Uncertainty Mapping of Upscaled Net Ecosystem Production in Grasslands of the Great Plains

Flux tower networks, such as AmeriFlux and FLUXNET, consist of a growing number of eddy covariance flux tower sites that provide a synoptic record of the exchange of carbon, water, and energy between the ecosystem and atmosphere at a 30 minute
of 8
All materials on our website are shared by users. If you have any questions about copyright issues, please report us to resolve them. We are always happy to assist you.
Related Documents
  REPORT Mapping carbon flux uncertainty and selecting optimallocations for future flux towers in the Great Plains Yingxin Gu  • Daniel M. Howard  • Bruce K. Wylie  • Li Zhang Received: 22 August 2011/Accepted: 14 December 2011/Published online: 28 December 2011   Springer Science+Business Media B.V. 2011 Abstract  Flux tower networks (e.g., AmeriFlux,Agriflux)providecontinuousobservationsofecosystemexchanges of carbon (e.g., net ecosystem exchange),water vapor (e.g., evapotranspiration), and energybetween terrestrial ecosystems and the atmosphere.Thelong-termtimeseriesoffluxtowerdataareessentialforstudyingandunderstandingterrestrialcarboncycles,ecosystem services, and climate changes. Currently,there are 13 flux towers located within the Great Plains(GP). The towers are sparsely distributed and do notadequately represent the varieties of vegetation covertypes,climateconditions,andgeophysicalandbiophys-ical conditions in the GP. This study assessed how wellthe available flux towers represent the environmentalconditionsor‘‘ecologicalenvelopes’’acrosstheGPandidentified optimal locations for future flux towers in theGP. Regression-based remote sensing and weather-drivennet ecosystemproduction (NEP) models derivedfrom different extrapolation ranges (10 and 50%) wereused to identify areas where ecological conditions werepoorly represented by the flux tower sites and yearspreviously used for mapping grassland fluxes. Theoptimal lands suitable for future flux towers within theGP were mapped. Results from this study provideinformation to optimize the usefulness of future fluxtowersintheGPandserveasaproxyfortheuncertaintyof the NEP map. Keywords  Net ecosystem production (NEP)   Regression tree models    Extrapolation   Mean absolute difference    Flux tower location   Great Plains Introduction Flux tower networks (e.g., AmeriFlux, Agriflux) haveprovided continuous observations of ecosystemexchanges of carbon (e.g., net ecosystem exchange),water vapor (e.g., evapotranspiration), and energybetween terrestrial ecosystems and the atmospheresince 1996 (Baldocchi et al. 2001; Law 2007; http:// ; (1) to provide ground truth information to validate Y. Gu ( & )ASRC Research & Technology Solutions, Contractorto US Geological Survey (USGS) Earth ResourcesObservation and Science (EROS) Center, 47914 252ndStreet, Sioux Falls, SD 57198, USAe-mail: ygu@usgs.govD. M. HowardStinger Ghaffarian Technologies, Inc., Contractorto USGS EROS, Sioux Falls, SD 57198, USAB. K. WylieUSGS EROS, Sioux Falls, SD 57198, USAL. ZhangKey Laboratory of Digital Earth, Center for EarthObservation and Digital Earth, Chinese Academyof Sciences, Beijing, China  1 3 Landscape Ecol (2012) 27:319–326DOI 10.1007/s10980-011-9699-7  and verify remote-sensing-derived products (e.g., netprimary productivity, evaporation, and energy absorp-tion) (Running et al. 1999; Sims et al. 2005; http://, (2) todevelop, test, and apply terrestrial ecosystem modelsand atmospheric models (Running et al. 1999; Baldoc-chi et al. 2001; Law 2006; Jung et al. 2009; http://;and (3) to integrate with remotely sensed data to quan-tify gross primary production and net ecosystem pro-duction (NEP) over large areas with high spatial andtemporalresolutions(Yangetal.2007;Xiaoetal.2010; Zhang et al. 2011). As a result, these long-term timeseries of flux tower data are essential for studying andunderstanding terrestrial carbon cycles, ecosystem ser-vices, and climate changes (Running et al. 1999; Bal-docchi et al. 2001; Gilmanov et al. 2003; Law 2006; Reichstein et al. 2007; Wylie et al. 2007; Xiao et al. 2008; Jung et al. 2009). The Great Plains (GP), which covers 14 states andincludes 17 ecoregions as defined by Omernik’s levelIII Ecoregions (Omernik  1987), includes a variety of vegetation cover types and a broad range of climateconditions and plant productivities (Tieszen et al.1997). 