designing systems to address outstanding issues in climate change betsy weatherhead

Designing Systems to Designing Systems to Address Outstanding Address Outstanding

Issues in Climate ChangeIssues in Climate Change

Betsy WeatherheadBetsy Weatherhead

SPM 1a

Variations of the Earth’s surface temperature for the past 140 years

M.S.U. Channel 2

Trends in Surface and Trends in Surface and Tropospheric TemperatureTropospheric Temperature

Existing Satellite and Surface Existing Satellite and Surface Measurements are not in agreement.Measurements are not in agreement.

Satellites have a difficult time Satellites have a difficult time measuring low in the atmosphere.measuring low in the atmosphere.

Interpretation of satellite Interpretation of satellite measurements requires assumptions measurements requires assumptions about the vertical and chemical about the vertical and chemical structure of the atmosphere.structure of the atmosphere.

NOME FED BLDG OME -165.4 64.5

Trend

-0.6 -0.2 0.2 0.4 0.6

1000

800

600

400

200

POINT BARROW BRW -156.78 71.3

Trend

-0.6 -0.2 0.2 0.4 0.6

1000

800

600

400

200

NAKNEK AKN -156.65 58.68

Trend

-0.6 -0.2 0.2 0.4 0.6

1000

800

600

400

200

LANDER LND -108.72 42.8

Trend

-0.6 -0.2 0.2 0.4 0.6

1000

800

600

400

200

GRAND JUNCTION GJT -108.53 39.12

Trend

-0.6 -0.2 0.2 0.4 0.6

1000

800

600

400

200

EL PASO ELP -106.4 31.82

Trend

-0.6 -0.2 0.2 0.4 0.6

1000

800

600

400

200

DAYTON/WRIGHT PATT DAY -84.12 39.87

Trend

-0.6 -0.2 0.2 0.4 0.6

1000

800

600

400

200

HUNTINGTON HTS -82.55 38.37

Trend

-0.6 -0.2 0.2 0.4 0.6

1000

800

600

400

200

PITTSBURGH/PITTSBG PIT -80.22 40.5

Trend

-0.6 -0.2 0.2 0.4 0.6

1000

800

600

400

200

Sonde Temperature Trends (C/decade) at 12 Z

Weatherhead Thu Nov 8 13:09:45 2001

NOME FED BLDG OME -165.4 64.5

Trend

-0.6 -0.2 0.2 0.4 0.6

1000

800

600

400

200

10

-0.6 -0.2 0.2 0.4 0.6

1000

800

600

400

200

POINT BARROW BRW -156.78 71.3

Trend

-0.6 -0.2 0.2 0.4 0.6

1000

800

600

400

200

10

-0.6 -0.2 0.2 0.4 0.6

1000

800

600

400

200

NAKNEK AKN -156.65 58.68

Trend

-0.6 -0.2 0.2 0.4 0.6

1000

800

600

400

200

10

-0.6 -0.2 0.2 0.4 0.6

1000

800

600

400

200

LANDER LND -108.72 42.8

Trend

-0.6 -0.2 0.2 0.4 0.6

1000

800

600

400

200

10

-0.6 -0.2 0.2 0.4 0.6

1000

800

600

400

200

GRAND JUNCTION GJT -108.53 39.12

Trend

-0.6 -0.2 0.2 0.4 0.6

1000

800

600

400

200

10

-0.6 -0.2 0.2 0.4 0.6

1000

800

600

400

200

EL PASO ELP -106.4 31.82

Trend

-0.6 -0.2 0.2 0.4 0.6

1000

800

600

400

200

10

-0.6 -0.2 0.2 0.4 0.6

1000

800

600

400

200

DAYTON/WRIGHT PATT DAY -84.12 39.87

Trend

-0.6 -0.2 0.2 0.4 0.6

1000

800

600

400

200

10

-0.6 -0.2 0.2 0.4 0.6

1000

800

600

400

200

