RESEARCH ARTICLE SUMMARY

CARBON CYCLE

OCO-2 advances photosynthesisobservation from space via solar-induced chlorophyll fluorescenceY Sun C Frankenberg J D Wood D S Schimel M Jung L GuanterD T Drewry M Verma A Porcar-Castell T J Griffis L Gu T S MagneyP Koumlhler B Evans K Yuen

INTRODUCTIONReliable estimation of grossprimary production (GPP) from landscape toglobal scales is pivotal to a wide range of eco-logical research areas such as carbon-climatefeedbacks and agricultural applications suchas crop yield and drought monitoring Howevermeasuring GPP at these scales remains a majorchallenge Solar-induced chlorophyll fluores-cence (SIF) is a signal emitted directly fromthe core of photosynthetic machinery SIF in-tegrates complex plant physiological functionsin vivo to reflect photosynthetic dynamics inreal time The advent of satellite SIF observa-tion promises a new era in global photosynthe-sis research The Orbiting Carbon Observatory-2(OCO-2) SIF product is a serendipitous but crit-ically complementary by-product of OCO-2rsquosprimary mission targetmdashatmospheric columnCO2 (XCO2 ) OCO-2 SIF removes some impor-tant roadblocks that prevent wide and in-depthapplications of satellite SIF data sets and offersnew opportunities for studying the SIF-GPPrelationship and vegetation functional gra-dients at different spatiotemporal scales

RATIONALE Compared with earlier satellitemissions with SIF capability the OCO-2 SIFproduct has substantially improved spatialresolution data acquisition and retrieval pre-cision These improvements allow satellite SIFdata to be validated for the first time directlyagainst ground and airborne measurementsand also used to investigate the SIF-GPP rela-tionship and terrestrial ecosystem functionaldynamics with considerably better spatiotem-poral credibility

RESULTS Coordinated airborne measure-ments of SIF with the Chlorophyll Fluores-cence Imaging Spectrometer (CFIS) were usedto validate OCO-2 retrievals The validation

shows close agreement between OCO-2 andCFIS SIF with a regression slope of 102 and R2

of 071 Landscape gradients in SIF emissioncorresponding to differences in vegetationtypes were clearly delineated by OCO-2 a ca-pability that was lacking in previous satellitemissions The SIF-GPP relationships at eddycovariance flux sites in the vicinity of OCO-2orbital tracks were found to be more consist-ent across biomes than previously suggestedFinally empirical orthogonal function (EOF)analyses on OCO-2 SIF and available GPPproducts show highly consistent spatiotem-poral correspondence in their leading EOF

modes across the globe suggesting that SIFand GPP are governed by similar dynamics andcontrolled by similar environmental and bi-ological conditions

CONCLUSION OCO-2 represents a majoradvance in satellite SIF remote sensing Ouranalyses suggest that SIF is a powerful proxyfor GPP at multiple spatiotemporal scales andthat high-quality satellite SIF is of central im-portance to studying terrestrial ecosystemsand the carbon cycle Although the possibility

of a universal SIF-GPP rela-tionship across differentbiome types cannot be dis-missed in-depth process-based studies are neededto unravel the true natureof covariations between

SIF and GPP Of critical importance in suchefforts are the potential coordinated dy-namics between the light-use efficienciesof CO2 assimilation and fluorescence emis-sion in response to changes in climate andvegetation characteristics Eventual syner-gistic uses of SIF with atmospheric CO2 en-abled by OCO-2 will lead to more reliableestimates of terrestrial carbon sources andsinksmdashwhen where why and how carbonis exchanged between land and atmospheremdashas well as a deeper understanding of carbon-climate feedbacks

RESEARCH | REMOTE SENSING

Sun et al Science 358 189 (2017) 13 October 2017 1 of 1

The list of author affiliations is available in the full article onlineCorresponding author Email ys776cornelledu (YS)cfrankencaltechedu (CF)Cite this article as Y Sun et al Science 358 eaam5747(2017) DOI 101126scienceaam5747

The marked ecological gradients depicted by OCO-2rsquos high-resolution SIF measurementsalong a transect of temperate deciduous forests crops and urban area from Indiana tosuburban Chicago Illinois

ON OUR WEBSITE

Read the full articleat httpdxdoiorg101126scienceaam5747

on March 30 2020

httpsciencesciencem

agorgD

ownloaded from

RESEARCH ARTICLE

CARBON CYCLE

OCO-2 advances photosynthesisobservation from space via solar-induced chlorophyll fluorescenceY Sun1dagger C Frankenberg21 J D Wood3 D S Schimel1 M Jung4 L Guanter5

D T Drewry16 M Verma7 A Porcar-Castell8 T J Griffis9 L Gu10 T S Magney1

P Koumlhler2 B Evans11 K Yuen1

Quantifying gross primary production (GPP) remains a major challenge in global carboncycle research Spaceborne monitoring of solar-induced chlorophyll fluorescence (SIF)an integrative photosynthetic signal of molecular origin can assist in terrestrial GPPmonitoring However the extent to which SIF tracks spatiotemporal variations in GPPremains unresolved Orbiting Carbon Observatory-2 (OCO-2)rsquos SIF data acquisition and finespatial resolution permit direct validation against ground and airborne observationsEmpirical orthogonal function analysis shows consistent spatiotemporal correspondencebetween OCO-2 SIF and GPP globally A linear SIF-GPP relationship is also obtainedat eddy-flux sites covering diverse biomes setting the stage for future investigations of therobustness of such a relationship across more biomes Our findings support the centralimportance of high-quality satellite SIF for studying terrestrial carbon cycle dynamics

