model-data comparison of mid-continental intensive field campaign atmospheric co 2 mixing ratios

54
Model-Data Comparison of Mid- Continental Intensive Field Campaign Atmospheric CO 2 Mixing Ratios Liza I. Díaz May 10, 2010

Upload: licia

Post on 23-Feb-2016

24 views

Category:

Documents


0 download

DESCRIPTION

Model-Data Comparison of Mid-Continental Intensive Field Campaign Atmospheric CO 2 Mixing Ratios. Liza I. Díaz May 10, 2010. Outline. Introduction Mid-Continental Intensive Field Campaign Motivation for this study Methods Results Conclusion. Carbon Dioxide (CO 2 ). - PowerPoint PPT Presentation

TRANSCRIPT

Page 1: Model-Data Comparison of Mid-Continental Intensive Field Campaign Atmospheric CO 2  Mixing Ratios

Model-Data Comparison of Mid-Continental Intensive Field Campaign Atmospheric CO2

Mixing Ratios

Liza I. DíazMay 10, 2010

Page 2: Model-Data Comparison of Mid-Continental Intensive Field Campaign Atmospheric CO 2  Mixing Ratios

Outline

• Introduction– Mid-Continental Intensive Field Campaign– Motivation for this study

• Methods• Results• Conclusion

2

Page 3: Model-Data Comparison of Mid-Continental Intensive Field Campaign Atmospheric CO 2  Mixing Ratios

Carbon Dioxide (CO2)

• Understanding the CO2 balance is necessary because of the impacts this greenhouse gas has on the climate.

• This motivated an interest to estimate the carbon budget.

• Not all the CO2 emissions remain in the atmosphere; a portion of these emissions are taken up by the oceans and the terrestrial biosphere.

3

Page 4: Model-Data Comparison of Mid-Continental Intensive Field Campaign Atmospheric CO 2  Mixing Ratios

Sinks• Several studies concluded

that a large net sink of CO2 must exist at temperate latitudes of the Northern Hemisphere (Tans et al., 1990; Ciais et al., 1995; Fan et al., 1998).

Tans et al., 19904

Observations

Models

Page 5: Model-Data Comparison of Mid-Continental Intensive Field Campaign Atmospheric CO 2  Mixing Ratios

North American Carbon Program (NACP)

• Goals include synthesizing models and observations, evaluating current modeling capability and investigating discrepancies between different flux estimates.

• One method of estimating fluxes is the “top-down” or atmospheric inversion.

5

Page 6: Model-Data Comparison of Mid-Continental Intensive Field Campaign Atmospheric CO 2  Mixing Ratios

Atmospheric Inversions• Atmospheric inversions are

highly variable.• This variability is caused by:

1. Lack of observations2. Transport models

NACP Interim SynthesisBy : Andy Jacobson 6

Page 7: Model-Data Comparison of Mid-Continental Intensive Field Campaign Atmospheric CO 2  Mixing Ratios

Mid-Continental Intensive (MCI) campaign

• The main goal is to reach a convergence between the “top-down” atmospheric budget and agricultural inventory.

• Tans et al.,(2005) proposed to study both approaches in a specific region and time so the information and credibility of each method could be maximized.

7

Page 8: Model-Data Comparison of Mid-Continental Intensive Field Campaign Atmospheric CO 2  Mixing Ratios

Mid-Continental Intensive (MCI) Region

MCI Region consists of:1. Eastern South Dakota2. Eastern Nebraska3. Eastern Kansas4. Northern Missouri5. Iowa6. Southern Minnesota7. Southern Wisconsin8. Illinois

Main Reason The agricultural production is recorded in detail which helps to provide accurate information on the carbon flux for the inventories.

8

Page 9: Model-Data Comparison of Mid-Continental Intensive Field Campaign Atmospheric CO 2  Mixing Ratios

Networks of CO2 measurements in the MCI Region

1. National Oceanic and Atmospheric Administration (NOAA) tall towers and trace gases sampling.

2. The Ring2 network managed by the Pennsylvania State University.

• Location had the highest density of CO2 measurements to date.

• A goal is to determine the density of measurements that is needed to understand the carbon budget.

9

Page 10: Model-Data Comparison of Mid-Continental Intensive Field Campaign Atmospheric CO 2  Mixing Ratios

Atmospheric Inversion

• Temporal and spatial patterns in atmospheric CO2 mixing ratios are combined with a transport model to infer surface fluxes (Gurney et al., 2002; Rödenbeck et al., 2003; Baker et al., 2006).

