Using AmeriFlux Observations in the NACP Site-level Interim Synthesis Kevin Schaefer NACP Site Synthesis Team Flux Tower PIs Modeling Teams.

Using AmeriFlux Observations in the NACP

Site-level Interim Synthesis

Kevin Schaefer

NACP Site Synthesis Team

Flux Tower PIs

Modeling Teams

Do models match observations? If not, why?

30 Models47 Flux Tower Sites

36 AmeriFlux11 Fluxnet Canada

24 submitted output10 runs per site

Num Model Num Model

1 Agro-IBIS 16 GTEC2 BEPS 17 ISAM3 Biome-BGC 18 ISOLSM4 Can-IBIS 19 LoTEC5 CLM-CASA' 20 LoTEC-DA6 CLM-CN 21 LPJ_wsl7 CN-CLASS 22 ORCHIDEE8 DAYCENT 23 ORCHIDEE-STICS9 DLEM 24 SiB310 DNDC 25 SiBCASA11 ecosys 26 SiBCrop12 ED2 27 SIPNET13 EDCM 28 SSiB214 EPIC 29 TECO15 GFDL LM3V 30 TRIPLEX-Flux

Analysis Projects

Published

Submitted

Num Title Lead author1 Agriculture Sites Erandi Lokupitiya2 Disturbance Peter Thornton3 Flux uncertainty Alan Barr4 Forest sites Bill Munger5 GPP Kevin Schaefer6 IAV Guerric Lemaire7 Interannual Variability Trevor Keenan8 Isotopes Chun-Ta Lai9 Latent and Sensible Heat Flux Alok Sahoo10 Model Parameters Hans Verbeeck11 Model uncertainty Ensheng Weng12 NEE Seasonal Cycle Christopher Schwalm13 NEE Spectral Error Michael Dietze14 NEE Wavelet Coherence Paul Stoy15 Phenology Andrew Richardson16 Regional vs. Sites Brett Raczka17 Weather Uncertainty Dan Ricciuto18 Wetland Sites Ankur Desai

Products Derived from Flux Data

• Gap-filled observed weather (Ricciuto et al.)

• BADM files (everyone)

• Gap-filled fluxes & Uncertainty (Barr et al.)• Random• U* threshold• Gap-filling Algorithm• Partitioning Algorithm

Random Uncertainty (Barr et al.)Needleleaf

ForestBroadleaf

ForestMixedwood

ForestJuvenileForest

WetlandGrasslandShrublandCropland

▲ USA ● Canada

Ann

ual

NE

P (

g C

m-2 y

-1)

Annual Re (g C m-2 y-1)

• Random NEP ~4% Re

• U*th NEP ~1.3% Re

BADM Files

• Extremely useful to modelers• Soil texture• Site history• Initial pools sizes• Leaf Area Index

• We strongly encourage more submissions

Weather Uncertainty (Ricciuto et al.)

Swdown (W m-2) Delta (%)Observed 65.38LaThu 65.74 0.55ECMWF 74.76 14.35NARR 83.98 28.45NCEP 90.7 38.73NCEP2 84.24 28.85

Princeton 69.18 5.81NLDAS 79.6 21.75

CRU-NCEP 62.45 -4.48

• Bias in radiation produces bias in GPP

Agriculture Sites (Lokupitiya et al.)

• Need crop specific parameterizations

Soybean CornCorn Soybean

US-Ne3

Wetland Sites (Desai et al.)

• Residuals correlate to water table depth

• Models should include water table dynamics

GPP residuals

Res

idua

ls

mol

m-2 s

-1

Reco residuals

Res

idua

ls

mol

m-2 s

-1

Water Table Depth (cm)

Spectral NEE Error (Dietze et al.)

Diurnal

Annual

Synoptic

Month

• Error peak at diurnal & annual time scales• Errors at synoptic & monthly time scales

Not Significant

NEE Wavelet Coherence (Stoy et al.)

Tim

e S

cale

(ho

urs)

Hour

Diurnal

Month

Synoptic

Annual

SiB at US-UMB

Sig

nifi

cant

• Models match observations only some of the time

NEE Seasonal Cycle (Schwalm et al.)

0.4

0.5

0.6

0 11 151 2 3 7 9 10Number Soil Layers

Tay

lor

Ski

ll

0.4

0.5

0.6

Semi-Prog PrescribedPrognostic

PhenologyImprove prognostic phenology

Add soil layers

0.4

0.5

0.6

8 90 3 4 6 7Number Veg Pools

Add vegetation

pools Tay

lor

Ski

llT

aylo

r S

kill

Phenology (Richardson et al.)

• Early/late uptake means positive GPP bias

• Models need better phenology

Regional vs. Site (Raczka et al.)

• Enzyme kinetic models biased high• LUE models biased low

Flux Towers

Light Use EfficiencyEnzyme Kinetic

GPP Annual Bias (Schaefer et al.)

• Slope of LUE Curve drives Annual bias

• Models need better Vmax, leaf-to-canopy scaling, …

Daily Average Shortwave Radiation (W m-2)

Dai

ly A

vera

ge G

PP (m

ol m

-2 s

-1)

ObservedSimulated

US-Me2 Light Use Efficiency Curve

Areas For Model Development

• Better Phenology

• More soil layers

• More vegetation pools

• Slopes to LUE curve

• Water table dynamics

• Crop parameterizations

Extra Slides

Annual GPP Bias due to PhenologyEvergreen sitesDeciduous sites

160±145

75±130

40±80

-5±65

Multi-Model wavelet CoherenceSc

ale

(hou

rs)

NEE Seasonal Cycle (Schwalm et al.)T

aylo

r S

kill

Normalized Mean Absolute Error Chi-squared

Our 1st published paper!

Perfect Model

Needleleaf Forest

Broadleaf Forest

Mixedwood Forest

JuvenileForest

WetlandGrasslandShrublandCropland

▲ USA ● Canada

U*t

h A

nnua

l N

EP

(g C

m-2 y

-1)

Random Annual NEP (g C m-2 y-1)

U*th vs. Random Uncertainty (Barr et al.)