Photosynthesis one of the most fundamen-tal biological processes on Earth providesfood and oxygen to all higher life forms andregulates the capacity of terrestrial eco-systems to offset anthropogenic CO2 emis-

sions The continuing increase of atmosphericCO2 and its impact on climate are expected toaffect photosynthesis in fundamental (1) but un-certain ways (2 3) This uncertainty constitutesa critical constraint in the projection of futurecrop production (4) and carbon-climate feedbacks(5 ) However we currently possess a limited ca-pability to directly measure photosynthesis atspatial and temporal scales relevant for under-standing and predicting future agricultural risksand roles of carbon-climate feedbacks in theEarth system The approach of observing solar-

induced chlorophyll fluorescence (SIF) globallyfrom space is an important step toward alleviat-ing this deficiencySIF an emission from chlorophyll a molecules

excited by absorbed photons is an optical sig-nal emanating from the core of photosyntheticmachinery and contains functional informationabout photosynthesis The terrestrial SIF emis-sion spectrum spans 660 to 850 nm with twopeaks centered at 685 and 740 Although ac-tively induced fluorescence has been used fordecades to probe photosynthesis in vivo at mo-lecular and leaf scales (6ndash10) passive SIF from avantage point in space became available onlyrecently (11ndash15) A major challenge of passive SIFremote sensing is to discern a small SIF emissionfrom a much higher background signalmdashreflectedsunlight This challenge is overcome by mea-suring the in-filling of narrow solar Fraunhoferlines using high-resolution spectrometers Withthis technique the confounding effects of sur-face reflectance as well as atmospheric scatteringand absorption are minimized thereby facilitat-ing the SIF retrieval The first spaceborne missionspossessing the spectral and radiometric sensitivityto enable SIF retrievals include the GreenhouseGases Observing Satellite (GOSAT) Global OzoneMonitoring Experiment-2 (GOME-2) onboardMetOp-A and the SCanning Imaging Absorp-tion spectroMeter for Atmospheric CHartographY(SCIAMACHY) onboard Envisat (11ndash14) None ofthese instruments were originally designed formeasuring SIF but they nevertheless providedinitial data sets that revealed the feasibility ofsatellite SIF retrievals albeit at very coarse spa-tial resolution

The Orbiting Carbon Observatory-2 (OCO-2)SIF product is also a serendipitous but criticalcomplementary by-product of OCO-2rsquos primarymission targetmdashatmospheric column CO2 Thedynamics of column CO2 reflects net fluxes ofproduction and consumption processes of bothnatural and anthropogenic origins Thus the ser-endipitous and primary products together formindependent constraints on the CO2 source andsink distributions from the land and atmosphericperspectives respectively Their joint availabilitywill help curb uncertainties in the estimate ofregional and global carbon budgets and advanceunderstanding terrestrial biosphere responses toclimate change which will eventually lead to bet-ter prediction of carbon-climate feedbacksIn recent years progress has been made in

applying spaceborne SIF to study large-scale ter-restrial ecosystem dynamics covering a varietyof topics such as high-latitude vegetation phenol-ogy tropical carbon cycle seasonality crop yieldmonitoring and detection of impending droughts(16ndash23) However it remains unclear to what ex-tent the spatiotemporal variations of SIF andgross primary production (GPP) are related toeach other as well as how and at which scales SIFcan be used to predict GPP Here we will first dis-cuss roadblocks in addressing these critical issuesand then show how OCO-2 SIF helps to removethem and how it can be used to shed lights onSIF-GPP relationships at different spatiotemporalscales In doing so a variety of potential researchopportunities enabled by this distinctive space-borne SIF data product will also be introduced

High-resolution studies of vegetationfunctional gradients with OCO-2SIF products

Studies of the relationship between vegetationand climate gradients are crucial to understand-ing biosphere dynamics It has been difficult touse SIF retrievals from previous satellite mis-sions to investigate heterogeneous landscapesbecause of their coarse spatial resolutions andlimited sampling strategies (Fig 1A) For ex-ample the native footprint of GOME-2 is 40 kmby 40 km (40 km by 80 km before 15 July 2013)(14) whereas GOSAT has a circular footprint of10 km in diameter with individual samplessparsely distributed across Earthrsquos surface (11 13)This severely masked out fine-scale variationsthat are key to studying vegetation functionalgradients In contrast a nominal OCO-2 footprintis 13 km by 225 km with eight independentacross-track measurements within its 10-km-swathwidth yielding ~105 clear-sky SIF observationsover land each day (24) This high data-acquisitiondensity allows the use of averaging to effectivelyminimize random measurement uncertainty andto produce high-precision SIF products at anunprecedented high spatial resolution eventhough the narrow swath reduces the extent ofglobal coverage For regions covered by orbitaltracks of OCO-2 its SIF data sets can be used toquantify the heterogeneity of ecosystem struc-ture and function and to detect fine-scale eco-physiological phenomena such as functional

RESEARCH | REMOTE SENSING

Sun et al Science 358 eaam5747 (2017) 13 October 2017 1 of 6

1Jet Propulsion Laboratory California Institute of TechnologyPasadena CA USA 2Division of Geological and PlanetarySciences California Institute of Technology Pasadena CAUSA 3School of Natural Resources University of MissouriColumbia MO USA 4Max Planck Institute for BiogeochemistryJena Germany 5Helmholtz Centre Potsdam GermanResearch Centre for Geosciences Potsdam Germany6Joint Institute for Regional Earth System Science andEngineering University of California Los Angeles CA USA7Consulting for Statistics Computing and Analytics ResearchUniversity of Michigan Ann Arbor MI USA 8Optics ofPhotosynthesis Laboratory Department of Forest SciencesUniversity of Helsinki Helsinki Finland 9Department ofSoil Water and Climate University of Minnesota Saint PaulMN USA 10Environmental Science Division Oak Ridge NationalLaboratory Oak Ridge TN USA 11Faculty of Agricultureand Environment School of Life and Environmental ScienceThe University of Sydney Sydney AustraliaCorresponding author Email ys776cornelledu (YS)cfrankencaltechedu (CF) daggerPresent address School ofIntegrative Plant Science Soil and Crop Sciences Section CornellUniversity Ithaca NY USA

