Canopy foliar nitrogen retrieved from airborne hyperspectral imagery by correcting for canopy structure effects

https://doi.org/10.1016/j.jag.2016.09.008Get rights and content

Highlights

  • %N can be accurately retrieved from the canopy scattering coefficient.

  • The structural influence on canopy BRF can be suppressed using DASF.

  • Wavelet features capture the narrow-range absorption features of nitrogen.

  • A wavelet based approach yields a more accurate estimation of nitrogen than PLSR.

Abstract

A statistical relationship between canopy mass-based foliar nitrogen concentration (%N) and canopy bidirectional reflectance factor (BRF) has been repeatedly demonstrated. However, the interaction between leaf properties and canopy structure confounds the estimation of foliar nitrogen. The canopy scattering coefficient (the ratio of BRF and the directional area scattering factor, DASF) has recently been suggested for estimating %N as it suppresses the canopy structural effects on BRF. However, estimation of %N using the scattering coefficient has not yet been investigated for longer spectral wavelengths (>855 nm). We retrieved the canopy scattering coefficient for wavelengths between 400 and 2500 nm from airborne hyperspectral imagery, and then applied a continuous wavelet analysis (CWA) to the scattering coefficient in order to estimate %N. Predictions of %N were also made using partial least squares regression (PLSR). We found that %N can be accurately retrieved using CWA (R2 = 0.65, RMSE = 0.33) when four wavelet features are combined, with CWA yielding a more accurate estimation than PLSR (R2 = 0.47, RMSE = 0.41). We also found that the wavelet features most sensitive to %N variation in the visible region relate to chlorophyll absorption, while wavelet features in the shortwave infrared regions relate to protein and dry matter absorption. Our results confirm that %N can be retrieved using the scattering coefficient after correcting for canopy structural effect. With the aid of high-fidelity airborne or upcoming space-borne hyperspectral imagery, large-scale foliar nitrogen maps can be generated to improve the modeling of ecosystem processes as well as ecosystem-climate feedbacks.

Introduction

Foliar nitrogen (N) is a primary regulator of physiological processes, such as photosynthesis, leaf respiration, and transpiration (Evans, 1989, Field and Mooney, 1986, Reich et al., 1998, Reich et al., 2006). Nitrogen is regarded as a limiting nutrient for plant growth (Heimann and Reichstein, 2008, LeBauer and Treseder, 2008), and is one of the key plant traits driving stand-level forest productivity (Reich, 2012). Nitrogen availability constrains carbon assimilation and, thereby, plays an important role in terrestrial ecosystem carbon dynamics, acting as potential climate feedback (Heimann and Reichstein, 2008, Lamarque et al., 2005, Ollinger et al., 2008, Sievering et al., 2000). As a consequence, foliar nitrogen content has recently been proposed as one of the key essential biodiversity variables (EBVs) for satellite monitoring of progress towards the Aichi Biodiversity Targets (Pereira et al., 2013, Skidmore et al., 2015).

Nitrogen has been retrieved with good accuracy using leaf- and canopy-level hyperspectral data despite the fact that it is only a relatively small constituent (0.2–6.4%) in leaves (Kokaly et al., 2009, Wright et al., 2004). Hyperspectral data provide a time-efficient and cost-effective solution for estimating foliar nitrogen compared to the traditional destructive sampling methods. Previous studies on nitrogen estimates used spectra of ground leaf powder, dry leaves and fresh leaves, as well as reporting results at canopy level (Curran, 1989, Kokaly and Clark, 1999, Martin et al., 2008). Nitrogen has been quantified in forest, grassland, and crops (Inoue et al., 2012, Singh et al., 2015, Skidmore et al., 2010). Empirical techniques are dominant in nitrogen retrieval using hyperspectral data, ranging from vegetation indices (Miphokasap et al., 2012), traditional regression techniques, such as stepwise multiple linear regression (Kokaly and Clark, 1999) and partial least square regression (Martin et al., 2008), to a number of artificial intelligence methods, such as support vector regression, neural network and Bayesian model averaging (Axelsson et al., 2013, Skidmore et al., 2010, Zhao et al., 2013).

