Global estimates of the land–atmosphere water flux based on monthly AVHRR and ISLSCP-II data, v.....IntroductionMethodsModel descriptionData: validation sitesData: global estimatesUncertainty analysisResultsFLUXNET validationUncertainty analysisGlobal analysisDiscussionAcknowledgmentsReferencesGlobal estimates of the land–atmosphere water flux based on monthlyAVHRR and ISLSCP-II data, validated at 16 FLUXNET sitesJoshua B. Fishera,⁎, Kevin P. Tub, Dennis D. BaldocchiaaDepartment of Environmental Science, Policy and Management, University of California at Berkeley, USAbDepartment of Integrative Biology, University of California at Berkeley, USAReceived 1 December 2005; received in revised form 12 June 2007; accepted 30 June 2007AbstractNumerous models of evapotranspiration have been published that range in data-driven complexity, but global estimates require a model that doesnot depend on intensive field measurements. The Priestley–Taylor model is relatively simple, and has proven to be remarkably accurate andtheoretically robust for estimates of potential evapotranspiration. Building on recent advances in ecophysiological theory that allow detection ofmultiple stresses on plant function using biophysical remote sensing metrics, we developed a bio-meteorological approach for translating Priestley–Taylor estimates of potential evapotranspiration into rates of actual evapotranspiration. Five model inputs are required: net radiation ( Rn), normalizeddifference vegetation index (NDVI), soil adjusted vegetation index (SAVI), maximum air temperature (Tmax), and water vapor pressure (ea). Ourmodel requires no calibration, tuning or spin-ups. The model is tested and validated against eddy covariance measurements (FLUXNET) from a widerange of climates and plant functional types—grassland, crop, and deciduous broadleaf, evergreen broadleaf, and evergreen needleleaf forests. Themodel-to-measurement r2was 0.90 (RMS=16 mm/month or 28%) for all 16 FLUXNET sites across 2 years (most recent data release). Globalestimates of evapotranspiration at a temporal resolution of monthly and a spatial resolution of 1° during the years 1986–1993 were determined usingglobally consistent datasets from the International Satellite Land-Surface Climatology Project, Initiative II (ISLSCP-II) and the Advanced Very HighResolution Spectroradiometer (AVHRR). Our model resulted in improved prediction of evapotranspiration across water-limited sites, and showedspatial and temporal differences in evapotranspiration globally, regionally and latitudinally.© 2007 Elsevier Inc. All rights reserved.Keywords: Evapotranspiration; Water flux; FLUXNET; AmeriFlux; Eddy flux; MODIS; International Land-Surface Climatology Project; ISLSCP; ISLSCP-II;Remote sensing; Model; Ecophysiology; Priestly–Taylor; Global1. IntroductionEvapotranspiration (LE) is a major component in theprocesses and models of global climate change, water balance,net primary product ivity, floods, droughts, and irrigation. LE isdifficult to measure and predict, however, especially at largespatial scales (Turner, 1989). Understanding the variability inwater cycle processes requires a spatially detailed analysis ofglobal land surface processes (Running et al., 2000). Closingthe water budget worldwide is of utmost importance to waterand energy cycle research; the overall goal of which is to deliverreliable estimates of precipitation and LE over the whole surfaceof the earth using a combination of measurements and modelestimates (Entekhabi et al., 1999).Global LE estimation in the literature has been marked bya struggle between realistic models that are hindered by com-plex parameterization and simple models that lack mechanisticrealism (Cleugh et al., 2007). The trend has been towardsincreasing complexity, as opposed to applicability (Federeret al., 1996). Yet, greater complexity requires detailed inputparameters that limit application to areas where the necessarydata are available (Federer et al., 2003; Kustas & Norman,1996). Before the widespread ecological application of re-mote sensing data, researchers estimated regional LE with in-terpolated data from thousands of meteorological stations(Baumgartner & Reichel, 1975; Budyko, 1978; Hare, 1980;Morton, 1983).Available online at www.sciencedirect.comRemote Sensing of Environment xx (2007) xxx– xxx+ MODELRSE-07011; No of Pages 19www.elsevier.com/locate/rse⁎Corresponding author.E-mail address: [email protected] (J.B. Fisher).0034-4257/$ - see front matter © 2007 Elsevier Inc. All rights reserved.doi:10.1016/j.rse.2007.06.025ARTICLE IN PRESSPlease cite this article as: Fisher, J. B., et al., Global estimates of the land–atmosphere water flux based on monthly AVHRR and ISLSCP-II data, validated at 16FLUXNET sites, Remote Sensing of Environment (2007), doi:10.1016/j.rse.2007.06.025LE methods – Thornthwaite (1948), Priestley and Taylor(1972), and Monteith (1965) – continue to be used withdifferent theoretical (and subsequent operational) modificationsto generate global patterns of LE (Choudhury, 1997; Choudhury& DiGirolamo, 1998; Choudhury et al., 1998; Cleugh et al.,2007; Gordon et al., 2005; Houborg & Soegaard, 2004; Mintz &Walker, 1993; Nishida et al., 2003; Tateishi & Ahn, 1996). ThePenman–Monteith equation is more theoretically accurate thanare the Priestley–Taylor or Thornthwaite meth ods, butrequires parameters that are difficult to characterize globallysuch as aerodynamic resistance, stomatal resistance, and windspeed. S till, the Penman–Monteith and Priestley–Taylormethods have been shown to give relatively low biases,particularly in comparison with the relativel y poor accuracyof the Tho rnthwaite method (Vörösmarty et al., 1998). Thepotential LE equations, however, must be reduced to actual LEbased on soil moisture (Federer et al., 2003; Maurer et al.,2002). Further constraints by temperature and soil-canopypartitioning may be implemented (McNaughton & Spriggs,1986).Two major datasets are being used to drive and validate globalLE estimates. A global network of eddy covariance towers –FLUXNET – provides measurements of water and energy fluxesover 0.5–5km2across a wide range of ecosystems and climates(Baldocchi et al., 2001). Nishida et al. (2003) validated theirNOAA/AVHRR-driven model of evaporative fraction across 13sites in the AmeriFlux network (r2=0.71).Houborg and Soegaard(2004) validated their MODIS/AVHRR-driven model with fluxmeasurements in Denmark