on March 30 2020

httpsciencesciencem

agorgD

ownloaded from

changes in ecologically and climatically sensitiveregions These regions include human-disturbedand heavily fragmented areas as well as ecotonessuch as southern and northern boreal tree lineswhere forest expansion and contraction has beenobserved due to climate changeOCO-2 SIF is capable of capturing the spatial

gradients in SIF emission across diverse ecosys-tems As an example Fig 1B shows a markedcontrast of SIF in an August orbital track alonga transect spanning forests crops and developedurban areas south of Chicago Here we find thatSIF of croplands far exceeds that of forests where-as urban green areas emit a very weak signal Thisexample demonstrates that with its current levelof retrieval precision OCO-2 SIF has the capa-bility to capture real-time photosynthetic activ-ities Although sharp contrasts in SIF betweendifferent vegetation types have been reported withground-based measurements (25) none of theprevious satellite missions were able to resolvesuch fine-scale functional transitions across eco-systems As the data records provided by OCO-2grow it will be possible to use them in conjunctionwith column CO2 concentration measurements toinvestigate how vegetation functional gradientsvary in response to changes in environmental con-ditions and how these gradients are related tospatial variations in the terrestrial carbon sink

Coordinated spaceborne and airborneSIF measurements for validation andoptimizing retrieval algorithms

So far various retrieval algorithms have beendeveloped and applied depending on different

spectrometer specifications These algorithmscan be classified into two broad categories Thefirst employs physically based approaches to fitnarrow spectral windows and is used for spec-trometers with very high spectral resolution (eglt005 nm GOSAT and OCO-2) The second usesstatistical approaches to fit relatively broad spec-tral windows and is for spectrometers with mod-erate spectral resolution (eg ~05 nm GOME-2)Because there is always a trade-off among spatialtemporal and spectral resolutions these two ap-proaches have pros and cons in terms of retrievalrobustness and sensitivity to atmospheric scatter-ing and absorption spatial resolution and revisittime (26) Cross-mission comparisons with differ-ent spaceborne instruments have been performedyet true validation of these retrieval algorithmswith independent airborne or ground measure-ments of SIF has lagged behind the developmentof these previous spaceborne SIF productsToward that goal the OCO-2 team developed

the airborne Chlorophyll Fluorescence ImagingSpectrometer (CFIS) CFIS combines high spectralresolution (lt01 nm) with a wide spectral cover-age (737 to 772 nm) (Fig 2A) which is optimallysuited for SIF retrievals and allows for testingboth retrieval strategies mentioned above Sev-eral validation campaigns with CFIS were carriedout in 2015 and 2016 across a range of differentecosystems (including crops grassland and forests)under-flying the OCO-2 orbital tracks in IllinoisIowa Colorado Nebraska Minnesota and Califor-nia Figure 2B shows the initial validation flightsin 2015 revealing a strong agreement betweenOCO-2 and CFIS SIF along latitudinal gradients

The spaceborne and airborne measurements werelinearly correlated with a regression slope of 102(R2 = 071 fig S1) indicating that the latitudinalvariation of SIF is well characterized by OCO-2retrievals Figure 2C shows fine-scale CFIS SIFretrievals over agricultural areas in southwest Min-nesota including a region affected by partial cloudcover The Fraunhofer line-based SIF retrievals areinsensitive to atmospheric scattering which canbe seen as the partial cloud cover appears not toaffect the SIF retrieval As a comparison a four-band [red-green-blue (RGB) + near-infrared (NIR)]context camera was used to derive RGB imageryand the normalized difference vegetation index(NDVI) at a much higher spatial resolution thanCFIS The reflectance-based NDVI is expectedlystrongly affected by clouds in general whereasSIF is only reduced in the optically thick parts ofthe cloud as well as over cloud shadows In theSIF imagery strong gradients between corn andsoybean plots can be discerned with soybean beingmore fluorescent in mid-August 2016 most likelyrelated to an offset in planting dates betweenthese two primary US Midwest cropsSuch coordinated spaceborne and airborne

measurements for the purpose of SIF validationhave not been carried out for any previous satelliteinstruments with SIF capabilities The CFIS val-idation data set will be an important step towardbridging the scale gaps between ground-basedspectrometersmdashwhich are now being developed anddeployed to examine fine temporal- and spatial-scale functional SIF dynamicsmdashand satellite-basedmeasurements fromOCO-2 and follow-onmissionssuch as the European Space Agency Fluorescence

Sun et al Science 358 eaam5747 (2017) 13 October 2017 2 of 6

Fig 1 High-resolution OCO-2 footprints and the marked ecologicalgradients depicted by OCO-2 SIF along a transect from Indiana tosuburban Chicago Illinois (A) Spatial resolution of OCO-2 SIF (nadirmode 13 km by 225 km) compared with existing products from GOME-2(40 km by 40 km) onboard MetOp-A and GOSATndashFourier TransformSpectrometer (10-km diameter) SIF is acquired during August overpass(es)in 2015 20 August for OCO-2 5 August for GOME-2 and entire month

of August for GOSAT (B) Visualization of the vegetation functionalgradients across a transect of temperate deciduous forests croplandsand urban area with OCO-2 SIF The National Land Cover Database2011 (NLDC 2011) (45) created by the Multi-Resolution Land Character-istics Consortium is used here (spatial resolution 30 m) The rural-urbancontrast is well characterized by the high-resolution OCO-2 SIF (thezoomed-in boxes) The asterisk indicates Alaska only

RESEARCH | RESEARCH ARTICLE | REMOTE SENSINGon M

arch 30 2020

httpsciencesciencemagorg

Dow

nloaded from

Explorer (FLEX) (27 ) In fact the airborne instru-ment HyPlant has been designed for preparingthe launch of FLEX and a number of campaignshave already been performed since 2012 (28 29)In the coming years these HyPlant campaigns willbe continued and potential synchronous flightswith OCO-2 will greatly advance the developmentand optimization of SIF retrieval algorithms