Mapping forest canopy foliar nitrogen concentration (%N) at a local level is feasible with the aid of imaging spectrometry. The current lack of higher spatial resolution satellite hyperspectral sensors impedes the mapping of nitrogen at larger scales (e.g. continental or global scale), though the launch of the EnMAP mission planned for 2018 may provide an opportunity. Ollinger et al. (2008) provided a continental-scale map of foliar nitrogen by utilizing a statistical relationship between %N and the canopy bidirectional reflectance factor (BRF) for near-infrared (NIR) wavelengths (800–850 nm) for temperate and boreal forests. The findings were encouraging for the remote sensing community, because they allow the estimation of foliar nitrogen across larger extents and provide the option for frequent updates using data from broadband satellite data such as MODIS (Ollinger et al., 2008).

Following the work of Ollinger et al. (2008), Knyazikhin et al. (2013a) applied the physically-based spectral invariants theory (Huang et al., 2007, Knyazikhin et al., 2011) to interpret the process of radiation transfer from leaves and canopies, and explicitly analyzed the coupling between canopy structure and leaf biochemistry in driving variations in canopy BRF. Knyazikhin et al. (2013a) claimed that the significant positive relationship between %N and BRF in the NIR domain (Ollinger et al., 2008) should in fact be attributed to the correlation between canopy structure and BRF. Knyazikhin et al. (2013a) further demonstrated that the BRF between 710 and 790 nm is critical for deriving a canopy structure parameter – the directional area scattering factor (DASF). The DASF is not considered to be a specific canopy structure parameter per se such as leaf area index (LAI), stem density, tree height etc., but rather DASF is a parameter that governs how the scattered radiation from a leaf is further transformed through multiple scattering processes, and can be interpreted as canopy BRF if it is assumed that foliage does not absorb radiation (Knyazikhin et al., 2013a). This was the first study to illustrate the physical interaction between leaf albedo and canopy BRF, which had been neglected in previous studies on hyperspectral remote sensing of leaf biochemical constituents (Knyazikhin et al., 2013a, Ustin, 2013).

Although estimation of canopy foliar %N has been tested by using the ratio of BRF and DASF spectra (canopy scattering coefficient, Wλ) after suppressing the impact of canopy structure (Knyazikhin et al., 2013a), the analyses were restricted to using information from each wavelength between 423 and 855 nm (Knyazikhin et al., 2013a, Knyazikhin et al., 2013b). Canopy foliar %N estimates using the visible spectral region rely on the well-known correlation between chlorophyll and nitrogen (Evans, 1989, Field and Mooney, 1986, Kokaly et al., 2009). However, the capability of estimating %N using the scattering coefficient based on longer wavelengths from 855 to 2500 nm was not addressed, despite the fact that many nitrogen absorption bands are located in the shortwave infrared regions (SWIR, >1100 nm) (Curran, 1989, Fourty et al., 1996, Kokaly et al., 2009).

Continuous wavelet analysis has emerged as an effective tool in remote sensing − being applied for image fusion, image segmentation, as well as quantifying leaf-level biochemical parameters such as leaf mass per area (Cheng et al., 2014a), leaf chlorophyll (Blackburn and Ferwerda, 2008), leaf water content (Cheng et al., 2011, Ullah et al., 2012), and leaf nitrogen (Ferwerda and Jones, 2006). The technique has rarely been tested using an imaging spectrometer except in a recent study on detecting water content (Cheng et al., 2014b). Therefore, the feasibility of using continuous wavelet analysis in quantifying canopy foliar %N from airborne hyperspectral data is an interesting challenge.

Here, we investigated to what extent canopy foliar %N may be retrieved by disengaging the canopy structural impact from the canopy bidirectional reflectance factor (BRF). We derived the directional area scattering factor (DASF) based on spectral invariant theory, and calculated the canopy scattering coefficient to correct for canopy structural effects. A continuous wavelet transformation was performed on the canopy scattering coefficient. We aimed to (1) test if the scattering coefficient derived from airborne hyperspectral data in the spectral range from 400 to 2500 nm retains information required to estimate %N once canopy structure effects are suppressed; (2) evaluate the performance of continuous wavelet analysis approach in %N prediction and compare the wavelet approach to the widely used partial least squares regression approach; (3) identify the spectral regions most sensitive to %N variations in the canopy scattering coefficient.

