Journal of Vegetation Science 13: 817-830, 2002 © IAVS; Opulus Press Uppsala. - A model for seasonality and distribution of leaf area index of forests - 817 A model for seasonality and distribution of leaf area index of forests and its application to China Luo, Tianxiang1*; Neilson, Ronald P.2; Tian, Hanqin3; Vörösmarty, Charles J.4; Zhu, Huazhong1 & Liu, Shirong5 1Institute of Geographical Sciences and Natural Resources Research, Chinese Academy of Sciences, P.O. Box 9717, Beijing 100101, China; 2USDA Forest Service, Oregon State University, Oregon 97333, USA; 3Department of Ecology and Evolutionary Biology, University of Kansas, KS 66045, USA; 4Institute for the Study of Earth, Oceans and Space, University of New Hampshire, Durham, NH 03824, USA; 5Institute of Forestry Ecology and Environment Protection, Chinese Academy of Forestry, Beijing 100091, China; Corresponding author; Fax: +861064889399; E-mail [email protected], [email protected] Introduction Abstract. We have constructed a phenological model of leaf area index (LAI) of forests based on biological principles of leaf growth. Field data of maximum LAI from 794 plots with Leaf area index (LAI) is defined as the assimilative mature or nearly mature stand ages over China were used to leaf area relative to the projected ground area for a plant parameterize and calibrate the model. New measurements of community (one-side area for broad-leaved trees and maximum LAI from 16 natural forest sites were used to vali- curve surface area exposed to sunlight for coniferous date the simulated maximum LAI. The predictions of seasonal trees). In recent global change modelling, LAI has been LAI patterns were compared with seasonal changes derived widely used as an important variable in relation to from the 1-km satellite AVHRR-NDVI data for nine undis- carbon, water and energy fluxes (Woodward 1987; Run- turbed forest sites in eastern China. Then, we used the model ning et al. 1989; Prentice et al. 1992; Neilson 1995). to map maximum LAI values for forests in China. Model results indicated that the PhenLAI model generally Much uncertainty in quantifying LAI exists due to the predicted maximum LAI well for most forest types, even when limitation of our understanding of biological processes maximum LAI is > 6. This suggests an ecological approach to that determine the magnitude and distribution of LAI. the saturation problem in satellite detection of high forest LAI Several approaches have been used to study the where the relationship between NDVI and LAI reaches an phenology of leaf mass and productivity, including: (1) asymptote near a projected LAI value of 5 or 6. Furthermore, length of the growing season, (2) life cycles and dura- the predictions of seasonal LAI patterns in timing and dynam- tion of phenophases, (3) growth analysis, (4) correla- ics were generally consistent with the satellite NDVI changes, tions between environmental factors and phenological except for monsoon forest and rain forest in south China where events, (5) modelling and monitoring of seasonal car- satellite detection of seasonal variation in leaf area is hardly possible. Compared with average projected LAI measure- bon dynamics. Traditional techniques are based on phe- ments of global forests from 809 field plots in literature data, nological observations and climate data, e.g. ‘pheno- our maximum LAI values were close to the global literature metry’ (Lieth 1970), ‘heat-units’ (Wang 1960), ‘degree- data for most of Chinese forests, but the average area-weighted days’ (Thomson & Moncrieff 1982; Hunter & Lechowicz maximum LAI for all forests of China (6.68 ± 3.85) was higher 1992) or ‘chill-days’ (Landsberg 1974; Cannell & Smith than the global mean LAI of the 809 field plots (5.55 ± 4.14). 1983; Murray et al. 1989). These techniques have been We believe that forest LAI in China is commonly > 6, espe- used for modelling development phases (phenophases) cially in tropical rainforest, subtropical evergreen broad-leaved in plant life cycles (e.g. phenological spectrum) and forest, temperate mixed forest, and boreal/alpine spruce-fir quantitative changes within one phenophase. More ad- forest where satellite detection of high LAI is hardly possible. vanced techniques for modelling onset and offset of greenness and seasonal carbon dynamics of regional/ Keywords: LAI; Leaf growth; Leaf mass; Modelling; NDVI; global vegetation include: (1) computer simulations of Phenology. carbon flux by maximizing plant carbon gain (Janecek et al. 1989; Kikuzawa 1995; Woodward et al. 1995) or site Abbreviations: LAI = Leaf Area Index; NDVI = Normalized water balance (Neilson 1995); (2) satellite monitoring Difference Vegetation Index; fPAR = Fraction of incoming based on the relation of LAI to NDVI (Normalized Differ- Photosynthetically Active Radiation absorbed by plant canopy; ence Vegetation Index) (Spanner et al. 1990; Curran et al. WBM = Water balance model. 