Using OCO-2 SIF as a shortcutto estimating GPP

A direct linkage between satellite observationand flux towerndashbased measurements is crucialfor using remotely sensed SIF to estimate GPPat larger scales This has not been possible in thepast because previous missions had coarse spa-tial resolutions low data-acquisition rates andhigh observation uncertainties which made itdifficult to apply spaceborne and ground-basedmeasurements in a synergistic and integratedfashion For example the spatial resolutions of

GOME-2 and SCIAMACHY (30 km by 60 km)are much larger than a typical eddy covariance(EC) flux tower footprint whereas GOSATrsquos sparsedata acquisition severely restricted the possibilityof reducing random retrieval errors via sampleaveraging These limitations prevented simulta-neous uses of spaceborne SIF and ground-basedmeasurements for in-depth analyses of biome-specific responses In contrast OCO-2rsquos fine spa-tial resolution (13 km by 225 km) together withhigh data-acquisition density alleviates such lim-itations Thus OCO-2 SIF data sets can be integratedwith ground-based measurements to investigatethe SIF-GPP relationship at least for regionscovered by OCO-2 orbital tracksWe have identified multiple FLUXNET sites

that are located in the direct or close vicinity ofOCO-2 orbits (30 31) These sites span structurallyand functionally diverse ecosystems includingcrops temperate deciduous forests and grass-lands We related OCO-2 SIF to GPP derived from

net ecosystem exchange measured with the ECtechnique (32) at these sites We found that theSIF-GPP relationship is consistent across dis-tinct vegetation types (Fig 3A) This finding isdifferent from previous reports that suggestedhighly variable biome-dependent relationships(13) There may be multiple explanations for thepotential divergence in their SIF-GPP relation-ships One possibility is the effect of the differencesacross biomes in plant physiology and canopystructure such as leaf angle orientation leaf clump-ing leaf area index and chlorophyll profiles (8)Such effect has also been demonstrated throughmodel simulations (33) Yet another possibility isthe potential systematic bias in either SIF or GPPproducts To investigate this latter possibility werepeatedouranalysesusingbothFLUXCOM(34)andModerate Resolution Imaging Spectroradiometer(MODIS) GPP products (35 36) at flux towerndashspecific pixels during the OCO-2 overpass datesFLUXCOM GPP is a data set derived from statis-tically upscaled EC measurements whereas MODISGPP is a data set modeled with the light-use ef-ficiency concept In these analyses the obtainedSIF-GPP relationships diverge among biomes(Fig 3 B and C) especially for MODIS productsThis indicates that the highly biome-dependentSIF-GPP relationships as found in previous studiesmay at least partly result from systematic biasesin GPP data sets whose production inevitably in-volves assumptions and models as GPP cannot bedirectly measured at regional or global scales (37)Although it is tempting to think about a uni-

versal SIF-GPP relationship across biomes (Fig3A) the existence of biome-specific relationshipscannot be ruled out at present Just a few EC sitesare located in the direct or close vicinity of OCO-2orbits and the OCO-2 lifetime is still short lim-iting our analyses Only continued research withmore land biomes and growing data records canreveal the true discrepancies or consistenciesunderlying this relationship and the responsiblemechanismsThe following simple equations however show

the possibility of variable SIF-GPP relationshipsamong biomes but also point to plausible mech-anisms if a universal relationship does exist Tradi-tionally the simplest way to model GPP is throughthe approach of light-use efficiency of CO2 assim-ilation (FCO2 ) (35) via the Monteith theory (38)

GPP frac14 FCO2aI eth1THORN

Here I is the photosynthetically active radia-tion (PAR) incident upon the canopy and a isthe canopy absorbance of PAR This approachhas been widely used particularly in the field ofvegetation remote sensing (36) To effectively usethis method measuring I is not enough one mustalso know bothFCO2 and a which are extreme-ly difficult to determine at large scales and canchange with environmental conditions and vege-tation types However the task becomes some-what easier if SIF is known For SIF a similarrelationship holds via the Berry equation (16)

SIF frac14 FFaIb eth2THORN

Sun et al Science 358 eaam5747 (2017) 13 October 2017 3 of 6

Fig 2 Instrument characteristics and SIF retrievals with CFIS (A) Spectral coverage and SIFretrieval windows for both OCO-2 (red) and CFIS (black) (B) Two initial validation flights forOCO-2 SIF on 13 and 16 August 2015 Error bars represent the SE of the OCO-2 SIF retrieval(C) Fine-scale CFIS SIF retrievals over agricultural areas in southwest Minnesota together with theRGB imagery derived from a four-band (RGB + NIR) context camera and NDVI at a finer spatialresolution (lt1 m as opposed to 10 to 20 m of CFIS) Note that different resolutions between CFISand context camera images should not be confused with each other as the longer exposure timeof the former results in elongated pixels along the flight track The region affected by partialcloud cover is highlighted by the red box

RESEARCH | RESEARCH ARTICLE | REMOTE SENSINGon M

arch 30 2020

httpsciencesciencemagorg

Dow

nloaded from

Here FF is the light-use efficiency of SIF and bis the probability of SIF photons escaping thecanopy Combining Eqs 1 and 2 leads to