13 grassland flux towers from Ameriflux,Agriflux networks, and non-affiliated tower siteslocated within the GP were used for developing NEPmapping algorithms (Zhang et al. 2011; Fig. 1). These flux towers are sparsely distributed throughout theregion and probably do not provide an adequaterepresentation of certain vegetation cover types,climate conditions, or geophysical and biophysicalconditions (e.g., soil conditions, vegetation phenol-ogy) of the GP. Strategic augmentation of the fluxtower networks will help capture a broader range of climate and biome conditions in the GP and makefuture carbon flux mapping more accurate and robust.In a previous study, Hargrove et al. (2003) used an‘‘unsupervised clustering’’ approach, which was pri-marily based on climatic and physiographic drivers, toidentify southern Texas as underrepresented by theAmeriFlux Network. In this study, we developed anew approach that focuses on the model input datagaps by using flux tower data, remote sensing,geophysical and biophysical data, and weather vari-ables. The main objective of this study is to assess the Fig. 1  Namesandlocationsof the 13 flux towers withintheGP. Theland cover typesas identified in the 2001NLCD are shown320 Landscape Ecol (2012) 27:319–326  1 3  uncertainty of the NEP maps developed by Zhanget al. (2011) and to identify optimal locations forfuture flux towers in the GP. Remote sensing andweather-driven NEP models were developed usingtwo different extrapolation strategies (labeled 10 and50% extrapolation) to assess areas and times whencurrentfluxtower datawell(orpoorly)representedtheenvironmental and climate conditions across the GPgrasslands. Results from this study provide usefulinformation for ensuring future flux tower locations inthe GP effectively complement current and historicalgrassland flux towers and provide a proxy for theuncertainty of the NEP maps produced by Zhang et al.(2011). Study area Our study area is the GP, where the main vegetationcover types are grassland and cultivated crops. Otherland cover types include shrubs, evergreen anddeciduous forests, urban, wetlands, open water, andbarren lands. Figure 1 is the 2001 National LandCover Database(NLCD2001)map forthe GP(Homeretal.2004).The average annual temperature generallyincreases from the Northern GP (less than 4  C) to theSouthern GP (exceeds 22  C) and the annual precip-itation increases from the Western GP (less than200 mm) to the Eastern GP (over 1100 mm) (Joyceet al. 2001). The soil available water capacity (AWC),which represents important soil and agronomic char-acteristics and was derived from the State SoilGeographic (STATSGO) database, is shown inFig. 2. The AWC values vary from 0 to 0.64 withinthe GP. Examples of phenological metrics (i.e., long-term median start of growing season time and long-term average of growing season integrated normalizeddifference vegetation index, NDVI), which representthe biophysical characteristics of the vegetation, arealso shown in Fig. 2. These phenological metrics werederived from 2000 to 2009 eMODIS (expeditedmoderate resolution imaging spectroradiometer)(Jenkerson et al. 2010) data and were based on Reedet al. (1994) method. Data and methods Method for building GP NEP modelsThe data-driven piecewise regression NEP modelswere developed from multiple flux tower observations(the gap-filled NEP data), satellite-derived NDVI,phenological metrics, precipitation and temperature,photosynthetically active radiation, and AWC (Zhanget al. 2011). The NEP gap-filling algorithms used the30 min step data and light response curve analysis aswell as relationships with flux tower ‘‘slow data’’ Fig. 2  Soil and phenological conditions in grassland areas of the GP.  a  available water capacity;  b  long-term median SOST; and c  long-term averaged time integrated NDVILandscape Ecol (2012) 27:319–326 321  1 3  (atmospheric andsoilvariables)tofillshort gapsinthecarbon flux estimates (Gilmanov et al. 2005). TheseNEP models were used to map weekly NEP for the GPgrasslands but with varying degrees of