HUNTINGTON HTS -82.55 38.37

Trend

-0.6 -0.2 0.2 0.4 0.6

1000

800

600

400

200

10

-0.6 -0.2 0.2 0.4 0.6

1000

800

600

400

200

PITTSBURGH/PITTSBG PIT -80.22 40.5

Trend

-0.6 -0.2 0.2 0.4 0.6

1000

800

600

400

200

10

-0.6 -0.2 0.2 0.4 0.6

1000

800

600

400

200

Sonde Temperature Trends (C/decade) at 12 Z


Future Temperature TrendsFuture Temperature Trends

Temperature trends Temperature trends are predicted by a are predicted by a number of different number of different models.models.

Can we identify Can we identify more accurate more accurate models?models?

Detection of TrendsDetection of Trends

Fundamentally: a signal to noise problem.Fundamentally: a signal to noise problem.• We don’t control the signal.We don’t control the signal.• We don’t control the noise.We don’t control the noise.

Equator

year

Te

mp

era

ture

1980 1985 1990 1995 2000

-4

-2

0

2

4

San Francisco

year

Te

mp

era

ture

1980 1985 1990 1995 2000

-4

-2

0

2

4

MSU Channel 2

Weatherhead Tue Dec 11 20:36:28 2001

1000 mb Hadley B2

-0.5 0.0 0.5 1.0

AVAILABLE TEMPERATURE SONDE TRENDS AT 0Z

-150 -100 -50 0

02

04

06

08

0

ABQ

ACC

ACK

ACM

ACY

ADK

ADQ

AHN

AKN

ALB

AMA

ANC

ANN

AQQAYS

BDI

BET

BIS

BLB

BNA

BOI

BRO

BRWBTI

BUF

BVE

CAR

CDB

CGU

CHH

CHSCKL

COU DAYDDC

DEN

DRA

DRTELP

ELY

EYW

FAI

FFR

FNT

FWH

GEG

GGG

GGW

GJT

GRB

GSO

GTF

GTL

GYM

HAT

HTG

HTS IAD

IGP

IKF

ILW

INL

INW

ITO

JANJAX

JSJKCRKJP

KPP

LCH

LIH

LIT

LND

MAF

MCG

MCVMDY

MEX

MFR

MGM

MIA

MID

MTY

MZTNQX

NSINTD

OAKOKC

OMA

OMEOTZ

OUN

PBI

PIA PIT

PWMRAP

SAN

SDQ

SEPSHV

SLC

SLE

SLO

SNP

SSMSTC

SWA

TBW

TUS

UCC

UGM

UIL

UMNVBG

VCT

VER

VPS

WMC

WOS

WSE

XKF

XMR

YAK

YAR

YBK

YCBYCO

YCY

YEU

YEV

YGM

YIC

YJT

YLT

YMD

YMO

YMW

YNI

YPH

YQD

YRB

YSA

YSM

YSY

YTL

YUX

YVP

YVQ

YVR

YXS

YXYYYE YYQ

YYR

YYT

YZS

YZT YZV


We can control only four aspects of We can control only four aspects of monitoring to detect trendsmonitoring to detect trends

Where we monitorWhere we monitor

What frequencyWhat frequency

What accuracyWhat accuracy

What we monitorWhat we monitor

Where do we monitor?Where do we monitor?

Some places are inherently better for Some places are inherently better for detecting trends than others.detecting trends than others.

Monitoring by satellite involves Monitoring by satellite involves averaging over height, longitude and averaging over height, longitude and latitude.latitude.• Measurement smoothing can damage Measurement smoothing can damage

our ability to detect trendsour ability to detect trends

MSU Channel 4Correlation with lat=0 and long=0

-0.4

-0.2

0.0

0.2

0.4

0.6

0.8

1.0


-0.4-0.20.00.20.40.60.81.0

MSU Channel 4Correlation with S.F.

