Modeling the influence of snow cover on low Arctic net ecosystem exchange

Modeling the influence of snow cover on low Arctic net ecosystem exchange

This content has been downloaded from IOPscience Please scroll down to see the full text

Download details

IP Address 19316022513

This content was downloaded on 04102013 at 0203

Please note that terms and conditions apply

2013 Environ Res Lett 8 035045

(httpiopscienceioporg1748-932683035045)

View the table of contents for this issue or go to the journal homepage for more

Home Search Collections Journals About Contact us My IOPscience

IOP PUBLISHING ENVIRONMENTAL RESEARCH LETTERS

Environ Res Lett 8 (2013) 035045 (10pp) doi1010881748-932683035045

Modeling the influence of snow cover onlow Arctic net ecosystem exchange

K A Luus12 R E J Kelly1 J C Lin23 E R Humphreys4 P M Lafleur5 andW C Oechel6

1 Geography University of Waterloo ON N2L 3C5 Canada2 Earth and Environmental Sciences University of Waterloo ON N2L 3C5 Canada3 Atmospheric Sciences University of Utah UT 84112 USA4 Geography and Environmental Studies Carleton University Ottawa ON K1S 5B6 Canada5 Geography Trent University Peterborough ON K9J 7B8 Canada6 Biology San Diego State University CA 95616 USA

E-mail kaluusuwaterlooca

Received 27 March 2013Accepted for publication 22 August 2013Published 11 September 2013Online at stacksioporgERL8035045

AbstractThe Arctic net ecosystem exchange (NEE) of CO2 between the land surface and theatmosphere is influenced by the timing of snow onset and melt The objective of this study wasto examine whether uncertainty in model estimates of NEE could be reduced by representingthe influence of snow on NEE using remote sensing observations of snow cover area (SCA)Observations of NEE and time-lapse images of SCA were collected over four locations at alow Arctic site (Daring Lake NWT) in MayndashJune 2010 Analysis of these observationsindicated that SCA influences NEE and that good agreement exists between SCA derivedfrom time-lapse images Landsat and MODIS MODIS SCA was therefore incorporated intothe vegetation photosynthesis respiration model (VPRM) VPRM was calibrated usingobservations collected in 2005 at Daring Lake Estimates of NEE were then generated overDaring Lake and Ivotuk Alaska (2004ndash2007) using VPRM formulations with and withoutexplicit representations of the influence of SCA on respiration andor photosynthesis Modelperformance was assessed by comparing VPRM output against unfilled eddy covarianceobservations from Daring Lake and Ivotuk (2004ndash2007) The uncertainty in VPRM estimatesof NEE was reduced when respiration was estimated as a function of air temperature whenSCA le 50 and as a function of soil temperature when SCA gt 50

Keywords net ecosystem exchange Arctic snow remote sensing modeling CO2cryosphere

1 Introduction

In low Arctic regions the initial onset and final melt ofsnow mark important transitions in net ecosystem exchange(NEE) (Olsson et al 2003 Grogan et al 2004 Bokhorst et al2010 Buckeridge and Grogan 2010) where NEE is defined

Content from this work may be used under the terms ofthe Creative Commons Attribution 30 licence Any further

distribution of this work must maintain attribution to the author(s) and thetitle of the work journal citation and DOI

as the net biospheric flux of CO2 into and out of the landsurface (Lovett et al 2006) NEE can be described as thesum of photosynthetic uptake by vegetation (GEE or grossecosystem exchange) and ecosystem respiration (R) NEE =minusGEE + R According to the sign convention used in thisstudy uptake of CO2 from the atmosphere is represented asnegative NEE and release of CO2 into the atmosphere isshown as positive NEE

Photosynthetic uptake by vegetation is maximized duringthe growing season when above-freezing air temperatures(Tair) and sunny conditions support plant growth During snow

11748-932613035045+10$3300 ccopy 2013 IOP Publishing Ltd Printed in the UK

Environ Res Lett 8 (2013) 035045 K A Luus et al

onset in autumn the land surface is cooling (Zhang et al1996) light availability is limited and most plants are insenescence resulting in diminished rates of photosynthesisand respiration (Billings and Mooney 1968 Carstairs andOechel 1978 Oquist and Huner 2003 Olsson et al 2003Euskirchen et al 2012) Snowpack development decouples Tairand soil temperature (Tsoil) (Bonan 2002) allowing subniveanrespiration to persist at low Tair (Zimov et al 1993 Olsson et al2003) Snow melt in the Arctic normally takes place within amonth of the solstice Light availability is therefore high andmelt is accompanied by warmer Tair soil thaw and greateravailability of nutrients As a result snowmelt is accompaniedby rapid increases in rates of respiration (Zimov et al 1996Mikan et al 2002 Oelbermann et al 2008 Elberling et al2008) Although the length of time between snowmelt andgreen-up varies by species (Humphreys and Lafleur 2011)the timing of snowmelt at a site influences the timing ofphotosynthetic uptake by vegetation (Morgner et al 2010Buckeridge and Grogan 2010)

Landscape rates of photosynthesis and respiration duringsnow onset and snow melt are also influenced by theproportion of the land surface which is snow covered at anygiven point in time (snow cover area or SCA) Comparisonsof NEE at plots with differing quantities of snow have founddiminished rates of photosynthesis and respiration duringsnow onsetmelt at plots with greater SCA (eg Buckeridgeand Grogan 2010 and Morgner et al 2010) Representing SCAin biospheric carbon flux models could therefore allow thesnow transitional and growing seasons to be clearly delimitedModel estimates of NEE during these time periods could thenbe generated by simulating the differing seasonal drivers ofNEE for each period Hence uncertainty in model estimatesof Arctic NEE might be reduced by explicitly representing theinfluence of SCA on NEE

Although most model estimates of Arctic NEE do notsimulate snow season influences on NEE several process-based approaches (eg McGuire et al 2000 Wania et al2009 and Gouttevin et al 2012) have represented snowpackproperties (eg density diffusivity) mechanistically and havethen generated estimates of subnivean respiration in lightof these snowpack properties Findings from McGuire et al(2000) indicate that model agreement with observations canbe improved by explicitly representing the influence of snowon heterotrophic respiration Since gt50 of annual lowArctic CO2 efflux can occur during the snow season (Mikanet al 2002 Sullivan et al 2008) the resulting improvementsin model performance can have important implications foraccuracy in estimates of the Arctic carbon cycle

