Fast track diagnostics: hyperspectral reflectance differentiates disease from drought stress in trees

Abstract

Plant pathogens and environmental stress can have a detrimental impact on forest health, biomass and growth. Early detection and identification of biotic and abiotic stresses, ideally before visible symptoms are present, are important for effective forest management practice and early control of pathogen, and to minimize damage. However, reliably detecting and identifying plant diseases and stress in the field are usually constrained by time and the large size of the area of interest. In this issue, Fallon et al. (2020) demonstrates how hyperspectral reflectance measurements can be used to detect fungal diseases and drought stress in two North American Oak species. The work of Fallon and co-authors exemplifies that fast and non-destructive high-throughput field phenotyping is not limited to crops and agricultural settings (Singh et al. 2016, Lowe et al. 2017), but is also promising to detect and classify the early onset of diseases and stress in forest tree species. Much focus has been on reflectance measurements in the visible, near-infrared (NIR) and short-wave infrared (SWIR), because changes in these wavelength areas are associated with several physiological and biochemical traits (Ge et al. 2019). Hyperspectral measurements are non-destructive, so repeated measurements can be taken of the same plant, and it takes only a few seconds to obtain multiple scans of a single leaf (Ge et al. 2019). In addition, multiple traits or properties can be modelled and estimated from a single leaf scan. All this greatly advances detection and monitoring of biological processes and stress responses at scales from single leaves to ecosystems, to assess for example, the relationship between water shortage and photosynthetic properties of crops (El-Hendawy et al. 2017), to identify species richness and biodiversity in prairie ecosystems (Wang et al. 2018), or for pre-visual detection of symptoms of pathogen infection (Zarco-Tejada et al. 2018). The physical basis for hyperspectral systems is the measurement of the leaf reflectance spectrum resulting from the interaction of the plant canopy with the incoming solar radiation (Mulla 2013). For plant and vegetation studies, the wavelength bands between 400–2500 nm are of particular interest (Curran 1989, Pauli et al. 2016). These bands provide information on the physiological status, structural properties and biochemical composition of plants (Figure 1). This includes the 400–700 nm or photosynthetically active region, where absorption of solar radiation by chlorophyll a, b and carotenoids determines the spectrum; the 700–1400 nm or NIR region, where healthy and non-stressed plants are highly reflective; and the 1400–2500 nm or SWIR region, which can be indicative of water content and the biochemical composition of leaves (Homolová et al. 2013). Spectral reflectance curve for healthy vegetation. Leaf pigment content and composition largely determine the spectrum in the visible region, e.g., absorption by chlorophylls at 430 and 460 nm and 640 and 660 nm cause the dip in spectral reflectance at these wavelengths bands. In the NIR and SWIR, protein content, water content and biochemical composition contribute to the shape of the curve, which then can be used as a diagnostic tool to identify vegetation type or infer plant physiological and health status (modified from after Curran (1989), Pauli et al. (2016)). In their paper, Fallon et al. (2020) identified several diagnostic wavelengths for the detection of oak wilt in asymptomatic leaves in the region of 820–1320 nm, and in the region of 780–2400 nm for the detection of oak wilt in symptomatic leaves caused by cell rupture and altered leaf cellular structure that are not correlated with drought stress and changes in leaf water potential. For the detailed wavelength information, see Fallon et al. (2020). The detection and classification of healthy versus stressed or diseased vegetation follows two approaches: the use of a combination of two or more key wavelengths in the spectrum, commonly known as vegetation indices, and the use of the entire reflectance spectrum. Numerous vegetation indices were developed for monitoring vegetation dynamics, stress and ecological conditions, including the widely used normalized difference vegetation index as a greenness index; the photochemical reflectance index and the chlorophyll/carotenoid index, both of which track carotenoid pigment dynamics (Ustin et al. 2009, Gamon et al. 2016, Wong et al. 2019, D’Odorico et al. 2020); or the plant water index for estimating plant water concentration (Peñuelas et al. 1997). These indices use only a narrow part of the spectra captured by hyperspectral sensors. An extensive body of literature has explored the usefulness of broader parts of the reflectance