Oecologia (2011) 165:289���299 DOI 10.1007/s00442-010-1800-4 123 PHYSIOLOGICAL ECOLOGY - ORIGINAL PAPER Predicting tropical plant physiology from leaf and canopy spectroscopy Christopher E. Doughty �� Gregory P. Asner �� Roberta E. Martin Received: 11 November 2009 / Accepted: 14 September 2010 / Published online: 21 October 2010 �� Springer-Verlag 2010 Abstract A broad regional understanding of tropical for- est leaf photosynthesis has long been a goal for tropical for- est ecologists, but it has remained elusive due to diYcult canopy access and high species diversity. Here we develop an empirical model to predict sunlit, light-saturated, tropi- cal leaf photosynthesis using leaf and simulated canopy spectra. To develop this model, we used partial least squares (PLS) analysis on three tropical forest datasets (159 species), two in Hawaii and one at the biosphere 2 labora- tory (B2L). For each species, we measured light-saturated photosynthesis (A), light and CO2 saturated photosynthesis (Amax), respiration (R), leaf transmittance and reXectance spectra (400���2,500 nm), leaf nitrogen, chlorophyll a and b, carotenoids, and leaf mass per area (LMA). The model best predicted A [r2 = 0.74, root mean square error (RMSE) = 2.9 mol m��2 s��1)] followed by R (r2 = 0.48), and Amax (r2 = 0.47). We combined leaf reXectance and transmittance with a canopy radiative transfer model to simulate top- of-canopy reXectance and found that canopy spectra are a better predictor of A (RMSE = 2.5 �� 0.07 mol m��2 s��1) than are leaf spectra. The results indicate the potential for this technique to be used with high-Wdelity imaging spectrometers to remotely sense tropical forest canopy photosynthesis. Keywords Photosynthesis �� Tropical forests �� Spectranomics �� Spectra �� Gas exchange �� Remote sensing �� CAO Introduction Tropical forest net primary production plays an important role in the global carbon cycle (Field et al. 1998), but there is much debate on whether tropical forests are net sources or sinks of CO2 to the atmosphere (Grace et al. 1995 Phil- lips et al. 1998 Saleska et al. 2003). It has been clear for some time that a better understanding of tropical forest physiology and leaf-level gas exchange could shed light on these questions, but the diYculty of accessing tropical for- est canopies, along with their high species diversity, has so far hindered the development of a broad regional under- standing. High-resolution remote sensing may be an ideal tech- nique for acquiring data that would improve our under- standing of tropical forest physiology over large geographic areas. However, remote sensing of physiology has been attempted at the leaf and canopy scales with mixed results. Sims and Gamon (2002) detected photosynthetic stress at the leaf level in multiple species via xanthophyll pigment changes with the photochemical reXectance index (PRI) (Sims and Gamon 2002). Asner et al. (2004) measured drought stress in an Amazon tropical forest by space-borne imaging spectroscopy of the canopy water content. These studies showed the potential of the systems to remotely detect physiology, but the sensors used were not designed to identify physiological diVerences between individual species. To understand the physiology of tropical forests, we Wrst must understand leaf chemistry. Although there may be Communicated by Gerardo Avalos. Electronic supplementary material The online version of this article (doi:10.1007/s00442-010-1800-4) contains supplementary material, which is available to authorized users. C. E. Doughty (&) �� G. P. Asner �� R. E. Martin Department of Global Ecology, Carnegie Institution, Stanford, CA 94305, USA e-mail: chris.doughty@ouce.ox.ac.uk

290 Oecologia (2011) 165:289���299 123 broad shifts in leaf chemistry with soil fertility, these tend to be modest in size and indeed perhaps smaller than the range of variation seen among co-occurring species in terms of their leaf chemistry. For example, the dominant landscape-level inXuence over foliar nitrogen (N) and phos- phorus (P) ratios in the humid tropics is species variability, not substrate (Townsend et al. 2007). Several physiologi- cally important chemicals, such as N, are directly expressed in the leaf spectra. Studies have shown that foliar concen- trations of N, chlorophyll (Chl) a and b, carotenoids, water, and leaf mass per area (LMA) can be predicted using the reXectance and transmittance of light from individual leaves and canopies (Curran 1989 Sims and Gamon 2002 Smith et al. 2003a, b). Laboratory and modeling studies have also demonstrated how these chemicals inXuence leaf spectral properties (Feret et al. 2008 Jacquemoud and Baret 