To date model estimates of NEE have not made useof remote sensing observations of snowpack characteristicsalthough a variety of remote sensing observations couldbe used to represent different influences of snow on NEE(Luus et al 2013) Specifically estimates of SCA can bemade from visible and infrared remote sensing observationsfrom satellites such as Landsat (Dozier 1989 Rosenthaland Dozier 1996) and MODIS (moderate resolution imagingspectroradiometer) (Hall et al 2002 Hall and Riggs 2007Riggs and Hall 2011) Remote sensing observations of SCA

could be applied to represent the influence of snow ondecoupling soil and air temperatures (Olsson et al 2003Bonan 2002) and causing snow season respiration to bedriven by soil temperature rather than air temperatureSimilarly rates of photosynthesis are greatly limited by thepresence of an overlying snowpack (Tieszen 1974) sincesnow scatters incoming solar radiation and therefore limits theamount of light that can reach subnivean vegetation (Warren1982 Zhou et al 2003) Estimates of SCA from satellites suchas Landsat and MODIS could therefore be incorporated intomodels of biospheric CO2 fluxes and applied to improve therepresentation of snow season (SCA gt 0) respiration andorphotosynthesis

Modeling the influence of snow on NEE using aremote sensing approach offers several advantages overa process-based approach Incorporating remote sensingobservations of snow characteristics in a model of NEE limitsthe propagation of meteorological biases into estimates ofsnow characteristics and allows spatial variability in snowdistributions to be captured without having to simulate themany land surface influences on snow accumulation and melt(ie topography aerodynamic roughness) In order for SCAto be included in models of NEE it is important to firstaddress the challenges which have previously prevented theirinclusion in models of NEE Specifically uncertainty hasexisted about the accuracy of remote sensing estimates ofSCA at high-latitude sites whether the influence of snow onNEE could be detected using SCA and whether uncertaintyin model estimates of NEE could be reduced by incorporatingSCA The research provided here therefore systematicallyaddresses these three challenges

The central goal of this research was to explore thepotential for incorporating SCA and its effects on biosphericcarbon fluxes into the vegetation photosynthesis respirationmodel (VPRM) a diagnostic remote sensing assimilationscheme designed to provide regional estimates of NEE(Mahadevan et al 2008) The specific objectives were toexamine the feasibility of assimilating MODIS SCA intoa model of biospheric carbon fluxes and to examinewhether uncertainty in VPRM estimates of NEE couldsubsequently be reduced by simulating the influences of SCAon photosynthesis andor respiration

2 Methodology

21 Study sites

The Daring Lake (DL) site is located in the southern portionof the Northwest Territories at 6452N 11134primeW asymp200 kmnortheast of Yellowknife Daring Lake receives an average of200ndash300 mm in precipitation annually and has a mean annualtemperature of minus125 to minus105 C (Lafleur and Humphreys2008) At this study site four time-lapse cameras wereautomated to capture thrice daily images of the land surfacein MayndashJune 2010 from towers overlooking mixed tundra(MT) fen (FN) low shrub mixed tundra (LK) and tall shrub(SB) vegetation Observations of NEE were simultaneouslyacquired from eddy covariance towers at MT and FN in

2


Figure 1 Camera images of fractional snow cover over four vegetation types at Daring Lake (from top to bottom mixed tundra fen lowshrub mixed tundra and tall shrub) on 9 May (DOY 129) 31 May (DOY 151) and 3 June (DOY 154) 2010

MayndashJune 2010 Measurements of NEE and meteorologicalvariables have been collected at the Daring Lake MT sitesince 2004 The MT site is underlain by sand to loamy sandtextured soil and is composed of shrub tussock tundra andmesic heath The FN site is a wet sedge meadow with 40to 70 cm of peat soil overlying silt loam textured soil Thedominant vegetation is sedges with minor amounts of dwarfbirch and a moss understory (Humphreys and Lafleur 2011Lafleur and Humphreys 2008)

In order to select a paired validation site for DaringLake mixed tundra (MT) all North American sites with eddycovariance and meteorological observations (2004ndash2007)were considered Of these Ivotuk (IV) was selected as thevalidation site because it is the most similar to Daring LakeMT in terms of vegetation precipitation and temperatureIvotuk is an AmeriFlux site located on the north Slope ofAlaska at 6829N 15544prime asymp300 km south of Barrow Theaverage temperature and liquid precipitation at Ivotuk havebeen reported within the minus89 to minus146 C and 123ndash221 mmranges by Laskowski (2010) The Ivotuk site has beenclassified as a moist acidic tussock tundra site dominatedby Eriophorum vaginatum and containing shrubs mosses andlichen (Thompson et al 2006 Laskowski 2010)

All eddy covariance observations of NEE were filteredto remove periods of time with low frictional velocity andtime periods with sensor malfunction (eg when windowson the open-path infrared gas analyzer were obscured) Nogap-filling was performed for observations from either site

and no additional data points were removed The samenon-gap-filled datasets were used throughout the entire study

22 Calculating snow cover area

Snow cover area was estimated from visible andor infraredobservations available at three resolutions from threedifferent sources time-lapse camera (lt10 m) Landsat (30 m)and MODIS (500 m) Although images from both 1 Mayndash30June 2010 and 30 Augustndash7 December 2010 were examinedwe focus the analysis of snow-NEE associations on thetime period when camera images of SCA at DL MTwere acquired simultaneously with meteorological and eddycovariance observations 1 Mayndash30 June 2010 All cameraimages acquired during this two month time period wereindividually classified in ENVIIDL using a combination ofsupervised parallelepiped and unsupervised isodata classifiersin ENVI Unsupervised isodata classifiers were first appliedAll images were then visually assessed to determine how wellthe isodata classifications captured SCA For images that werenot well classified using the isodata approach a supervisedparallelepiped classification was instead applied Figure 1shows a selection of time-lapse images from the four locationsat Daring Lake at the start middle and end of snow melt in2010 From these analyses a percentage of fractional snowcover was calculated three times per day

