Using hyperspectral plant traits linked to photosynthetic efficiency to assess N and P partition
Introduction
The prediction of foliage nutrient concentration across broad spatial scale has considerable utility for assessing health and productivity of ecosystems (McNeil et al., 2007a, McNeil et al., 2007b, Townsend et al., 2003). Significant growth responses within forest stands have been noted through the application of fertiliser, mainly in the form of nitrogen (N) and phosphorus (P) (Albaugh et al., 2003, Allen et al., 2005, Liechty and Fristoe, 2013). However, determining the optimal application of fertiliser is often very difficult as nutritional deficiencies vary widely across the landscape (Campion, 2008, Fox et al., 2007).
Foliage nutrition is also a key determinant of rates of carbon assimilation. Many studies have investigated the use of N and P to account for variation in photosynthetic capacity (Walker et al., 2014) and the importance of these elements on the photosynthesis process is well documented. Nitrogen is a major component of Rubisco (Niinemets and Tenhunen, 1997) and P has an impact on many important aspects of photosynthesis including membrane solubility, ATP, and NADPH production (Marschner, 1995, Taiz et al., 2015). Given these strong links it is important to understand how N and P regulate photosynthesis when developing models of these elements that will be subsequently used to predict photosynthesis.
Hyperspectral imagery has been successfully used to predict foliar concentrations of N for a wide range of broadleaf and coniferous species at both the leaf level and canopy level (for an overview see Hill et al., 2019, Watt et al., 2019). Nitrogen shows marked absorption features in both the visible near infrared (VNIR) and short-wave infrared (SWIR) ranges (Curran, 1989, Kokaly, 2001, Kokaly et al., 2009) and models are able to capitalise on these features to predict N (Watt et al., 2019). At the leaf level the precision of models of N has been found to range widely (R2 range of 0.37–0.99) but generally N has been predicted with moderate to high levels of precision, with mean R2 of 0.77, recorded across a wide range of species (Asner and Martin, 2008, Asner et al., 2011, Curran et al., 2001, Dechant et al., 2017, Gillon et al., 1999, Luther and Carroll, 1999, Masaitis et al., 2014, Petisco et al., 2005, Schlerf et al., 2010, Serbin et al., 2014, Stein et al., 2014, Tsay et al., 1982, Wang et al., 2018, Wang et al., 2015, Yoder and Pettigrew-Crosby, 1995). Despite the difficulties in scaling to the canopy, moderate to strong predictions of N have also been made at this level with values of R2 ranging from 0.48 to 0.98 (Coops et al., 2003, Huang et al., 2004, Knyazikhin et al., 2013, Martin et al., 2008, Martin et al., 2018a, Martin et al., 2018b, Ollinger et al., 2008, Singh et al., 2015, Smith et al., 2003).
Although many studies have developed models of P, the lack of spectral absorption features associated with P is likely to limit their generality as these relationships are indirect. Leaf level hyperspectral data have generally been found to predict P empirically with moderate to high precision (mean R2 of 0.74) using proxies correlated with P, with R2 varying from 0.32 to 0.95 (Asner and Martin, 2008, Asner et al., 2011, Curran et al., 2001, Gillon et al., 1999, Masaitis et al., 2014, Petisco et al., 2005, Stein et al., 2014). However, as P does not directly absorb energy in the shortwave spectrum, these predictions of P are actually due to the correlations with N found under most conditions (Asner and Martin, 2008, Gillon et al., 1999, Porder et al., 2005). As a consequence, these models may not be as robust when applied to conditions where ratios of N/P deviate from typical ranges. Development of models of P using datasets with little correlation between N and P are likely to reveal the true precision between hyperspectral-based traits and P, likely resulting in predictions that have greater generality.
Although little research considers whether N and P co-limit or independently limit photosynthesis, this is a key assumption that is likely to affect model precision, generality and applicability. Datasets used generally assume that photosynthesis is co-limited by N and P as both elements are predicted across their complete range. Although it has not been widely investigated, a number of studies suggest that N and P independently limit growth and photosynthesis in many species (Bown et al., 2007, Domingues et al., 2010, Ingestad, 1971, Ingestad, 1979, Ingestad and Lund, 1986). The premise underlying this research is that a mass based N/P ratio of 10 is optimal with values of N/P ≤ 10 leading to N limitations and N/P > 10 resulting in P limitations (Aerts and Chapin, 2000, Marschner, 1995, Reich and Schoettle, 1988). Previous research supports this suggestion through showing that N and P independently influence the key biochemical limitations to photosynthesis that include the maximum rate of carboxylation (Vcmax) and electron transport (Jmax) (Domingues et al. 2010) and that a stoichiometric ratio of 10 can be used to partition N from P limitations for these variables (Bown et al. 2007). If photosynthesis is independently regulated by N and P it follows that Vcmax and Jmax will exhibit positive relationships with both elements within their respective limiting ranges and these relationships will be weaker when they are constructed using the entire dataset under the assumption of co-limitation.
