) might be very useful for identifying process issues…

Comparisons of Aerosol Climatology

• Annual means• Seasonality

• Model reproduces the annual mean absorption quite well at BRW and NIM• Model does not simulate MLO and WLG as well

• MLO and WLG are situated in complex mountainous terrain

Aerosol Climatology: Annual Mean Absorption

Station font=white for better visibility?

• Arctic is complex aerosol environment

• Model/measurement discrepancy high at some Arctic sites

• No systematic variability in discrepancy (e.g., model underestimates absorption at TIK and ZEP, and overestimates absorption at ALT and SUM)

Focus on the Arctic

Aerosol Climatology: Annual Mean Absorption

Comparison of in-situ and modeled single scattering albedo (SSA) for four models and 9 sites in NOAA’s federated aerosol network for 2006

• GOCART and TM5 most similar to in-situ• Best agreement in Arctic (ALT, BRW) and free troposphere (MLO)• Worst agreement for US continental sites (BND, SGP) – is it a grid size issue?

Aerosol Climatology: Annual Mean SSA

GOCART simulates the seasonal changes at BRW fairly well, but…• Modeled springtime Arctic haze peak is broader • Modeled summertime absorption values are higher

Discrepancies in seasonality may help identify issues with model emissions, transport and/or atmospheric processing

Aerosol Climatology: Seasonality

• Model predicts darker aerosol (lower SSA) than suggested by in-situ observations• Model predicts seasonal variation which is not observed by in-situ measurements• Compare 2006 model output with in-situ data for different years (for SSA only?)

Much AeroCom model output focuses on 2006; many in-situ sites start after 2006

Aerosol Climatology: Seasonality

Comparisons of Aerosol Characteristics & Behavior

• Systematic Relationships• Lag-Autocorrelation/Persistence

• Systematic relationships observed between SSA and extinction in in-situ data are not necessarily reproduced by model output

• Similar relationship observed at some AERONET sites

Systematic variability can provide information about aerosol processes and sources

In-situ: Lower loading corresponds to darker (and smaller) particles

preferential scavenging of large, scattering aerosol by clouds/precipitation?

Aerosol Behavior: Systematic Variability

ArcticContinentalClean MarinePolluted Marine

• Relationship between aerosol loading and aerosol size distribution changes with location

• Currently no model output to evaluate this sort of systematic variability for surface, low RH conditions

Aerosol Behavior: Systematic Variability

Arctic

Continental

Clean Marine

Polluted Marine

Aerosol Behavior: Systematic Variability

• Relationship between aerosol loading and aerosol size distribution changes with location

• Currently no model output to evaluate this sort of systematic variability for surface, low RH conditions

• Constrain comparisons by identification of expected ‘best case’ agreement between data sources with different temporal/spatial spacing

• Provides information about atmospheric processes, especially for higher frequency data (e.g., NPF, uplope/downslope…)

Lag is the time between measurements being compared (Dt)‘r’ is the lag autocorrelation statistic.

Dt=1 h, r=0.96 Dt=3 h, r=0.86

Dt=12 h, r=0.68 Dt=24 h, r=0.57

Scattering at t=t0

Scatt

erin

g at

t=t 0+

Dt

Aerosol Behavior: Lag-Autocorrelation

• TM5 captures the persistence and seasonality of aerosol extinction at some sites quite well (BRW, WLG).

• In general, in-situ aerosol extinction appears to be less persistent than predicted by the model.

Lag-autocorrelation will vary from site to site as a function of sources, processes and transport affecting the aerosol at that location

Aerosol Behavior: Lag-Autocorrelation

CN absorption scattering Anderson

Lag

Coeffi

cien

t ‘r’

hours

High frequency data (at least hourly!) + parameter co-variance can highlight atmospheric processes

BRW – no diurnal effects, very persistent aerosolBND – CN and absorption co-vary diurnally (local source + NPF?)MLO – CN, absorption and scattering co-vary (upslope/downslope)CPT – CN exhibits diurnal variability (NPF?)

Depending on site, aerosol properties may co-vary

Aerosol Behavior: Lag-Autocorrelation

Where do we go from here?