Landsat images were collected over the 2004ndash2007and 2010 periods in MayndashJune These images wereclassified in terms of snow presenceabsence using the

3


normalized difference snow index (NDSI) The NDSI hasbeen used to identify snow on the basis that snow reflectsless middle-infrared (rMIDIR 155ndash175 microm) than visible(rGREEN 052ndash060 microm) radiation

NDSI =rGREEN minus rMIDIR

rGREEN + rMIDIR (1)

Regions with NDSI gt 04 and r076minus090 microm gt 11 wereclassified as snow covered whereas other regions wereclassified as non-snow covered as according to Hall et al(1995) The MODIS 10A1 fractional SCA product was usedas the 500 m estimate of SCA As MOD10A1 and Landsat areavailable on 7ndash9 day intervals basic linear interpolation wasused to generate three-hourly estimates of snow cover

23 Incorporating the influence of snow cover in VPRM

VPRM has previously been used to simulate NEE over mid-latitude (south of 56N) sites in North America (Mahadevanet al 2008) Photosynthetic uptake by vegetation (grossecosystem exchange GEE) has been calculated accordingto incoming photosynthetically active radiation (PAR) thefraction of PAR which can be absorbed by photosyntheticallyactive vegetation (FAPARPAV) and scalar values representingthe limiting influences of air temperature (Tscale) land surfacewater (Wscale) and phenology (Pscale) on GEE PAR0 and λhave represented empirically calibrated parameters describingvegetation light use efficiency (Mahadevan et al 2008)In this formulation snow implicitly limits GEE when airtemperatures PAR and FAPARPAV decline

GEE = (λTscaleWscalePscale)FAPARPAV1

1+ PARPAR0

PAR (2)

In VPRM NEE has been calculated according to thesum of GEE and respiration NEE = minusGEE + RESP andrespiration has been calculated using a piecewise approachWhen air Tair has been warmer than a threshold temperature(Tlow) respiration has been calculated as a linear functionof air temperature (RESP = αTair + β) When Tair lt Tlowrespiration has been set to a low baseline value independentof temperature Tlow α and β have been calculated accordingto the relationships found between NEE and Tair usingtower observations In this original formulation cold seasonrespiration has been assumed to remain at a constant ratethroughout the time period when Tair lt Tlow regardless offluctuations over time in snowpack properties or subniveantemperatures

VPRM driver data was composed entirely of remote-sensing-based estimates of land surface characteristics andmeteorology by MODIS and North American regionalreanalysis (NARR) PAR TairTsoil and Tscale were acquiredfrom NARR downward shortwave radiation 0ndash10 cm Tsoiland 2 m Tair datasets NARR estimates were generatedby a model which assimilated a variety of meteorologicalobservations (Mesinger et al 2006) At Daring Lake(2004ndash2007 2010) and Ivotuk (2004ndash2007) NARR estimateshad good agreement (R2 gt 08) with all meteorological towerobservations of air temperature soil temperature at 5 cm

and PAR (shortwave radiation) Preliminary investigationsindicated that NARR estimates of these variables had betteragreement with field observations than those generated byother leading approaches WscalePscale and FAPARPAV werederived from MODIS MODIS offers established moderateresolution estimates of land surface characteristics fromvisible and infrared observations MODIS observations havebeen used extensively including in previous versions ofVPRM (Mahadevan et al 2008) Both NARR and MODISinputs therefore represent the best available driver datasets forVPRM at high latitudes

In this study daily estimates of NEE in 2004ndash2007 atDaring Lake and Ivotuk were generated using six modelformulations based on VPRM some of which represent theinfluence of snow on photosynthesis (GEEs) andor respiration(RESPs) according to MODIS SCA (0ndash100) and someof which do not The RESPs amp GEE0 model formulationwhich calculates snow season respiration according toNARR Tsoil and growing season respiration according toNARR Tair was found to have the best agreement witheddy covariance observations To determine whether thisbenefit arose from the calculation of respiration accordingto Tsoil rather than Tair alone (and not due to the useof SCA) two model formulations that calculate respirationyear-round as a linear (RESPTsoil linear) or piecewiselinear (RESPTsoil pwl) function of soil temperature were alsoevaluated The α βTlow αs and βsTTsoil-low αTsoil andβTsoil parameters were all individually calibrated to eddycovariance observations collected at Daring Lake in 2005 Inall of the model formulations below GEE0 refers to GEE ascalculated in equation (2) and NEE = minusGEE+ RESP

(i) RESP0 amp GEE0

GEE = GEE0

R =

Tair ge Tlow αTair + β

Tair lt Tlow RESPlow

(ii) RESP0 amp GEEs

GEE = GEE0(100minus SCA)

R =



(iii) RESPs amp GEE0

GEE = GEE0

R =

SCA lt 50 αTair + β

SCA ge 50 αsTsoil + βs

(iv) RESPs amp GEEs


R =



(v) RESPTsoil linear amp GEE0

GEE = GEE0

R = αTsoilTsoil + βTsoil

4


Figure 2 Fractional snow cover over time (25 Mayndash11 June2010) from classified thrice daily camera images from four sites atDaring Lake NWT MT (blue) FN (green) LK (yellow) and SB(red) Landsat derived estimates of SCA appear as straight undottedlines Landsat SCA is shown in red for the pixel containing SB andwith a solid black line for FN LK and MT MODIS SCA isindicated with a black dotted line for all four sites

(vi) RESPTsoil pwl amp GEE0

GEE = GEE0

R =

Tsoil ge TTsoil-low αTsoilTsoil + βTsoil

Tsoil lt TTsoil-low RESPTsoil-low

VPRM performance with these six model formulationswas assessed both qualitatively and statistically The errorε or difference between predicted (VPRM) and observed(non-gap-filled eddy covariance) daily average values of NEE(εi = prediminusobsi) was evaluated using two metrics the meanabsolute error (MAE) and root mean squared error (RMSE)(Willmott and Matsuura 2005)

MAE = nminus1nsum

i=1

|εi| (3)

RMSE =

[nminus1

nsumi=1

|εi|2

] 12

(4)

Results were evaluated over two time periods 1 Mayndash7 Juneof 2004ndash2007 (MJ) and the portion of years 2004ndash2007 whenMOD10A1 reported gt0 SCA (snow season or SS)