The assumption of independent limitations has important implications for the development of foliage nutrition models. In order to identify the type of required nutrient additions it would be useful to be able to partition N from P limitations at scales ranging from the tree to the forest level. Following this partitioning, the use of N and P models developed using data from their respective limiting ranges could be applied to estimate the severity of any deficiency and impact on photosynthesis within these two ranges. This approach may provide a means of improving spatial accuracy when characterising the type and extent of nutrient deficiencies. Understanding the nature of relationships, within each limiting range, is also likely to provide considerable insight into the mechanistic link between plant functional traits quantified from hyperspectral imagery and foliage nutrition and, as a consequence, growth and photosynthesis. Despite this, we are unaware of any research that has used this approach for developing models of N and P for tree species.
Traditional methods used to track changes in plant nutrition from remote sensing have most often targeted chlorophyll content (Ca+b) as chlorophyll and other pigments such as carotenoids, xanthophyll and anthocyanins are important indicators of plant photosynthetic status (Baret et al., 2007, Evans, 1989, Yoder and Pettigrew-Crosby, 1995). Nitrogen is a major component of chlorophyll, and nitrogen and chlorophyll deficiencies are directly related to reductions in photosynthetic rates (Evans, 1989). It therefore follows that Ca+b has been the focus of remote sensing research as a proxy for nitrogen within agriculture and forestry (Baret et al., 2007, Yoder and Pettigrew-Crosby, 1995).
Remote sensing research carried out in the 1980s identified the red-edge and green spectral regions as potential targets for estimating Ca+b that were linked to nitrogen content (Carter, 1994, Gitelson and Merzlyak, 1996, Rock et al., 1988). Further research developed specific narrow-band hyperspectral indices (Haboudane et al., 2002) including the red-edge Chlorophyll Index (Zarco-Tejada et al., 2001). Combined indices were also developed such as the Transformed Chlorophyll Absorption in Reflectance Index, TCARI (Haboudane et al., 2002) normalized by the Optimized Soil-Adjusted Vegetation Index, OSAVI (Rondeaux et al., 1996) to form the TCARI/OSAVI proxy for chlorophyll and nitrogen.
In addition to Ca+b, recent research has identified other pigments, such as xanthophylls, that are more dynamically related to rapid changes in photosynthesis and are potentially more useful for tracking nutritional impacts on photosynthesis. The changes observed in the green spectral region through the Photochemical Reflectance Index (PRI) (Gamon et al., 1992) have been demonstrated to be linked to the xanthophyll cycle, and this index has been successfully used to predict photosynthetic rate (Drolet et al., 2008, Fuentes et al., 2006, Gamon et al., 1997, Guo and Trotter, 2004, Hilker et al., 2008, Middleton et al., 2009, Nichol et al., 2000, Penuelas et al., 1995b, Stylinski et al., 2000) and the photosynthetic response of plants to a range of stresses (Buddenbaum et al., 2015, Dobrowski et al., 2005, Hernández-Clemente et al., 2011, Scholten et al., 2019, Suarez et al., 2008).
During the last 50 years (see review by Mohammed et al., 2019) considerable research has demonstrated the utility and feasibility of Solar-Induced Chlorophyll Fluorescence (SIF) in predicting photosynthetic activity at both the leaf and the canopy scales from a range of remote sensing platforms (Cendrero-Mateo et al., 2015, Zarco-Tejada et al., 2013, Zarco-Tejada et al., 2016). Given the strong relationship between chlorophyll pigments, the xanthophyll dynamics (PRI) and photosynthesis (through SIF) it also follows that these indicators might be significantly related to N and P as these two elements are key determinants of photosynthesis rate.