extrapolationsallowed. Zhang et al. (2011) constrained the NEP prediction to 10% beyond the training data range. Inthis study, a second map was created with predictionconstrained to 50% beyond the training data range.The piecewise regression approach divides themulti-dimensional domain of environmental variablesinto many segments and derives a multiple regressionequation for each segment. The training data for themodel (sets of environmental data for the flux towersite-year combinations) are then modeled, and theresulting extreme values (NEP max  and NEP min ) arerecorded for each segment. Two extrapolated rangesarethendefinedforeachsegment:a10%extrapolation(by adding 10% of the range to the NEP max  andsubtracting 10% from the NEP min ) and a 50% range(by adding 50% to the NEP max  and subtracting 50%from the NEP min ). Two maps (10 and 50% extrapo-lations) were generated by applying the piecewiseregressions to the full set of data inputs. If the NEP fora pixel was beyond the limited extrapolation range, itwas reset to the limit of the extrapolation range.Identify optimal locations for future flux towersin the GPModel interpolation occurswhenpredictionsare madewithin the range of the training data that are used tocreate the model (Fig. 3, blue box). Modelextrapolation, which is inherently less reliable thaninterpolation, occurs when model predictions extendbeyond the training data range (Fig. 3, red and yellowareas). Theoretically, a pixel that incorporates envi-ronmental input variables (e.g., NDVI, weather, soils,and phenology) that conform to the characteristicsobserved in the flux tower training data would beassigned a similar NEP value in both the 10 and 50%extrapolationmodels(blueandyellowareasinFig. 3).On the other hand, the environmental variables of apixel that are poorly represented by any flux tower inthe network would likely have different NEP valuescomputed using the 10 and 50% extrapolation models(red areas in Fig. 3). In a sense, the degree of difference between the two NEP (10 and 50%extrapolation) for a given pixel (the Y-axis inFig. 3) are used to identify the degree to which theenvironmental conditions of the pixel (the X-axis inFig. 3) are different than the training data. If a lot of pixels in an area are different than the training data,that is an indication that a new flux tower in that areawould help explain and model that set of environ-mental conditions.The absolute differences between the two NEP datasets (10 and 50% extrapolation limits) were calculatedfor weekly NEP estimates from 2000 to 2008. Theweekly absolute differences were averaged to yearlytime steps, referred to as mean absolute difference(MAD), to quantify interannual variations in modelextrapolation.YearlyMADvalueswerethenaveragedto a multi-year (2000–2008) MAD (MMAD). Areaswith high MMAD imply that these regions are poorly Fig. 3  A hypotheticalexample of quantifying thedegree of extrapolation inthe NEP model. ModeledNEP will be on the heavy blue line  if environmentalconditions are interpolatedwithin the domain of thetraining data, on the heavy  yellow line  if the 10%constraint is used, or theheavy  red line  if the 50%constraint is used322 Landscape Ecol (2012) 27:319–326  1 3  represented by conditions at flux tower sites in theyears used by Zhang et al. (2011). The high MMAD areas have a lower prediction reliability of NEP andwouldbeconsideredascandidatesfornewfluxtowers.Unique environmental grassland conditions mayoccur infrequently or sporadically in areas with highlyvariable annual MAD. Therefore, we excluded areaswhere the interannual coefficient of variation (CV) forMMAD was high (i.e., CV of the 9-year MAD isgreater than 20%). This will ensure the proposed fluxtower locations have consistently different environ-mental conditions than existing flux towers, not justfrom one or two outlier years. Results and discussion Figure 4 is the 2000–2008 MMAD map and shows the13 flux towers in the GP. Areas with high MMADimply that the remote sensing, soil, climate, andweather conditions are poorly represented by the fluxtowers sites and years used by Zhang et al. (2011). These high MMAD areas would be suitable forpotential new flux towers. High MMAD areas (i.e.,MMAD [ 0.75 g C m - 2 day - 1 ) are mainly locatedin the following three regions (within the red ovals inFig. 