3 Results and discussion

31 Landsat and MODIS estimates of local snow cover

Time-lapse camera observations of fractional SCA agreedwell with linearly interpolated Landsat NDSI and MODISobservations of whether the pixel containing each camerawas snow covered or snow-free (figure 2) Time-lapse cameraobservations and Landsat derived estimates of SCA showedsnow depletion to occur within a seven day time periodwith no substantial melts before snow depletion and no snowaccumulation following depletion

To assess the agreement between time-lapse Landsatand MODIS derived estimates of the timing of snowmelt

Table 1 The ordinal date at which SCA is first below 50according to time-lapse camera observations and interpolatedestimates of SCA derived from Landsat and MODIS over fourvegetation types at Daring Lake

FN LK MT SB

Camera 150 151 151 150Landsat 155 155 155 148MODIS 149 149 149 149

MODIS and Landsat SCA were linearly interpolatedbetween asympweekly acquisitions Comparisons indicated aslight advance or delay of one to five days in Landsatand MODIS estimates of depletion relative to time-lapsecamera observations 1 A slight (lt7 day) discrepancy isacceptable considering the asympweekly temporal resolution ofLandsat and MOD10A1 SCA Overall these results indicategood agreement between ground-based and remote sensingobservations of SCA

32 Associations between NEE and SCA

Preliminary investigation of SCA TsoilTair and NEE overtime at the Daring Lake MT site indicated that SCA mayhave several important effects on NEE (figure 3) Althoughthe low thermal conductivity of snow enables Tsoil to remainwarmer than Tair throughout midwinter (figure 4) in the timeperiod immediately preceding snow melt air temperaturesare warmer than soil temperatures As Tair rises Tsoil slowlyfollows to reach a temperature of 0 C at which point snowmelt begins

A decline in SCA during snowmelt initially increases therate of respiration causing an increase in CO2 efflux Oncesnow melt is complete vegetation begins to green-up leadingto an increase in the rate of photosynthetic uptake of CO2 Airand soil temperatures also become more closely synchronizedfollowing snow melt In summary these findings suggest thatthe timing of snowmelt coincides closely with an increasein respiration and subsequent increase in photosynthesis atDaring Lake MT It therefore appears feasible that remotesensing observations of SCA could be incorporated intoVPRM to help describe the influences of SCA on respirationandor photosynthesis

33 VPRM estimates of NEE with and without MOD10A1SCA

All VPRM formulations which estimated growing seasonrespiration according to Tair generated reasonable estimatesof NEE over Ivotuk (IV) and Daring Lake MT (DL) in years2004ndash2007 (MAE = 02ndash05 micromol mminus2 sminus1 and RMSE =06ndash18 micromol mminus2 sminus1) MAEs were similar between DL andIV RMSE values tend to be greater at IV than DL becausea larger portion of observations at IV were collected inmidwinter a time of year in which uncertainty in observationsof NEE is greatest (Amiro 2010)

Both the GEEs and GEE0 formulations accuratelyrepresented the lack of photosynthesis during the long Arctic

5


Figure 3 Landsat SCA MODIS SCA time-lapse camera SCA NEE 5 cm Tsoil and Tair from 25 May to 11 June 2010 at the Daring Lakemixed tundra (MT) site

Figure 4 Tair (green) and 10 cm Tsoil (yellow) over time at DaringLake MT (2005)

midwinter Snow melt was accompanied by a slight increase inphotosynthesis (figure 3) which was best simulated by GEE0models An increase in photosynthetic uptake at the end ofthe snow season was implicitly simulated by GEE0 accordingto rising air temperatures increased sunlight and an eventualslight increase in EVI Simulating further reductions inphotosynthesis when 0 lt SCA lt 100 caused VPRM tounderestimate GEE The RMSEs and MAEs from RESP0 ampGEEs and RESPs amp GEEs therefore exceeded those from theformulations which did not further suppress GEE as a linearfunction of SCA (RESP0 amp GEE0 and RESPs amp GEE0)

It is possible that GEE0 outperformed GEEs becauseunlike GEEs GEE0 does not explicitly specify that nosubnivean photosynthesis can take place Indeed subniveanphotosynthesis has been observed to occur prior to snowmeltover a sub-Arctic moss-dominated heath (Larsen et al2007) Similarly when Tair asymp 0 and snowcover is thin orpatchy the following Arctic species have been observedto conduct photosynthesis at lt25 of the rate observedin the peak of the growing season Carex aquatilisDupontia fisheri Eriophorum angustifolium (Tieszen 1974)Eriophorum vaginatum Ledum palustre Vaccinium vitis-idaea and Cassiope tetragona (Starr and Oberbauer 2003) Acentral reason why subnivean photosynthesis is a very gradualprocess is because snowpacks act as an optical scatteringmedium (Warren 1982) lt5 of incoming solar radiation hasbeen observed to penetrate gt10 cm snowpacks at BarrowAlaska (Tieszen 1974) Yet an examination of snowpackobservations from Daring Lake (Nobrega and Grogan 2007Rees et al 2010) and surrounding regions (Derksen et al2009) in relation to outputted thresholds from a penetrationdepth model by Zhou et al (2003) indicates that a portion ofincoming light most likely penetrates thin (lt5 cm) or patchy

snowpacks during snowmelt at DL Presently subniveanphotosynthesis has not been quantified over the vast majorityof Arctic vegetation species or low Arctic sites The explicitrepresentation of subnivean photosynthesis within VPRM istherefore beyond the scope of this study especially since GEE0describes snow and growing season GEE at both DL and IVwith low RMSEs and MAEs

Uncertainty in VPRM estimates of NEE was reducedwhen snow season respiration was calculated as a functionof Tsoil (RESPs) (table 2) When considering MayndashJuneestimates of NEE RESPs amp GEE0 had mean MAEs andRMSEs of 020 and 075 micromol mminus2 sminus1 whereas RESP0amp GEE0 had mean MAEs and RMSEs of 031 and080 micromol mminus2 sminus1 Likewise MAEs and RMSEs werediminished throughout the snow season with RESPs amp GEE0relative to RESP0 amp GEE0