Pinus radiata D. Don (radiata pine) is the most widely planted plantation species within the southern hemisphere (Lewis and Ferguson, 1993). Over 4.1 M ha of this species has been established in New Zealand, Chile and Australia, where the species comprises, respectively, 90% (NZFOA, 2018), 62% (Salas et al., 2016) and 39% (Downham and Gavran, 2019) of the total plantation area. Pinus radiata frequently suffers from nutrient limitations, particularly during mid-rotation when nutrient demand often exceeds supply. The key elements that limit productivity of P. radiata are N and P (Watt et al., 2005). A shortage of these elements can result in significant reductions in growth (Raison and Myers, 1992, Sheriff et al., 1986) and also limits the key processes that control the rate of photosynthesis (Bown et al., 2009).
In this study, measurements of hyperspectral imagery, photosynthesis and foliage nutrition were taken from an experiment that included a factorial combination of N and P treatments applied to P. radiata. Using these data, the initial objective of this research was to (i) identify whether trees were co-limited or independently limited by N and P, and then use hyperspectral imagery to (ii) partition N and P limited trees, (iii) build models of N and P from a range of hyperspectral indices and (iv) explore links between key plant traits and both Vcmax and Jmax.
Section snippets
Experimental set up
The experiment was undertaken within the Scion nursery, located in Rotorua, New Zealand. A total of 120 P. radiata seedlings were transplanted into 15 L pots during October 2018. The medium into which the plants were transplanted consisted of a mixture of perlite and vermiculite which are silica-based products without any nutritional content. Plants were grown in a thermostatically controlled greenhouse where temperature in spring fluctuated between 10 and 24 °C during the day and between 10
Foliar nutrition
The applied treatments resulted in a wide range in N and P (Fig. 1). Values of N ranged from 0.41 to 2.12% when expressed on a mass basis and 11.1–49.0 µg cm−2 on an area basis while P varied from respectively 0.052–0.330% and 1.33–7.66 µg cm−2. The relationship between N and P was insignificant when data was expressed on a mass (R2 = 0.0073; P = 0.52) or area basis (R2 = 0.0006; P = 0.85). Of the 60 plants, 43 were categorised as N limited while 17 were P limited and the separation of these
Discussion
This study advances our understanding of the functional relationships between SIF, PRI and both foliage nutrition and photosynthesis. The strongest relationships between N and P and both Vcmax and Jmax were developed through splitting the data at N/P = 10 suggesting that photosynthesis is independently limited by these elements. The model that was used to accurately partition N from P limited trees demonstrated the importance of PRI and SIF in separating these two groups. Within each of these
Declaration of Competing Interest
The authors declare that they have no known competing financial interests or personal relationships that could have appeared to influence the work reported in this paper.
Acknowledgements
We are grateful to Kendra Newick who assisted with the preparation of the nutrient solutions. The project was partly funded through the Resilient Forests programme, which is funded through Scion SSIF as well as the Forest Grower’s Levy Trust. Funding was also received from the National Institute for Forest Products Innovation (Project Number NIF073-1819), which comprised contributions from the Australian Government, Australasian Forestry Companies and South Australian and Tasmanian State
References (159)
- et al.
Carnegie Airborne Observatory-2: Increasing science data dimensionality via high-fidelity multi-sensor fusion
Remote Sens. Environ.
(2012) - et al.
Spectral and chemical analysis of tropical forests: Scaling from leaf to canopy levels
Remote Sens. Environ.
(2008) - et al.
Spectroscopy of canopy chemicals in humid tropical forests
Remote Sens. Environ.
(2011) - et al.
A reappraisal of the use of DMSO for the extraction and determination of chlorophylls a and b in lichens and higher plants
Environ. Exp. Bot.
(1992) - et al.
Comparing prediction power and stability of broadband and hyperspectral vegetation indices for estimation of green leaf area index and canopy chlorophyll density
Remote Sens. Environ.
(2001) - et al.
High-resolution airborne hyperspectral and thermal imagery for early detection of Verticillium wilt of olive using fluorescence, temperature and narrow-band spectral indices
Remote Sens. Environ.
(2013) - et al.
The applicability of empirical vegetation indices for determining leaf chlorophyll content over different leaf and canopy structures
Ecol. Complexity
(2014) Remote sensing of foliar chemistry
Remote Sens. Environ.
(1989)- et al.
Estimating the foliar biochemical concentration of leaves with reflectance spectrometry: testing the Kokaly and Clark methodologies
Remote Sens. Environ.
(2001) - et al.
Modeling the impact of spectral sensor configurations on the FLD retrieval accuracy of sun-induced chlorophyll fluorescence
Remote Sens. Environ.
(2011)