Advertisement

Surveys in Geophysics

, Volume 40, Issue 3, pp 553–588 | Cite as

Imaging Spectroscopy of Forest Ecosystems: Perspectives for the Use of Space-borne Hyperspectral Earth Observation Systems

  • Joachim HillEmail author
  • Henning Buddenbaum
  • Philip A. Townsend
Article

Abstract

The emerging challenges in preserving and managing forest ecosystems are multiscale in terms of space and time, and therefore require spatially and temporally contiguous information sources. Imaging spectroscopy has the potential to contribute information that cannot be raised by other Earth Observation Systems. In particular, the spectral capacity to monitor the distributions of chemical traits, such as canopy foliar nitrogen distribution, and to track changes in water content or the percentage water in plants, has already opened novel pathways toward assessing the global variability of ecosystem functions and services. However, there is an ongoing debate on how to best extract this type of information from the spectral measurements. Empirical approaches have demonstrated their efficiency in a multitude of local studies, but are criticized with respect to poor generalization capacities. Alternative strategies, such as the use of physically based models of leaf or canopy reflectance, or hybrid approaches, have the potential advantage to be more widely applicable. This paper attempts to assess achievements and shortcomings of these strategies and finds that the often-cited disadvantages of using empirical approaches are becoming less pronounced in the light of recent research results. While retrievals based on physically based models on leaf/needle level are close to laboratory quality, results on canopy level available to date still have considerable deficits. Owing to improved instrumental designs, better data calibration, new approaches for compensating canopy effects, and the use of increasingly efficient methods for establishing data-driven models, the scope of empirical approaches has considerably widened and they have been successfully applied to large areas. The future availability of regularly acquired hyperspectral imagery from Earth orbits will substantially contribute to their generalizability.

Keywords

Imaging spectroscopy Forest ecosystems Biochemical traits Retrieval strategies 

1 Introduction

Forest ecosystems provide numerous environmental goods and services, such as conservation of biological diversity and climatic control, and, of course, timber and non-timber products (FAO 2010). However, forests and forested ecosystems face numerous risks under increasing pressure due to global warming (Birdsey and Pan 2011; Peng et al. 2011; Swann et al. 2018) and expanding human populations and economies (Hansen et al. 2008). Deforestation associated with conversion of forests to agricultural land and development, legal and illegal logging, drought stress, biotic stress, and increasing wildfire frequency, extent and severity are some of the most important processes which affect forested landscapes (Anderegg et al. 2013a, b; Bond et al. 2010; Ciais et al. 2005).

The European Environmental Agency expects forests in Europe to be among the ecosystems most affected by climate change (EEA 2012). Examples of recent die-offs have been reported from southern and central parts of Europe (e.g., Adams et al. 2017; Bréda 2003). It is expected that forests will become increasingly affected by drought, either during exceptional events or under a long-term drift toward more arid conditions (Ma et al. 2012). Extreme climatic events, such as the heat waves and dry spells experienced during the European summers 2003 and 2018, are expected to occur at increased frequencies (Beniston 2004; Fink et al. 2004). Diffuse tree mortality within forest stands, frequently reported after such extreme events, may be exacerbated by soil properties (mainly local water storage capacity) or decreases in genetic diversity among and within tree species (Lindner et al. 2010; Rennenberg et al. 2006). Due to the direct impact of heat waves and dry spells, forest ecosystems are also becoming more susceptible to pests and diseases. The combined effects bear the risk of a reduction of important ecosystem services, such as carbon sequestration, water retention, soil protection, preserving biodiversity and, last but not least, biomass and timber production (Lindner et al. 2010; Schelhaas et al. 2003; Seidl et al. 2011).

In order to keep track of these effects, monitoring of forest ecosystems is a necessity, both in situ and through remote sensing. While in situ terrestrial forest monitoring usually employs surveys at plot level, logistically it is not feasible to make measurements everywhere that data are needed. Remote sensing offers the potential to sample with reasonable accuracy in inaccessible locations (Asner et al. 2015) while also providing wall-to-wall information at multiple temporal and spatial scales. The emerging challenges in forest management, intricately linked to society’s needs to preserve multiple forest values and continue to benefit from all services of forest ecosystems, are multiscale in terms of space and time and therefore need spatially and temporally contiguous information sources. Moreover, since sustainable forest management aims at reconciling economic interests with ecologic concerns, remote sensing must serve both, economically oriented assessments and management requirements, as well as studies of ecological processes and functions (Franklin 2001).

Lausch et al. (2018, 2016, and 2017) summarize this wide scope under the term “forest health,” whose proper assessment requires utilitarian as well as ecosystem indicators to be implemented. In their review papers, they provide a detailed summary of indicator methods and strategies to assess forest health. These include various types of Earth Observation systems.

Applications of remote sensing contributing to sustainable forest management are generally presented in four categories that include classification of forest type (i.e., tree species), the assessment of forest structure and available resources (e.g., timber volume, height, age, crown closure), modeling forest ecosystem processes, and forest change detection. For each category, measurable indicators are needed to quantify the effects of management activities and natural phenomena on the sustainability of forest ecosystems on large spatial scales. Some of these indicators can be derived best by imaging spectroscopy data.

We are currently on the brink of a new age of space-borne imaging spectrometers. While the first space-borne imaging spectrometers, CHRIS–Proba and Hyperion, were not able to create a sustainable impact on operational forest remote sensing, the next generation is expected to more reliably deliver a higher data quality. Recently, the imaging spectrometer DESIS (Eckardt et al. 2015) has been installed in the International Space Station. In the near future, several space-borne imaging spectrometers, such as PRISMA (Labate et al. 2009), SHALOM (Feingersh and Ben Dor 2015), HISUI (Matsunaga et al. 2017), EnMAP (Guanter et al. 2015), HYPXIM (Carrere et al. 2013), Surface Biology and Geology (SBG, formerly known as HyspIRI) (Lee et al. 2015), or the recently announced Earth Surface Mineral Dust Source Investigation (EMIT) will be launched. In the Copernicus program, the hyperspectral mission CHIME is in consideration (Nieke and Rast 2018). During the preparation programs to these sensors, especially EnMAP and SBG, large-scale airborne campaigns, have been carried out. So now a few large-scale data sets are available, including some at multiple points in time. Given the unprecedented range of future remote sensing systems in space, an essential question is which of the spectral traits can be better characterized with high spectral, medium spatial resolution instruments. The first instruments will cover only limited areas and are thus expected to be used primarily in scientific contexts. Products developed with scientific instruments may be of general interest for forest practitioners once operational wall-to-wall space-borne imaging spectroscopy becomes available with SBG and CHIME. In the following, we will first highlight which role imaging spectroscopy (IS) data is likely to play in three key application fields of forestry and then outline the scope of this study.

1.1 Tree Species Composition

Spatially explicit information on tree species composition of managed and natural forests is relevant for biomass or timber volume and quality estimation, habitat quality assessment, and biodiversity characterization. It thus provides valuable information for nature conservationists as well as for forest managers and is frequently required over large spatial extents. Assuming the same spatial resolution, hyperspectral imaging certainly has the potential to provide higher classification accuracies for forest species mapping than multispectral data (e.g., Asner and Martin 2009; Fassnacht et al. 2014; Foster and Townsend 2004), especially when implemented at local scales. As well, the range of advanced spectral classifiers has been substantially expanded so that it is now possible to make the best use of the available data (e.g., Ghamisi et al. 2017). However, a high degree of species diversity does not necessarily lead to an equally high diversity of traits (Lausch et al. 2016), and large area species mapping is rather boosted by phenological dynamics observed by operational multi- or super-spectral observation systems (Somers and Asner 2013; Stoffels et al. 2015). Considering this and the limited data availability, it is not surprising that only few examples for using hyperspectral data for successful tree species classifications over large and diverse geographic areas have been presented (Fassnacht et al. 2016; Sommer et al. 2016). On the one hand, this might change as soon as space-borne imaging spectrometers will regularly cover large areas. On the other hand, the spatial resolution of proposed sensors may not be sufficient for species identification, especially if species in mixed stands are to be mapped, thus requiring mapping of ecological communities (e.g., Foster and Townsend 2004).

1.2 Structural Traits

2D or 3D structural spectral traits depending on leaf arrangement and geometry, tree height or area, density, size, the shape of forest patches as well as fragmentation, complexity, and homogeneity are important characteristics of forests that strongly influence the spectral response of forest canopies. Moreover, forest canopy height, forest extent, and the vertical and horizontal vegetation structure of forests are important proxies for assessing standing timber volume (Nink et al. 2015). There is evidence that multi-resolution multispectral imagery can be used to estimate forest structure based on texture (Wolter et al. 2009) which likely also applies to hyperspectral data, but not with the accuracy and physical realism of active remote sensing data. Given the physical limitations of addressing structure and biomass information over closed canopies with mono-directional, passive optical systems, it is not surprising that hyperspectral data alone will contribute little that goes beyond the information provided by active remote sensing such as airborne laser scanning and synthetic aperture radar (Swatantran et al. 2011).

1.3 Detection of Stress and Biochemical Traits

Many studies emphasize the importance of hyperspectral data for characterizing the physiological state of canopies, for example those under stress (e.g., Abdullah et al. 2018; Asner et al. 2004; Dotzler et al. 2015; Lausch et al. 2016; Paz-Kagan et al. 2018). As such, one of the most important application domains of hyperspectral imaging is the observation of biochemical traits. Alterations to biochemical traits such as cellulose, protein, and lignin, changes in the composition and configuration of photosynthetically active pigments in leaves (chlorophyll, xanthophyll, carotenoids) are highly relevant indicators for climate- and site-specific stress conditions, disturbances or resource limitations (Asner et al. 2015; Féret et al. 2017; Singh et al. 2015), although their detection may be hampered by leaf structural changes such as rolling (Chávez et al. 2013). Accurate retrieval of foliar nitrogen, for example, could be used to detect nutrient deficiencies, but also enhance ecosystem process models that describe ecosystem functioning, since nitrogen is an important input parameter of these models. Foliar N is a primary regulator of physiological processes (Wang et al. 2017) and an important indicator of photosynthetic and growth rates (Townsend et al. 2003) since it reflects the investment in photosynthetic enzymes, structure, and light-harvesting complexes of a plant (Singh et al. 2015). Most foliar N (70–80%) is stored in proteins, and less in chlorophyll (Hank et al. 2019; Huber et al. 2008; Schlerf et al. 2010).

The spectral capacity to monitor the distributions of biochemical traits, such as canopy foliar N distribution (Knyazikhin et al. 2013; Schlerf et al. 2010; Serbin et al. 2014; Singh et al. 2015; Townsend et al. 2003), and changes in the water content (Asner et al. 2016; Paz-Kagan et al. 2018), has already revealed the potential to assess indicators for ecosystem functioning on extended spatial dimensions. Pigment-related trait variations also support the observation of shifts in the photosynthesis system (Gamon et al. 1992; Garbulsky et al. 2011; Peñuelas et al. 1995; Suárez et al. 2008), although unveiling the causal relationships behind diurnal, seasonal, and spatial variations of photosynthesis will require instruments with high repetition rates and exceptionally high spectral resolution, capable of tracking changes in the solar-induced chlorophyll fluorescence (Drusch et al. 2017; Rascher et al. 2015).

1.4 Scope of the Paper

To make the best use of space-borne imaging spectroscopy data, robust retrieval algorithms that work over large geographic and temporal scales, are needed. Ideally, these should also output uncertainties for all values retrieved. While in the past many imaging spectroscopy measurements were singular studies, in the age of operational space-borne sensors, operational algorithms are required. The way to estimate spectral traits and their variation is intricately linked to the fundamental debate about empirical and physically based retrieval strategies, discussed in the following section.

Since neither tree species classification (with present technologies and data) nor the determination of structural traits is substantially improved by imaging spectroscopy, we will focus on chemical traits. A main scope will be on the comparison of empirical, physically based, and hybrid modeling on the leaf and canopy scale and the question whether empirical approaches can also be generalized to an extent that they become applicable to a wider range of conditions. There are a number of reasons to focus on biochemical traits and the generalizability of retrieval approaches: the availability of radiometrically well-calibrated data and the level-2 processing capacities have constantly improved (Frantz et al. 2016; Masek et al. 2006; Storch et al. 2013; Thompson et al. 2018, 2019; Vermote et al. 2016). In addition, the hyperspectral observation capacities have been increasing during the past years and will further increase with the advent of hyperspectral satellites. Last, but not least, the efficiency of algorithms for building adequate models has been boosted by a range of powerful machine learning algorithms (Fassnacht et al. 2014; Hultquist et al. 2014; Paz-Kagan et al. 2018; Waske et al. 2012).

Here, we provide a review of the literature, with some points illustrated with measurements recently conducted by the authors. The second section provides a comparison of empirical versus physically based modeling; Sect. 3 deals with studies on the estimation of biochemical traits on the leaf and canopy scale, with Sect. 4 providing discussion and conclusions.

2 Empirical Versus Physically Based Modeling

2.1 Empirical Modeling

In remote sensing, quantitative forest variables like structure parameters and foliar chemical content are estimated either empirically, physically based, or in a hybrid approach (Fig. 1, Féret et al. 2017; Li et al. 2018; Verrelst et al. 2019). Hybrid approaches combine both strategies, e.g., by developing a spectral index based on artificial data created with forward simulations of a radiative transfer model (le Maire et al. 2004, 2008). Verrelst et al. (2019) give a comprehensive overview on retrieval methods.
Fig. 1

Principal workflows of empirical and physically based approaches for estimating biochemical leaf traits on leaf and canopy level

Empirical models, both parametric and nonparametric, aim at building a statistical relationship between measured traits and an experimental reflectance database. However, these empirical relationships are considered limited in transferability and robustness because of their dependency on the particular conditions under which they were established (field data collection strategy, species diversity, instrumental characteristics, date of acquisition, phenological state of the vegetation, atmospheric, and lighting conditions) (Féret et al. 2017; Li and Wang 2011; Verrelst et al. 2015). Establishing empirical relationship builds on regression algorithms (e.g., multivariate linear regression, principal component regression, partial least-squares regression), machine learning techniques (support vector machines, import vector machines, artificial neural networks, deep learning networks, random forests, or Gaussian processes regression), or unmixing approaches, such as multiple endmember spectral mixture analysis (Roberts et al. 1998) or iterative spectral mixture analysis (Rogge et al. 2006).

In recent years, partial least-squares regression (PLSR) has emerged as the most popular parametric empirical method that can be used to estimate variables of interest even if the explaining variables (spectral reflectance) are redundant and collinear (Atzberger et al. 2010; Wold et al. 2001). The popularity of PLSR is related to its long history (Wold et al. 1984), ease of implementation, relative robustness to new data, and operational computational speed relative to deep learning methods. PLSR generates comparable results to machine learning methods such as Gaussian processes (Wang et al. 2019), but uncertainties need to be assessed through permutation rather than direct calculation (Singh et al. 2015). Asner et al. (2011) and Martin et al. (2018a), for example, used PLSR to measure a wide range of canopy chemicals using field and laser-guided airborne spectroscopy. On the machine learning side (nonparametric empirical models), deep learning networks are increasing in popularity due to improved computational capacities, but other machine learning approaches are also successfully employed. For example, Hultquist et al. (2014) assessed the post-fire burn severity with airborne MASTER data and found that random forest, support vector regression, and Gaussian process regression outperformed conventional multiple regression by 48%, 29%, and 27%, respectively.