Modeling subnivean respiration as a function of Tsoilrather than Tair prevented the magnitude of respiration frombeing overestimated at the end of the snow season whenTair was consistently warmer than Tsoil (figures 3 and 5)This is consistent with previous findings that freeze-thawtemperature fluctuations accompanying snow melt do notsubstantially influence effluxes of CO2 (Grogan et al 2004Buckeridge et al 2010) Calculating subnivean respiration as afunction of Tsoil allowed VPRM to simulate both the gradualsteady increase in respiration accompanying snowmelt andmidwinter declines in soil respiration (Bokhorst et al 2010)

However when respiration was modeled year-round asa function of Tsoil (RESPTsoil) the RMSE and MAE errorswere substantially larger than the errors in either RESP0 orRESPs (table 3) The errors in RESPTsoil were largest whenSCA waslt50 As snow cover diminished the contributionsof aboveground biomass to respiration increased accordinglyThe rate of respiration by vegetation is driven primarily by Tairrather than Tsoil and is therefore best modeled as a functionof Tair As a result unreliable estimates of growing seasonNEE would likely be generated using RESPTsoil Thereforealthough estimates of subnivean respiration were improvedwhen subnivean respiration was calculated as a function ofTsoil uncertainty in VPRM estimates of NEE at the startand end of the growing seasons were reduced by calculatingrespiration as a function of Tair (figure 6) These reductionsin model uncertainty are important because the cold seasonis so long in the Arctic that even small biases in daily

6


Table 2 Uncertainty in 1 Mayndash30 June (MJ) and snow season (SS) estimates of NEE by VPRM both with (s) and without (0)representations of the influences of snow on respiration and GEE Results are indicated for the Daring Lake MT calibration site as well asthe Ivotuk validation site for years 2004ndash2007 Mean absolute error (MAE) values are indicated first followed by root mean squared error(RMSE) values in brackets All error rates were calculated by comparing daily average eddy covariance NEE to daily average model NEE

Time Site VPRM form 2004 2005 2006 2007

MJ DL RESP0 amp GEE0 0320 (0366) 0408 (0587) 0415 (0742) 0212 (0405)MJ DL RESP0 amp GEEs 0356 (0417) 0594 (0850) 0422 (0748) 0566 (0727)MJ DL RESPs amp GEE0 0223 (0278) 0122 (0392) 0404 (0737) 0083 (0501)MJ DL RESPs amp GEEs 0240 (0288) 0303 (0428) 0413 (0740) 0209 (0354)

MJ IV RESP0 amp GEE0 0201 (0968) 0194 (0627) 0517 (0826) 0228 (1845)MJ IV RESP0 amp GEEs 0247 (1042) 0315 (0623) 0502 (0824) 0103 (1658)MJ IV RESPs amp GEE0 0085 (0913) 0045 (0589) 0415 (0741) 0194 (1832)MJ IV RESPs amp GEEs 0141 (0935) 0137 (0595) 0417 (0743) 0195 (1832)

SS DL RESP0 amp GEE0 0475 (0672) 0307 (0791) 0283 (0603) 0295 (0597)SS DL RESP0 amp GEEs 0517 (0723) 0341 (0837) 0350 (0618) 0364 (0717)SS DL RESPs amp GEE0 0425 (0649) 0251 (0767) 0258 (0610) 0206 (0613)SS DL RESPs amp GEEs 0433 (0652) 0286 (0771) 0260 (0611) 0269 (0591)

SS IV RESP0 amp GEE0 0270 (0722) 0063 (1163) 0244 (0789) 0045 (1840)SS IV RESP0 amp GEEs 0207 (0775) 0269 (137) 0229 (0832) 0139 (1801)SS IV RESPs amp GEE0 0179 (0656) 0185 (1198) 0178 (0808) 0055 (1837)SS IV RESPs amp GEEs 0193 (0664) 0175 (1199) 0179 (0809) 0055 (1836)

Table 3 Uncertainty in 1 Mayndash30 June (MJ) and snow season (SS) estimates of NEE by VPRM at Daring Lake and Ivotuk usingformulations through which respiration is calculated year-round as a linear or piecewise linear (pwl) function of Tsoil Mean absolute error(MAE) values are indicated first followed by root mean squared error (RMSE) values in brackets MAE and RMSE were calculated fromdaily average observations


MJ DL RESPTsoillinear 0662 (1042) 0558 (0830) 0806(1126) 0583 (0800)MJ DL RESPTsoilpwl 0623 (0991) 0520 (0754) 0736(1021) 0532 (0732)

MJ IV RESPTsoillinear 1161 (1753) 1550 (2310) 0619(0874) 1510 (2073)MJ IV RESPTsoilpwl 1110 (1714) 1536 (2293) 0598(0864) 1512 (2076)

SS DL RESPTsoillinear 0329 (0476) 0650 (0883) 0367(0532) 0457 (0659)SS DL RESPTsoilpwl 0309 (0475) 0605 (0820) 0371(0528) 0423 (0609)

SS IV RESPTsoillinear 0438 (0717) 0802 (1393) 0527(0819) 1390 (2000)SS IV RESPTsoilpwl 0422 (0718) 0785 (1378) 0512(0809) 1379 (1993)

Table 4 Estimates of annual net carbon exchange (gC mminus2 yrminus1)by different VPRM formulations

Site VPRM form 2004 2005 2006 2007

DL RESP0 amp GEE0 008 161 427 minus086DL RESPs amp GEE0 minus352 minus237 057 minus349

IV RESP0 amp GEE0 537 522 419 489IV RESPs amp GEE0 136 149 063 103

average NEE can have a substantial effect on annual estimatesof net carbon exchange Comparisons of the cumulativecarbon exchange for each site indicated that RESP0 amp GEE0consistently estimated greater quantities of net carbon effluxthan RESPs amp GEE0 (table 4 figure 7)

Respiration was best simulated according to Tair duringthe growing season because diurnal variability in Tairinfluences the magnitude and timing of both photosynthesisand respiration Conversely the low thermal conductivityof an overlying snowpack decouples Tsoil and Tair andrespiration persists throughout the snow season according

to Tsoil Due to the rapidity of this transition and the factthat spring snowmelt usually occurs only once per year inArctic regions the use of a 50 threshold did not introduceany discontinuities in estimates of NEE Differentiating thesnow and growing seasons according to SCA and simulatingrespiration as a function of the dominant temperature for eachseason therefore allowed for reduced uncertainty in estimatesof NEE at both sites (2004ndash2007)