-0.4-0.20.00.20.40.60.81.0


-0.4-0.20.00.20.40.60.81.0

MSU Channel 2Correlation with S.F.

-0.4-0.20.00.20.40.60.81.0

How does spatial How does spatial redundancy affect our redundancy affect our

ability to detect trends?ability to detect trends?

82 Station Subset of HCN Network(1.75º “Distance” Factor)

225 Station Subset of HCN Network225 Station Subset of HCN Network

Where do we monitor: global Where do we monitor: global coveragecoverage

Interpretation of raw signals can be Interpretation of raw signals can be difficult.difficult.

Inversion methods can be dependent Inversion methods can be dependent on all other parameters not on all other parameters not changing.changing.

Footprint size as well as vertical Footprint size as well as vertical resolution are critical to detection of resolution are critical to detection of trends.trends.

MSU - 2 Weighting Function

0.0 0.002 0.004 0.006

1000

800

600

400

200

0

Where do we monitor: global coverage

deseasonalized data with predicted trend addedsurface

goodx

de

gre

es

2000 2020 2040 2060 2080 2100

-10

-5

0

5

10

-10

-5

0

5

10

deseasonalized data with predicted trend added400mb

goodx

de

gre

es

2000 2020 2040 2060 2080 2100

-10

-5

0

5

10

-10

-5

0

5

10

Topeka

Formula offers 45 and 25 years respectively Weatherhead Wed Aug 14 15:08:42 2002

Years Saved by Monitoring Free Troposphere 0Z

-180 -160 -140 -120 -100 -80 -60

20

40

60

80

5

10

15

20

25

Weatherhead Wed Aug 14 11:20:30 2002 True Range of Data: (3 to 36)


Well designed in situ measurements Well designed in situ measurements can offercan offer• Monitoring in critical, unmonitored Monitoring in critical, unmonitored

areas;areas;• Unprecedented accuracy;Unprecedented accuracy;• Critical information on climate Critical information on climate

processes.processes.

What frequency?What frequency?

Inherent memory in environmental Inherent memory in environmental data results in redundancy of data results in redundancy of measurements.measurements.

Daily data may be more than Daily data may be more than needed.needed.

Less than daily measurements may Less than daily measurements may obscure diurnal trendsobscure diurnal trends

STERLING(WASH DULL 0 Z tempLat. = 38.98 Long. = -77.47

Surf.

1000

900

800

700

600

500

400

300

200

100

1960 1970 1980 1990

-80

-60

-40

-20

0

20

40

Weatherhead Wed Mar 13 13:30:52 2002

STERLING(WASH DULL 0 ZLat. = 38.98 Long. = -77.47

Surf.

1000

900

800

700

600

500

400

300

200

100

1960 1970 1980 1990

-4

-2

0

2

4

Weatherhead Wed Mar 13 13:36:18 2002

How do the trends change How do the trends change when we take data less when we take data less

frequently than every day?frequently than every day?

500 mb Temperature Trend, Dulles

Measurements per Month

Tre

nd

(d

eg

ree

s p

er

de

cad

e)

5 10 15 20 25 30

-0.2

0.0

0.2

0.4

0.6

0.8

1.0

Weatherhead Wed Jun 5 12:52:10 2002

How long will it take to How long will it take to detect trends?detect trends?

Years to Detect 0.2 degrees per DecadeDulles 0Z 500 mb

measurements per month

yea

rs

0 5 10 15 20 25 30

30

40

50

60

Weatherhead Wed Jun 5 12:45:30 2002

Decreasing the data frequencyDecreasing the data frequency

We can optimize data collection We can optimize data collection frequency to assure efficiency.frequency to assure efficiency.

Decreasing the data frequency can Decreasing the data frequency can reduce our ability to:reduce our ability to:• Detect extreme eventsDetect extreme events• Detect diurnal (or perhaps seasonal) Detect diurnal (or perhaps seasonal)

signalssignals

What accuracy?What accuracy?