Section snippets

Field data

The study area is located in the southern region of the Bavarian Forest National Park (BFNP) (49° 3′ 19″ N, 13° 12′ 9″ E) in southeastern Germany (Fig. 1). The BFNP has a total area of 24,218 ha. The bedrock is primarily composed of gneiss and granite. Soils weathered from these parent materials are naturally acidic and low in nutrients. The main soil types are brown soils, loose brown soils and podzol brown soils. Elevation ranges from 600 to 1453 m. The Bavarian Forest lies in the temperate

Characteristics of canopy foliar %N

Variation in canopy foliar %N was observed in broadleaf, needle leaf and mixed forest plots (Table 1). The mean canopy foliar %N in these three types of plots was 2.77, 1.55, and 1.69, respectively, similar to measurements from the Bavarian Forest National Park agency (unpublished data). The pure broadleaf plots had the largest range in %N, namely from 2.18 to 3.29, and largest coefficient of variance (CV), 0.139. The %N in needle leaf plots showed the smallest range (0.24), and consequently,

Discussion

Hyperspectral remote sensing of foliar nitrogen may be confounded by canopy structure, causing variation in the canopy BRF (Knyazikhin et al., 2013a). This study derived the canopy scattering coefficient by correcting for canopy structure effects on the canopy BRF over the full spectrum from 400 to 2500 nm using airborne hyperspectral data. Our results confirmed the feasibility of accurately estimating canopy foliar %N (R2 = 0.65, RMSE = 0.33, Table 2) from the scattering coefficient using a wavelet

Conclusions

To estimate canopy foliar nitrogen the canopy BRF was evaluated across all wavelengths from 400 to 2500 nm, correcting for canopy structural impacts using the spectral invariants parameter, DASF. DASF is derived based on physical laws and serves as a practical way of characterizing canopy structure. Canopy foliar nitrogen can be accurately estimated from the canopy scattering coefficient, which suppresses the impact of canopy structure. Continuous wavelet analysis led to a more accurate

Acknowledgments

This work of the first author was supported by the China Scholarship Council under Grant 201204910232 and co-funded by ITC Research Fund under Grant 93003032 from the Faculty of Geo-Information Science and Earth Observation (ITC), University of Twente, the Netherlands. We are grateful to the “Applied spectroscopy” team of the German Aerospace Center (DLR) and Bavarian Forest National Park for assistance in the fieldwork. We acknowledge the support of the “Data Pool Forestry” data-sharing

References (85)

  • S. Foley et al.

    Foliar spectral properties following leaf clipping and implications for handling techniques

    Remote Sens. Environ.

    (2006)
  • T. Fourty et al.

    Leaf optical properties with explicit description of its biochemical composition: direct and inverse problems

    Remote Sens. Environ.

    (1996)
  • J.A. Gamon et al.

    A narrow-waveband spectral index that tracks diurnal changes in photosynthetic efficiency

    Remote Sens. Environ.

    (1992)
  • P. Geladi et al.

    Partial least-squares regression: a tutorial

    Anal. Chim. Acta

    (1986)
  • Z. Huang et al.

    Estimating foliage nitrogen concentration from HYMAP data using continuum removal analysis

    Remote Sens. Environ.

    (2004)
  • D. Huang et al.

    Canopy spectral invariants for remote sensing and model applications

    Remote Sens. Environ.

    (2007)
  • Y. Inoue et al.

    Diagnostic mapping of canopy nitrogen content in rice based on hyperspectral measurements

    Remote Sens. Environ.

    (2012)
  • S. Jacquemoud et al.

    PROSPECT: a model of leaf optical properties spectra

    Remote Sens. Environ.

    (1990)
  • S. Jacquemoud et al.

    Estimating leaf biochemistry using the PROSPECT leaf optical properties model

    Remote Sens. Environ.

    (1996)
  • Y. Knyazikhin et al.