4 Conclusions

In Arctic regions the timing of snow onset and melt influencethe rates of photosynthesis and respiration The importanceof snow cover transitions for NEE suggests that insights intothe northern carbon cycle and its response to changing snowconditions may be gained by representing the influence ofsnow on NEE The feasibility of incorporating remote sensingobservations of snow into models of NEE was demonstratedby findings showing (1) good agreement between time-lapsecamera (lt10 m) and remote sensing estimates of SCA from

7


Figure 5 NEE from 1 May to 7 June of years 2004ndash2007 at Daring Lake MT (left) and Ivotuk (right) as observed using the eddycovariance technique (black) and as estimated by the RESP0 amp GEE0 (orange) and RESPs amp GEE0 (blue) Within each plot the date whereestimates from the two models appear to merge represents the day at which SCA initially decreases below 50

Landsat (30 m) and MODIS (500 m) and (2) associationsbetween in situ NEE and SCA at Daring Lake NWT(MayndashJune 2010)

Uncertainty in VPRM estimates of NEE at two lowArctic sites was reduced by representing the decouplingeffects of a snowpack on Tsoil and Tair Estimating subniveanrespiration as a function of Tsoil prevented respiration frombeing overestimated when it was limited by cool Tsoil atthe startend of the snow season and enabled variabilityin cold season NEE to be simulated The timing andmagnitude of photosynthesis at the start and end of thesnow season were best captured by GEE0 which usedan implicit approach to simulate the influences of coldtemperature senescent vegetation and diminished sunlight onhindering photosynthesis The resulting VPRM formulationcontaining an implicit representation of the effects of SCA on

photosynthesis and an explicit representation of the influenceof SCA on respiration had diminished RMSEs and MAEsacross both sites and all years

Climate change is predicted to both increase Arcticsnow accumulation and diminish the length of the Arcticsnow season (AMAP 2011) Previous studies at DaringLake have found that natural inter-annual variability in snowmelt timing did not markedly affect early or total growingseason CO2 flux (Humphreys and Lafleur 2011) but thatplots with artificially increased snow depth and durationshowed altered CO2 fluxes upon snow melt (Buckeridgeand Grogan 2010) Incorporating satellite observations ofSCA into biospheric carbon flux models could thereforeallow the snow and growing seasons to be delineated enablesnow season influences on respiration to be representedand permit reduced uncertainty in estimates of the Arctic

8


Figure 6 Mean absolute error of all model formulations at Ivotukfor the snow seasons of years 2004 (dark gray) to 2007 (light gray)

Figure 7 Net carbon (tC haminus1) predicted in years 2004 (dark gray)to 2007 (light gray) by RESP0 amp GEE0 and RESPs amp GEE0

carbon cycle Insights could therefore be gained into theregional scale response of the Arctic carbon cycle to alteredbiological meteorological and cryospheric conditions Futurework will consist of applying the snow season formulationto a variety of sub-Arctic low Arctic and high Arctic sitesin order to determine if model error may be reduced in allcases

Acknowledgments

We thank Michael Treberg for assistance in setting upMoultrie I-65 cameras and gathering in situ observationsof NEE and snow cover Logistical assistance from Steve

Matthews at Daring Lake was greatly appreciated We alsowish to thank all researchers and field assistants involvedin the collection of eddy covariance observations at Ivotukand Daring Lake Funding from NSERC through a DiscoveryGrant and a Vanier Canada Graduate Scholarship is gratefullyacknowledged The eddy covariance data obtained fromthe Daring Lake sites was supported by the CanadianFoundation for Climate and Atmospheric Sciences NaturalScience and Engineering Council of Canada InternationalPolar YearmdashGovernment of Canada Polar Continental ShelfProject and Canada Foundation for Innovation and OntarioInnovation Trust

References

AMAP 2011 Snow Water Ice and Permafrost in the Arctic (SWIPA)(Oslo Arctic Monitoring and Assessment Programme)

Amiro B 2010 Estimating annual carbon dioxide eddy fluxes usingopen-path analyzers for cold forest sites Agric ForestMeteorol 150 1366ndash72

Billings W and Mooney H 1968 The ecology of Arctic and Alpineplants Biol Rev 43 481ndash529

Bokhorst S et al 2010 Impacts of extreme winter warming events onlitter decomposition in a sub-Arctic heathland Soil BiolBiochem 42 611ndash7

Bonan G 2002 Ecological Climatology Concepts and Applications(New York Cambridge University Press)

Buckeridge K and Grogan P 2010 Deepened snow increases latethaw biogeochemical pulses in mesic low Arctic tundraBiogeochemistry 101 105ndash21

Buckeridge K et al 2010 Soil biogeochemistry during the earlyspring in low Arctic mesic tundra and the impacts of deepenedsnow and enhanced nitrogen availability Biogeochemistry99 127ndash41

Carstairs A and Oechel W 1978 Effects of several microclimaticfactors and nutrients on net carbon dioxide exchange inCladonia AlpesTris (L) Rabh in the sub-Arctic Arct Alp Res10 81ndash94

Derksen C et al 2009 Northwest territories and nunavut snowcharacteristics from a subarctic traverse implications forpassive microwave remote sensing J Hydrometeorol10 448ndash63

Dozier J 1989 Spectral signature of Alpine snow cover from theLandsat thematic mapper Remote Sens Environ 28 9ndash22

Elberling B et al 2008 High-Arctic soil CO2 and CH4 productioncontrolled by temperature water freezing and snow Adv EcolRes 40 441ndash72

Euskirchen E et al 2012 Seasonal patterns of carbon dioxide andwater fluxes in three representative tundra ecosystems innorthern Alaska Ecosphere 3 4

Gouttevin I et al 2012 How the insulating properties of snow affectsoil carbon distribution in the continental pan-Arctic areaJ Geophys ResmdashBiogeosci 117 G02020