10

Page 11: Model-Data Comparison of Mid-Continental Intensive Field Campaign Atmospheric CO 2  Mixing Ratios

Atmospheric Inversion

Fluxes + Uncertainties

Predicted [CO2]

11

Transport Model + Uncertainties

Obs.[CO2]

- Predicted [CO2]

Page 12: Model-Data Comparison of Mid-Continental Intensive Field Campaign Atmospheric CO 2  Mixing Ratios

Atmospheric Inversion

• The steps involved in an inverse model are:1. Atmospheric CO2 concentrations are predicted by a

forward model (a combination of a vegetation model with an atmospheric transport model).

2. Concentrations predicted by the forward model are compared to the observations.

3. Fluxes are adjusted to minimize the difference.

12

Page 13: Model-Data Comparison of Mid-Continental Intensive Field Campaign Atmospheric CO 2  Mixing Ratios

Objective

• Evaluate the performance of a global atmospheric inversion in the MCI region.

• Evaluate the statistical characteristics of the model-data differences needed to conduct an atmospheric inversion.

• Examine atmospheric CO2 mixing ratio.

13

Page 14: Model-Data Comparison of Mid-Continental Intensive Field Campaign Atmospheric CO 2  Mixing Ratios

Methodology

• Observations– PSU Ring of Towers– NOAA Tall Towers

• Carbon Tracker Model– Transport Model (TM5)– Carnegie Ames Stanford Approach (CASA)

• Data Selection• Statistics

14

Page 15: Model-Data Comparison of Mid-Continental Intensive Field Campaign Atmospheric CO 2  Mixing Ratios

Observation Location

• Midwest agricultural belt in the northern U.S.• Distance between sites range from 125 to 370 km

(Miles et al., to be submitted).• Three of these sites are located in the “corn belt”.

15

Page 16: Model-Data Comparison of Mid-Continental Intensive Field Campaign Atmospheric CO 2  Mixing Ratios

Ring2

• PSU Ring2– Centerville, IA– Galesville, WI– Kewanee, IL– Mead, NE– Round Lake, MN

• Sampling heights: 30 and 110-140 m AGL• In operation between April 2007 and November 2009

16

Page 17: Model-Data Comparison of Mid-Continental Intensive Field Campaign Atmospheric CO 2  Mixing Ratios

NOAA Tall Towers

• NOAA Tall Towers– LEF, WI

• Sampling heights: 11,30,76,122,244,396 m AGL• In operation: 1994-current

– WBI, IA• Sampling heights: 31, 99, 379 m AGL • In operation: July 2007-current

17

Page 18: Model-Data Comparison of Mid-Continental Intensive Field Campaign Atmospheric CO 2  Mixing Ratios

Carbon Tracker Model

• Calculates biogenic CO2 fluxes by integrating daily daytime average CO2 observations with an atmospheric transport model and a first guess of the biogenic fluxes.

• Biogenic fluxes are optimized by minimizing the difference between observed and modeled atmospheric CO2.

18

Page 19: Model-Data Comparison of Mid-Continental Intensive Field Campaign Atmospheric CO 2  Mixing Ratios

Transport Model Version 5 (TM5)

• Meteorological data is provided by the model of European Centre for Medium-Range Weather Forecast (ECMWF).

• For this study TM5 is run at a horizontal resolution of 6°× 4° (longitude × latitude), with nested regions over:– North America (3°× 2°)– United States (1°× 1°)

• Vertical resolution includes 25 vertical levels.

19

Page 20: Model-Data Comparison of Mid-Continental Intensive Field Campaign Atmospheric CO 2  Mixing Ratios

Carnegie Ames Stanford Approach (CASA)

• Produces fluxes in a monthly time resolution and global 1°× 1° spatial resolution.

• To calculate global fluxes CASA uses input from weather models and satellite- observed Normalized Difference Vegetation Index (NDVI).

20

Page 21: Model-Data Comparison of Mid-Continental Intensive Field Campaign Atmospheric CO 2  Mixing Ratios

Data Selection

• Average mid-day CO2 observations 1800-2200 UTC.

• Sampling levels of the observations around or above 100 m AGL.

• Ignored nighttime data.Convective Boundary Layer

Daytime

21

Page 22: Model-Data Comparison of Mid-Continental Intensive Field Campaign Atmospheric CO 2  Mixing Ratios

Choosing Model Level

• First level of Carbon Tracker is approximately 200 to 400 m AGL but behaves like a surface layer.