    Canopy spectral invariants. Part 1: a new concept in remote sensing of vegetation

    J. Quant. Spectrosc. Radiat. Transf.

    (2011)
  • R.F. Kokaly et al.

    Spectroscopic determination of leaf biochemistry using band-depth analysis of absorption features and stepwise multiple linear regression

    Remote Sens. Environ.

    (1999)
  • R.F. Kokaly et al.

    Characterizing canopy biochemistry from imaging spectroscopy and its application to ecosystem studies

    Remote Sens. Environ.

    (2009)
  • G. le Maire et al.

    Calibration and validation of hyperspectral indices for the estimation of broadleaved forest leaf chlorophyll content, leaf mass per area, leaf area index and leaf canopy biomass

    Remote Sens. Environ.

    (2008)
  • S.G. Leblanc et al.

    Methodology comparison for canopy structure parameters extraction from digital hemispherical photography in boreal forests

    Agric. For. Meteorol.

    (2005)
  • P. Lewis et al.

    Spectral invariants and scattering across multiple scales from within-leaf to canopy

    Remote Sens. Environ.

    (2007)
  • C. Macfarlane

    Classification method of mixed pixels does not affect canopy metrics from digital images of forest overstorey

    Agric. For. Meteorol.

    (2011)
  • Z. Malenovsky et al.

    Sentinels for science: potential of Sentinel-1, −2, and −3 missions for scientific observations of ocean, cryosphere and land

    Remote Sens. Environ.

    (2012)
  • M.E. Martin et al.

    A generalizable method for remote sensing of canopy nitrogen across a wide range of forest ecosystems

    Remote Sens. Environ.

    (2008)
  • O. Mutanga et al.

    Predicting in situ pasture quality in the Kruger National Park, South Africa: using continuum-removed absorption features

    Remote Sens. Environ.

    (2004)
  • P.A. Pellissier et al.

    Remote sensing of foliar nitrogen in cultivated grasslands of human dominated landscapes

    Remote Sens. Environ.

    (2015)
  • M. Schlerf et al.

    Remote sensing of forest biophysical variables using HyMap imaging spectrometer data

    Remote Sens. Environ.

    (2005)
  • M.A. Schull et al.

    Canopy spectral invariants, Part 2: application to classification of forest types from hyperspectral data

    J. Quant. Spectrosc. Radiat. Transf.

    (2011)
  • W. Shi et al.

    Wavelet-based image fusion and quality assessment

    Int. J. Appl. Earth Obs. Geoinform.

    (2005)
  • A.K. Skidmore et al.

    Forage quality of savannas—simultaneously mapping foliar protein and polyphenols for trees and grass using hyperspectral imagery

    Remote Sens. Environ.

    (2010)
  • S. Smolander et al.

    Simple parameterizations of the radiation budget of uniform broadleaved and coniferous canopies

    Remote Sens. Environ.

    (2005)
  • S. Ullah et al.

    An accurate retrieval of leaf water content from mid to thermal infrared spectra using continuous wavelet analysis

    Sci. Total Environ.

    (2012)
  • Z. Wang et al.

    Applicability of the PROSPECT model for estimating protein and cellulose + lignin in fresh leaves

    Remote Sens. Environ.

    (2015)
  • W. Woodgate et al.

    Understanding the variability in ground-based methods for retrieving canopy openness, gap fraction, and leaf area index in diverse forest systems

    Agric. For. Meteorol.

    (2015)
  • P.J. Zarco-Tejada et al.

    Needle chlorophyll content estimation through model inversion using hyperspectral data from boreal conifer forest canopies

    Remote Sens. Environ.

    (2004)
  • K. Zhao et al.

    Hyperspectral remote sensing of plant biochemistry using Bayesian model averaging with variable and band selection

    Remote Sens. Environ.

    (2013)
  • C. Axelsson et al.

    Hyperspectral analysis of mangrove foliar chemistry using PLSR and support vector regression

    Int. J. Remote Sens.

    (2013)
  • J. Bring

    How to standardize regression coefficients

    Am. Statistician

    (1994)
  • Cited by (49)

    View all citing articles on Scopus
    View full text