Grogan P et al 2004 Freeze-thaw regime effects on carbon andnitrogen dynamics in sub-Arctic heath tundra mesocosms SoilBiol Biochem 36 641ndash54

Hall D and Riggs G 2007 Accuracy assessment of MODIS snowproducts Hydrol Process 21 1534ndash47

Hall D Riggs G and Salomonson V 1995 Development of methodsfor mapping global snow cover using moderate resolutionimaging spectroradiometer data Remote Sens Environ54 127ndash40

Hall D Riggs G Salomonson V DiGirolamo N and Bayr K 2002MODIS snow-cover products Remote Sens Environ83 181ndash94

9


Humphreys E and Lafleur P 2011 Does earlier snowmelt lead togreater CO2 sequestration in two low Arctic tundraecosystems Geophys Res Lett 38 L09703

Lafleur P and Humphreys E 2008 Spring warming and carbondioxide exchange over low Arctic tundra in central CanadaGlob Change Biol 14 740ndash56

Larsen K et al 2007 Respiration and microbial dynamics in twosubarctic ecosystems during winter and spring thaw effects ofincreased snow depth Arct Antarct Alp Res 39 268ndash76

Laskowski C 2010 Seasonal annual and interannual carbondynamics of a remote tussock tundra ecosystem in IvotukAlaska PhD Thesis University of California Davis and SanDiego State University

Lovett G Cole J and Pace M 2006 Is net ecosystem productionequal to ecosystem carbon accumulation Ecosystems 9 152ndash5

Luus K Lin J Kelly R and Duguay C 2013 Subnivean Arctic ampsub-Arctic net ecosystem exchange (NEE) towardsrepresenting snow season processes in models of NEE usingcryospheric remote sensing Prog Phys Geogr 37 483ndash514

Mahadevan P et al 2008 A satellite-based biosphere parametrizationfor net ecosystem CO2 exchange Vegetation Photosynthesisand Respiration Model (VPRM) Glob Biogeochem Cycles22 GB2005

Mesinger F et al 2006 North American regional reanalysis Bull AmMeteorol Soc 87 343ndash60

McGuire A et al 2000 Modeling the effects of snowpack onheterotrophic respiration across northern temperate and highlatitude regions comparison with measurements ofatmospheric carbon dioxide in high latitudes Biogeochemistry48 91ndash114

Mikan C Schimel J and Doyle A 2002 Temperature controls ofmicrobial respiration in Arctic tundra soils above and belowfreezing Soil Biol Biochem 34 1785ndash95

Morgner E et al 2010 The importance of winter in annual ecosystemrespiration in the high Arctic effects of snow depth in twovegetation types Polar Res 29 58ndash74

Nobrega S and Grogan P 2007 Deeper snow enhances winterrespiration from both plant-associated and bulk soil carbonpools in birch hummock tundra Ecosystems 10 419ndash31

Oelbermann M English M and Schiff S 2008 Evaluating carbondynamics and microbial activity in Arctic soils under warmertemperatures Can J Soil Sci 88 31ndash44

Olsson P et al 2003 Five stages of the Alaskan Arctic cold seasonwith ecosystem implications Arct Antarct Alp Res 35 74ndash81

Oquist G and Huner N 2003 Photosynthesis of overwinteringevergreen plants Annu Rev Plant Biol 54 329ndash55

Rees A et al 2010 Observed and modelled effects of ice lensformation on passive microwave brightness temperatures oversnow covered tundra Remote Sens Environ 114 116ndash26

Riggs G and Hall D 2011 MODIS snow and ice products and theirassessment and applications Land Remote Sensing and GlobalEnvironmental Change Remote Sensing and Digital ImageProcessing vol 11 ed B Ramachandran C Justice andM Abrams (New York Springer) pp 681ndash707

Rosenthal W and Dozier J 1996 Automated mapping of montanesnow cover at subpixel resolution from the landsat thematicmapper Water Resour Res 32 115ndash30

Starr G and Oberbauer S 2003 Photosynthesis of Arctic evergreensunder snow implications for tundra ecosystem carbon balanceEcology 84 1415ndash20

Sullivan P et al 2008 Continuous estimates of CO2 efflux fromArctic and boreal soils during the snow-covered season inAlaska J Geophys Res G 113 G04009

Thompson C McGuire A Clein J Chapin F and Beringer J 2006Net carbon exchange across the Arctic tundrandashboreal foresttransition in Alaska 1981ndash2000 Mitig Adapt Strateg GlobChange 11 805ndash27

Tieszen L L 1974 Photosynthetic competence of the subniveanvegetation of an Arctic tundra Arct Alp Res 6 253ndash6

Wania R Ross I and Prentice I C 2009 Integrating peatlands andpermafrost into a dynamic global vegetation model 1Evaluation and sensitivity of physical land surface processesGlob Biogeochem Cycles 23 GB3014

Warren S 1982 Optical properties of snow Rev Geophys 20 67ndash89Willmott C and Matsuura K 2005 Advantages of the mean absolute

error (MAE) over the root mean square error (RMSE) inassessing average model performance Clim Res 30 79

Zhang T Osterkamp T and Stamnes K 1996 Influence of the depthhoar layer of the seasonal snow cover on the ground thermalregime Water Resour Res 32 2075ndash86

Zhou X Li S and Stamnes K 2003 Effects of vertical inhomogeneityon snow spectral albedo and its implication for optical remotesensing of snow J Geophys ResmdashAtmos 108 4378

Zimov S et al 1993 Wintertime CO2 emission from soils ofNortheastern Siberia Arctic 46 197ndash204

Zimov S et al 1996 Siberian CO2 efflux in winter as a CO2 sourceand cause of seasonality in atmospheric CO2 Clim Change35 111ndash20

10

IOP PUBLISHING ENVIRONMENTAL RESEARCH LETTERS

Environ Res Lett 8 (2013) 035045 (10pp) doi1010881748-932683035045

Modeling the influence of snow cover onlow Arctic net ecosystem exchange

K A Luus12 R E J Kelly1 J C Lin23 E R Humphreys4 P M Lafleur5 andW C Oechel6

1 Geography University of Waterloo ON N2L 3C5 Canada2 Earth and Environmental Sciences University of Waterloo ON N2L 3C5 Canada3 Atmospheric Sciences University of Utah UT 84112 USA4 Geography and Environmental Studies Carleton University Ottawa ON K1S 5B6 Canada5 Geography Trent University Peterborough ON K9J 7B8 Canada6 Biology San Diego State University CA 95616 USA