1992; White et al. 1997) or the relation of fPAR (the 818 Luo, T. et al. fraction of incoming photosynthetically active radiation in E. China. Observed data of annual actual evapotrans- absorbed by the canopy) to NDVI or other satellite piration by the catchment technique (CT) in some typi- vegetation index (Prince 1991a; Ruimy et al. 1994; cal forests of E China were used to validate the esti- Goetz & Prince 1996); (3) combinations of these first mated actual evapotranspiration in the PhenLAI model. two approaches (e.g. Field et al. 1995; Hunt et al. 1996). Then, we use the model to map LAImax of forests. Measurements of LAI by satellite data (NDVI or fPAR) have been largely successful in croplands and grasslands (e.g. Prince 1991a; Daughtry et al. 1992), but some Leaf Area Index variations in PhenLAI problems remain where LAI is high, such as forests where stand LAI is commonly > 6. These problems exist be- Predictions of LAI are based on three general tempo- cause the regression equations of NDVI or fPAR to LAI ral and spatial patterns over a range of forests: (1) LAI reach an asymptote around a projected LAI value of 5 or 6 increases initially with forest age, then decreases slightly; (Sellers 1985; Spanner et al. 1990; Prince 1991b; Curran et (2) it generally has one seasonal peak; (3) it changes al. 1992). Because of the saturation problem, satellite detec- regionally with temperature and precipitation variables. tion of seasonal variation in leaf area of southern evergreen forests is hardly possible (White et al. 1997). Considerable Growth of leaf mass with increasing stand age seasonality can exist in evergreen forests, but satellite obser- vations of vegetation phenology almost always focus on There is evidence that leaf mass of a forest commu- deciduous forests or annual grasslands and croplands. nity reaches a more or less steady state early in succes- Although the most sophisticated models are based on sion (Grier & Running 1977). In general, for a young simulations of ecosystem processes, they are not satisfac- stand, LAI increases as stand age increases, and reaches tory for application at a global scale due to the lack of a maximum in the nearly mature or mature stages. This parameters for most of the world ecosystems (Ruimy et maximum can be maintained if sufficient resources are al. 1994). One of the critical gaps in developing a dy- available (Tadaki 1977; Waring & Schlesinger 1985). In namic biosphere model is a poor understanding of how to many even-aged forests, competition among trees re- scale up and the inability to validate such a model (Tian et sults in some mortality before LAI approaches a maxi- al. 1998). LAI is the most important variable for measur- mum, so that LAI actually peaks and then decreases ing vegetation structure over large areas, and for relating slightly, maintaining a plateau until the trees die. Maxi- it to energy and mass exchanges. In recent large-scale mum sustained LAI is often 10-20% less than peak LAI ecosystem modelling, LAI has been treated as an impor- (Waring & Schlesinger 1985). tant variable. However, most of the published biogeogra- phy models and biogeochemistry models essentially con- Seasonal changes of leaf mass sider LAI as an uncalibrated, internal state variable that is used to produce calibrated variables, such as vegetation Generally, boreal and temperate forests are seasonal carbon or vegetation types. The LAI outputs from various where leafing patterns seem to be governed mainly by models differ greatly in magnitude and distribution (R.J. seasonally changing temperature and radiation. Warm- Drapek, R.P. Neilson & VEMAP Members pers. comm.). temperate forests are moderately seasonal. Tropical and In this study, we present a phenological model of subtropical forests have variable phenological patterns leaf area index (here called PhenLAI, the ‘Phenology of throughout the year but with a major pulse in the summer. Leaf Area Index’), based on the morphological charac- Then, seasonality of leaf mass for a site can be quantita- teristics of leaf growth and development instead of tively described as a function of time only (Fig. 1), such as physiological processes and biogeochemical cycles. The the symmetrical sine curve (Waggoner 1974) or Gaussian field data of maximum LAI (LAImax) from 794 plots with distribution (Lieth 1970). In temperate forests, Kikuzawa mature or nearly mature stand ages collected by Luo (1983, 1984) found for deciduous tree and deciduous/ever- (1996) are used to parameterize and calibrate the model. green shrub species in Japan three leafing patterns over a New measurements of LAImax for 13 natural forest sites in growing season: (1) flush type, (2) intermediate type, and (3) the Tibetan Alpine Vegetation Transects (TAVT) from succeeding type. Most of the leaf survivorship curves look 1999-2000, and three natural forest sites in eastern China like an inverse U-shape or a bell shape. Taylor (1974), from the Chinese literature (1998-2001) are used to vali- Larcher (1975), Sharik & Barnes (1976), and Bormann & date the simulated LAImax. The predictions of seasonal Likens (1979) observed similar leafing patterns for decidu- LAI patterns in timing and dynamics instead of the LAI ous trees.