Because of the usually high number of inter-correlated spectral channels, it is beneficial for most empirical modeling approaches to reduce the number of spectral features (feature reduction), either by selecting a subset of important bands (feature selection) or by transforming original bands into new features with maximal correlation to the dependent variable of interest (feature extraction) (Fassnacht et al. 2014; Held et al. 2015). Main feature extraction strategies include spectral indices (Féret et al. 2011), wavelets (Cheng et al. 2014; Wang et al. 2017), or linear transformations such as principal components or minimum noise fractions (Green et al. 1998), and singular value decomposition (Rogge et al. 2014). In feature selection approaches, only significant bands are used before establishing a regression model. Popular techniques include stepwise multiple regression (Schlerf et al. 2010), competitive adaptive reweighted sampling (CARS) (Vohland et al. 2016), and band selection based on variable importance in the projection (VIP) score (Abdullah et al. 2018; Heim et al. 2015). Empirical models may also be improved by spectral pre-treatment techniques like first and second derivative, transformation from reflectance to absorption, z-transformation, or Savitzky–Golay smoothing (Buddenbaum and Steffens 2012; Serbin et al. 2014).

2.2 Physically Based Modeling

Physically based modeling approaches aim at overcoming the limitations in transferability and robustness by simulating the radiative transfer between light source, canopy, and sensor (Feret et al. 2008). Leaf and canopy reflectance models predict the directional spectral reflectance based on radiation principles and a mathematical description of absorption and scattering processes in leaves, needles, or complete canopies. Foliar or canopy traits are then retrieved from spectral measurements by inverting the models (Combal et al. 2002; Jacquemoud and Baret 1990; Jacquemoud et al. 2009; Koetz et al. 2004; Schlerf and Atzberger 2006; Wang et al. 2018).

The physically based approach involves numerous strategies of varying complexity. There are models for leaf or needle optical properties, 1-D models for homogeneous canopies, and 3-D models for heterogeneous stands. Among the models of leaf optical properties (Miller et al. 2005) the by far the most widely used representative is PROSPECT. While early versions of PROSPECT only considered Chlorophyll a and b (Jacquemoud et al. 2000; Jacquemoud and Baret 1990; Jacquemoud et al. 1996), PROSPECT-4 and -5 added carotenoid content (Feret et al. 2008), and PROSPECT-D added anthocyanins (Féret et al. 2017). The PROSPECT model family has been designed for broadleaf species, but can also be recalibrated for needles (Malenovský et al. 2006a). A popular alternative for needles is LIBERTY (Dawson et al. 1998).

The most simplistic model to up-scale leaf measurements to canopy scale is the Lillesaeter model (Lillesaeter 1982). The model calculates canopy reflectance as ρ/(1–τ2), where ρ is leaf reflectance and τ is leaf transmittance. Leaf-scale reflectance models are often combined with canopy models to account for effects like leaf inclination, multiple scattering of several leaf layers, background effects, and illumination and observation geometry. The SAIL (Scattering by Arbitrarily Inclined Leaves) model (Verhoef 1984) is commonly combined with PROSPECT to simulate reflectance of homogeneous canopies (Berger et al. 2018; Jacquemoud et al. 2009). Since forest canopies tend to be more complicated, further model extensions like GeoSail (Huemmrich 2001), PROFLAIR (Omari et al. 2013), and InFoRM (Atzberger 2000; Schlerf and Atzberger 2006) add realism related to discontinuous canopies, varying tree geometry and density and multiple scattering. More advanced forest reflectance models such as FRT (Kuusk et al. 2014), FLIGHT (Barton and North 2001; North 1996), and 3D ray tracing models like DART (Gastellu-Etchegorry et al. 2004) permit even more differentiated and detailed parametrization of canopy architectures.

The inversion of these models is based either on numerical algorithms, or on searching through lookup tables (LUT) which have been created by using the model in forward mode. Owing to the structural complexity of forest canopies with their diverse architectures, the inversion problem is complex and becomes frequently ill-posed with increasingly complicated models, i.e., different combinations of input parameters may produce almost identical spectra in the forward mode (Combal et al. 2002).

A typical strategy (e.g., Bicheron and Leroy 1999) is to fix some the model parameters to “reasonable” values, and to determine the remaining (free) parameters by minimizing a merit function, i.e., a function of difference between observed and modeled reflectance.

The LUT-based approach identifies the best-fitting spectrum from the collective of pre-computed solutions; the corresponding parameters represent the inversion result. The LUT has to be created just once, and only the relatively simple task of finding the best match has to be repeated for each spectrum. This can be much faster, especially for large data sets (e.g., a hyperspectral image) and computationally complex models. But again, if a large number of free parameters need to be determined, LUTs tend to become too large and this may as well lead to ill-posedness. Since the best-fitting spectrum does not always yield the best set of parameters, a range of similarly good fitting spectra can be selected and their mean or median accepted as the inversion result; their standard deviation then provides a measure of uncertainty (Koetz et al. 2007; Locherer et al. 2015).

Figure 1 illustrates the alternative workflows of estimating biochemical traits on leaf and canopy level via empirical or physically based methods. The desired estimations of leaf biochemistry can be obtained directly from the measured canopy reflectance through inversion of a physically based model. If empirical models on canopy levels are to be used, measured leaf biochemistry values are needed that have to be up-scaled to canopy level. It may also be advisable to transform the spectral data using approaches like double ratios or directional area scattering factor (DASF, Knyazikhin et al. 2013, see Sect. 4). A different pathway to estimating leaf biochemistry on canopy level can be taken through measurements of leaf reflectance. Since reflectance measurements are faster and cheaper than chemical analyses they can be used to create a large ensemble of leaf biochemical values that can be up-scaled to canopy level.

3 Identifying Biochemical Traits on Leaf and Canopy Level

Spectroscopic studies on the leaf level serve a number of purposes: first, they can supplement or even replace laboratory analyses and help creating reference data sets for studies on canopy level. Second, they provide a basis to understand interactions between plants and radiation without confounding effects of the canopy structure. Generally, identifying biochemical traits is much more straightforward on leaf level than on canopy level (Ustin et al. 2009), although the structural contributions at the canopy level can reinforce empirical relationships at the expense of physical realism (Townsend et al. 2013). Third, results on leaf level can be up-scaled to canopy level (Gara et al. 2018; Malenovský et al. in review).

Reflectance measurements of broadleaves are ideally acquired using an integrating sphere (Asner et al. 2011; Jay et al. 2016; Kokaly et al. 2009), although measurements with a leaf clip are fast and popular. Needle reflectance is a little harder to determine in a reliable way. Malenovský et al. (2006a), Yáñez-Rausell et al. (2014a, b), and Hovi et al. (2017) have presented studies on how to measure needle reflectance with integrating spheres. Alternative ways to determine needle reflectance include measuring needle stacks or mats (Dawson et al. 1998; Schlerf et al. 2010).

A considerable range of biochemical and structural forest variables have been examined using (imaging) spectroscopy (Homolová et al. 2013). They are either expressed as concentrations (per leaf mass) or as content (per leaf area) (Homolová et al. 2013). Since the content of a light-absorbing leaf constituent can be linked to leaf reflectivity via Lambert–Beer’s law, this measure is preferred in remote sensing contexts, especially if the value is to be up-scaled to canopy level via leaf area index. Figure 2 shows some of the specific absorption coefficients of pigments and other main optically active leaf ingredients. In the visible part of the spectrum (Fig. 2a), chlorophyll is usually the dominant pigment with strong absorption in the blue and red spectral regions that leads to leaves’ green color. Carotenoids (yellow pigments) and anthocyanins (red pigments) dominate in senescent leaves (Féret et al. 2017). The near-infrared region has weak water absorption features around 975 and 1200 nm, strong water absorption features occur around 1450, 1930, and 2600 nm (Asner 1998) (Fig. 2b). Wang et al. (2015) have separated the absorption coefficients of dry matter into cellulose + lignin and nitrogen-containing protein (Fig. 2c), and Curran (1989) has an extensive listing of absorption features associated with specific molecular bonds.
Fig. 2

a, b Specific absorption coefficients of main absorbing leaf constituents according to the PROSPECT-D leaf optical model. Inset in subfigure b shows weak water absorption bands near 975 and 1200 nm. Note that the absorption coefficients for brown pigments are given in arbitrary units. c Absorption coefficients of protein and lignin + cellulose according to Wang et al. (2015)

Both empirical models and the inversion of physically based reflectance models for analyzing biochemical traits on the level of leaves and needles have already reached quite a high level of accuracy, in particular for pigments, water content, leaf mass per area, but not for cellulose, lignin, and protein in fresh leaves (Gitelson et al. 2006; Jay et al. 2016; Li et al. 2018; Malenovský et al. 2006a; Wang et al. 2015, 2018).

Asner et al. (2012, 2017) propose seven leaf traits that are functionally relevant to forest diversity and mostly uncorrelated (Asner et al. 2017). Foliar nitrogen (N), water, and LMA, which play important roles in photosynthesis and primary production, have been characterized remotely in various studies (Ustin et al. 2004). Polyphenols act as foliar defense compounds (Kokaly et al. 2009), while lignin is an important leaf structural compound that is recalcitrant to decomposition. Foliar phosphorus (P) and calcium (Ca) are related to species community turnover in tropical forests and have also been mapped by IS (Asner and Martin 2016; Chadwick and Asner 2016), although relationships with spectra appear correlative and unrelated to currently known absorption features in the VSWIR. As considering the full range of traits is beyond the scope of this paper, we will focus on foliar pigment, water, and nitrogen content.

On the level of tree canopies, either the integrated content of a biochemical trait per ground area can be measured (e.g., canopy water content, CWC, or canopy chlorophyll content, CCC), or the respective content per leaf area (e.g., EWT or leaf chlorophyll content, LCC). These are related by the leaf area index (LAI), i.e., \( {\text{LCC}} \cdot {\text{LAI}} = {\text{CCC}} \) and \( {\text{EWT}} \cdot {\text{LAI}} = {\text{CWC}} \) (Hank et al. 2019). Where LAI is not estimated, estimates generally refer to traits at the top of the canopy (Singh et al. 2015); note also that some traits vary with canopy position (especially LMA, but not so much N), so that any calibration or validation of an integrated canopy estimate must take into consideration within-canopy variation. The retrieval of biochemical traits of leaves is also substantially compromised through structural properties of the crown layer and the ground reflectance that lead to nonlinear signal mixing through multiple scattering (Ewald et al. 2018). Various approaches for compensating these effects have been presented. These include inversion of physically based reflectance models and approaches like the normalization of spectra through a spectral index correlated with leaf area (double-ratio approach, Colombo et al. 2008) or by physically based normalization factors such as the directional area scattering factor (DASF, Knyazikhin et al. 2013). The selection of illuminated crown pixels is a valid strategy to be applied to either approach, provided the spatial resolution is sufficiently detailed (Asner et al. 2015; Dotzler et al. 2015).

Dotzler et al. (2015) demonstrated the importance of canopy-related effects when investigating the potential of EnMAP and Sentinel-2 to detect early drought stress effects in deciduous forests. Differences between stands on different soil moisture regimes that were statistically significant in moisture stress index (MSI), normalized difference water index (NDWI), and chlorophyll index (CI) disappeared after these indices were normalized to leaf level by using the double-ratio approach of Colombo et al. (2008). The comparison with data from airborne laser scanning suggested that the moisture and pigment differences detected on canopy level were actually due to variations in crown density and volume. The detection of drought-related stress symptoms on leaf level therefore requires that canopy effects be addressed and removed, in particular when spectral information from spectral regions with high transmission and scattering characteristics is used (especially near infrared).

On canopy level, the retrievals of biochemical and structural traits through inversion of physically based models are so far less successful than on leaf level. While most geometric-optical canopy reflectance models [e.g., InFoRM (Atzberger 2000; Schlerf and Atzberger 2006)] are surprisingly good in reproducing measured canopy reflectance as long as important canopy properties (leaf area index, leaf angle distribution, etc.) are known (forward mode), their performance tends to break down in the inversion (where these parameters must be retrieved simultaneously with the biochemical traits).

We summarize accuracies of cited studies in tables. Each study reports the accuracies in a different way and has been conducted under different conditions, so these studies should not be compared with each other directly. Since the coefficient of determination (R2) is the only accuracy measure reported in nearly all studies, we give R2 values of all studies cited although they are quite dependent on the range of values considered. If the root-mean-square error (RMSE) or the RMSE relative to the mean value or the range of values is also reported, that can be found in the respective table. From a statistical standpoint, achievement of a low RMSE and seemingly poor R2 is common when the data range is limited (e.g., often seen with the prediction of carbon concentration, Singh et al. 2015). The resulting maps may exhibit reasonable uncertainty based on RMSE, but the models will be unlikely to be extensible.

3.1 Pigment Content Retrieval

Pigment content drives light harvesting to provide the energy used in photosynthesis to drive carbon fixation of plants and is a primary controller of leaf reflectance and transmittance in visible wavelengths, and thus, stand reflectance (Ustin et al. 2009). Independent of using empirical or physically based approaches, most pigment contents on leaf level are retrieved from spectral measurements with moderate to high precision. The accuracies of the studies cited here are summarized in Table 1.
Table 1

Accuracies and further information of pigment estimation studies

Study

Pigment

Unit

Range

Accuracy

Species

Method

Min

Max

Mean

R 2

RMSE

RMSE%

Leaf level

Gitelson and Merzlyak (1997)

Chlorophyll a + b

µg/cm2

0.27

62.9

 

≥ 0.9

≤ 5.3a

 

8 Different species

Indices

Gitelson et al. (2003)

Chlorophyll a + b

µmol/cm2

1

832

 

≥ 0.95

≤ 75

 

8 Different species

Indices

Gitelson et al. (2006)

Chlorophyll a + b

mg/m2

  

83–408

≥ 0.91

≤ 51

< 25

6 Different species

Indices

Gitelson et al. (2006)

Carotenoids

mg/m2

16

166

50–96

≥ 0.70

≤ 17.2

≤ 19

3 Different species

Indices

Gitelson et al. (2006)

Anthocyanins

mg/m2

   

> 0.93

  

2 Different species

Indices

Fassnacht et al. (2015)

Carotenoids

mg/m2

15.96

137.2

63.38

0.88

 

8.12

3 Tree species

Index fusion

Asner et al. (2011)

Chlorophyll a

mg/g

0.83

15.99

4.78b

0.83

 

6.2

Many tropical tree species

PLSR on transm

Asner et al. (2011)

Chlorophyll b

mg/g

0.29

5.83

1.78b

0.82

 

6.2

Many tropical tree species

PLSR on transm

Asner et al. (2011)

Carotenoids

mg/g

0.35

4.32

1.41b

0.79

 

8.1

Many tropical tree species

PLSR on transm

Abdullah et al. (2018)

Chlorophyll a + b

mg/g

0.4

1.1

 

0.64

 

0.24c

Healthy Spruce Needles

PLSR

Abdullah et al. (2018)

Chlorophyll a + b

mg/g

0.2

1

 

0.55

 

0.75c

Infested Spruce Needles

PLSR

Malenovský et al. (2006a)

Chlorophyll a + b

µg/cm2

25.57

85.1

  

8.05

 

Norway spruce

Modified PROSPECT-3 inv.