• Differences between levels 2,3 and 4 are less than 2 ppm.

• Level 3 is used in Carbon Tracker assimilation system.• Therefore we use LEVEL 3.

22

Page 23: Model-Data Comparison of Mid-Continental Intensive Field Campaign Atmospheric CO 2  Mixing Ratios

Statistical Analysis

• Two periods1. June through December 2007

This period lets us evaluate the seasonal cycle over the year of 2007.

2. Growing Season 2007 (June through August) This period eliminates any seasonality and leaves day-

to-day variability. During this season the variability of the mixing ratios are large because the fluxes are large.

23

Page 24: Model-Data Comparison of Mid-Continental Intensive Field Campaign Atmospheric CO 2  Mixing Ratios

Statistical Analysis• Time Series analysis • Model Performance

– Taylor Diagram• Residual distributions

– Gaussian fit• Temporal Correlations

– Autocorrelations • Power Spectrum• Spatial Correlations • Relation between meteorological data and residuals

24

Page 25: Model-Data Comparison of Mid-Continental Intensive Field Campaign Atmospheric CO 2  Mixing Ratios

Time Series CO2 Observations

• “Corn belt” sites with the highest drawdown.• Seasonal amplitude:

• “corn belt” - 40 ppm• Sites with different vegetation – 24-29 ppm 25

Page 26: Model-Data Comparison of Mid-Continental Intensive Field Campaign Atmospheric CO 2  Mixing Ratios

Simulated and Observed Times Series

Carbon Tracker Observation – Carbon Tracker

• Carbon Tracker does not simulate these three “corn belt” sites drawdown correctly.• Possible causes to these differences:

1. Uptake is underestimated in the “corn belt” region.2. Vertical mixing is too strong.

26

Page 27: Model-Data Comparison of Mid-Continental Intensive Field Campaign Atmospheric CO 2  Mixing Ratios

Statistical Analysis• Time Series analysis • Model Performance

– Taylor Diagram• Residual distributions

– Gaussian fit• Temporal Correlations

– Autocorrelations • Power Spectrum• Spatial Correlations • Relation between meteorological data and residuals

27

Page 28: Model-Data Comparison of Mid-Continental Intensive Field Campaign Atmospheric CO 2  Mixing Ratios

June through December Taylor Diagram

• Model is highly correlated with observations (R > 0.8).

• The model underestimates the day-to-day variability for all the sites.

28

Page 29: Model-Data Comparison of Mid-Continental Intensive Field Campaign Atmospheric CO 2  Mixing Ratios

Growing Season Taylor Diagram

• Model is less correlated with observations (R < 0.8),except Centerville site (R > 0.8).

• The model underestimates the day-to-day variability for all the sites, except Centerville.

29

Page 30: Model-Data Comparison of Mid-Continental Intensive Field Campaign Atmospheric CO 2  Mixing Ratios

Taylor Diagram

• Model can better simulate the seasonal cycle than synoptic variability.

• Model tends to underestimate the amplitude of both the seasonal cycle and the day-to-day variability.

30

Page 31: Model-Data Comparison of Mid-Continental Intensive Field Campaign Atmospheric CO 2  Mixing Ratios

Statistical Analysis• Time Series analysis • Model Performance

– Taylor Diagram• Residual distributions

– Gaussian fit• Temporal Correlations

– Autocorrelations • Power Spectrum• Spatial Correlations • Relation between meteorological data and residuals

31

Page 32: Model-Data Comparison of Mid-Continental Intensive Field Campaign Atmospheric CO 2  Mixing Ratios

June through December Distributions

• “Corn belt” sites have the highest standard deviation.• All of the sites are negatively skewed, except

Centerville.

Galesville Round Lake

32

Page 33: Model-Data Comparison of Mid-Continental Intensive Field Campaign Atmospheric CO 2  Mixing Ratios

June through December Distributions

Sites Mean (ppm) Standard Deviation (ppm) SkewnessCenterville -0.98 4.94 0.95Galesville 0.37 5.86 -0.37Kewanee -4.42 7.65 -0.72

Mead -0.63 5.52 -0.76Round Lake -1.48 6.70 -1.07

WBI -0.42 6.77 -0.32LEF -1.27 4.43 -0.08

33

Page 34: Model-Data Comparison of Mid-Continental Intensive Field Campaign Atmospheric CO 2  Mixing Ratios

Growing Season Distribution

• In general “Corn belt” sites have the highest mean and standard deviation.