GPP frac14 FCO2

bFFSIF eth3THORN

which relates SIF to GPPEquations 1 to 3 may not be the most mecha-

nistic way to describe the relationship betweenSIF and GPP and its potential variations with amultitude of biotic and abiotic factors across dif-ferent climates and biomes A more mechanisticalternative involves the description of processessuch as energy partitioning between photosys-tem I and II canopy structure stoichiometry andfluorescing properties of these two photosystemsphotorespiration photosynthetic pathways (C3versus C4) linear and cyclic electron transportsand fluorescence radiative transfer modeling incanopies (8) Although such a complex approachwill be important for developing the knowledgebase needed to bridge the gap between bottom-up biophysical modeling and empirical top-downconstraints Eqs 1 to 3 offer a convenient frame-work for presenting and evaluating arguments andcounterarguments for the SIF-GPP relationshipAn important difference between Eqs 3 and

1 is that the former involves a ratio of two po-tentially covarying terms of energy-use efficiencywhereas the latter uses the product of two in-dependent variables The variability of b likelydepends on canopy geometry solar elevationview angle and other conditionsFCO2 andFF arealso not constants at the leaf scale (9) FCO2varieswith photosynthetic capacities and environmen-tal conditions (such as light atmospheric CO2and humidity) in a way that is typically predicted

with the Farquharndashvon CaemmererndashBerry modelof photosynthesis (39) FF changes with envir-onmental conditions that affect photochemicaland nonphotochemical quenching (8) Thus itseems natural to assume that the slope of theGPP-SIF relationship will vary across biomesA universal SIF-GPP linear relationship wouldat least require interbiome variations in FCO2

andFF to cancel each other a scenario that seemsdifficult to realize Clearly more in-depth process-based studies are needed to understand the na-ture of the SIF-GPP relationship A particularemphasis should be placed on the covariation be-tweenFCO2 andFF at different spatiotemporalscales which will be key to using SIF as a short-cut to estimating GPP at large scales (40)

Consistent spatiotemporal variationsin SIF and GPP revealed by OCO-2

So far we have focused on the characteristics ofOCO-2 SIF and its relationship with GPP at finescales We now address the question of to whatextent SIF can be used to predict the spatiotem-poral dynamics of GPP To fully address thisissue both SIF and GPP products will have to beimproved considerably Nevertheless this doesnot prevent us from using data already availableto gain initial insights We therefore employedthe empirical orthogonal function (EOF) methodto decompose the complex spatial and temporalvariability of SIF and GPP into various orthog-onal components This analysis allows us to iden-tify common patterns and discrepancies acrossnoisy data sets that are usually characterized bynonlinearity and high dimensionality We per-formed an EOF analysis on monthly data sets ofOCO-2 SIF as well as FLUXCOM and MODIS GPP

products and investigated their temporal andspatial coherences for each orthogonal compo-nent Figure 4A shows the four leading EOFmodes for all variables ordered by how muchvariance in the data set each mode explains (figS2) For the first leading mode all three varia-bles closely match each other except in tropicalSouth America explaining 63 74 and 66 of thetotal variance in OCO-2 SIF FLUXCOM GPP andMODIS GPP respectively The corresponding timeseries depicts the seasonal dynamics with all var-iables in good agreement with each other (Fig4B) The Pearson correlation coefficients betweenOCO-2 SIF and FLUXCOM GPP quantifying theirspatial similarity are consistently high across allbiomes in this first mode (Fig 4C) Comparedwith the FLUXCOM product MODIS GPP tendsto have a lower correlation with OCO-2 SIF espe-cially in the tropical evergreen broadleaf forestsFrom the second mode onward interesting dis-crepancies emerge between MODIS GPP and theother two data sets in different regions of theworld In South America for example the secondmode of OCO-2 SIF and FLUXCOM GPP identi-fies a northeast-to-southwest stretch (the greenpositive phase) which is absent in MODIS GPPdata Similar contrast exists in the northern edgeof tropical Africa In the third and fourth modesMODIS GPP shows a profound dipole in the cen-tral Amazon which is not present in the othertwo data sets These discrepancies have led to alower correlation between OCO-2 SIF and MODISGPP than with the FLUXCOM product (Fig 4C)especially for the grassland and savanna systemsThe time series of the second to fourth compo-nents also display a closer similarity betweenOCO-2 SIF and FLUXCOMGPP than withMODIS

Sun et al Science 358 eaam5747 (2017) 13 October 2017 4 of 6

Fig 3 SIF-GPP relationships The relationship between GPP and OCO-2SIF (daily mean value denoted as SIF converted from instantaneousmeasurements) at three flux tower sites representative of three differentbiomes crops (Minnesota Tall Tower KCMP) (30) grass (Stuart Plain inAustralia) (31) and deciduous temperate forests [Missouri Ozark site(US_MOz)] The first two sites are selected because they are in the directunderpass of OCO-2 orbital tracks for the US_MOz site OCO-2 SIFretrievals are obtained from representative forests in the vicinity of thetower The KCMP footprint covers a mixture of corn soybean and grassesbut is dominated by the two major crops Error bars represent the SE of

the OCO-2 SIF retrieval Daily GPP in the 2015 growing season is obtainedduring the OCO-2 overpasses from (A) eddy covariance measurements(B) FLUXCOM products and (C) MODIS products sampled at these threeflux sites Both FLUXCOM and MODIS GPP are 8-day products and arelinearly interpolated to the OCO-2 overpass dates The site-specificFLUXCOM GPP value is extracted from the grid cell (0083deg by 0083deg) thatcorresponds to the latitude and longitude of the tower location The site-specific MODIS (MOD17A2) GPP value is the average of nine adjacentpixels (1 km by 1 km) centered at the tower location Both are roughlyequivalent to ~9-km2 area

RESEARCH | RESEARCH ARTICLE | REMOTE SENSINGon M

arch 30 2020

httpsciencesciencemagorg

Dow

nloaded from

GPP (Fig 4B) The highly consistent correspon-dences both in space and time in the leadingEOF modes of OCO-2 SIF and FLUXCOM GPPprovide robust proof that SIF and GPP are gov-erned by similar dynamics and controlled bysimilar environmental and biological conditionsIn addition these findings emphasize that SIFprovides additional information compared withgreenness-based productivity estimates as thespatiotemporal variations are markedly differ-ent from the MODIS GPP product