Féret et al. (2011)

Chlorophyll a + b

µg/cm2

0.3

106.72

32.81

 

5.38d

 

Wide range of species

PLSR

Féret et al. (2011)

Carotenoids

µg/cm2

0.04

25.3

8.51

 

1.9d

 

Wide range of species

PLSR

Buddenbaum et al. (2011)

Chlorophyll a + b

µg/cm2

28.73

82.35

57.75

0.7764

15.4592

26.77

European Beech

PROSPECT-3 inversion

Buddenbaum et al. (2011)

Chlorophyll a + b

µg/cm2

48.13

102.11

78.87

0.5351

8.951

11.35

Oak

PROSPECT-3 inversion

Buddenbaum et al. (2012)

Chlorophyll a + b

µg/cm2

7

29

 

0.73

 

20.9

European Beech

PROSPECT-5b inversion

Schlerf et al. (2010)

Chlorophyll a + b

mg/g

1.8

4.94

3.21

0.83

0.26

 

Norway spruce

SMR, CM

Canopy level

Malenovský et al. (2006b)

Chlorophyll a + b

µg/cm2

28

94

 

0.72

9.53

 

Norway spruce

ANMB605-725 index

Schlerf et al. (2010)

Chlorophyll a + b

mg/g

2.71

3.62

3.21

0.8

0.13

 

Norway spruce

SMR, CM

Asner et al. (2015)

Chlorophyll a + b

mg/g

2.65

8.53

 

0.58 ± 0.05

1.5 ± 0.71

28.14

Many tropical tree species

PLSR

Asner et al. (2015)

Carotenoids

mg/g

0.69

1.84

 

0.49 ± 0.06

0.25 ± 0.08

21.2

Many tropical tree species

PLSR

Martin et al. (2018a)

Chlorophyll a + b

mg/g

2

10.3

5.6

0.21

1.68

29.8

Many tropical tree species

PLSR

Buddenbaum et al. (2015)

Chlorophyll a + b

µg/cm2

15

40

 

0.325

4.059

16.85

European Beech

PLSR

Omari et al. (2013)

Chlorophyll a + b

µg/cm2

39

56

 

0.2553

4.46

 

Trembling aspen, balsam poplar, jack pine

PROFLAIR inversion

aEstimation error

bMedian

cNormalized

dPLS-PRESS, LOOCV

Gitelson and Merzlyak (1997) and Gitelson et al. (2003) showed that simple spectral indices can be used for empirical pigment estimations. Gitelson et al. (2006) further proposed spectral three-band indices for estimating chlorophyll, carotenoids, and anthocyanin contents from leaf reflectance. Fassnacht et al. (2015) introduced a method for merging vegetation indices for carotenoid estimation over a large range of values. Carotenoid contents of the test data set ranged from 16 to 137 mg/m2 and the merged index could explain 88% of the variance. These index-based pigment estimations reach quite high accuracies and have been applied over a wide range of vegetation types (e.g., Beamish et al. 2018; Fawcett et al. 2018; Hernández-Clemente et al. 2012; Sonobe and Wang 2018). Alternatively, Asner et al. (2011) used full-spectrum empirical regression models (PLSR) to estimate numerous foliar traits of many tropical tree species on leaf level using reflectance and transmittance spectra. The highest accuracies for pigments were achieved using transmittance in the full spectral range of 400–2500 nm with R2 values of 0.83, 0.82, and 0.79 for chlorophyll a, chlorophyll b, and carotenoids, respectively. Feilhauer et al. (2015) conducted a multi-method ensemble selection of spectral bands and estimated chlorophyll concentrations of several sets of leaf measurements with large variations in R2 values. The highest accuracies were achieved using an optimized PLSR. Abdullah et al. (2018) used PLSR to estimate chlorophyll content of spruce needles from the reflectance in the 400 to 790 nm range to detect bark beetle attacks. In healthy needles, chlorophyll content was estimated with R2 = 0.64, in infested needles with R2 = 0.55.

High to very high leaf-level estimation accuracies are also achieved by inverting the PROSPECT model or by using it for hybrid approaches. Le Maire et al. (2004) used PROSPECT simulated spectra to derive universal leaf chlorophyll content spectral indices. In order to test the applicability of PROSPECT for spruce needles, Malenovský et al. (2006a) applied a constrained inversion of PROSPECT-3.01 and retrieved chlorophyll content with an RMSE of 8.05 µg/cm2. Féret et al. (2011) compared the performance of spectral indices, PLSR, and inversion of PROSPECT-5 for the estimation of chlorophylls and carotenoids on synthetic and real leaf reflectance and transmittance data. They found that the inversion approach is only reliable when both leaf reflectance and transmittance data are available, in which case, the accuracies were higher than using regression. Using PROSPECT-3 inversion, Buddenbaum et al. (2011) explained 0.78 and 0.54 of the variance of SPAD chlorophyll measurements of beech and oak leaves, respectively, that were collected by tree climbers from the crowns of old and young trees. In a greenhouse experiment, Buddenbaum et al. (2012) yielded R2 = 0.73 for leaf chlorophyll content of potted young beech trees by inverting PROSPECT-5B based on spectra measured with a leaf clip. Dechant et al. (2017) also measured reflectance spectra with a leaf clip and inverted PROSPECT-5B to determine chlorophyll, carotenoid, and water content of leaves instead of chemical analyses, assuming they achieve laboratory level precision.

Figure 3 shows the relationship between chemically determined chlorophyll a + b and carotenoid content of beech (Fagus sylvatica) leaves that have been collected by tree climbers in the Donnersberg area in Germany and PROSPECT-D inversion results from spectra recorded using a Spectral Evolution PSR-3500 spectroradiometer with a leaf clip (unpublished data from the authors of this paper). The inversion was done following the approach of Jay et al. (2016) using a numerical minimization in MATLAB (least-squares curve fit with a trust region reflective algorithm). While the chlorophyll inversion results are relatively unbiased, carotenoid content is overestimated by a factor of nearly 2. Whether the reason for this kind of bias lies in the laboratory analysis or in the inversion process is subject of ongoing research.
Fig. 3

PROSPECT-D inversion results versus laboratory measurements of chlorophyll a + b (left) and carotenoids (right)

On canopy level, the results from published studies are inconsistent and in some cases conflicting. Several studies report reasonably high estimation accuracies of foliar pigment content on canopy level (Malenovský et al. 2006b; Schlerf et al. 2010), while others report low accuracies. Malenovský et al. (2006b) reached R2 = 0.72 for canopy chlorophyll content estimation in a Norway spruce stand with a high variability of chlorophyll contents (27.5-–93.5 µg/cm2) using their ANMB605–725 index and only R2 = 0.17 using the established TCARI/OSAVI index that was originally designed for agricultural crops. Schlerf et al. (2010) used stepwise multiple regression for chlorophyll estimation on stand level in Norway spruce stands with chlorophyll concentrations of 2.7 to 3.6 mg/g dry matter and reached cross-validated R2 values between 0.46 and 0.9, depending on spectral transformation before the regression step. Asner et al. (2015) used airborne high-fidelity visible-to-shortwave infrared spectroscopy to estimate multiple foliar chemicals in tropical forests with PLSR to establish chemical fingerprints of numerous tree species. Independent field validation gave a mean R2 of 0.58 for chlorophyll estimation for several tropical tree species with chlorophyll concentrations of 2.65–8.53 mg/g and an R2 of 0.49 for carotenoid estimation in a range of 0.69–1.84 mg/g. Martin et al. (2018a) evaluated the methods of Asner et al. (2015) on data and field measurements of a wide array of canopies in Bornean forests of Sabah, Malaysia. Despite the strong chlorophyll absorption, the retrieval accuracy of chlorophyll concentration was among the lowest of all variables with R2 values of 0.43, 0.17, and 0.21, respectively, for model calibration, validation, and test for a range of 2–10.3 mg/g. Buddenbaum et al. (2015) used field-level IS and PLSR to estimate leaf chlorophyll content of beech seedlings with chlorophyll contents of 15 to 40 µg/cm2. The R2 only reached 0.325, but an interesting finding was that the regression coefficients were significantly different from zero only in the chlorophyll-dominated region of the spectrum, which showed that PLSR was able to determine important bands and ignore the bands outside chlorophyll absorption features.

A physically based model inversion approach for chlorophyll retrieval from Hyperion data is reported in Omari et al. (2013). They inverted the PROFLAIR model (Omari et al. 2009; White et al. 2001) and yielded a low RMSE but only R2 = 0.26 for LCC in a quite limited range of 40–55 µg/cm2. They ascribed the low R2 value to heterogeneity within the site, uncertainties in preprocessing, the sampling scheme, and the temporal shift between sampling and satellite data acquisition. Due to large LAI variations, CCC also had a larger range and was estimated with R2 = 0.60 after successful LAI retrieval.

3.2 Water Content

Water dominates total vegetation reflectance in the shortwave infrared region of the spectrum. Foliar water can be quantified relative to full turgor, to fresh or dry leaf mass or relative to leaf area. Since water content per area is directly related to light absorption in the leaf, this is the most common way of describing water content in spectroscopy-related studies (Colombo et al. 2008), e.g., use of equivalent water thickness (EWT, g/cm2 or cm) in the PROSPECT model. The fuel moisture content (FMC) is the ratio of EWT and dry matter content and is often used to quantify the susceptibility to fire (Riaño et al. 2005). Further information on the studies cited here is given in Table 2.
Table 2

Accuracies and further details of water content estimation studies

Study

Unit

Range

Accuracy

Species

Method

Min

Max

Mean

R 2

RMSE

RMSE%

Leaf level

Riaño et al. (2005)

EWT (cm)

0.001

0.05

 

0.94

  

LOPEX data set

PROSPECT inversion

Malenovský et al. (2006a)

EWT (cm)

0.0419

0.0701

  

0.0006

55

Norway spruce

PROSPECT-3.01.S inversion

Buddenbaum et al. (2011)

EWT (cm)

0.00263

0.0173

 

0.75

  

Beech and Oak

PROSPECT-3 inversion

Buddenbaum et al. (2012)

EWT (cm)

0.00125

0.005

0.004

0.87

 

11.3

Beech

PROSPECT-5b inversion

Colombo et al. (2008)

EWT (cm)

0.0091

0.0154

0.0122

0.65

  

Poplar

PROSPECT inversion

Cheng et al. (2011)

LWCD (%)

32.31

418.2

143.6

0.71

 

26.04

Lianas and trees

Continuous wavelet analysis

Cheng et al. (2011)

LWCF (%)

24.42

80.7

57.23

0.75

 

4.34

Lianas and trees

Continuous wavelet analysis

Fang et al. (2017)

EWT (cm)

0.0037

0.0525

0.0115

0.95

 

3.63

LOPEX data set

Spectral index (SWI Double SAC)

Fang et al. (2017)

EWT (cm)

0.0038

0.0255

0.0136

0.8

 

7.9

PANAMA data set

Spectral index (SWI Double SAC)

Canopy level

Riaño et al. (2005)

FMC (%)

0

200

 

0.81

  

Gall oak, rosemary, Rock Rose

PROSPECT–Lillesaeter inversion

Colombo et al. (2008)

mean leaf EWT (cm)

0.0117

0.0136

 

0.57

 

2.264

Poplar

Double ratio index MSI/SR

Colombo et al. (2008)

EWTcanopy (g m−2)

80

400

 

0.818

  

Poplar

PROSPECT–SAILH inversion

Buddenbaum et al. (2015)

EWT (cm)

0.001

0.008

 

0.7979

0.0007

9.951

European Beech

PLSR

Buddenbaum et al. (2012)

EWT (cm)

0.00125

0.005

0.004

0.88

  

European Beech

PLSR

Physically based model retrievals of EWT usually yield high accuracies on leaf level. Riaño et al. (2005) estimated EWT and FMC by inversion of radiative transfer models to quantify fire danger. When applied to the LOPEX (Hosgood et al. 1994) leaf data set, EWT and FMC were estimated with R2 = 0.94 and 0.89, respectively, by inverting PROSPECT. Malenovský et al. (2006a) used unconstrained and constrained inversions of both PROSPECT-3.01 and an adapted version called PROSPECT-3.01.S to retrieve EWT of spruce needles from VNIR spectra. While the unconstrained inversions yielded quite low accuracies, the constrained inversion results were very close to the measured values with RMSE = 0.0006 (R2 not reported). Buddenbaum et al. (2011) inverted PROSPECT-3 to estimate EWT of beech and oak leaves with R2 = 0.75, and Buddenbaum et al. (2012) achieved R2 = 0.87 with an inversion of PROSPECT-5B for beech leaves. The inversion results were used as a plausibility check to verify the reference measurements of EWT and the leaf spectra.

Similarly, Colombo et al. (2008) tested several spectral indices and regression approaches to estimate EWT on leaf level. Reduced major axis regression exploiting the 1200 nm water absorption feature performed best among the index-based approaches and explained 0.61 of variance, while the inversion of PROSPECT produced a coefficient of determination of 0.65. Cheng et al. (2011) estimate leaf water concentration relative to fresh (LWCF) and dry (LWCD) leaf mass using continuous wavelet analysis (CWA) and several spectral indices. While the spectral indices failed at estimating LWC with R2 values around 0.1, LWCF and LWCD were estimated with R2 = 0.68 and 0.75, respectively, using a combination of six wavelet features. Fang et al. (2017) tested spectral similarity water indices on two freely available leaf data sets (LOPEX93 an PANAMA) and on PROSPECT simulations and explained up to 0.95 and 0.80 of the variance of LOPEX93 and PANAMA EWT, respectively.