Relative accuracy is all that’s needed Relative accuracy is all that’s needed for trend detection.for trend detection.

Relative accuracy is extremely hard Relative accuracy is extremely hard to maintain for decades without to maintain for decades without absolute accuracy.absolute accuracy.

Improved accuracy may save Improved accuracy may save decades in monitor or may be decades in monitor or may be irrelevant.irrelevant.

Case Example

• Uncertainty: ±2% ; Trend: 4% per decade

• Result: – First ten years of data are still unsubstantial

• Improving Accuracy to ±1% saves five years of monitoring

years

0 2 4 6 8 10 12 14 16 18 20

-8

-4

0

4

8

Measurement Uncertainty is Not Measurement Uncertainty is Not Generally RandomGenerally Random

Trends generally require decades to Trends generally require decades to detectdetect

Reference instruments and Reference instruments and calibration mechanisms often change calibration mechanisms often change over the period of several decadesover the period of several decades

Most materials for both Most materials for both instrumentation and calibration drift instrumentation and calibration drift or shift preferentially in one directionor shift preferentially in one direction

GSFC Predictions - without climate change

2000 2010 2020 2030 2040 2050

350360370380390400410

GSFC Predictions with SBUV Lowess Residuals

2000 2010 2020 2030 2040 2050

350360370380390400410

with +-1% error plus +-1% drift

2000 2010 2020 2030 2040 2050

350

360

370

380

390

400

Weatherhead Fri Nov 2 11:53:23 2001

GSFC 2d Predictions with SBUV Residuals of Total Col. Ozone (d.u.) 40N

Accuracy directly influences our Accuracy directly influences our ability to detect trendsability to detect trends

In some cases, our measurement In some cases, our measurement uncertainty is considerably larger uncertainty is considerably larger than the signal we want to detect.than the signal we want to detect.

Estimating appropriate measurement Estimating appropriate measurement uncertainty over decades of uncertainty over decades of monitoring is extremely difficult. monitoring is extremely difficult.


Changes are predicted for a large Changes are predicted for a large variety of parameters:variety of parameters:• Temperature, humidity, cloud cover, Temperature, humidity, cloud cover,

tropopause height, precipitation, tropopause height, precipitation, mesopause height, sea ice, snow cover mesopause height, sea ice, snow cover extent, extrememe events, ENSO, extent, extrememe events, ENSO, tropical cyclones.tropical cyclones.

Is there a canary parameter?Is there a canary parameter?

Proposed CanariesProposed Canaries

The Arctic – change may be greatestThe Arctic – change may be greatest

The subtropics – small changes are easy to The subtropics – small changes are easy to detectdetect

The stratosphere – very responsiveThe stratosphere – very responsive

The tropopause height – integrative responseThe tropopause height – integrative response

The ionosphere – changes can be very largeThe ionosphere – changes can be very large

Is there a canary parameter?Is there a canary parameter?

What is meant by this?What is meant by this?A parameter where the signal is A parameter where the signal is considerably larger than the variability.*considerably larger than the variability.*

A parameter where change can only imply A parameter where change can only imply anthropogenic influenceanthropogenic influence- - this requires considerably understanding over this requires considerably understanding over long time scales.long time scales.

A parameter where a change can imply A parameter where a change can imply significant changes at the Earth’s surface.significant changes at the Earth’s surface.

* and measurement uncertainty?* and measurement uncertainty?


Tropospheric parameters, particularly Tropospheric parameters, particularly temperature are canary parameters.temperature are canary parameters.

Free troposphere is considerably Free troposphere is considerably better for detecting trends than the better for detecting trends than the surface.surface.

Explanatory variables can offer Explanatory variables can offer insight to mechanisms.insight to mechanisms.

IntegrationIntegration

We make choices about all four of the We make choices about all four of the parameters we control.parameters we control.

