The New Zealand kauri trees (Agathis australis
(D. Don) Lindl.) are a key species of New Zealand’s northern indigenous forests [1
] and are of high cultural [2
] and ecological significance. The conifers are threatened by the deadly kauri dieback disease (Phytophthora agathidicida
(PA)). The soil-borne disease was first officially confirmed by Beever (2009) [3
] in the Waitakere Ranges, although it might have been in New Zealand for decades already [3
]. Meanwhile, it has been verified over major parts of the kauri distribution area [5
]. To date, the monitoring of kauri dieback symptoms has relied on fieldwork and the manual interpretation of aerial images and photos taken from aircraft and helicopters [6
]. There is a need for a cost-efficient, objective approach for the monitoring of stress symptoms which allows for the coverage of large areas [8
1.1. Kauri and Kauri Dieback Disease
The New Zealand kauri is an endemic conifer with a natural distribution in the upper North Island. The existing stands of mature kauri are what remained from extensive logging by European settlers in the 19th and early 20th centuries [9
]. Young kauri have a small conical shape with dense foliage. Older kauri emerge over the surrounding vegetation and develop a massive trunk and a large dome-shaped crown [10
] with measured diameters of over 30 m and heights of up to 40 m in the study areas. The lanceolate leaves of kauri are broad needle-shaped, ca. 2 to 5 cm long, with a smooth leather-like surface [10
] and form a spiky foliage surface. While the foliage of small kauri is dense and evenly spread over the crown, the leaves of medium and large size kauri are arranged in clusters (Figure 1
) which expose gaps, shadows and visible branch material, even in the non-symptomatic stages. The kauri foliage occurs in colour variations from darker yellow-green to lighter blue-green (Figure 2
Infection with PA causes lesions in the trunk and roots, which block the transport of water and nutrients [1
]. The first visible signs in the canopy are yellowing of the leaves and leaf loss in the top of the crowns, which exposes bare branches. In some crowns, the symptoms impair only parts of the upper crown if the transport system is partially blocked. With progressing decline, the foliage becomes sparse, bare branches become exposed, and the influence of woody material, internal shadows, visible undergrowth and ground litter in the canopy reflectance increases. Weakened kauri are less effective in shedding off climbers and epiphytes, which, again, add green plant material in the canopy. In the final stage, the remaining foliage turns brown before it falls off and small branches drop until only a bare skeleton remains. Dead and dying kauri trees are again quickly overgrown by undergrowth, neighbouring trees, epiphytes and climbers.
A variety of factors can accelerate the progress and intensify the symptoms of an existing PA infection, including drought conditions, difficult growing conditions on shallow soil and the exposition to strong and salty winds from the sea. These factors alone and also infections with other pathogens, e.g., Phytophthora cinnamomi
], can cause similar canopy stress symptoms.
1.2. Remote Sensing for Stress Monitoring
Various authors have found that there is a good relationship between spectrally derived indicators extracted from remotely acquired optical imagery and stress symptoms in tree canopies caused by forest diseases [15
Airborne hyperspectral images have been used successfully in many studies to analyse tree canopy health in conifers [22
] and broad-leaved species [28
]. The full spectral range allows for the processing of a spectral continuum and the identification of important bands from a large range of narrow bands [31
] that are sensitive to subtle reflectance changes for early stress detection [32
]. However, high costs for the data acquisition and maintenance of the sensors, elaborate calibration and processing, and small swath widths qualify airborne hyperspectral sensors for the time being more for analytical research tasks than regular large-area forest monitoring.
Spaceborne imagery is the most cost-efficient option to cover larger areas, as it is comparably easy to process and has been widely used to monitor stress responses in tree canopies [35
]. However, satellite images often lack the spatial and spectral resolution for an assessment on the individual tree crown level and are bound to certain over-flight times [19
]. Acquisitions with crewed aircraft are a more expensive option than satellite imagery but provide a more flexible timing and higher spatial and spectral resolutions. Unlike unmanned aerial vehicle (UAV) sensors [43
], multispectral sensors for crewed aircraft are well suited to large-area coverage with a large swath width, a low noise-to-signal ratio and a robust sensor setup [44
]. However, the spectral limitation of multispectral sensors to usually four to six bands requires previous knowledge about the best band combinations to detect the target features. The approach in this study combines the strengths of both the hyperspectral and multispectral platforms. We utilized the high spectral resolution imagery from a hyperspectral sensor to define band and index combinations that are suitable to be mounted on a multispectral sensor on a crewed aircraft for large-area stress monitoring in kauri canopies. The spectral ranges used in this study are defined in Table 1
Vegetation indices (VI) for stress monitoring usually combine bands that are sensitive to the stress parameter(s) with insensitive bands [49
]. An ideal VI for stress analysis shows a linear relationship with the targeted symptoms, is equally sensitive for all levels of stress, independent of the scale, and shows minimal saturation effects [19
]. VIs were developed for all levels of stress from the first, even pre-visible, reactions on leaf level to obscured canopies of dead crowns.