The high compatibility between empirical and physically based model retrievals of EWT is also demonstrated using results from an experiment with oak leaves that were successively dried in an oven (Fig. 4). Panel (a) shows the reflectance spectra of the leaves with colors depending on the EWT. In addition to SWIR, the near-infrared plateau is heavily affected when drying out due to cell plasmolysis, i.e., the leaf structure is changing, while only the shortwave infrared region is affected if only the EWT is changed in a PROSPECT simulation. In panel (b), the result of a PROSPECT-D inversion (numerical inversion as proposed by Jay et al. (2016)) is highly correlated with the laboratory measurements, although the estimated values are systematically underestimated. Panels (c) and (d) show two empirical estimations. For these, the data set was split into a training and a validation subset. All possible combinations of two-band ratio indices were calculated, and their coefficient of determination for estimating EWT was determined with the training subset. The optimal index found was the ratio of wavelengths 1745 and 1455 nm; the result is shown in panel (c). A PLSR was trained on the training subset and applied on both subsets; the result is shown in panel (d). These empirical estimates are unbiased, but may not be transferable to different species.
Fig. 4

Drying experiment of oak leaves. a Reflectance spectra colored by EWT, b EWT estimation through PROSPECT-D inversion, c EWT estimation using the best ratio index, d EWT estimation using PLSR, data set split into training and validation sample

In contrast to the other variables discussed here, canopy water content can be derived from IS data during atmospheric correction when absorption bands at 980 and 1160 nm of gaseous, liquid, and solid water are fit to the spectrum (Green et al. 2006; Thompson et al. 2019). Asner et al. (2016) used this approach and characterized the progressive forest canopy water loss during the 2012–2015 drought in California (see also Martin et al. 2018b; Paz-Kagan et al. 2018). Paz-Kagan et al. (2018) mapped the canopy water content of giant Sequoia trees using a random forest machine learning approach. Their results show that IS can contribute to large-scale mapping of environmental influences like drought on forest ecosystems.

Riaño et al. (2005) inverted the PROSPECT–Lillesaeter model to estimate EWT and FMC on canopy level with R2 = 0.75 for a range of 0–0.05 g/cm2. FMC estimations improved from R2 = 0.62 to 0.81 when dry matter content from dried leaf samples measured in a laboratory were taken into account. Koetz et al. (2004) used radiative transfer modeling for forest fire fuel estimation. A GeoSail (Huemmrich 2001) inversion yielded low RMSE for EWT retrieval, but also a very low R2. The estimation of CWC showed a better result, but no R2 is reported. Koetz et al. (2007) inverted the GeoSail model on synthetic data and could retrieve EWT values between 0.025 and 0.065 g/cm2 with R2 = 0.92.

Colombo et al. (2008) used Multispectral Infrared and Visible Imaging Spectrometer (MIVIS) data acquired over poplar plantations in Italy for landscape level estimations of EWT. A double-ratio index (MSI/SR) was used and yielded a cross-validated R2 of 0.57 for the value range of 0.0116–0.0136 g/cm2. Normalizing the moisture stress index (MSI) with the LAI-dependent simple ratio (SR) reduces the influence of LAI on the EWT estimation. An inversion of the PROSPECT + SAILH canopy reflectance model resulted in a relative RMSE of 27% (R2 = 0.818) for EWTcanopy, while the mean leaf EWT was only estimated with R2 = 0.048. This can be explained by a very small range of leaf EWT values, while LAI ranged from 0.5 to 3.5.

White et al. (2007) calculated moisture indices from Hyperion data to detect mountain pine beetle red attack damage. The moisture stress index (MSI) had highest correlation to red attack fraction. They suggest that space-borne hyperspectral imagery may be able to detect pine beetle damages at lower levels of infestation than Landsat imagery. Zhang et al. (2012) used the physically based coupled canopy-leaf radiative transfer model PROSAIL2 to derive the fraction of photosynthetically active radiation absorbed by chlorophyll and the leaf water content of a coniferous forest canopy from Hyperion data. Estimation accuracies are not provided, but the authors stress that their products provide unique information and that using the model inversion strategy, a Markov Chain Monte Carlo (MCMC) method, the globally optimal solution is found.

Buddenbaum et al. (2015, 2012) used field IS to monitor drought-stressed beech seedlings grown in pots in two different years, showing that after water supply was stopped, soil moisture declined first, followed by leaf water content and leaf chlorophyll content (Fig. 6). The drought stress is obvious in the spectra for most plants after some weeks of water deprivation (Fig. 6, red curves). For 2011 data, a PLSR in the visible/near-infrared range yielded R2 = 0.88 for EWT values from 0.001 to 0.005 g/cm2 (Buddenbaum et al. 2012); for 2012 data, a PLSR in the shortwave infrared range yielded R2 = 0.80 for a range of 0.005–0.009 g/cm2 (Buddenbaum et al. 2015).

3.3 Leaf Nitrogen Concentration

A variable of special interest is the nitrogen (N) concentration of leaves. Since the protein absorption features in fresh leaves are mostly masked by water absorption, foliar N retrieval is less straightforward than pigment or water content retrieval, but still, some very high accuracies have been reported both on leaf and on canopy level. In some studies (e.g., Serbin et al. 2014), foliar N has been retrieved from dry leaf reflectance, so that the confounding effect of water was removed. Dry spectral approaches based on ground material appear to have good transferability (Fig. 5a), as discussed in more detail below. The accuracies of the studies cited here are listed in Table 3.
Fig. 5

Spectra of drought-stressed beech seedlings over the course of 2 weeks

(from Buddenbaum et al. 2015)

Table 3

Accuracies and further details of nitrogen concentration estimation studies

Study

Unit

Range

Accuracy

Species

Method

Min

Max

Mean

R 2

RMSE

RMSE%

Leaf level

Schlerf et al. (2010)

% dry

0.88

1.46

 

0.59

0.073

6.1

Norway Spruce

FD-Stepw

Serbin et al. (2014)

%

0.3

6.4

 

0.97

0.13

4

Northern temperate and boreal tree species

PLSR

Wang et al. (2015)

g Protein/cm2

0.000292

0.002

0.001

0.47

0.000275

0.17a

LOPEX data

PROSPECT-5N inversion (fresh leaves)

Wang et al. (2015)

g Protein/cm2

0.000292

0.002

0.001

0.66

0.000202

0.12a

LOPEX data

SMLR

Dechant et al. (2017)

g/m2

0.52

3.16

1.53

0.8808

0.1757

11.5

Mixed

PLSR

Wang et al. (2018)

g/m2

1.43

3.68

2.78

0.51

0.851

17

Mixed

Prospect-5/Inform inversion

Canopy level

Townsend et al. (2003)

%

1.53

4.24

 

0.978

0.25

 

Mixed forest

PLSR Hyperion

Townsend et al. (2003)

%

1.53

4.24

 

0.849

  

Mixed forest

PLSR AVIRIS

Coops et al. (2003)

%

0.77

1.58

1.22

0.62

 

13

Eucalypt

PLSR Hyperion

Smith et al. (2003)

%

1

2.48

 

0.79

0.19

 

Mixed forest

PLSR AVIRIS

Smith et al. (2003)

%

1

2.48

 

0.6

0.25

 

Mixed forest

PLSR Hyperion

Huang et al. (2004)

mg/g

9.8

17.8

14.35

0.8532

  

Eucalypt

Modified PLSR, HySpex

Martin et al. (2008)

%

0.7

2.4

 

0.83

0.19b

 

Mixed forest

PLSR AVIRIS

Martin et al. (2008)

%

0.7

2.43

 

0.82

0.25b

 

Mixed forest

PLSR Hyperion

Ollinger et al. (2008)

%

0.75

2.24

 

0.79

  

Mixed

Univariate regression

Schlerf et al. (2010)

% dry

0.88

1.46

 

0.57

0.055

4.6

Norway spruce

FD-Stepw

Knyazikhin et al. (2013)

%

0.75

2.24

 

0.607

  

Mixed

CSC (DASF)-Univariate

Singh et al. (2015)

%

0.96

3.31

2.2

0.84

0.23

10.5

Northern temperate and boreal tree species

PLSR

Asner et al. (2015)

%

1.28

3.33

 

0.48

0.31

15.19

Mixed tropical

PLSR (cal values)

Wang et al. (2017)

%

1.45

3.29

1.69?

0.65

0.33

19

Mixed

CSC (DASF)-wavelets

Wang et al. (2018)

g/m2

1.43

3.68

2.78

0.64

1.9

18

Mixed

Prospect-5/Inform inversion

Martin et al. (2018a, b)

%

0.6

4.46

1.79

0.54

0.43

24.4

Mixed tropical

PLSR (model test)

anRMSE

bStandard error of cross-validation

Wang et al. (2015) extended the physically based leaf reflectance model PROSPECT-5 by adding protein and cellulose + lignin absorption (see Fig. 2c). They were able to invert the model for fresh and dry leaves with moderate to good accuracies for fresh leaves (R2 = 0.70 for cellulose + lignin and R2 = 0.47 for protein) and dry leaves (R2 = 0.79 for cellulose + lignin and R2 = 0.57 for protein). A stepwise multiple linear regression approach gave slightly better results (R2 = 0.83 and 0.66 for cellulose + lignin and protein in fresh leaves, respectively), but may have limited transferability.

Dechant et al. (2017) measured reflectance, leaf mass per area, nitrogen content, and traits linked to photosynthesis, maximum carboxylation capacity (Vcmax), and maximum electron transport rate (Jmax) of leaves of several tree species. They found clear relationships of the photosynthesis traits with leaf nitrogen content. The nitrogen content per leaf area was estimated via PLSR with R2 = 0.88. Studying the effects of bark beetle green attack on foliar reflectance and biochemical properties, Abdullah et al. (2018) also employed a PLSR to estimate N concentration and reached R2 = 0.76 and 0.68 for healthy and bark beetle-infested spruce needles, respectively. Schlerf et al. (2010) compared several spectral preprocessing techniques for estimating N concentration in Norway spruce (Picea abies) needles with a stepwise multiple regression approach and could explain 0.59 of the variance (cross-validated) with first derivative spectra, although the range of N in this single-species data set was quite limited (0.88 to 1.46%). Asner et al. (2011) successfully (R2 = 0.81) estimated foliar N concentration of thousands of tropical species from around the world with a PLSR using visible-to-shortwave infrared transmittance spectra, and Serbin et al. (2014) established PLS regressions for estimating seven biochemical traits (N, C, δ15N, SLA, cellulose, ADF, ADL) from ground and dried leaf samples for a large number of North American tree species. The regression models were able to explain between 60% (δ15N) and 97% (N) of the variance in the validation data set. These large samples are an important step toward the development of transferable models. Serbin et al. (2014) have published their PLSR coefficients so that they can be applied to any data set.

In an experiment recently performed by the authors of this paper, leaf samples of Norway spruce, Douglas fir (Pseudotsuga menziesii), and European beech (Fagus sylvatica) were collected by professional tree climbers at two sites in Germany, the Nationalpark Hunsrück-Hochwald (NP) and near Gerolstein (Ge). The samples were dried, ground, chemically analyzed, and spectrally measured. Figure 6 shows estimations of N content using the PLSR coefficients by Serbin et al. (2014), and a PLSR-based estimation of foliar N content trained on a subsample of our data with R2 values for the total samples and all subsamples. While the locally adapted PLSR yields better results, the use of coefficients from the American data set suggests that transfers are possible.
Fig. 6

Estimation of leaf nitrogen content from dried and powdered samples collected in Germany (NP = Nationalpark Hunsrück-Hochwald, Ge = Gerolstein area). a Application of PLSR coefficients from Serbin et al. (2014), b local PLSR, small squares represent calibration subsample, large circles are the validation sample

Large area assessments of ecosystem productivity and health are improved tremendously by reliable estimations of foliar nitrogen content (McNeil et al. 2007a, b; Ollinger et al. 2008; Townsend et al. 2003), just as applications of fertilizers to tree plantations (Albaugh et al. 2003; Allen et al. 2005; Liechty and Fristoe 2013). Large efforts have been undertaken to collect data sets that enable transferable estimations of foliar nitrogen on canopy level.

Townsend et al. (2003) used space-borne EO-1 Hyperion and airborne AVIRIS IS data to map canopy N in Green Ridge State Forest in Maryland, USA, where leaf samples were collected in 27 ground plots. PLSR models explained 97.9% of the variation of N for Hyperion and 84.9% for AVIRIS for N concentrations between 1.2 and 4.1%. Similarly, Smith et al. (2003) explained 79% and 60% of variance of validation data in the range between 1 and 2.5% N using AVIRIS and Hyperion, respectively. Coops et al. (2003) also used Hyperion data. They mapped N concentration of Eucalypt foliage with R2 = 0.62 for a range of 0.77–1.58%N using PLSR. Huang et al. (2004) also worked in an Eucalypt forest. They estimated N concentrations from continuum-removed HyMap airborne data with stepwise regression, PLSR, and neural networks and reached R2 = 0.85 for a range of 0.98–1.78% N. Martin et al. (2008) used PLSR to estimate foliar N content on airborne (AVIRIS) and space-borne (Hyperion) hyperspectral data from 137 forested sites in North America, Costa Rica, and Australia. They were able to retrieve N concentrations in the range of 0.7–3.2% with a cross-validated R2 of 0.83 and 0.82 with AVIRIS and Hyperion, respectively, for all sites combined. Singh et al. (2015) built on the leaf-level study by Serbin et al. (2014) and established respective regression models for airborne hyperspectral data. 145 AVIRIS classic data sets from the Northern USA and data from 237 field plots were used to create transferable algorithms for retrieving foliar traits and their uncertainties. They calculated mean stand values of the leaf traits by weighting them by leaf biomass of the trees. Between 0.49 (cellulose content) and 0.88 (SLA) of the variance could be explained by the canopy-level PLSR. N concentration was estimated with R2 = 0.85 for a range of 0.96–3.31% using the average of 500 prediction models.

Ollinger et al. (2008) developed a regression model for foliar N concentration across several North American mixed forests that exploits a high positive correlation between foliar N and near-infrared (NIR) reflectance. The model explained 79 percent of the variance for a data range of 0.75–2.35% N. The regression model was applied to all forested areas in North America using MODIS satellite data. However, Knyazikhin et al. (2013) pointed out that the positive correlation between reflectance and the content of an absorbing material is counter-intuitive, contending that the observed relationship must be spurious. They showed that the NIR reflectance of closed forest canopies can be explained to a very high degree (R2 = 0.81) by the broadleaf fraction of a canopy and that the N content is also highly dependent on the broadleaf fraction (R2 = 0.89). This corresponds to the common knowledge that (a) leaves contain significantly higher amounts of N than needles and (b) that conifers appear darker than broadleaf trees. The lower reflectivity of coniferous trees is mostly due to structural effects: needle shoots have a much higher photon recollision probability due to their higher complexity compared to broadleaves. Knyazikhin et al. (2013) propose a normalizing factor, the DASF, that takes this complexity into account and can be easily computed from remote sensing data. If the reflectance of closed forest stands is divided by the DASF, the structural differences are removed from the signal. The resulting signal is termed canopy scattering coefficient (CSC). By adopting this normalization strategy, Wang et al. (2017) were able to estimate foliar N concentration of mixed forest stands in the Bavarian forest national park from airborne hyperspectral imagery using continuous wavelet analysis with a coefficient of determination of 0.65 for a range of 1.45–3.29% N. Ewald et al. (2018) also stress the influence of forest structure on the prediction of canopy N and P concentrations. They integrated predictor variables derived from airborne laser scanning to improve PLSR models (\( R_{cv}^{2} \) of N estimation was improved from 0.31 to 0.41 for a range of 14–25 g/kg N). The relatively low accuracies are attributed to the structural differences between and within broadleaved and coniferous species and the relatively low range of values.