These choices have direct impact on how These choices have direct impact on how long we will likely need to monitor in order long we will likely need to monitor in order to detect trends.to detect trends.

Optimal choices exist.Optimal choices exist. All choices will affect our ability to detect All choices will affect our ability to detect

trends and the scientific questions we may trends and the scientific questions we may ask of the emerging data.ask of the emerging data.

ConclusionConclusion1.1. Trends are difficult to detect:Trends are difficult to detect:

• Predicted trends are smallPredicted trends are small• natural variability is largenatural variability is large• Measurement uncertainty can be largeMeasurement uncertainty can be large

2.2. We can control only four aspects to We can control only four aspects to detect trends:detect trends:

• What we monitor; Where we monitor; What we monitor; Where we monitor; • What frequency; What accuracyWhat frequency; What accuracy

3.3. We can optimize systems to detect trends We can optimize systems to detect trends most efficiently with the following most efficiently with the following benefits:benefits:

• Answering scientific questions earlierAnswering scientific questions earlier• Confirming, improving modelsConfirming, improving models• Allowing for earliest policy decisionsAllowing for earliest policy decisions• Maintaining prudent use of available fundsMaintaining prudent use of available funds

Understanding the climate system Understanding the climate system is more important than detecting is more important than detecting

trends.trends.

Next StepsNext Steps

We can work to identify true canaries.We can work to identify true canaries. We can examine existing networks for We can examine existing networks for

efficiency.efficiency. We can determine savings due to:We can determine savings due to:

• Improved accuracyImproved accuracy• Improved spatial informationImproved spatial information• Improved temporal informationImproved temporal information

Optimization of existing networks can Optimization of existing networks can allow scientific, environmental and policy allow scientific, environmental and policy relevant results earlier.relevant results earlier.

New networks can be established in a New networks can be established in a defensible, efficient manner.defensible, efficient manner.

Visual ExampleVisual Example

How many years does it take to How many years does it take to detect a trend in ozone?detect a trend in ozone?

Use our understanding of variability;Use our understanding of variability; Use our understanding of the Use our understanding of the

predicted trendspredicted trends Estimate visually how long it will take Estimate visually how long it will take

to detect a trend.to detect a trend.

Original Monthly Averaged Data

80 82 84 86 88 90

280

300

320

340

360

380


SBUV OZONE TOTAL COLUMN OZONE - 40N


80 82 84 86 88 90

280

300

320

340

360

380



Monthly Means Removed, Lowess Line Fit Superimposed

80 82 84 86 88 90

-20

-10

0

10

20


80 82 84 86 88 90

280

300

320

340

360

380



Monthly Means Removed, Lowess Line Fit Superimposed

80 82 84 86 88 90

-20

-10

0

10

20

Residuals From Lowess Line Fit

80 82 84 86 88 90

-20

-10

0

10

20


2000 2010 2020 2030 2040 2050

350360370380390400410


2000 2010 2020 2030 2040 2050

350360370380390400410

GSFC Predictions with SBUV Lowess Residuals

2000 2010 2020 2030 2040 2050

350360370380390400410

Changing local observation time leads to aliasing of diurnal signal into long term trends

Corrected Global Time Series Uncorrected Global Time Series

Difference (expanded scale): 0.15K over 20 years

Courtesy Frank Wertz

Effect of Diurnal Correction on MSU Channel 2Effect of Diurnal Correction on MSU Channel 2

Years to Detect at the Surface 0Z

-180 -160 -140 -120 -100 -80 -60

20

40

60

80

30

35

40

45

50

Weatherhead Wed Aug 14 11:27:03 2002 True Range of Data: (27 to 60)

designing systems to address outstanding issues in climate change betsy weatherhead

Documents

monitorwhat accuracy

relative accuracy

improved accuracy

absolute accuracy

data collection frequency

years of data

daily data

surface measurements