Synergistic uses of OCO-2 SIF withother remote sensing products

Past satellite observations (eg NDVI) have beenlargely based on surface spectral reflectance whichcontains a mixture of information about greenvegetation elements (eg contents of differentpigments leaf amount and spatial arrangement)nongreen biomass elements and underlying soil(41) They represent the best structural descrip-tions of the land surface at large scales and canbe used synergistically with the SIF functionalquantification of vegetation to understand andpredict terrestrial ecosystem dynamics underchanging environmental conditions For exam-ple phenology which is a key driver of terres-trial carbon uptake (42) is routinely monitoredby conventional reflectance-based remote sensing(eg MODIS) Therefore remotely sensed phenol-

ogy data sets can be combined with OCO-2 SIFproducts to study how vegetation phenology af-fects the SIF dynamics which will help to illumi-nate the causes for the potential biome-dependentSIF-GPP relationships at different phenologicalstages Conventional vegetation indices and SIFproducts could also be used to compare effici-encies of different farming practices around theworld and to monitor drought stress (18) In theseaspects the OCO-2 SIF data sets will be particu-larly valuable for synergistic use with conventionalvegetation indices obtained globally at spatialresolutions of 30 m and finer (eg Landsat andSentinel-2)Another potential application is to integrate

OCO-2 SIF with CO2 concentration measurementsto disentangle the net fluxes derived from atmo-spheric CO2 data into process-specific and thusinformation-rich gross fluxesmdashnamely GPP eco-system respiration and anthropogenic compo-nents Additionally SIF could also be used as aprior constraint on GPP for atmospheric trans-port inversion models or in data assimilationframeworks for such purposes (43)With growing data records OCO-2 SIF will

provide a robust benchmark for upcoming mis-sions such as OCO-3 FLEX the TROPOsphericMonitoring Instrument (TROPOMI) (26) andthe Geostationary Carbon Cycle Observatory(GeoCARB) (44) In addition tracking the sub-

daily variation of SIF will be possible by combin-ing OCO-2 products with products from theseupcoming satellite missions These endeavorswill lead to valuable data sets that can be usedin a variety of ecological research predictionand management activities Despite the lack ofcontiguous global coverage OCO-2rsquos measure-ments can be considered a cornerstone towardmore optimal SIF data sets that can fulfill evengreater requirements in terms of accuracy spa-tial coverage and resolution and can thus unleashthe power of SIF in revealing global photosyn-thetic activities

Summary

OCO-2 provides a high-resolution SIF data setthat allows direct validation against ground-based and airborne SIF observations With thisdata set it is possible to resolve diverse heterog-eneous landscape patterns in SIF emission Thesecapabilities have been challenging for previoussatellite missions that produced SIF retrievalsThe EOF analyses of the validated OCO-2 SIF dataset and GPP products revealed that the spatio-temporal variations of SIF and GPP are highlyconsistent suggesting that SIF is a powerful proxyfor GPP We also found that the relationship be-tween SIF and GPP when the latter is directly ob-tained at eddy flux sites in the vicinity of OCO-2orbits is more consistent across biomes than

Sun et al Science 358 eaam5747 (2017) 13 October 2017 5 of 6

Fig 4 Spatiotemporal patterns of EOF decomposition of OCO-2 SIFFLUXCOM GPP and MODIS GPP for the first four leading modesin descending order (A) Spatial maps of EOFs from monthly datasets (B) Corresponding time series of each EOF (C) Quantification ofthe spatial resemblance of OCO-2 SIF with FLUXCOM (green) andMODIS GPP (white) across biomes denoted as NF (needleleaf forests)

EBF (evergreen broadleaf forests) DBF (deciduous broadleaf forests)SHR (shrublands) SAV (savannas) GRA (grasslands) and CRO(croplands) The land cover data are from the International SatelliteLand Surface Climatology Project Initiative II biome classificationproducts using the International Geosphere-Biosphere Programme (IGBP)scheme following (11)

RESEARCH | RESEARCH ARTICLE | REMOTE SENSINGon M

arch 30 2020

httpsciencesciencemagorg

Dow

nloaded from

was suggested by earlier studies However fu-ture investigations with expanded data sets willneed to test the robustness of this finding acrossall land biomes Despite this uncertainty it isclear that high-quality satellite SIF is of centralimportance to studying terrestrial ecosystemdynamics and carbon cycle Eventual synergisticuses of SIF with atmospheric CO2 enabled byOCO-2 will lead to more reliable estimates of ter-restrial carbon sources and sinks and a deeperunderstanding of carbon-climate feedbacks

REFERENCES AND NOTES

1 R J Norby et al Forest response to elevated CO2 isconserved across a broad range of productivity Proc NatlAcad Sci USA 102 18052ndash18056 (2005) doi 101073pnas0509478102 pmid 16330779

2 P Friedlingstein et al Climate-carbon cycle feedback analysisResults from the C4MIP model intercomparison J Clim 193337ndash3353 (2006) doi 101175JCLI38001

3 C Huntingford et al Simulated resilience of tropicalrainforests to CO2-induced climate change Nat Geosci 6268ndash273 (2013) doi 101038ngeo1741

4 C Rosenzweig et al Assessing agricultural risks of climatechange in the 21st century in a global gridded crop modelintercomparison Proc Natl Acad Sci USA 111 3268ndash3273(2014) doi 101073pnas1222463110 pmid 24344314

5 D Schimel et al Observing terrestrial ecosystems and thecarbon cycle from space Glob Change Biol 21 1762ndash1776(2015) doi 101111gcb12822 pmid 25472464

6 G C Papageorgiou Govindjee in Non-PhotochemicalQuenching and Energy Dissipation in Plants Algae andCyanobacteria B Demmig-Adams G Garab W Adams IIIGovindjee Eds (Springer 2014) pp 1ndash44