Foliar N retrievals through physically based model inversions are still rare, partially because there are a range of spectral features associated with N bonds. When coupling the canopy reflectance model InFoRM with their revised leaf-level model PROSPECT-5N (Wang et al. 2015), the achieved accuracy in estimating foliar N concentration was quite limited (R2 = 0.46), while canopy N content was estimated with R2 = 0.64 (Wang et al. 2018). The inversion procedure consisted of creating a LUT with 200 000 entries, then iteratively applying three filters to delete improbable parameter combinations from the LUT, and lastly identifying the best-fitting spectra in several different spectral subsets. The mean parameters from the 10, 20, or 100 best-fitting spectra were taken as inversion result. In total, 48 inversion results (4 differently filtered LUTs × 4 spectral subsets × 3 numbers of best fits) were obtained.

Schlerf et al. (2010) reached the highest estimation accuracy for foliar N concentration on Norway spruce (Picea abies) canopy level by using a stepwise multiple regression on first derivative spectra from the HyMap sensor with a cross-validated R2 of 0.56. As an extension of the study by Schlerf et al. (2010) that also includes data on beech trees, PLSR models with 2 latent variables for estimating foliar N content were established on reflectance data and on CSC data. Although the PLSR for CSC could not rely on the sharp contrast between coniferous and broadleaf spectra in the NIR, the regression accuracy was even better for CSC data (Fig. 7). Similar to other studies, the high overall coefficient of determination is once more driven by the contrast of N concentrations between spruce and beech. The simple PLSR models were not able to capture as much variation within the species as models with larger numbers of latent variables would. In cases like this, mapping the fraction of coniferous and broadleaved species may suffice to characterize N distribution in forests.
Fig. 7

BRF and CSC spectra of beech and spruce trees. Lower row: scatter plots of PLSR accuracies for estimating foliar nitrogen content from the respective spectra

In a further analysis step, the PLSR coefficients for canopy-level N estimation from Singh et al. (2015) were applied to the HyMap data. Since vegetation spectra contain no sharp features, resampling them to a different sensor with the same spectral range should theoretically not pose a problem. But although these coefficients had been trained on a very large ensemble of measurements, reliable N estimations were not achieved with them (Fig. 8a). Figure 8b shows the AVIRIS classic spectra used for establishing the PLSR coefficients, and the HyMap spectra these coefficients were applied to. There are clear differences between the atmospherically corrected spectra, especially in the near-infrared region, that may explain the failure of estimating N content with the AVIRIS-trained PLSR coefficients. In addition, the HyMap spruce spectra are significantly darker than the AVIRIS needleleaf tree spectra. Townsend et al. (2017) highlight some of the difficulties of transferring empirical models to different sensors and ecoregions, e.g., different bandwidths, instrument calibration approaches, atmospheric correction schemes, methods of handling variable illumination due to sun-sensor-target geometry (BRDF), and finally variable stand structures.
Fig. 8

a Foliar nitrogen content of spruce and beech trees in Germany estimated using coefficients by Singh et al. (2015). b AVIRIS Classic spectra used by Singh et al. (2015) (gray), and HyMap spectra of the trees in Germany

4 Conclusions

Imaging spectroscopy of forest ecosystems has come a long way since the first analyses of foliar biochemistry by Wessman et al. (1989, 1988), and abundant literature on experiments and applications has in principle demonstrated its value for assessing forest ecosystems. It is obvious that mapping biochemical traits, especially those with less dominant absorption features like carotenoids, anthocyanins, or proteins, are one of the specific contributions of IS which cannot be adequately replaced by multi- or super-spectral remote sensing systems, nor with radar systems. The question whether biochemical traits are more efficiently extracted based on carefully established empirical models, through the inversion of physically based reflectance models, or in a hybrid approach, still cannot be answered definitely.

Physically based approaches offer substantial conceptual advantages and most of the available models are capable of accurately reproducing the spectral characteristics of leaves or forest canopies as long as the required parameters are known (forward mode). However, the inversion of reflectance models is more critical. As long as the number of model parameters is limited and their impact on the reflectance signal not overlapping with other parameters, inversion provides reasonably accurate estimates (Omari et al. 2013). Models of optical leaf properties, such as the PROSPECT family, have thus been successfully used in many retrieval studies and are meanwhile accepted as replacement of traditional laboratory analysis (i.e., for measuring leaf water or pigment concentrations) (Dechant et al. 2017). Establishing empirical models for estimating leaf properties requires abundant high-quality calibration data covering the range of trait variation; once calibrated, they can build on a range of powerful regression algorithms able to deal with high dimensionality spectral data and equally reproduce laboratory quality (Martin et al. 2008; Serbin et al. 2014; Singh et al. 2015).

On canopy level, the use of both strategies becomes more difficult. Physically based models require a substantially higher number of parameters to adequately reproduce the scattering and absorption mechanisms within structurally complex forest canopy architectures. The behavior of canopy reflectance models in the forward simulation mode is convincing for a range of models, and increasingly powerful computational resources support the development of even more differentiated simulation concepts. However, difficulties arise when the physically based forest canopy reflectance models are inverted. Although the general principle has been demonstrated, the reported levels of accuracy in retrieving leaf biochemical traits from canopy observations are still limited (e.g., Koetz et al. 2004; Omari et al. 2013; Wang et al. 2018), owing to the ill-posedness of the problem. Strategically, this limitation has been treated by constraining the model inversion, either through including external measurements (e.g., by LAI estimation through LIDAR measurements), fixing some of the model parameters to “reasonable” values, or by imposing constraints (rules) on the inversion algorithm while retrieving the remaining (free) parameters. Wang et al. (2018), for example, filtered their lookup table and restricted the spectral range until acceptable results were found; Riaño et al. (2005) used species-specific dry matter content from measurements of dried leaves to constrain the inversion and get better EWT estimations, and Colombo et al. (2008) manually set different weights for different variables in the cost function of the inversion. The problem with these strategies is that physically based models then may become so much tuned to local conditions that they lose their inherent generalization capacities.

Although powerful retrieval strategies exist, employing empirical approaches for retrieving forest canopy properties becomes demanding, in particular with respect to collecting reference data from elevated tree crowns, and because of the conceptual difficulties in upscaling sampled leaf properties to the level of multi-storey forest canopies. The necessity to collect leaf samples across extended areas and diverse forest types adds further complications. However, a number of studies have demonstrated that this is feasible for a range of canopy traits and within considerably large areas (e.g., Asner et al. 2011; Martin et al. 2008; Ollinger et al. 2008; Singh et al. 2015). Hyperspectral systems or sample collectors mounted on unmanned aerial vehicles may become important tools for collecting reference data to span a bridge between ground-based measurements and space-borne spectroscopy.

Since canopy observations inherently combine leaf biochemical traits with the amount of leaves to estimate a bulk quantity, additional complexity emerges when the objective is to retrieve biochemical traits on leaf level from canopy reflectance data. Strategies for solving this problem that have been proposed (Colombo et al. 2008; Dotzler et al. 2015; Knyazikhin et al. 2013) could demonstrate that spatial differences in pigment and water concentrations identified on canopy scale are not necessarily a consequence of differences at leaf level. However, if the objective of the analysis of imaging spectroscopy data is to generate a map to be used for management or to characterize spatial patterns of important biological quantities, then it may make sense to exploit emergent relationships (e.g., differences between conifer and broadleaf species for estimating canopy N) to produce accurate maps. However, such maps will require good independent validation data and will only be valid within the range of measured conditions. Such work provides a basis for ongoing research and may yield results to enable better understanding of ecological patterns (Asner et al. 2017), but will require more study for associated models to have biophysical realism or to not be ill-posed (Townsend et al. 2013).

Sensu stricto, empirical approaches can only be applied to data sets that are comparable to the training data. First attempts in transferring empirically defined models across large distances to similarly structured forests were contradictory. However, it was demonstrated in this paper that applying PLSR coefficients determined by Serbin et al. (2014) to leaf and needle samples from Western Europe was satisfactory, although outperformed by training a PLSR model on the samples themselves. Transferring PLSR coefficients for canopy estimates (Singh et al. 2015) was clearly less successful, likely due to spectral calibration differences between the AVIRIS and HySpex images and different strategies in scaling leaf traits to canopy level. Free access to collections of spectra and biochemical properties from forests around the world, similar to the Spectranomics database (Carnegie Institution for Science), will be a key element for enhancing satellite-based studies of forest ecosystems in general, and for the development of transferable relationships that go beyond locally restricted studies.

In case the existing technical and methodological incompatibilities can be reduced, e.g., through a more consistent data quality provided by space-borne instruments, it is expected that empirical models can be established which are sufficiently generalizable and transferable between similar forest ecosystems. Since they are computationally fast and therefore suited to be applied in large area monitoring schemes, empirical approaches may establish themselves a valuable alternative to computationally more complex inversions of physically based models.

Hybrid approaches complement pure empirical or physically based approaches. When physically based reflectance models can reproduce canopy reflectance well in the forward mode, these simulated spectra can feed a parametric or nonparametric empirical model to be applied to image data. If too little reference data could be collected in the field to establish a robust statistical relationship, the field data can be used to validate the hybrid model instead. However, a major prerequisite is that globally distributed regular observations with space-borne imaging spectrometers become available to foster the required cross-comparison exercises.

Successful applications of Hyperion data to map canopy chlorophyll content (Omari et al. 2013), canopy water content (White et al. 2007; Zhang et al. 2012), and foliar N (Coops et al. 2003; Martin et al. 2008; Smith et al. 2003; Townsend et al. 2003) prove the great potential of space-borne imaging spectroscopy for forest ecosystem studies. Also the derivation of canopy liquid water content during atmospheric correction, a truly spectroscopic approach, shows the unique value of imaging spectroscopy (Asner et al. 2016; Green et al. 2006; Thompson et al. 2019).

In this perspective, the Italian PRISMA (to be launched in 2019) and the German EnMAP missions (to be launched in 2020) constitute cornerstones in the efforts to regularly map diverse forest ecosystems from space with hyperspectral capacities. Their availability will provide an essential contribution toward optimizing the approaches discussed in this paper. The US-American SBG mission (to be launched in 2025) and the European CHIME mission (under design) will extend the regional coverage of previous satellite systems toward a truly global scope. While their scientific value is not in doubt, the question whether hyperspectral satellite observations will provide added value for many operational forestry applications in comparison with the enormous observation capacities of multispectral sensor systems like Sentinel-2 or Landsat remains open.

Notes

Acknowledgements

The study was supported within the framework of the EnMAP project (Contract No. 50 EE 1530) by the German Aerospace Center (DLR) and the Federal Ministry of Economic Affairs and Energy, and the CalTech Jet Propulsion Laboratory (Contracts 1579654 and 1590148). The authors thank Willy Werner, Dorothee Krieger, Bernhard Backes, Martin Schlerf, Johannes Stoffels, Sandra Dotzler, Barbara Paschmionka, Marion Lusseau, Max Gerhards, and many others who helped gather the data presented here. We also thank the two anonymous reviewers for highly constructive comments that helped improve the manuscript.