spectra, including the NIR and SWIR, showing that these regions can reveal chemistry and foliar traits in forest canopies related to photosynthesis, pigments, protein, nitrogen and lignin contents non-destructively (Curran 1989, Wessman et al. 1989, Ustin et al. 2009). In order to handle and analyse broader parts or entire spectra from hyperspectral remote sensing, machine learning algorithms such as partial least squares (PLS) regression (Haaland and Thomas 1988, Wold et al. 2001) have been used for studying leaf physiological and chemical traits in agricultural systems (Ge et al. 2019). In addition, partial least squares discriminant analysis (PLS-DA) is increasingly used in remote sensing-based classification of plant communities and species identification (Peerbhay et al. 2013, Wang et al. 2018). Beyond determining foliar traits and leaf chemical composition, there are considerable efforts underway to exploit hyperspectral reflectance for the early detection of symptoms of plant pathogens to be able to minimize the enormous losses to agricultural or damage to tree yields (Zarco-Tejada et al. 2018). Both pathogen infections and water stress have been shown to cause changes in physiological function such as a decrease in photosynthesis (Cotrozzi et al. 2017, Zarco-Tejada et al. 2018). In agricultural systems, diagnosing and differentiating among multiple pathogens as well as differentiating pathogens and nutrient deficiencies have been quite successful (Abdulridha et al. 2019). However, in natural mixed forest stands differentiating between stress responses that cause very similar symptoms such as wilting due to drought or fungal infection is far more complex and elusive. Nevertheless, Fallon et al. (2020) show how hyperspectral reflectance measurements can differentiate between different fungal diseases and drought stress in North American Oak species. They subjected seedlings of Quercus ellipsoidales and Quercus macrocarpa to drought or inoculated the seedlings with the oak wilt fungus (Bretziella fagacerarum) or the bur oak blight fungus (Tubakia iowensis). The authors then compared instantaneous photosynthesis and stomatal conductance and measured leaf and seedling canopy spectral reflectance to assess physiological changes and symptom appearance. Spectral models developed using PLS-DA were able to predict treatment effects from reflectance data. Most importantly, the spectral models were able to reliably diagnose fungal disease before visible symptoms were apparent. The specific spectral regions that provided the diagnostic information for early disease detection were all in the NIR and SWIR. These bands were significantly different in oak wilt-infected seedlings compared with other treatments, whereas in the visible spectral region oak wilt infection was indistinguishable from drought stress. Interestingly, the diagnostic wavelengths in the NIR and SWIR region are also associated with water content and leaf water potential, and hence the physiological measurements complementing the spectral reflectance data revealed a strong correlation with the mechanisms that contribute to differences between plant disease response and fungal infection. Fallon and co-authors argue that because oak wilt fungus causes cell rupture in the leaves and alters leaf cellular structure, water availability in the leaves is permanently altered and results in much greater reduction in stomatal conductance than in drought stressed plants. Nevertheless, the authors also observed that the accuracy of the spectral models to detect fungal infection differed between the two oak species, and argued this might be caused by differences in the progression of the fungal diseases and differences in the systemic responses of the two oak species to a fungal infection. Fallon et al. (2020) clearly demonstrate that spectral reflectance is a strong diagnostic tool that can reflect different physiological changes associated with stress and fungal diseases response mechanisms; however, the authors also demonstrate that the ability of spectral models as a diagnostic tool can vary from species to species. The results presented by Fallon’s study are solely based on experiments with seedlings, but eventually the goal is to scale this approach to larger trees. It will be interesting to see how accurately this approach will diagnose and differentiate fungal disease and drought stress in complex, whole canopies using versatile hyperspectral sensors on, i.e., drones, where canopy architecture, viewing geometries and solar angle pose additional technical challenges that need to be addressed. None declared.

Plant phenotyping relevance

植物の病害・乾燥ストレスを hyperspectral reflectance で診断する手法を中心に解説する方法論的レビューであり、植物フェノタイピング手法が主要内容である。

Code and data availability

公開本文の所在を確認できませんでした。非公開または購読が必要な可能性があります。

No evidence-backed public reproduction asset is currently recorded.