7 N R Baker Chlorophyll fluorescence A probe ofphotosynthesis in vivo Annu Rev Plant Biol 59 89ndash113(2008) doi 101146annurevarplant59032607092759pmid 18444897

8 A Porcar-Castell et al Linking chlorophyll a fluorescence tophotosynthesis for remote sensing applications Mechanismsand challenges J Exp Bot 65 4065ndash4095 (2014)doi 101093jxberu191 pmid 24868038

9 B Genty J-M Briantais N R Baker The relationship betweenthe quantum yield of photosynthetic electron transport andquenching of chlorophyll fluorescence Biochim BiophysActa Gen Subj 990 87ndash92 (1989) doi 101016S0304-4165(89)80016-9

10 G H Krause E Weis Chlorophyll fluorescence andphotosynthesis The basics Annu Rev Plant Physiol PlantMol Biol 42 313ndash349 (1991) doi 101146annurevpp42060191001525

11 C Frankenberg et al New global observations of the terrestrialcarbon cycle from GOSAT Patterns of plant fluorescencewith gross primary productivity Geophys Res Lett 38 L17706(2011) doi 1010292011GL048738

12 J Joiner et al First observations of global and seasonalterrestrial chlorophyll fluorescence from space Biogeosciences8 637ndash651 (2011) doi 105194bg-8-637-2011

13 L Guanter et al Retrieval and global assessment of terrestrialchlorophyll fluorescence from GOSAT space measurementsRemote Sens Environ 121 236ndash251 (2012) doi 101016jrse201202006

14 J Joiner et al Global monitoring of terrestrial chlorophyllfluorescence from moderate-spectral-resolution near-infraredsatellite measurements Methodology simulations andapplication to GOME-2 Atmos Meas Tech 6 2803ndash2823(2013) doi 105194amt-6-2803-2013

15 P Koumlhler L Guanter J Joiner A linear method for the retrievalof sun-induced chlorophyll fluorescence from GOME-2 and

SCIAMACHY data Atmos Meas Tech 8 2589ndash2608 (2015)doi 105194amt-8-2589-2015

16 J-E Lee et al Forest productivity and water stress inAmazonia Observations from GOSAT chlorophyll fluorescenceProc Biol Sci 280 20130171 (2013) doi 101098rspb20130171 pmid 23760636

17 N C Parazoo et al Interpreting seasonal changes in thecarbon balance of southern Amazonia using measurements ofXCO2 and chlorophyll fluorescence from GOSAT Geophys ResLett 40 2829ndash2833 (2013) doi 101002grl50452

18 Y Sun et al Drought onset mechanisms revealed by satellitesolar-induced chlorophyll fluorescence Insights from twocontrasting extreme events J Geophys Res Biogeosci 1202427ndash2440 (2015) doi 1010022015JG003150

19 Y Yoshida et al The 2010 Russian drought impact on satellitemeasurements of solar-induced chlorophyll fluorescenceInsights from modeling and comparisons with parametersderived from satellite reflectances Remote Sens Environ 166163ndash177 (2015) doi 101016jrse201506008

20 K Guan et al Improving the monitoring of crop productivityusing spaceborne solar-induced fluorescence Glob ChangeBiol 22 716ndash726 (2016) doi 101111gcb13136pmid 26490834

21 J Joiner et al The seasonal cycle of satellite chlorophyllfluorescence observations and its relationship to vegetationphenology and ecosystem atmosphere carbon exchangeRemote Sens Environ 152 375ndash391 (2014) doi 101016jrse201406022

22 S Walther et al Satellite chlorophyll fluorescencemeasurements reveal large-scale decoupling of photosynthesisand greenness dynamics in boreal evergreen forests GlobChange Biol 22 2979ndash2996 (2016) doi 101111gcb13200pmid 26683113

23 L Guanter et al Global and time-resolved monitoring ofcrop photosynthesis with chlorophyll fluorescence Proc NatlAcad Sci USA 111 E1327ndashE1333 (2014) doi 101073pnas1320008111 pmid 24706867

24 C Frankenberg et al Prospects for chlorophyll fluorescenceremote sensing from the Orbiting Carbon Observatory-2Remote Sens Environ 147 1ndash12 (2014) doi 101016jrse201402007

25 M Rossini et al Analysis of red and far-red sun-inducedchlorophyll fluorescence and their ratio in different canopiesbased on observed and modeled data Remote Sens 8412 (2016) doi 103390rs8050412

26 L Guanter et al Potential of the TROPOspheric MonitoringInstrument (TROPOMI) onboard the Sentinel-5 Precursor for themonitoring of terrestrial chlorophyll fluorescence Atmos MeasTech 8 1337ndash1352 (2015) doi 105194amt-8-1337-2015

27 M Drusch et al The FLuorescence EXplorer Mission ConceptmdashESArsquos Earth Explorer 8 IEEE Trans Geosci Remote Sens55 1273ndash1284 (2017) doi 101109TGRS20162621820

28 U Rascher et al Sun-induced fluorescence ndash a new probeof photosynthesis First maps from the imagingspectrometer HyPlant Glob Change Biol 21 4673ndash4684(2015) doi 101111gcb13017 pmid 26146813

29 S Wieneke et al Airborne based spectroscopy of red and far-redsun-induced chlorophyll fluorescence Implications for improvedestimates of gross primary productivity Remote Sens Environ184 654ndash667 (2016) doi 101016jrse201607025

30 J D Wood et al Multiscale analyses of solar-inducedflorescence and gross primary production Geophys Res Lett44 533ndash541 (2017) doi 1010022016GL070775

31 M Verma et al Effect of environmental conditions on therelationship between solar-induced fluorescence and grossprimary productivity at an OzFlux grassland site J Geophys ResBiogeosci 122 716ndash733 (2017) doi 1010022016JG003580