Various other methods exist for scoring model/measurement comparisons (e.g., Glackler et al., 2008; Murphy and Epstein, 1989).

Quantifying Model/Measurement Agreement

Taylor diagrams provide a way of graphically summarizing how closely a model matches observations. • Correlation (R)• Root-mean-square difference • Standard deviation

Modelled daily scattering (GOCART) tends to under-predict observed 2006 scattering variability at several sites. R<0.8 for all sites.

• Point measurement vs Area prediction• “…sites dominated by local pollution or sites near mountains are

expected to introduce unwanted biases with respect to the regional average” (Kinne et al., 2006)

• Meteorological adjustments • e.g., Measurement to ambient conditions (T, P, RH)

• Averaging • In-situ daily: 0 UTC-24 UTC, start of average• Model daily: ??

Potential Issues for In-situ/Model Comparisons

• Vertical profiles (BND and SGP) • Dry vs ambient – what does it tell us about f(RH) and how does that

compare with sparse in-situ humidograph measurements• IMPROVE network measures scattering at ambient RH; only in US

In-situ Profiles over BND Modelled Profiles over BNDModel data from P. Ginoux in 2008 (AM2 model)

Beyond surface observations at low RH…

Model Output Wishlist

• Spectral scattering and absorption (dry, surface aerosol)• Indicator of phase function (e.g., asymmetry parameter, backscatter fraction

or upscatter fraction) for dry, surface aerosol• RH (only some models output daily specific humidity)• Output for specific locations (i.e., GAW sites)• Higher frequency (hourly!) data

• Potential for lots of measurement/model comparisons

• Climatological comparisons tell us how models are doing now and may identify regions of difficulty for models

• Behavioral comparisons may indicate discrepancies in aerosol modules in terms of atmospheric processing

Takeaways

Questions? Comments? Let’s Discuss!

Extra slidesAdd these?:--systematic variability plots for other models?--AAE vs SAE plot with aerosol type matrix overlaid?--map of locations where we have long term (at least ~1 yr) f(RH) data? (AMF sites, GSN, THD, SGP, BRW, UGR, APP) plus paul zieger’s sites (ZEP, JFJ, MHD, CES, MPZ) [if we were really cool we could add field campaign f(RH) sites (CBG, HLM, KCO, Carrico sites), but that’s lots of work!!]

• Currently, have not included aethalometer data sets due to correction scheme issues

• Including aethalometer data increases number of sites with in-situ absorption data

Aethalometers

Aethalometer (Mm-1)

CLAP

(Mm

-1)

What Site?

Preliminary analyses suggest properly corrected historical(?) aethalometer data are in good agreement with better characterized aerosol absorption instruments.

y=0.999x+0.071, R2=0.97

Lag-Autocorrelation

Differences in lag-autocorrelation amongst models may be due to grid size, grid boundaries, differences in atmospheric processes and/or some combination.

GOCART is amazing in Arctic!

All models have difficulties with KPS (Hungary)

All models look good at CPR (Carribean) and WLG (China)

Plots courtesy of Jim Sherman (Sherman et al., ACPD, 2014)

Scattering

Single Scattering Albedo

All Data (1997-2013)First 4 years (1997-2000)Last 4 years (2010-2013) Fairly large inter-

annual variability in scattering

Little inter-annual variability in single scattering albedo

Inter-annual variability

1.0

0.8

0.6

0.4

0.2

0.0

-0.2

s

1.00.80.60.40.20.0Fo = POM / (POM+SO4)

ACE Asia s= 0.9 - 0.7Fo r2 = 0.66

ICARTT s= 0.8 - 0.5Fo r2 = 0.42

INDOEX s= 0.5 - 0.3Fo r2 = 0.36

ACE Asia y=0.9-0.7x INDOEX y=0.5-0.3x Nova Scotia y=0.8-0.5x Holme Moss y=0.6-0.3x Point Reyes y=0.7-0.1x

OC/(OC+SO4)

g

Quinn et al., GRL, 2005

No OC All OC

.

Mor

e H

2O u

ptak

e

Can we relate modelled aerosol water to Quinn in-situ parameterization?

Hyg

rosc

opic

gro

wth

fact

or