References

  1. Abdullah H, Darvishzadeh R, Skidmore AK, Groen TA, Heurich M (2018) European spruce bark beetle (Ips typographus, L.) green attack affects foliar reflectance and biochemical properties. Int J Appl Earth Obs Geoinf 64:199–209.  https://doi.org/10.1016/j.jag.2017.09.009 Google Scholar
  2. Adams HD et al (2017) A multi-species synthesis of physiological mechanisms in drought-induced tree mortality. Nat Ecol Evol 1:1285–1291.  https://doi.org/10.1038/s41559-017-0248-x Google Scholar
  3. Albaugh TJ, Allen HL, Zutter BR, Quicke HE (2003) Vegetation control and fertilization in midrotation Pinus taeda stands in the southeastern United States. Ann For Sci 60:619–624.  https://doi.org/10.1051/forest:2003054 Google Scholar
  4. Allen HL, Fox TR, Campbell RG (2005) What is ahead for intensive pine plantation silviculture in the south? South J Appl For 29:62–69.  https://doi.org/10.1093/sjaf/29.2.62 Google Scholar
  5. Anderegg LDL, Anderegg WRL, Berry JA (2013a) Not all droughts are created equal: translating meteorological drought into woody plant mortality. Tree Physiol 33:672–683.  https://doi.org/10.1093/treephys/tpt044 Google Scholar
  6. Anderegg WRL, Kane JM, Anderegg LDL (2013b) Consequences of widespread tree mortality triggered by drought and temperature stress. Nat Clim Change 3:30–36.  https://doi.org/10.1038/nclimate1635 Google Scholar
  7. Asner GP (1998) Biophysical and biochemical sources of variability in canopy reflectance. Remote Sens Environ 64:234–253.  https://doi.org/10.1016/S0034-4257(98)00014-5 Google Scholar
  8. Asner GP, Martin RE (2009) Airborne spectranomics: mapping canopy chemical and taxonomic diversity in tropical forests. Front Ecol Environ 7:269–276.  https://doi.org/10.1890/070152 Google Scholar
  9. Asner GP, Martin RE (2016) Convergent elevation trends in canopy chemical traits of tropical forests. Glob Change Biol 22:2216–2227.  https://doi.org/10.1111/gcb.13164 Google Scholar
  10. Asner GP, Nepstad D, Cardinot G, Ray D (2004) Drought stress and carbon uptake in an Amazon forest measured with spaceborne imaging spectroscopy. Proc Natl Acad Sci USA 101:6039–6044.  https://doi.org/10.1073/pnas.0400168101 Google Scholar
  11. Asner GP et al (2011) Spectroscopy of canopy chemicals in humid tropical forests. Remote Sens Environ 115:3587–3598.  https://doi.org/10.1016/j.rse.2011.08.020 Google Scholar
  12. Asner GP et al (2012) Carnegie Airborne Observatory-2: increasing science data dimensionality via high-fidelity multi-sensor fusion. Remote Sens Environ 124:454–465.  https://doi.org/10.1016/j.rse.2012.06.012 Google Scholar
  13. Asner GP, Martin RE, Anderson CB, Knapp DE (2015) Quantifying forest canopy traits: imaging spectroscopy versus field survey. Remote Sens Environ 158:15–27.  https://doi.org/10.1016/j.rse.2014.11.011 Google Scholar
  14. Asner GP, Brodrick PG, Anderson CB, Vaughn N, Knapp DE, Martin RE (2016) Progressive forest canopy water loss during the 2012–2015 California drought. Proc Natl Acad Sci 113:E249–E255.  https://doi.org/10.1073/pnas.1523397113 Google Scholar
  15. Asner GP et al (2017) Airborne laser-guided imaging spectroscopy to map forest trait diversity and guide conservation. Science 355:385.  https://doi.org/10.1126/science.aaj1987 Google Scholar
  16. Atzberger C (2000) Development of an invertible forest reflectance model: the INFOR-model. In: Buchroithner M (ed) A decade of trans-European remote sensing cooperation. Proceedings of the 20th EARSeL symposium, Dresden, Germany, 14–16 June 2000, pp 39–44Google Scholar
  17. Atzberger C, Guérif M, Baret F, Werner W (2010) Comparative analysis of three chemometric techniques for the spectroradiometric assessment of canopy chlorophyll content in winter wheat. Comput Electron Agric 73:165–173.  https://doi.org/10.1016/j.compag.2010.05.006 Google Scholar
  18. Barton CVM, North PRJ (2001) Remote sensing of canopy light use efficiency using the photochemical reflectance index: model and sensitivity analysis. Remote Sens Environ 78:264–273.  https://doi.org/10.1016/S0034-4257(01)00224-3 Google Scholar
  19. Beamish AL, Coops NC, Hermosilla T, Chabrillat S, Heim B (2018) Monitoring pigment-driven vegetation changes in a low-Arctic tundra ecosystem using digital cameras. Ecosphere 9:e02123.  https://doi.org/10.1002/ecs2.2123 Google Scholar
  20. Beniston M (2004) The 2003 heat wave in Europe: a shape of things to come? An analysis based on Swiss climatological data and model simulations. Geophys Res Lett.  https://doi.org/10.1029/2003GL018857 Google Scholar
  21. Berger K, Atzberger C, Danner M, D’Urso G, Mauser W, Vuolo F, Hank T (2018) Evaluation of the PROSAIL model capabilities for future hyperspectral model environments: a review study. Remote Sens 10:85.  https://doi.org/10.3390/rs10010085 Google Scholar
  22. Bicheron P, Leroy M (1999) A method of biophysical parameter retrieval at global scale by inversion of a vegetation reflectance model. Remote Sens Environ 67:251–266.  https://doi.org/10.1016/S0034-4257(98)00083-2 Google Scholar
  23. Birdsey R, Pan Y (2011) Ecology: drought and dead trees. Nat Clim Change 1:444–445.  https://doi.org/10.1038/nclimate1298 Google Scholar
  24. Bond I, Chambwera M, Jones B, Nhantumbo I, Chundama M (2010) REDD+ in dryland forests: issues and prospects for pro-poor REDD in the miombo woodlands of southern Africa, natural resource issues, vol 21. International Institute for Environment and Development, LondonGoogle Scholar
  25. Bréda NJJ (2003) Ground-based measurements of leaf area index: a review of methods, instruments and current controversies. J Exp Bot 54:2403–2417.  https://doi.org/10.1093/jxb/erg263 Google Scholar
  26. Buddenbaum H, Steffens M (2012) The effects of spectral pretreatments on chemometric analyses of soil profiles using laboratory imaging spectroscopy. Appl Environ Soil Sci 2012:12.  https://doi.org/10.1155/2012/274903 Google Scholar
  27. Buddenbaum H, Pueschel P, Stellmes M, Werner W, Hill J (2011) Measuring water and chlorophyll content on the leaf and canopy scale. EARSeL eProc 10:66–72Google Scholar
  28. Buddenbaum H, Stern O, Stellmes M, Stoffels J, Pueschel P, Hill J, Werner W (2012) Field imaging spectroscopy of beech seedlings under dryness stress. Remote Sens 4:3721–3740.  https://doi.org/10.3390/rs4123721 Google Scholar
  29. Buddenbaum H et al (2015) Using VNIR and SWIR field imaging spectroscopy for drought stress monitoring of beech seedlings. Int J Remote Sens 36:4590–4605.  https://doi.org/10.1080/01431161.2015.1084435 Google Scholar
  30. Carrere V et al (2013) HYPXIM: a second generation high spatial resolution hyperspectral satellite for dual applications. In: 5th Workshop on hyperspectral image and signal processing: evolution in remote sensing (WHISPERS), 26–28 June 2013, pp 1–4.  https://doi.org/10.1109/whispers.2013.8080685
  31. Chadwick K, Asner G (2016) Organismic-scale remote sensing of canopy foliar traits in lowland tropical forests. Remote Sens 8:87.  https://doi.org/10.3390/rs8020087 Google Scholar
  32. Chávez R, Clevers J, Herold M, Acevedo E, Ortiz M (2013) Assessing water stress of desert tamarugo trees using in situ data and very high spatial resolution remote sensing. Remote Sens 5:5064.  https://doi.org/10.3390/rs5105064 Google Scholar
  33. Cheng T, Rivard B, Sánchez-Azofeifa A (2011) Spectroscopic determination of leaf water content using continuous wavelet analysis. Remote Sens Environ 115:659–670.  https://doi.org/10.1016/j.rse.2010.11.001 Google Scholar
  34. Cheng T, Rivard B, Sánchez-Azofeifa AG, Féret J-B, Jacquemoud S, Ustin SL (2014) Deriving leaf mass per area (LMA) from foliar reflectance across a variety of plant species using continuous wavelet analysis. ISPRS J Photogramm Remote Sens 87:28–38.  https://doi.org/10.1016/j.isprsjprs.2013.10.009 Google Scholar
  35. Ciais P et al (2005) Europe-wide reduction in primary productivity caused by the heat and drought in 2003. Nature 437:529–533.  https://doi.org/10.1038/nature03972 Google Scholar
  36. Colombo R, Meroni M, Marchesi A, Busetto L, Rossini M, Giardino C, Panigada C (2008) Estimation of leaf and canopy water content in poplar plantations by means of hyperspectral indices and inverse modeling. Remote Sens Environ 112:1820–1834.  https://doi.org/10.1016/j.rse.2007.09.005 Google Scholar
  37. Combal B et al (2002) Retrieval of canopy biophysical variables from bidirectional reflectance using prior information to solve the ill-posed inverse problem. Remote Sens Environ 84:1–15.  https://doi.org/10.1016/S0034-4257(02)00035-4 Google Scholar
  38. Coops NC, Smith M-L, Martin ME, Ollinger SV (2003) Prediction of eucalypt foliage nitrogen content from satellite-derived hyperspectral data. IEEE Trans Geosci Remote Sens 41:1338–1346.  https://doi.org/10.1109/TGRS.2003.813135 Google Scholar
  39. Curran PJ (1989) Remote sensing of foliar chemistry. Remote Sens Environ 30:271–278.  https://doi.org/10.1016/0034-4257(89)90069-2 Google Scholar
  40. Dawson TP, Curran PJ, Plummer SE (1998) LIBERTY—modeling the effects of leaf biochemical concentration on reflectance spectra. Remote Sens Environ 65:50–60.  https://doi.org/10.1016/S0034-4257(98)00007-8 Google Scholar
  41. Dechant B, Cuntz M, Vohland M, Schulz E, Doktor D (2017) Estimation of photosynthesis traits from leaf reflectance spectra: correlation to nitrogen content as the dominant mechanism. Remote Sens Environ 196:279–292.  https://doi.org/10.1016/j.rse.2017.05.019 Google Scholar
  42. Dotzler S, Hill J, Buddenbaum H, Stoffels J (2015) The potential of EnMAP and Sentinel-2 data for detecting drought stress phenomena in deciduous forest communities. Remote Sens 7:14227–14258.  https://doi.org/10.3390/rs71014227 Google Scholar
  43. Drusch M et al (2017) The FLuorescence EXplorer Mission Concep—ESA’s Earth Explorer 8. IEEE Trans Geosci Remote Sens 55:1273–1284.  https://doi.org/10.1109/tgrs.2016.2621820 Google Scholar
  44. Eckardt A, Horack J, Lehmann F, Krutz D, Drescher J, Whorton M, Soutullo M (2015) DESIS (DLR earth sensing imaging spectrometer for the ISS-MUSES platform). In: IEEE international geoscience and remote sensing symposium (IGARSS), 26–31 July 2015, pp 1457–1459.  https://doi.org/10.1109/igarss.2015.7326053
  45. EEA (2012) Annual environmental indicator report 2012—ecosystem resilience and resource efficiency in a green economy in Europe. European Environment Agency, Copenhagen.  https://doi.org/10.2800/487 Google Scholar
  46. Ewald M et al (2018) LiDAR derived forest structure data improves predictions of canopy N and P concentrations from imaging spectroscopy. Remote Sens Environ 211:13–25.  https://doi.org/10.1016/j.rse.2018.03.038 Google Scholar
  47. Fang M, Ju W, Zhan W, Cheng T, Qiu F, Wang J (2017) A new spectral similarity water index for the estimation of leaf water content from hyperspectral data of leaves. Remote Sens Environ 196:13–27.  https://doi.org/10.1016/j.rse.2017.04.029 Google Scholar
  48. FAO (2010) Guidelines on sustainable forest management in drylands of sub-Saharan Africa. FAO, RomeGoogle Scholar
  49. Fassnacht F et al (2014) Comparison of feature reduction algorithms for classifying tree species with hyperspectral data on three central European test sites. IEEE J Sel Top Appl Earth Obs Remote Sens 7:2547–2561.  https://doi.org/10.1109/jstars.2014.2329390 Google Scholar
  50. Fassnacht FE, Stenzel S, Gitelson AA (2015) Non-destructive estimation of foliar carotenoid content of tree species using merged vegetation indices. J Plant Physiol 176:210–217.  https://doi.org/10.1016/j.jplph.2014.11.003 Google Scholar
  51. Fassnacht FE et al (2016) Review of studies on tree species classification from remotely sensed data. Remote Sens Environ 186:64–87.  https://doi.org/10.1016/j.rse.2016.08.013 Google Scholar
  52. Fawcett D, Verhoef W, Schläpfer D, Schneider FD, Schaepman ME, Damm A (2018) Advancing retrievals of surface reflectance and vegetation indices over forest ecosystems by combining imaging spectroscopy, digital object models, and 3D canopy modelling. Remote Sens Environ 204:583–595.  https://doi.org/10.1016/j.rse.2017.09.040 Google Scholar
  53. Feilhauer H, Asner GP, Martin RE (2015) Multi-method ensemble selection of spectral bands related to leaf biochemistry. Remote Sens Environ 164:57–65.  https://doi.org/10.1016/j.rse.2015.03.033 Google Scholar
  54. Feingersh T, Ben Dor E (2015) SHALOM—a commercial hyperspectral space mission. In: Qian S-E (ed) Optical payloads for space missions. Wiley, Chichester.  https://doi.org/10.1002/9781118945179.ch11 Google Scholar
  55. Feret J-B et al (2008) PROSPECT-4 and 5: advances in the leaf optical properties model separating photosynthetic pigments. Remote Sens Environ 112:3030–3043.  https://doi.org/10.1016/j.rse.2008.02.012 Google Scholar
  56. Féret J-B et al (2011) Optimizing spectral indices and chemometric analysis of leaf chemical properties using radiative transfer modeling. Remote Sens Environ 115:2742–2750.  https://doi.org/10.1016/j.rse.2011.06.016 Google Scholar
  57. Féret JB, Gitelson AA, Noble SD, Jacquemoud S (2017) PROSPECT-D: towards modeling leaf optical properties through a complete lifecycle. Remote Sens Environ 193:204–215.  https://doi.org/10.1016/j.rse.2017.03.004 Google Scholar
  58. Fink AH, Brücher T, Krüger A, Leckebusch GC, Pinto JG, Ulbrich U (2004) The 2003 European summer heatwaves and drought—synoptic diagnosis and impacts. Weather 59:209–216.  https://doi.org/10.1256/wea.73.04 Google Scholar
  59. Foster JR, Townsend PA (2004) Linking hyperspectral imagery and forest inventories for forest assessment in the Central Appalachians. In: Yaussy DA, Hix DM, Long RP, Goebel PC (eds) 14th Central hardwood forest conference. U.S. Department of Agriculture, Forest Service, Northeastern Research Station, pp 76–86Google Scholar
  60. Franklin SE (2001) Remote sensing for sustainable forest management. Lewis Publishers, Boca RatonGoogle Scholar
  61. Frantz D, Röder A, Stellmes M, Hill J (2016) An operational radiometric landsat preprocessing framework for large-area time series applications. IEEE Trans Geosci Remote Sens.  https://doi.org/10.1109/tgrs.2016.2530856 Google Scholar
