Comparing Simulated and Observed Gross Primary Productivity Kevin Schaefer, Altaf Arain, Alan Barr, Jing Chen, Ken Davis, Dimitre Dimitrov, Ni Golaz, Timothy.

Comparing Simulated and Observed Gross Primary

Productivity

Kevin Schaefer, Altaf Arain, Alan Barr, Jing Chen, Ken Davis, Dimitre Dimitrov, Ni Golaz, Timothy Hilton, David Hollinger,

Elyn Humphreys, Benjamin Poulter, Brett Raczka, Andrew Richardson, Alok Sahoo, Christopher Schwalm, Peter Thornton,

Rodrigo Vargas, Hans Verbeeck, Chris Williams

NACP Synthesis Management Team

Ameriflux and Fluxnet Canada Investigators

Modeling Team Investigators

ObjectivesObjectives

• Quantify how well Quantify how well models simulate GPPmodels simulate GPP

• Identify sources of errorIdentify sources of error

32 Flux Tower Sites32 Flux Tower Sites

AGROIBIS LPJBEPS MODIS_algBIOMEBGC MODIS_C5CAN-IBIS MODIS_C5.1CNCLASS ORCHIDEEDLEM SIBDNDC SIBCASAECLUEEDCM SIBCROPECOSYS SSIB2ED2 TECOISAM TRIPLEXISOLSM Mean (all)LOTEC Mean (diurnal)

AGROIBIS LPJBEPS MODIS_algBIOMEBGC MODIS_C5CAN-IBIS MODIS_C5.1CNCLASS ORCHIDEEDLEM SIBDNDC SIBCASAECLUEEDCM SIBCROPECOSYS SSIB2ED2 TECOISAM TRIPLEXISOLSM Mean (all)LOTEC Mean (diurnal)

AGROIBIS LPJBEPS MODIS_algBIOMEBGC MODIS_C5CAN-IBIS MODIS_C5.1CNCLASS ORCHIDEEDLEM SIBDNDC SIBCASAECLUEEDCM SIBCROPECOSYS SSIB2ED2 TECOISAM TRIPLEXISOLSM Mean (all)LOTEC Mean (diurnal)

21 Models, 3 MODIS, 2 Model Mean21 Models, 3 MODIS, 2 Model Mean

Model RunsModel Runs

• Gap-filled observed weatherGap-filled observed weather

• Steady stateSteady state

• Observed NEE partitioned into GPP Observed NEE partitioned into GPP & respiration& respiration

• GPP UncertaintyGPP Uncertainty• RandomRandom• U* filteringU* filtering• Gap-fillingGap-filling• Partitioning Partitioning

Model-Data ComparisonModel-Data Comparison

• Daily average GPPDaily average GPP

• Performance MeasuresPerformance Measures• Chi-squared statisticChi-squared statistic• Root Mean Squared ErrorRoot Mean Squared Error• Normalized Mean Absolute ErrorNormalized Mean Absolute Error• BiasBias

Overall Model PerformanceOverall Model Performance

Model MeanMODIS

Optimized Model

C5C5.1

Daily GPP Taylor PlotSt

anda

rd D

evia

tion

Rat

ioG

PP

Rat

io

Typical GPPTypical GPP

Date (Year)

GP

P (m

ol m

-2 s

-1)

Monthly Average GPP for CA-Ca1

ObservedSimulated

Daily Average Shortwave Radiation (W m-2)

Light Use Efficiency CurvesLight Use Efficiency CurvesG

PP

(m

ol m

-2 s

-1)

Daily Average GPP for CA-Ca1

ObservedSimulated

Daily Average Temperature (°C)

Temperature Response CurvesTemperature Response CurvesG

PP

(m

ol m

-2 s

-1)

Daily Average GPP for CA-Ca1

ObservedSimulated

ConclusionsConclusions

• Models don’t simulate GPP wellModels don’t simulate GPP well

• Bias in seasonal amplitudeBias in seasonal amplitude

• Improve LUEImprove LUE

• Improve Temperature ResponseImprove Temperature Response

AcknowledgmentsAcknowledgments

• Funding provided by Funding provided by NASA, NOAA, and NSFNASA, NOAA, and NSF

StatisticsStatistics

2

2 1

GPP

estmod GPPGPP

nΧ

Chi-squared

21estmod GPPGPP

nRMSE

Root Mean Square Error

estmod GPPGPPn

B1

Bias

BB > 0 model greater than data > 0 model greater than data

RMSERMSE 0 perfect fit with data 0 perfect fit with data

22 ~ 1~ 1 model matches data within uncertainty

Normalized Mean Absolute Error

NMAENMAE 0 perfect fit with data 0 perfect fit with data estmod

est

GPPGPPnGPP

NMAE1

Overall Performance by ModelOverall Performance by Model

Ratio of Annual GPP AmplitudeRatio of Annual GPP Amplitude

Simulated:Observed Amplitude Ratio (-)

Per

cent

Yea

rs (

)

GPP Total Uncertainty for CA-Ca1GPP Total Uncertainty for CA-Ca1

Date (year)

GP

P U

ncer

tain

ty (m

ol m

-2 s

-1)

Daily Average GPP

GPP Uncertainty for CA-Ca1GPP Uncertainty for CA-Ca1

Date (year)

GP

P U

ncer

tain

ty (m

ol m

-2 s

-1)