Galesville Round Lake

34

Page 35: Model-Data Comparison of Mid-Continental Intensive Field Campaign Atmospheric CO 2  Mixing Ratios

Growing Season Distribution

Sites Mean (ppm) Standard Deviation (ppm) SkewnessCenterville -2.85 5.11 0.22Galesville -2.71 7.22 0.16Kewanee -8.49 7.08 -0.36

Mead -4.31 5.60 -0.95Round Lake -5.86 7.43 -0.42

WBI -6.38 7.28 0.66LEF -3.85 4.35 -0.22

35

Page 36: Model-Data Comparison of Mid-Continental Intensive Field Campaign Atmospheric CO 2  Mixing Ratios

Gaussian Distribution

• A χ2 test was performed to determine if distributions are Gaussian.• Limited number of samples make our χ2 inconsistent, but will be applied

to hourly residuals.

36

June through December 2007 Growing Season 2007

Page 37: Model-Data Comparison of Mid-Continental Intensive Field Campaign Atmospheric CO 2  Mixing Ratios

Data Independence

• Determine correlation in time:– Autocorrelations– Power Spectrum

• Determine correlation in space:– Estimation of correlation across sites

Motivation: Evaluate independence of the data in the inverse system.

37

Page 38: Model-Data Comparison of Mid-Continental Intensive Field Campaign Atmospheric CO 2  Mixing Ratios

Statistical Analysis• Time Series analysis • Model Performance

– Taylor Diagram• Residual distributions

– Gaussian fit• Temporal Correlations

– Autocorrelations • Power Spectrum• Spatial Correlations • Relation between meteorological data and residuals

38

Page 39: Model-Data Comparison of Mid-Continental Intensive Field Campaign Atmospheric CO 2  Mixing Ratios

Temporal Correlation

• Centerville and LEF sites show the lowest correlation of residuals in time.

• Centerville behaves completely different than the rest of the sites.

Centerville LEF

39

Page 40: Model-Data Comparison of Mid-Continental Intensive Field Campaign Atmospheric CO 2  Mixing Ratios

Temporal Correlation

• Sites located in “corn belt” tend to show correlation of the residuals in time for a period of 40 to 50 days.

• This autocorrelation shows an error in the seasonal cycle, because growing season lasts about 50 to 60 days in a year.

Round Lake WBI

40

Page 41: Model-Data Comparison of Mid-Continental Intensive Field Campaign Atmospheric CO 2  Mixing Ratios

Statistical Analysis• Time Series analysis • Model Performance

– Taylor Diagram• Residual distributions

– Gaussian fit• Temporal Correlations

– Autocorrelations • Power Spectrum• Spatial Correlations • Relation between meteorological data and residuals

41

Page 42: Model-Data Comparison of Mid-Continental Intensive Field Campaign Atmospheric CO 2  Mixing Ratios

Power Spectrum

• This suggests that residuals have a maximum every 4 to 5 days for all the sites.

• One possible cause for this is weather events. • Therefore, transport might be causing these changes

in residuals.

Round Lake WBI

42

Page 43: Model-Data Comparison of Mid-Continental Intensive Field Campaign Atmospheric CO 2  Mixing Ratios

Statistical Analysis• Time Series analysis • Model Performance

– Taylor Diagram• Residual distributions

– Gaussian fit• Temporal Correlations

– Autocorrelations • Power Spectrum• Spatial Correlations • Relation between meteorological data and residuals

43

Page 44: Model-Data Comparison of Mid-Continental Intensive Field Campaign Atmospheric CO 2  Mixing Ratios

Spatial Correlation

• Two sites located at the “corn belt” are the highest correlated.

• The second highest spatial correlation is WBI (corn belt) and Mead (mixed vegetation).

Annual Centerville Galesville Kewanee MeadRound Lake WBI LEF

Centerville 1 0.2 0.4 0.4 0.2 0.4 0.2Galesville 1 0.6 0.4 0.4 0.5 0.4Kewanee 1 0.5 0.5 0.8 0.4

Mead 1 0.6 0.6 0.4Round Lake 1 0.5 0.5

WBI 1 0.4LEF 1

June through December 2007

44

Page 45: Model-Data Comparison of Mid-Continental Intensive Field Campaign Atmospheric CO 2  Mixing Ratios

Spatial Correlation

• Two sites located at the “corn belt” are the highest correlated.

• The second highest spatial correlation is Kewanee (corn belt) and Galesville (mixed vegetation).