1990). Relationships between leaf spectral and chemical prop- erties have been developed using partial least squares (PLS) regression analysis. This technique does not focus on indi- vidual bands but uses the entire leaf spectral continuum. Previous studies using this technique showed that LMA, Chl a and b, and N concentrations are highly correlated with the reXectance and transmittance spectra of tropical species (Asner et al. 2009). Moreover, each spectral band is weighted to determine its relative importance in predicting chemical constituents, and these band weights indicate the spectral features (involving contiguous wavelength regions) that are most important to predict a variable (Haaland and Thomas 1988). For example, Asner et al. (2009) found that PLS weightings for LMA were strong throughout the 750- to 2,500-nm range because this region is dominated by variations in leaf water content and leaf thickness (Asner et al. 2009), both of which are related to LMA (Jacquemoud and Baret 1990). N weightings were associated with wavelengths absorbed by Chl a and b in the visible (400���700 nm), the spectral red edge (700���760 nm), and proteins in the shortwave infrared range (1,300��� 2,500 nm) (Gitelson and Merzlyak 1997 Kokaly 2001 Smith et al. 2003a). Leaf photosynthesis is generally correlated with leaf chemistry. A worldwide foliar dataset indicates that 82% of all variation in photosynthetic capacity (by mass) can be explained by LMA and N (on a log���log scale) (Wright et al. 2004). A similar study used 53 tropical species and found that LMA predicted mass-based rates of assimilation and respiration and that leaf life span predicted many other traits (Poorter and Bongers 2006). Despite these advances, few studies have considered the three-way connection between the spectroscopic, chemical, and physiological properties of foliage, although many have considered vari- ous aspects of two out of three of these areas, particularly as to how spectral band ratios [e.g., normalized diVerence vegetation index (NDVI)] scale with leaf and canopy chem- istry and physiology (e.g., Sellers 1985 Verma 1993). However, the full spectroscopic signatures of species have not been considered in this context. We combined leaf physiological, chemical, and spectral data from 159 tropical species gathered on three Weld cam- paigns, two in Hawaii and one at the biosphere 2 laboratory (B2L) in Arizona. We tested our ability to predict leaf physiological properties from leaf spectra using empirically based models. We then used observed relationships between leaf chemical/spectral and chemical/physiological properties to understand the relationships between the spec- tral/physiological properties. Finally, we scaled the leaf spectra to the canopy level using a radiative transfer model (Asner and Martin 2008) and developed models predicting physiological properties using simulated canopy spectra. We addressed the following questions: Can tropical leaf photosynthetic capacity and other physiological properties be predicted from full spectral range (400���2,500 nm) leaf spectroscopy? What is the relationship among spectral, chemical, and physiological properties, in the context of spectroscopic remote sensing? Are canopy spectra more or less useful than leaf spectra for the prediction of photosyn- thetic capacity and other physiological properties? Methods Study sites We collected foliar material and measured the physiologi- cal properties of 149 species in two tropical forest sites on Hawaii island in January and April of 2009 and ten species in the tropical rainforest biome and the display center of B2L (Dempster 1999) in October 2004. We emphasized collecting data from many species versus repeating measurements on individual species because much of the variability in leaf nutrient concentrations can be explained by species variability (Townsend et al. 2007). These species represent a wide variety of regeneration strategies. In the Wrst Hawaii campaign, we chose a site near to the Institute for PaciWc Islands Forestry in Hilo, Hawaii, located on the windward, wet side of the island (approx. 3,500 mm year��1). This site consists of young basalt sub- strate (approx. 220 years old) with Hawaiian endemic and common exotic species. We measured sunlit intact leaves from 11 species. For the second collection, we measured 138 tropical tree and palm species at sites spread across the windward side of the island (mean annual precipitation 2,000���4,000 mm). These samples were taken mainly from non-native, full-grown trees grown in botani- cal gardens near Hilo. At B2L, we measured ten canopy rainforest tree species, sampling ten leaves still connected