  62. Gamon JA, Peñuelas J, Field CB (1992) A narrow-waveband spectral index that tracks diurnal changes in photosynthetic efficiency. Remote Sens Environ 41:35–44.  https://doi.org/10.1016/0034-4257(92)90059-s Google Scholar
  63. Gara TW, Skidmore AK, Darvishzadeh R, Wang T (2018) Leaf to canopy upscaling approach affects the estimation of canopy traits. GISci Remote Sens.  https://doi.org/10.1080/15481603.2018.1540170 Google Scholar
  64. Garbulsky MF, Peñuelas J, Gamon J, Inoue Y, Filella I (2011) The photochemical reflectance index (PRI) and the remote sensing of leaf, canopy and ecosystem radiation use efficiencies: a review and meta-analysis. Remote Sens Environ 115:281–297.  https://doi.org/10.1016/j.rse.2010.08.023 Google Scholar
  65. Gastellu-Etchegorry JP, Martin E, Gascon F (2004) DART: a 3D model for simulating satellite images and studying surface radiation budget. Int J Remote Sens 25:73–96.  https://doi.org/10.1080/0143116031000115166 Google Scholar
  66. Ghamisi P, Plaza J, Chen Y, Li J, Plaza AJ (2017) Advanced spectral classifiers for hyperspectral images: a review. IEEE Geosci Remote Sens Mag 5:8–32.  https://doi.org/10.1109/mgrs.2016.2616418 Google Scholar
  67. Gitelson AA, Merzlyak MN (1997) Remote estimation of chlorophyll content in higher plant leaves. Int J Remote Sens 18:2691–2697.  https://doi.org/10.1080/014311697217558 Google Scholar
  68. Gitelson AA, Gritz Y, Merzlyak MN (2003) Relationships between leaf chlorophyll content and spectral reflectance and algorithms for non-destructive chlorophyll assessment in higher plant leaves. J Plant Physiol 160:271–282.  https://doi.org/10.1078/0176-1617-00887 Google Scholar
  69. Gitelson AA, Keydan GP, Merzlyak MN (2006) Three-band model for noninvasive estimation of chlorophyll, carotenoids, and anthocyanin contents in higher plant leaves. Geophys Res Lett.  https://doi.org/10.1029/2006gl026457 Google Scholar
  70. Green AA, Berman M, Switzer P, Craig MD (1998) A transformation for ordering multispectral data in terms of image quality with implications for noise removal. IEEE Trans Geosci Remote Sens 26:65–74Google Scholar
  71. Green RO, Painter TH, Roberts DA, Dozier J (2006) Measuring the expressed abundance of the three phases of water with an imaging spectrometer over melting snow. Water Resour Res.  https://doi.org/10.1029/2005wr004509 Google Scholar
  72. Guanter L et al (2015) The EnMAP spaceborne imaging spectroscopy mission for earth observation. Remote Sens 7:8830–8857.  https://doi.org/10.3390/rs70708830 Google Scholar
  73. Hank TB, Berger K, Bach H, Clevers JGPW, Gitelson A, Zarco-Tejada P, Mauser W (2019) Spaceborne imaging spectroscopy for sustainable agriculture: contributions and challenges. Surv Geophys.  https://doi.org/10.1007/s10712-018-9492-0 Google Scholar
  74. Hansen MC, Roy DP, Lindquist E, Adusei B, Justice CO, Altstatt A (2008) A method for integrating MODIS and Landsat data for systematic monitoring of forest cover and change in the Congo Basin. Remote Sens Environ 112:2495–2513.  https://doi.org/10.1016/j.rse.2007.11.012 Google Scholar
  75. Heim R, Jürgens N, Große-Stoltenberg A, Oldeland J (2015) The effect of epidermal structures on leaf spectral signatures of ice plants (Aizoaceae). Remote Sens 7:15862Google Scholar
  76. Held M, Rabe A, Senf C, Svd Linden, Hostert P (2015) Analyzing hyperspectral and hypertemporal data by decoupling feature redundancy and feature relevance. IEEE Geosci Remote Sens Lett 12:983–987.  https://doi.org/10.1109/lgrs.2014.2371242 Google Scholar
  77. Hernández-Clemente R, Navarro-Cerrillo RM, Zarco-Tejada PJ (2012) Carotenoid content estimation in a heterogeneous conifer forest using narrow-band indices and PROSPECT + DART simulations. Remote Sens Environ 127:298–315.  https://doi.org/10.1016/j.rse.2012.09.014 Google Scholar
  78. Homolová L, Malenovský Z, Clevers JGPW, García-Santos G, Schaepman ME (2013) Review of optical-based remote sensing for plant trait mapping. Ecol Complex 15:1–16.  https://doi.org/10.1016/j.ecocom.2013.06.003 Google Scholar
  79. Hosgood B, Jacquemoud S, Andreoli G, Verdebout J, Pedrini G, Schmuck G (1994) Leaf optical properties experiment 93 (LOPEX93). European Commission, Joint Research Center, IspraGoogle Scholar
  80. Hovi A, Raitio P, Rautiainen M (2017) A spectral analysis of 25 boreal tree species. Silva Fenn.  https://doi.org/10.14214/sf.7753 Google Scholar
  81. Huang Z, Turner BJ, Dury SJ, Wallis IR, Foley WJ (2004) Estimating foliage nitrogen concentration from HYMAP data using continuum removal analysis. Remote Sens Environ 93:18–29.  https://doi.org/10.1016/j.rse.2004.06.008 Google Scholar
  82. Huber S, Kneubühler M, Psomas A, Itten K, Zimmermann NE (2008) Estimating foliar biochemistry from hyperspectral data in mixed forest canopy. For Ecol Manag 256:491–501.  https://doi.org/10.1016/j.foreco.2008.05.011 Google Scholar
  83. Huemmrich KF (2001) The GeoSail model: a simple addition to the SAIL model to describe discontinuous canopy reflectance. Remote Sens Environ 75:423–431.  https://doi.org/10.1016/S0034-4257(00)00184-X Google Scholar
  84. Hultquist C, Chen G, Zhao K (2014) A comparison of Gaussian process regression, random forests and support vector regression for burn severity assessment in diseased forests. Remote Sens Lett 5:723–732.  https://doi.org/10.1080/2150704x.2014.963733 Google Scholar
  85. Jacquemoud S, Baret F (1990) PROSPECT: a model of leaf optical properties spectra. Remote Sens Environ 34:75–91.  https://doi.org/10.1016/0034-4257(90)90100-Z Google Scholar
  86. Jacquemoud S, Ustin SL, Verdebout J, Schmuck G, Andreoli G, Hosgood B (1996) Estimating leaf biochemistry using the PROSPECT leaf optical properties model. Remote Sens Environ 56:194–202.  https://doi.org/10.1016/0034-4257(95)00238-3 Google Scholar
  87. Jacquemoud S, Bacour C, Poilvé H, Frangi J-P (2000) Comparison of four radiative transfer models to simulate plant canopies reflectance: direct and inverse mode. Remote Sens Environ 74:471–481.  https://doi.org/10.1016/S0034-4257(00)00139-5 Google Scholar
  88. Jacquemoud S et al (2009) PROSPECT + SAIL models: a review of use for vegetation characterization. Remote Sens Environ 113:S56–S66.  https://doi.org/10.1016/j.rse.2008.01.026 Google Scholar
  89. Jay S, Bendoula R, Hadoux X, Féret J-B, Gorretta N (2016) A physically-based model for retrieving foliar biochemistry and leaf orientation using close-range imaging spectroscopy. Remote Sens Environ 177:220–236.  https://doi.org/10.1016/j.rse.2016.02.029 Google Scholar
  90. Knyazikhin Y et al (2013) Hyperspectral remote sensing of foliar nitrogen content. Proc Natl Acad Sci 110:E185–E192.  https://doi.org/10.1073/pnas.1210196109 Google Scholar
  91. Koetz B, Schaepman M, Morsdorf F, Bowyer P, Itten K, Allgöwer B (2004) Radiative transfer modeling within a heterogeneous canopy for estimation of forest fire fuel properties. Remote Sens Environ 92:332–344.  https://doi.org/10.1016/j.rse.2004.05.015 Google Scholar
  92. Koetz B, Sun G, Morsdorf F, Ranson KJ, Kneubühler M, Itten K, Allgöwer B (2007) Fusion of imaging spectrometer and LIDAR data over combined radiative transfer models for forest canopy characterization. Remote Sens Environ 106:449–459.  https://doi.org/10.1016/j.rse.2006.09.013 Google Scholar
  93. Kokaly RF, Asner GP, Ollinger SV, Martin ME, Wessman CA (2009) Characterizing canopy biochemistry from imaging spectroscopy and its application to ecosystem studies. Remote Sens Environ 113(Supplement 1):S78–S91.  https://doi.org/10.1016/j.rse.2008.10.018 Google Scholar
  94. Kuusk A, Kuusk J, Lang M (2014) Modeling directional forest reflectance with the hybrid type forest reflectance model FRT. Remote Sens Environ 149:196–204.  https://doi.org/10.1016/j.rse.2014.03.035 Google Scholar
  95. Labate D et al (2009) The PRISMA payload optomechanical design, a high performance instrument for a new hyperspectral mission. Acta Astronaut 65:1429–1436.  https://doi.org/10.1016/j.actaastro.2009.03.077 Google Scholar
  96. Lausch A, Erasmi S, King D, Magdon P, Heurich M (2016) Understanding forest health with remote sensing-part I—a review of spectral traits, processes and remote-sensing characteristics. Remote Sens 8:1029.  https://doi.org/10.3390/rs8121029 Google Scholar
  97. Lausch A, Erasmi S, King D, Magdon P, Heurich M (2017) Understanding forest health with remote sensing-part II—a review of approaches and data models. Remote Sens 9:129.  https://doi.org/10.3390/rs9020129 Google Scholar
  98. Lausch A et al (2018) Understanding forest health with remote sensing, part III: requirements for a scalable multi-source forest health monitoring network based on data science approaches. Remote Sens 10:1120.  https://doi.org/10.3390/rs10071120 Google Scholar
  99. le Maire G, François C, Dufrêne E (2004) Towards universal broad leaf chlorophyll indices using PROSPECT simulated database and hyperspectral reflectance measurements. Remote Sens Environ 89:1–28.  https://doi.org/10.1016/j.rse.2003.09.004 Google Scholar
  100. le Maire G et al (2008) 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 112:3846–3864.  https://doi.org/10.1016/j.rse.2008.06.005 Google Scholar
  101. Lee CM, Cable ML, Hook SJ, Green RO, Ustin SL, Mandl DJ, Middleton EM (2015) An introduction to the NASA Hyperspectral InfraRed Imager (HyspIRI) mission and preparatory activities. Remote Sens Environ 167:6–19.  https://doi.org/10.1016/j.rse.2015.06.012 Google Scholar
  102. Li P, Wang Q (2011) Retrieval of leaf biochemical parameters using PROSPECT inversion: a new approach for alleviating ill-posed problems. IEEE Trans Geosci Remote Sens 49:2499–2506.  https://doi.org/10.1109/tgrs.2011.2109390 Google Scholar
  103. Li D et al (2018) PROCWT: coupling PROSPECT with continuous wavelet transform to improve the retrieval of foliar chemistry from leaf bidirectional reflectance spectra. Remote Sens Environ 206:1–14.  https://doi.org/10.1016/j.rse.2017.12.013 Google Scholar
  104. Liechty HO, Fristoe C (2013) Response of midrotation pine stands to fertilizer and herbicide application in the western Gulf coastal plain. South J Appl For 37:69–74Google Scholar
  105. Lillesaeter O (1982) Spectral reflectance of partly transmitting leaves: laboratory measurements and mathematical modeling. Remote Sens Environ 12:247–254.  https://doi.org/10.1016/0034-4257(82)90057-8 Google Scholar
  106. Lindner M et al (2010) Climate change impacts, adaptive capacity, and vulnerability of European forest ecosystems. For Ecol Manag 259:698–709.  https://doi.org/10.1016/j.foreco.2009.09.023 Google Scholar
  107. Locherer M, Hank T, Danner M, Mauser W (2015) Retrieval of seasonal leaf area index from simulated EnMAP data through optimized LUT-based inversion of the PROSAIL model. Remote Sens 7:10321Google Scholar
  108. Ma Z et al (2012) Regional drought-induced reduction in the biomass carbon sink of Canada’s boreal forests. Proc Natl Acad Sci 109:2423–2427.  https://doi.org/10.1073/pnas.1111576109 Google Scholar
  109. Malenovský Z et al (in review) Variability and uncertainty challenges in upscaling imaging spectroscopy observations from leaves to vegetation canopies. Surv GeophysGoogle Scholar
  110. Malenovský Z, Albrechtová J, Lhotáková Z, Zurita-Milla R, Clevers JGPW, Schaepman ME, Cudlín P (2006a) Applicability of the PROSPECT model for Norway spruce needles. Int J Remote Sens 27:5315–5340.  https://doi.org/10.1080/01431160600762990 Google Scholar
  111. Malenovský Z, Ufer C, Lhotáková Z, Clevers JGPW, Schaepman ME, Albrechtová J, Cudlín P (2006b) A new hyperspectral index for chlorophyll estimation of a forest canopy: area under curve normalised to maximal band depth between 650–725 nm. EARSeL eProc 5:161–172Google Scholar
  112. Martin ME, Plourde LC, Ollinger SV, Smith ML, McNeil BE (2008) A generalizable method for remote sensing of canopy nitrogen across a wide range of forest ecosystems. Remote Sens Environ 112:3511–3519.  https://doi.org/10.1016/j.rse.2008.04.008 Google Scholar
  113. Martin R, Chadwick K, Brodrick P, Carranza-Jimenez L, Vaughn N, Asner G (2018a) An approach for foliar trait retrieval from airborne imaging spectroscopy of tropical forests. Remote Sens 10:199.  https://doi.org/10.3390/rs10020199 Google Scholar
  114. Martin RE et al (2018b) Remote measurement of canopy water content in giant sequoias (Sequoiadendron giganteum) during drought. For Ecol Manag 419–420:279–290.  https://doi.org/10.1016/j.foreco.2017.12.002 Google Scholar
  115. Masek JG et al (2006) A Landsat surface reflectance dataset for North America, 1990–2000. IEEE Geosci Remote Sens Lett 3:68–72.  https://doi.org/10.1109/lgrs.2005.857030 Google Scholar
  116. Matsunaga T et al (2017) Current status of Hyperspectral Imager Suite (HISUI) onboard International Space Station (ISS). In: IEEE international geoscience and remote sensing symposium (IGARSS), 23–28 July 2017, pp 443–446.  https://doi.org/10.1109/igarss.2017.8126989
  117. McNeil BE, de Beurs KM, Eshleman KN, Foster JR, Townsend PA (2007a) Maintenance of ecosystem nitrogen limitation by ephemeral forest disturbance: an assessment using MODIS, Hyperion, and Landsat ETM+. Geophys Res Lett.  https://doi.org/10.1029/2007gl031387 Google Scholar
  118. McNeil BE, Read JM, Driscoll CT (2007b) Foliar nitrogen responses to elevated atmospheric nitrogen deposition in nine temperate forest canopy species. Environ Sci Technol 41:5191–5197.  https://doi.org/10.1021/es062901z Google Scholar
  119. Miller J et al (2005) Development of a vegetation fluorescence canopy model—ESTEC contract no. 16365/02/NL/FF—final reportGoogle Scholar
  120. Nieke J, Rast M (2018) Towards the Copernicus Hyperspectral Imaging Mission for the Environment (CHIME). In: IGARSS 2018—IEEE international geoscience and remote sensing symposium, 22–27 July 2018, pp 157–159.  https://doi.org/10.1109/igarss.2018.8518384
  121. Nink S, Hill J, Buddenbaum H, Stoffels J, Sachtleber T, Langshausen J (2015) Assessing the suitability of future multi- and hyperspectral satellite systems for mapping the spatial distribution of Norway spruce timber volume. Remote Sens 7:12009–12040.  https://doi.org/10.3390/rs70912009 Google Scholar