Annual Centerville Galesville Kewanee MeadRound Lake WBI LEF

Centerville 1 0.1 0.3 0.3 -0.02 0.1 0.01Galesville 1 0.4 0.2 0.3 0.4 0.1Kewanee 1 0.2 0.2 0.6 0.1

Mead 1 0.3 0.4 0.2Round Lake 1 0.3 0.4

WBI 1 0.2LEF 1

Growing Season 2007

45

Page 46: Model-Data Comparison of Mid-Continental Intensive Field Campaign Atmospheric CO 2  Mixing Ratios

Spatial Correlation

• The only sites which residuals are highly correlated in space are WBI and Kewanee.

• Both sites are part of the “corn belt” region.• However, spatial correlation of the residuals are

weaker among the other “corn belt” sites.

Distance: 125 kmVegetation: corn

46

Page 47: Model-Data Comparison of Mid-Continental Intensive Field Campaign Atmospheric CO 2  Mixing Ratios

Statistical Analysis• Time Series analysis • Model Performance

– Taylor Diagram• Residual distributions

– Gaussian fit• Temporal Correlations

– Autocorrelations • Power Spectrum• Spatial Correlations • Relation between meteorological data and residuals

47

Page 48: Model-Data Comparison of Mid-Continental Intensive Field Campaign Atmospheric CO 2  Mixing Ratios

Wind Direction and Temperature• Some of the sites like

Round Lake shows a relationship between northerly winds and high residuals, but it is not clear.

• High residuals are also correlated with high temperature. Basically this period is the growing season, in which not only are the temperatures high but also the fluxes are large.

June through December 2007

48

Page 49: Model-Data Comparison of Mid-Continental Intensive Field Campaign Atmospheric CO 2  Mixing Ratios

Wind Direction and Temperature• Centerville does not

show any correlation between residuals and both temperature and wind direction.

• Round Lake shows again weak relationship between high residuals and wind coming from the north. But, there is no correlation between temperature and residuals.

• No persistent correlation between residuals and meteorological data.

Growing Season 2007

49

Page 50: Model-Data Comparison of Mid-Continental Intensive Field Campaign Atmospheric CO 2  Mixing Ratios

Conclusions• Carbon Tracker does not well simulate “corn belt” draw down.• Carbon Tracker simulates better the seasonal cycle than it

does the day-to-day variability.• Gaussian assumption might be violated and need to test what

effect could cause in the inversion.• Residual differences or error are repeated through the

growing season period.• Carbon Tracker shows a possible synoptic error.• Residuals show a weak spatial correlation amongst the sites,

suggesting some independence of the error. • Residuals are not correlated to temperature, but some sites

shows a weak correlation between residuals and wind coming from the North.

50

Page 51: Model-Data Comparison of Mid-Continental Intensive Field Campaign Atmospheric CO 2  Mixing Ratios

Thank You• This research was supported by the Department of Energy’s

Terrestrial Carbon Processes program and fellowships from Penn State’s Earth and Environmental Science Institute and Bunton-Waller Program.

• To my advisor, Dr. Kenneth J. Davis and to my committee members Dr. Natasha Miles, Dr. Chris E. Forest and Dr. Anne M. Thompson.

• Davis group for all the support (Thomas Lauvaux, Martha Butler, Brett Raczka, Yuning Shi, Kelly Cherrey, and Scott Richardson).

• To my mom and dad for all the support and my siblings too.• To all my friend and family.

51

Page 52: Model-Data Comparison of Mid-Continental Intensive Field Campaign Atmospheric CO 2  Mixing Ratios

Observation Location

• Midwest agricultural belt in the northern U.S.• Distance between sites range from 125 to 370 km

(Miles et al., to be submitted).• Three of these sites are located in the “corn belt”.

52

Page 53: Model-Data Comparison of Mid-Continental Intensive Field Campaign Atmospheric CO 2  Mixing Ratios

Time Series CO2 Observations

• “Corn belt” sites with the highest drawdown.• Seasonal amplitude:

• “corn belt” - 40 ppm• Sites with different vegetation – 24-29 ppm 53

Page 54: Model-Data Comparison of Mid-Continental Intensive Field Campaign Atmospheric CO 2  Mixing Ratios

Spatial Correlation

• The only sites which residuals are highly correlated in space are WBI and Kewanee.

• Both sites are part of the “corn belt” region.• However, spatial correlation of the residuals are

weaker among the other “corn belt” sites.

Distance: 125 kmVegetation: corn

54