Relationships between Leaf Area Index andLandsat TM Spectral Vegetation Indicesacross Three Temperate Zone SitesDavid P. Turner,* Warren B. Cohen,†Robert E. Kennedy,*Karin S. Fassnacht,‡and John M. Briggs§Mapping and monitoring of leaf area index (LAI) is ity, additional corrections were made for Sun–surface–sensor geometry. Across all sites, a strong general rela-important for spatially distributed modeling of vegetationproductivity, evapotranspiration, and surface energy bal- tionship was preserved, with SVIs increasing up to LAIance. Global LAI surfaces will be an early product of thevalues of 3 to 5. For all but the coniferous forest site,MODIS Land Science Team, and the requirements forsensitivity of the SVIs was low at LAI values above 5. InLAI validation at selected sites have prompted interest inconiferous forests, the SVIs decreased at the highest LAIaccurate LAI mapping at a more local scale. While spec-values because of decreasing near-infrared reflectance as-tral vegetation indices (SVIs) derived from satellite remotesociated with the complex canopy in these mature to old-sensing have been used to map LAI, vegetation type, andgrowth stands. The cross-site LAI–SVI relationships basedrelated optical properties, and effects of Sun–surface–on atmospherically corrected imagery were stronger thansensor geometry, background reflectance, and atmosphericthose based on DN, radiance, or top of atmosphere reflec-quality can limit the strength and generality of empiricaltance. Topographic corrections at the conifer site alteredLAI–SVI relationships. In the interest of a preliminary as-the SVIs in some cases but had little effect on the LAI–SVIsessment of the variability in LAI–SVI relationships acrossrelationships. Significant effects of vegetation properties onvegetation types, we compared Landsat 5 Thematic Map-SVIs, which were independent of LAI, were evident. Theper imagery from three temperate zone sites with on-sitevariability between and around the best fit LAI–SVI re-LAI measurements. The sites differed widely in location,lationships for this dataset suggests that for local accu-vegetation physiognomy (grass, shrubs, hardwood forest,racy in development of LAI surfaces it will be desirableand conifer forest), and topographic complexity. Compar-to stratify by land cover classes (e.g., physiognomic typeisons were made using three different red and near-infra-and successional stage) and to vary the SVI. Elsevierred-based SVIs (NDVI, SR, SAVI). Several derivations ofScience Inc., 1999the SVIs were examined, including those based on rawdigital numbers (DN), radiance, top of the atmosphere re-flectance, and atmospherically corrected reflectance. ForINTRODUCTIONone of the sites, which had extreme topographic complex-Leaf area index (LAI), or projected leaf area per unitground surface area, is a key biophysical variable influ-encing land surface photosynthesis, transpiration, and en-* Forest Science Department, Oregon State University, Corvallis† USDA, Forest Service, Pacific Northwest Research Station, For-ergy balance (Running, 1990; Bonan, 1995). As LAI canest Sciences Laboratory, Corvallisbe related to remotely sensed data, an LAI surface has‡Department of Forest Ecology and Management, University ofbeen an important driver to some ecosystem productivityWisconsin, Madison§ Division of Biology, Kansas State University, Manhattanmodels applied at landscape to global scales (Running etAddress correspondence to David P. Turner, Forest Scienceal., 1989; Milner et al., 1996; Hunt et al., 1996; MartinDepartment, Oregon State University, Corvallis, OR 97331. E-mail:and Aber, 1997), and in the biosphere–atmosphere [email protected] 25 April 1998; revised 15 April 1999.actions component of some general circulation modelsREMOTE SENS. ENVIRON. 70:52–68 (1999)Elsevier Science Inc., 1999 0034-4257/99/$–see front matter655 Avenue of the Americas, New York, NY 10010 PII S0034-4257(99)00057-7Monitoring Leaf Area Index53(Chase et al., 1996). The NASA-sponsored Earth Observ- and NIR to increases in LAI from an unvegetated condi-ing System (EOS), more specifically the Moderate Reso-tion, and to a somewhat compensating effect on variationslution Imaging Spectroradiometer (MODIS) sensor, isin reflectance caused by differences in Sun–surface–sensorintended to refine our ability to monitor land surfacegeometry (Hall et al., 1995; Chen, 1996).properties such as LAI, and the MODIS LAND ScienceThe alternative forms of band ratioing in SR andTeam (MODLAND) is charged with global mapping ofNDVI result in differential sensitivity to variations in RLAI at a grain size of 1 km (Running et al., 1994).and NIR reflectance (Chen and Chilar, 1996). UnderThe MODLAND LAI mapping approach will rely onconditions of low LAI, where R is relatively high andMODIS imagery in conjunction with a canopy reflectanceNIR relatively low, a small change in R produces a largermodel (Myneni et al., 1997, Knyazikhin et al., 1999).proportional change in NDVI than SR. With higher LAI,This approach is based on the absence of a comprehen-for which NIR is generally higher and R lower, a changesive database on LAI–reflectance relationships, and thein NIR will induce a larger proportional change in SRrequirement for an LAI estimation algorithm which isthan NDVI. The optimum SVI for a particular applica-globally applicable. Validation of sensor data products istion may thus depend on the local LAI range.an important component of the EOS program, and oneVariation in background (soil and litter) reflectanceoption in the case of MODIS LAI surfaces is invertingcan affect R–NIR SVIs. This has led to development oflocal, or generalized, empirical LAI–reflectance relation-alternative formulations which include correction factorsships. This article presents results from a step towardsor constants introduced to account for, or minimize, theMODIS LAI validation, with a cross-site comparisonvarying background reflectance (Huete, 1988; Huete andof relationships between field-measured LAI and Land-Tucker, 1991; Huete et al., 1994). In applications focusedsat Thematic Mapper (TM) spectral vegetation indiceson overstory LAI, variability in reflectance from under-(SVIs).story vegetation can likewise be problematical (Franklin,1986; Spanner et al., 1990) and has inspired alternativeSVIs incorporating mid-IR reflectance (Nemani et al.,BACKGROUND1993). SVIs involving transformations based on