  122. North PRJ (1996) Three-dimensional forest light interaction model using a Monte Carlo method. IEEE Trans Geosci Remote Sens 34:946–956.  https://doi.org/10.1109/36.508411 Google Scholar
  123. Ollinger SV et al (2008) Canopy nitrogen, carbon assimilation, and albedo in temperate and boreal forests: functional relations and potential climate feedbacks. Proc Natl Acad Sci 105:19336–19341.  https://doi.org/10.1073/pnas.0810021105 Google Scholar
  124. Omari K, White HP, Staenz K (2009) Multiple scattering within the FLAIR model incorporating the photon recollision probability approach. IEEE Trans Geosci Remote Sens 47:2931–2941.  https://doi.org/10.1109/tgrs.2009.2014466 Google Scholar
  125. Omari K, White HP, Staenz K, King DJ (2013) Retrieval of forest canopy parameters by inversion of the PROFLAIR leaf-canopy reflectance model using the LUT approach. IEEE J Sel Top Appl Earth Obs Remote Sens 6:715–723.  https://doi.org/10.1109/jstars.2013.2240264 Google Scholar
  126. Paz-Kagan T et al (2018) Landscape-scale variation in canopy water content of giant sequoias during drought. For Ecol Manag 419–420:291–304.  https://doi.org/10.1016/j.foreco.2017.11.018 Google Scholar
  127. Peng C et al (2011) A drought-induced pervasive increase in tree mortality across Canada’s boreal forests. Nat Clim Change 1:467–471.  https://doi.org/10.1038/nclimate1293 Google Scholar
  128. Peñuelas J, Baret F, Filella I (1995) Semi-empirical indices to assess carotenoids/chlorophyll a ratio from leaf spectral reflectance. Photosynthetica 31:221–230Google Scholar
  129. Rascher U et al (2015) Sun-induced fluorescence—a new probe of photosynthesis: first maps from the imaging spectrometer HyPlant. Glob Change Biol 21:4673–4684.  https://doi.org/10.1111/gcb.13017 Google Scholar
  130. Rennenberg H, Loreto F, Polle A, Brilli F, Fares S, Beniwal RS, Gessler A (2006) Physiological responses of forest trees to heat and drought. Plant Biol 8:556–571.  https://doi.org/10.1055/s-2006-924084 Google Scholar
  131. Riaño D, Vaughan P, Chuvieco E, Zarco-Tejada PJ, Ustin SL (2005) Estimation of fuel moisture content by inversion of radiative transfer models to simulate equivalent water thickness and dry matter content: analysis at leaf and canopy level. IEEE Trans Geosci Remote Sens 43:819–826.  https://doi.org/10.1109/tgrs.2005.843316 Google Scholar
  132. Roberts DA, Gardner M, Church R, Ustin S, Scheer G, Green RO (1998) Mapping chaparral in the Santa Monica Mountains using multiple endmember spectral mixture models. Remote Sens Environ 65:267–279.  https://doi.org/10.1016/S0034-4257(98)00037-6 Google Scholar
  133. Rogge DM, Rivard B, Jinkai Z, Jilu F (2006) Iterative spectral unmixing for optimizing per-pixel endmember sets. IEEE Trans Geosci Remote Sens 44:3725–3736.  https://doi.org/10.1109/tgrs.2006.881123 Google Scholar
  134. Rogge D, Bachmann M, Rivard B, Nielsen AA, Feng J (2014) A spatial–spectral approach for deriving high signal quality eigenvectors for remote sensing image transformations. Int J Appl Earth Obs Geoinf 26:387–398.  https://doi.org/10.1016/j.jag.2013.09.007 Google Scholar
  135. Schelhaas M-J, Nabuurs G-J, Schuck A (2003) Natural disturbances in the European forests in the 19th and 20th centuries. Glob Change Biol 9:1620–1633.  https://doi.org/10.1046/j.1365-2486.2003.00684.x Google Scholar
  136. Schlerf M, Atzberger C (2006) Inversion of a forest reflectance model to estimate structural canopy variables from hyperspectral remote sensing data. Remote Sens Environ 100:281–294.  https://doi.org/10.1016/j.rse.2005.10.006 Google Scholar
  137. Schlerf M, Atzberger C, Hill J, Buddenbaum H, Werner W, Schüler G (2010) Retrieval of chlorophyll and nitrogen in Norway spruce (Picea abies L. Karst.) using imaging spectroscopy. Int J Appl Earth Obs Geoinf 12:17–26.  https://doi.org/10.1016/j.jag.2009.08.006 Google Scholar
  138. Seidl R, Schelhaas M-J, Lexer MJ (2011) Unraveling the drivers of intensifying forest disturbance regimes in Europe. Glob Change Biol 17:2842–2852.  https://doi.org/10.1111/j.1365-2486.2011.02452.x Google Scholar
  139. Serbin SP, Singh A, McNeil BE, Kingdon CC, Townsend PA (2014) Spectroscopic determination of leaf morphological and biochemical traits for northern temperate and boreal tree species. Ecol Appl 24:1651–1669.  https://doi.org/10.1890/13-2110.1 Google Scholar
  140. Singh A, Serbin SP, McNeil BE, Kingdon CC, Townsend PA (2015) Imaging spectroscopy algorithms for mapping canopy foliar chemical and morphological traits and their uncertainties. Ecol Appl 25:2180–2197.  https://doi.org/10.1890/14-2098.1 Google Scholar
  141. Smith M-L, Martin ME, Plourde L, Ollinger SV (2003) Analysis of hyperspectral data for estimation of temperate forest canopy nitrogen concentration: comparison between an airborne (AVIRIS) and a spaceborne (Hyperion) sensor. IEEE Trans Geosci Remote Sens 41:1332–1337.  https://doi.org/10.1109/tgrs.2003.813128 Google Scholar
  142. Somers B, Asner GP (2013) Multi-temporal hyperspectral mixture analysis and feature selection for invasive species mapping in rainforests. Remote Sens Environ 136:14–27.  https://doi.org/10.1016/j.rse.2013.04.006 Google Scholar
  143. Sommer C, Holzwarth S, Heiden U, Heurich M, Müller J, Mauser W (2016) Feature-based tree species classification using hyperspectral and lidar data in the Bavarian Forest National Park. EARSeL eProc 14:49–70.  https://doi.org/10.12760/02-2015-2-05 Google Scholar
  144. Sonobe R, Wang Q (2018) Nondestructive assessments of carotenoids content of broadleaved plant species using hyperspectral indices. Comput Electron Agric 145:18–26.  https://doi.org/10.1016/j.compag.2017.12.022 Google Scholar
  145. Stoffels J et al (2015) Satellite-based derivation of high-resolution forest information layers for operational forest management. Forests 6:1982–2013.  https://doi.org/10.3390/f6061982 Google Scholar
  146. Storch T, Habermeyer M, Eberle S, Muhle H, Mueller RM (2013) Towards a critical design of an operational ground segment for an Earth observation mission. J Appl Remote Sens 7:075381.  https://doi.org/10.1117/1.JRS.7.073581 Google Scholar
  147. Suárez L, Zarco-Tejada PJ, Sepulcre-Cantó G, Pérez-Priego O, Miller JR, Jiménez-Muñoz JC, Sobrino J (2008) Assessing canopy PRI for water stress detection with diurnal airborne imagery. Remote Sens Environ 112:560–575.  https://doi.org/10.1016/j.rse.2007.05.009 Google Scholar
  148. Swann ALS et al (2018) Continental-scale consequences of tree die-offs in North America: identifying where forest loss matters most. Environ Res Lett 13:055014.  https://doi.org/10.1088/1748-9326/aaba0f Google Scholar
  149. Swatantran A, Dubayah R, Roberts D, Hofton M, Blair JB (2011) Mapping biomass and stress in the Sierra Nevada using lidar and hyperspectral data fusion. Remote Sens Environ 115:2917–2930.  https://doi.org/10.1016/j.rse.2010.08.027 Google Scholar
  150. Thompson DR, Natraj V, Green RO, Helmlinger MC, Gao B-C, Eastwood ML (2018) Optimal estimation for imaging spectrometer atmospheric correction. Remote Sens Environ 216:355–373.  https://doi.org/10.1016/j.rse.2018.07.003 Google Scholar
  151. Thompson DR, Guanter L, Berk A, Gao B-C, Richter R, Schläpfer D, Thome KJ (2019) Retrieval of atmospheric parameters and surface reflectance from visible and shortwave infrared imaging spectroscopy data. Surv Geophys.  https://doi.org/10.1007/s10712-018-9488-9 Google Scholar
  152. Townsend PA, Foster JR, Chastain RA, Currie WS (2003) Application of imaging spectroscopy to mapping canopy nitrogen in the forests of the central Appalachian Mountains using Hyperion and AVIRIS. IEEE Trans Geosci Remote Sens 41:1347–1354.  https://doi.org/10.1109/tgrs.2003.813205 Google Scholar
  153. Townsend PA, Serbin SP, Kruger EL, Gamon JA (2013) Disentangling the contribution of biological and physical properties of leaves and canopies in imaging spectroscopy data. Proc Natl Acad Sci 110:E1074.  https://doi.org/10.1073/pnas.1300952110 Google Scholar
  154. Townsend PA, Wang Z, Singh A (2017) Prospects for universal foliar trait-retrieval algorithms from imaging spectroscopy: cross-site and cross-platform. Paper presented at the 10th EARSeL SIG imaging spectroscopy workshop, Zurich, SwitzerlandGoogle Scholar
  155. Ustin SL, Roberts DA, Gamon JA, Asner GP, Green RO (2004) Using imaging spectroscopy to study ecosystem processes and properties. Bioscience 54:523–534.  https://doi.org/10.1641/0006-3568(2004)054%5b0523:uistse%5d2.0.co;2 Google Scholar
  156. Ustin SL, Gitelson AA, Jacquemoud S, Schaepman M, Asner GP, Gamon JA, Zarco-Tejada P (2009) Retrieval of foliar information about plant pigment systems from high resolution spectroscopy. Remote Sens Environ 113:S67–S77.  https://doi.org/10.1016/j.rse.2008.10.019 Google Scholar
  157. Verhoef W (1984) Light scattering by leaf layers with application to canopy reflectance modeling: the SAIL model. Remote Sens Environ 16:125–141.  https://doi.org/10.1016/0034-4257(84)90057-9 Google Scholar
  158. Vermote E, Justice C, Claverie M, Franch B (2016) Preliminary analysis of the performance of the Landsat 8/OLI land surface reflectance product. Remote Sens Environ 185:46–56.  https://doi.org/10.1016/j.rse.2016.04.008 Google Scholar
  159. Verrelst J, Camps-Valls G, Muñoz-Marí J, Rivera JP, Veroustraete F, Clevers JGPW, Moreno J (2015) Optical remote sensing and the retrieval of terrestrial vegetation bio-geophysical properties—a review. ISPRS J Photogramm Remote Sens 108:273–290.  https://doi.org/10.1016/j.isprsjprs.2015.05.005 Google Scholar
  160. Verrelst J et al (2019) Quantifying vegetation biophysical variables from imaging spectroscopy data: a review on retrieval methods. Surv Geophys.  https://doi.org/10.1007/s10712-018-9478-y Google Scholar
  161. Vohland M, Ludwig M, Harbich M, Emmerling C, Thiele-Bruhn S (2016) Using variable selection and wavelets to exploit the full potential of visible/near infrared spectra for predicting soil properties. J Near Infrared Spectrosc 24:255–269.  https://doi.org/10.1255/jnirs.1233 Google Scholar
  162. Wang Z, Skidmore AK, Wang T, Darvishzadeh R, Hearne J (2015) Applicability of the PROSPECT model for estimating protein and cellulose + lignin in fresh leaves. Remote Sens Environ 168:205–218.  https://doi.org/10.1016/j.rse.2015.07.007 Google Scholar
  163. Wang Z et al (2017) Canopy foliar nitrogen retrieved from airborne hyperspectral imagery by correcting for canopy structure effects. Int J Appl Earth Obs Geoinf 54:84–94.  https://doi.org/10.1016/j.jag.2016.09.008 Google Scholar
  164. Wang Z, Skidmore AK, Darvishzadeh R, Wang T (2018) Mapping forest canopy nitrogen content by inversion of coupled leaf-canopy radiative transfer models from airborne hyperspectral imagery. Agric For Meteorol 253–254:247–260.  https://doi.org/10.1016/j.agrformet.2018.02.010 Google Scholar
  165. Wang Z, Townsend PA, Schweiger AK, Couture JJ, Singh A, Hobbie SE, Cavender-Bares J (2019) Mapping foliar functional traits and their uncertainties across three years in a grassland experiment. Remote Sens Environ 221:405–416.  https://doi.org/10.1016/j.rse.2018.11.016 Google Scholar
  166. Waske B, van der Linden S, Oldenburg C, Jakimow B, Rabe A, Hostert P (2012) imageRF—a user-oriented implementation for remote sensing image analysis with Random Forests. Environ Model Softw 35:192–193.  https://doi.org/10.1016/j.envsoft.2012.01.014 Google Scholar
  167. Wessman CA, Aber JD, Peterson DL, Melillo JM (1988) Remote sensing of canopy chemistry and nitrogen cycling in temperate forest ecosystems. Nature 335:154.  https://doi.org/10.1038/335154a0 Google Scholar
  168. Wessman CA, Aber JD, Peterson DL (1989) An evaluation of imaging spectrometry for estimating forest canopy chemistry. Int J Remote Sens 10:1293–1316.  https://doi.org/10.1080/01431168908903969 Google Scholar
  169. White HP, Miller JR, Chen JM (2001) Four-scale linear model for anisotropic reflectance (FLAIR) for plant canopies. I. Model description and partial validation. IEEE Trans Geosci Remote Sens 39:1072–1083.  https://doi.org/10.1109/36.921425 Google Scholar
  170. White JC, Coops NC, Hilker T, Wulder MA, Carroll AL (2007) Detecting mountain pine beetle red attack damage with EO-1 Hyperion moisture indices. Int J Remote Sens 28:2111–2121.  https://doi.org/10.1080/01431160600944028 Google Scholar
  171. Wold S, Ruhe A, Wold H, Dunn IW (1984) The collinearity problem in linear regression. The partial least squares (PLS) approach to generalized inverses. SIAM J Sci Stat Comput 5:735–743.  https://doi.org/10.1137/0905052 Google Scholar
  172. Wold S, Sjöström M, Eriksson L (2001) PLS-regression: a basic tool of chemometrics. Chemometr Intell Lab Syst 58:109–130.  https://doi.org/10.1016/S0169-7439(01)00155-1 Google Scholar
  173. Wolter PT, Townsend PA, Sturtevant BR (2009) Estimation of forest structural parameters using 5 and 10 meter SPOT-5 satellite data. Remote Sens Environ 113:2019–2036.  https://doi.org/10.1016/j.rse.2009.05.009 Google Scholar
  174. Yáñez-Rausell L, Malenovský Z, Clevers JGPW, Schaepman ME (2014a) Minimizing measurement uncertainties of coniferous needle-leaf optical properties. Part II: experimental setup and error analysis. IEEE J Sel Top Appl Earth Obs Remote Sens 7:406–420.  https://doi.org/10.1109/jstars.2013.2292817 Google Scholar
  175. Yáñez-Rausell L, Schaepman ME, Clevers JGPW, Malenovský Z (2014b) Minimizing measurement uncertainties of coniferous needle-leaf optical properties, part I: methodological review. IEEE J Sel Top Appl Earth Obs Remote Sens 7:399–405.  https://doi.org/10.1109/jstars.2013.2272890 Google Scholar
  176. Zhang Q, Middleton EM, Gao B, Cheng Y (2012) Using EO-1 Hyperion to simulate HyspIRI products for a coniferous forest: the fraction of PAR absorbed by chlorophyll (fAPAR_chl) and leaf water content (LWC). IEEE Trans Geosci Remote Sens 50:1844–1852.  https://doi.org/10.1109/tgrs.2011.2169267 Google Scholar

Copyright information

© Springer Nature B.V. 2019

Authors and Affiliations

  1. 1.Environmental Remote Sensing and GeoinformaticsTrier UniversityTrierGermany
  2. 2.Department of Forest and Wildlife EcologyUniversity of WisconsinMadisonUSA

Personalised recommendations