32 D Baldocchi et al FLUXNET A new tool to study thetemporal and spatial variability of ecosystem-scale carbondioxide water vapor and energy flux densities Bull AmMeteorol Soc 82 2415ndash2434 (2001) doi 1011751520-0477(2001)082lt2415FANTTSgt23CO2

33 J Verrelst et al Evaluating the predictive power ofsun-induced chlorophyll fluorescence to estimate netphotosynthesis of vegetation canopies A SCOPE modelingstudy Remote Sens Environ 176 139ndash151 (2016)doi 101016jrse201601018

34 G Tramontana M Jung G Camps-valls K Ichii B RadulyPredicting carbon dioxide and energy fluxes across globalFLUXNET sites with regression algorithms Biogeosciences 134291ndash4313 (2016) doi 105194bg-13-4291-2016

35 S W Running et al A continuous satellite-derived measureof global terrestrial primary production Bioscience 54547ndash560 (2004) doi 1016410006-3568(2004)054[0547ACSMOG]20CO2

36 M Zhao S W Running Drought-induced reduction inglobal terrestrial net primary production from 2000through 2009 Science 329 940ndash943 (2010) doi 101126science1192666 pmid 20724633

37 A Anav et al Spatiotemporal patterns of terrestrial grossprimary production A review Rev Geophys 53 785ndash818 (2015)doi 1010022015RG000483

38 J L Monteith Solar radiation and productivity in tropicalecosystems Source J Appl Ecol 9 747ndash766 (1972)doi 1023072401901

39 G D Farquhar S von Caemmerer J A Berry A biochemicalmodel of photosynthetic CO2 assimilation in leaves of C3species Planta 149 78ndash90 (1980) doi 101007BF00386231pmid 24306196

40 T S Magney et al Connecting active to passive fluorescencewith photosynthesis A method for evaluating remotesensing measurements of Chl fluorescence New Phytol 2151594ndash1608 (2017) doi 101111nph14662 pmid 28664542

41 A Huete et al Overview of the radiometric and biophysicalperformance of the MODIS vegetation indices RemoteSens Environ 83 195ndash213 (2002) doi 101016S0034-4257(02)00096-2

42 T Keenan et al Net carbon uptake has increased throughwarming-induced changes in temperate forest phenologyNat Clim Chang 4 598ndash604 (2014) doi 101038nclimate2253

43 N C Parazoo et al Terrestrial gross primary productioninferred from satellite fluorescence and vegetationmodels Glob Change Biol 20 3103ndash3121 (2014)doi 101111gcb12652 pmid 24909755

44 I N Polonsky D M OrsquoBrien J B Kumer C W OrsquoDellthe geoCARB Team Performance of a geostationary missiongeoCARB to measure CO2 CH4 and CO column-averagedconcentrations Atmos Meas Tech 7 959ndash981 (2014)doi 105194amt-7-959-2014

45 C G Homer et al Completion of the 2011 National Land CoverDatabase for the conterminous United States-Representing adecade of land cover change information PhotogrammEng Remote Sens 81 345ndash354 (2015)

ACKNOWLEDGMENTS

A portion of this research was carried out at the Jet PropulsionLaboratory California Institute of Technology under a contractwith NASA AP-C is funded by the Academy of Finland (ResearchFellow grant 1288039) L Gu and JDW are supported by theUS Department of Energy Biological and Environmental Researchprogram through Oak Ridge National Laboratoryrsquos TerrestrialEcosystem Science (TES) Science Focus Area (SFA) MJacknowledges support from the European Union H2020 BiosphereAtmosphere Change Index project (grant 640176) We thank the dataproviders of FLUXCOM and MODIS for their contribution The OCO-2SIF data set is publicly available at httpsco2jplnasagov The CFISSIF retrieval and GPP inferred at flux towers are available upon request

SUPPLEMENTARY MATERIALS

wwwsciencemagorgcontent3586360eaam5747supplDC1Figs S1 and S2

11 December 2016 accepted 7 July 2017101126scienceaam5747

Sun et al Science 358 eaam5747 (2017) 13 October 2017 6 of 6

RESEARCH | RESEARCH ARTICLE | REMOTE SENSINGon M

arch 30 2020

httpsciencesciencemagorg

Dow

nloaded from

fluorescenceOCO-2 advances photosynthesis observation from space via solar-induced chlorophyll

Griffis L Gu T S Magney P Koumlhler B Evans and K YuenY Sun C Frankenberg J D Wood D S Schimel M Jung L Guanter D T Drewry M Verma A Porcar-Castell T J

DOI 101126scienceaam5747 (6360) eaam5747358Science

ARTICLE TOOLS httpsciencesciencemagorgcontent3586360eaam5747

MATERIALSSUPPLEMENTARY httpsciencesciencemagorgcontentsuppl201710123586360eaam5747DC1

CONTENTRELATED

httpsciencesciencemagorgcontentsci3586360eaam5782fullhttpsciencesciencemagorgcontentsci3586360eaam5690fullhttpsciencesciencemagorgcontentsci3586360eaam5776fullhttpsciencesciencemagorgcontentsci3586360eaam5745fullhttpsciencesciencemagorgcontentsci3586360186full

REFERENCES

httpsciencesciencemagorgcontent3586360eaam5747BIBLThis article cites 44 articles 4 of which you can access for free

PERMISSIONS httpwwwsciencemagorghelpreprints-and-permissions

Terms of ServiceUse of this article is subject to the

is a registered trademark of AAASScienceScience 1200 New York Avenue NW Washington DC 20005 The title (print ISSN 0036-8075 online ISSN 1095-9203) is published by the American Association for the Advancement ofScience

Science No claim to original US Government WorksCopyright copy 2017 The Authors some rights reserved exclusive licensee American Association for the Advancement of

on March 30 2020

httpsciencesciencem

agorgD

ownloaded from