4): (1) western Kansas (within the Central GP);(2) central Texas (mainly located within the South-western Tablelands, Central GP, Central Oklahoma/ Texas plains); and (3) southern Texas (mainly in theSouthern Texas Plains and Western Gulf CoastalPlains). Areas with high interannual CV of MMAD(black in Fig. 4), mainly located in the middle andwestern parts of the GP, are not recommended forfuture new flux towers because of the variableenvironmental conditions.The optimal locations (within the red ovals inFig. 4) derived from this study represent a variety of climate, geophysical, and biophysical conditions. Forexample,thereisonlyonefluxtower(FreemanRanch,TX) located in the southern part of the GP. The soilcondition (AWC) in the southern part of the GP variesfrom 0.18 to  * 0.48 cm/cm. The growing seasonintegrated NDVI (GSN), which represents a proxy of ecosystem productivity (Tieszen et al. 1997), variesfrom7to * 30.Thestartofseasontime(SOST),whichisanimportant variable forvegetation phenology, alsovaries from Julian day 40 (early February) to * Julianday 95 (early April). Therefore, it is necessary to buildnew fluxtowers tobetter represent the vegetation,soil,and climate conditions of this southern part of the GP.This optimal area is also in good agreement with theprevious research results from Hargrove et al. (2003),whoidentifiedtheSouthernGPasanareathatispoorlyrepresented by flux towers.In contrast, although there is only one flux towerlocated within the Nebraska Sandhills ecoregion(Gudmundsen Ranch site, a grassland site that is notin an upland part of the Sandhills), the MMAD is verylow ( \ 0.5 g C m - 2 day - 1 ) in this ecoregion. Ourstudy suggests that it would not be a priority to buildnewfluxtowersintheSandhillsecoregioneventhoughit contains a very large area and there is only one fluxtower. The unique dry climate and biophysical condi-tions (e.g., sandy soil with low AWC, similar SOSTfrom early March to early April) suggest that theenvironmental conditions are similar in the Sandhillsecoregion, and an additional flux tower would not benecessary. This demonstrates that our approach candistinguish areas that are poorly represented by thecurrent flux tower network from those that are wellrepresented in terms of climate conditions (e.g.,precipitation and temperature), soil conditions, andvegetation phenology. In this study, we used a coarseresolutionAWCmapderivedfromtheSTATSGOdataset in order to be consistent with data sets used byZhang et al. (2011). This coarse AWC map does not capture the fine scale differences; this could beimproved by using the Soil Survey Geographic(SSURGO) AWC data.We also found that there was a dramatic MMADdifference, occurring over short distances, in thesoutheastern edge of the Sandhills and the northernportion of the Central GP ecoregion (Figs. 2a, 4). The environmental conditions of this region include anearlier SOST and a higher productivity (as proxied byGSN) than the Sandhills ecoregion. This region wouldbe a candidate for a new flux tower.This study provides a general indication of theoptimal locations for the future flux towers at a 250-mspatial resolution based on the climate, geophysical,and biophysical conditions. Subsequent assessmentson the local (high resolution) environmental condi-tions (e.g., micrometeorological conditions necessaryfor the eddy covariance, vegetation homogeneity andlow relief within the fetch area, reasonably accessiblefor maintenance) of the optimal locations are neededbefore building the new flux towers. Landscape Ecol (2012) 27:319–326 323  1 3
Similar documents
View more...
Related Search
We Need Your Support
Thank you for visiting our website and your interest in our free products and services. We are nonprofit website to share and download documents. To the running of this website, we need your help to support us.

Thanks to everyone for your continued support.

No, Thanks

We need your sign to support Project to invent "SMART AND CONTROLLABLE REFLECTIVE BALLOONS" to cover the Sun and Save Our Earth.

More details...

Sign Now!

We are very appreciated for your Prompt Action!