E-mail kaluusuwaterlooca

Received 27 March 2013Accepted for publication 22 August 2013Published 11 September 2013Online at stacksioporgERL8035045

AbstractThe Arctic net ecosystem exchange (NEE) of CO2 between the land surface and theatmosphere is influenced by the timing of snow onset and melt The objective of this study wasto examine whether uncertainty in model estimates of NEE could be reduced by representingthe influence of snow on NEE using remote sensing observations of snow cover area (SCA)Observations of NEE and time-lapse images of SCA were collected over four locations at alow Arctic site (Daring Lake NWT) in MayndashJune 2010 Analysis of these observationsindicated that SCA influences NEE and that good agreement exists between SCA derivedfrom time-lapse images Landsat and MODIS MODIS SCA was therefore incorporated intothe vegetation photosynthesis respiration model (VPRM) VPRM was calibrated usingobservations collected in 2005 at Daring Lake Estimates of NEE were then generated overDaring Lake and Ivotuk Alaska (2004ndash2007) using VPRM formulations with and withoutexplicit representations of the influence of SCA on respiration andor photosynthesis Modelperformance was assessed by comparing VPRM output against unfilled eddy covarianceobservations from Daring Lake and Ivotuk (2004ndash2007) The uncertainty in VPRM estimatesof NEE was reduced when respiration was estimated as a function of air temperature whenSCA le 50 and as a function of soil temperature when SCA gt 50

Keywords net ecosystem exchange Arctic snow remote sensing modeling CO2cryosphere

1 Introduction

In low Arctic regions the initial onset and final melt ofsnow mark important transitions in net ecosystem exchange(NEE) (Olsson et al 2003 Grogan et al 2004 Bokhorst et al2010 Buckeridge and Grogan 2010) where NEE is defined

Content from this work may be used under the terms ofthe Creative Commons Attribution 30 licence Any further

distribution of this work must maintain attribution to the author(s) and thetitle of the work journal citation and DOI

as the net biospheric flux of CO2 into and out of the landsurface (Lovett et al 2006) NEE can be described as thesum of photosynthetic uptake by vegetation (GEE or grossecosystem exchange) and ecosystem respiration (R) NEE =minusGEE + R According to the sign convention used in thisstudy uptake of CO2 from the atmosphere is represented asnegative NEE and release of CO2 into the atmosphere isshown as positive NEE

Photosynthetic uptake by vegetation is maximized duringthe growing season when above-freezing air temperatures(Tair) and sunny conditions support plant growth During snow

11748-932613035045+10$3300 ccopy 2013 IOP Publishing Ltd Printed in the UK









2 Methodology

21 Study sites


2










3




rGREEN + rMIDIR (1)





1+ PARPAR0

PAR (2)





(i) RESP0 amp GEE0

GEE = GEE0

R =



(ii) RESP0 amp GEEs


R =




GEE = GEE0

R =



(iv) RESPs amp GEEs


R =




GEE = GEE0


4




GEE = GEE0

R =




MAE = nminus1nsum

i=1

|εi| (3)

RMSE =

[nminus1

nsumi=1

|εi|2

] 12

(4)







FN LK MT SB









5










6















Site VPRM form 2004 2005 2006 2007






4 Conclusions


7







8





Acknowledgments



References


















9































10









2 Methodology

21 Study sites


2










3




rGREEN + rMIDIR (1)





1+ PARPAR0

PAR (2)





(i) RESP0 amp GEE0

GEE = GEE0

R =



(ii) RESP0 amp GEEs


R =




GEE = GEE0

R =



(iv) RESPs amp GEEs


R =




GEE = GEE0


4




GEE = GEE0

R =




MAE = nminus1nsum

i=1

|εi| (3)

RMSE =

[nminus1

nsumi=1

|εi|2

] 12

(4)







FN LK MT SB









5










6















Site VPRM form 2004 2005 2006 2007






4 Conclusions


7







8





Acknowledgments



References


















9































10










3




rGREEN + rMIDIR (1)





1+ PARPAR0

PAR (2)





(i) RESP0 amp GEE0

GEE = GEE0

R =



(ii) RESP0 amp GEEs


R =




GEE = GEE0

R =



(iv) RESPs amp GEEs


R =




GEE = GEE0


4




GEE = GEE0

R =




MAE = nminus1nsum

i=1

|εi| (3)

RMSE =

[nminus1

nsumi=1

|εi|2

] 12

(4)







FN LK MT SB









5










6















Site VPRM form 2004 2005 2006 2007






4 Conclusions


7







8





Acknowledgments



References


















9































10




rGREEN + rMIDIR (1)





1+ PARPAR0

PAR (2)





(i) RESP0 amp GEE0

GEE = GEE0

R =



(ii) RESP0 amp GEEs


R =




GEE = GEE0

R =



(iv) RESPs amp GEEs


R =




GEE = GEE0


4




GEE = GEE0

R =




MAE = nminus1nsum

i=1

|εi| (3)

RMSE =

[nminus1

nsumi=1

|εi|2

] 12

(4)







FN LK MT SB









5










6















Site VPRM form 2004 2005 2006 2007






4 Conclusions


7







8





Acknowledgments



References


















9































10




GEE = GEE0

R =




MAE = nminus1nsum

i=1

|εi| (3)

RMSE =

[nminus1

nsumi=1

|εi|2

] 12

(4)







FN LK MT SB









5










6















Site VPRM form 2004 2005 2006 2007






4 Conclusions


7







8





Acknowledgments



References


















9































10










6















Site VPRM form 2004 2005 2006 2007






4 Conclusions


7







8





Acknowledgments



References


















9































10















Site VPRM form 2004 2005 2006 2007






4 Conclusions


7







8





Acknowledgments



References


















9































10







8





Acknowledgments



References


















9































10





Acknowledgments



References


















9































10































10

Modeling the influence of snow cover on low Arctic net ecosystem exchange

Documents

Transcript of Modeling the influence of snow cover on low Arctic net ecosystem exchange