Pre-visible stress reactions in tree foliage have been successfully detected with thermal sensors [51
] and narrow optical bands in the visible (VIS) part of the spectrum [53
]. The first stress symptoms are often a reaction to leaf pigment alteration and reduced canopy water content. VIs that provide a direct measure for canopy water content, like the Moisture Stress Index (MSI) [57
], the Normalized Difference Water Index (NDWI) [58
] and the Water Band Index (WBI) [59
], are based on water absorption bands in the near-infrared (NIR) and shortwave infrared (SWIR) regions. The yellowing of leaves as an early stress symptom is related to biochemical changes in the pigment concentrations, especially leaf chlorophyll [60
]. Absorption coefficients of chlorophyll are strongest in the blue and red region, around 450 and 680 nm, respectively, where green leaves absorb more than 80% of incident light [61
]. While indices in these bands ”saturate” rapidly, Gitelson (2003) [61
] found that ratios with narrow bands in the green and early red-edge regions are more sensitive to changes in chlorophyll, also at higher chlorophyll concentrations.
So-called “greenness” indices describe the reduction in chlorophyll based on its absorption in the red spectrum in combination with bands in the NIR region around 850 nm that are influenced by strong photon scattering in leaf air–cell–wall interfaces [62
]. Since these indices are correlated to the amount of photosynthetic active material, they also capture a reduction in leaf area and changes in leaf angle and thereby, structural changes in the canopy [50
]. Increased stress leads to a decline in red absorption and a narrowing of the red absorption region, which again causes a blue shift of the red-edge point. The red-edge region is very responsive to changes in chlorophyll content [61
]. Narrowband index combinations with red-edge bands have been successfully used to detect early signs of water stress [67
], dying material in Pinus radiata
] and early stress symptoms in conifer woodland [66
]. Further indicators of plant stress include a relative increase in carotenoid pigments and a reduction in leaf nitrogen content, which can be detected with indices that contain characteristic absorption bands in the blue region (445 nm) for carotenoids [68
] and 1510 nm for protein-bound nitrogen [69
Higher amounts of visible dry litter and dead branches are expressed by subtle reflectance characteristics of cellulose and lignin in the SWIR regions [70
]. However, these characteristics are easily obscured by water absorption features and require dry conditions for the most accurate results [62
]. Several studies found a close relationship between bands in the NIR and SWIR region and structural changes in the canopy due to foliage loss [72
]. Although these relationships are non-linear and therefore difficult to interpret [64
], indices in the NIR and SWIR regions were successfully used by Schlerf et al. (2005) [74
] to estimate the Leaf Area Index (LAI) in Norway spruce forests.
The reflectance characteristics of conifers—like a higher absorption, lower transmittance and higher backscattering—are most distinct in the NIR bands [70
]. A high optical depth in the NIR spectral region allows a maximum interaction of photons with crown elements in the lower canopy. It thereby enhances the influence of woody material and understory vegetation [57
When upscaling from leaf-scale responses to crown-scale, a range of crown characteristics need to be taken into account, such as foliage condition, canopy structure, the influence of non-photosynthetic branch and stem material, epiphytes and climbers and, depending on the gap fraction, understory vegetation, soils and ground litter as well as illumination conditions, viewing geometry and reflectance from neighbouring trees [48
]. These attributes have different cumulative effects depending on the spatial scale, such that successful indices at leaf-level often show lower performance at crown level [77
]. Several studies proved that leaf optical properties play only an inferior role in reflectance on canopy scales unless the foliage is dense with a more horizontal orientation and a high Leaf Area Index (LAI) [75
1.3. Approach and Objectives
A method for the detection of kauri trees with indices in the VIS to NIR2 range has already been presented in Meiforth et al. [11
]. This study focuses on the spectral analysis of stress responses in kauri crowns over the full hyperspectral range (437–2435 nm). The aim was to identify the best index combinations that are suitable for large-area stress monitoring with a multispectral sensor.
To account for the spectral characteristics on crown level and the assessment scale of whole crowns in the reference data, we use a crown-based scale for the analysis of canopy stress symptoms over the full spectral range from visible to shortwave infrared. We chose to use a combination of indices rather than one single index to account for the wide spectral range of stress symptoms and phenological characteristics of kauri in different growth and stand situations.
With regards to the practical implementation, we pay special attention to the performance of the recommended multispectral setup for kauri detection in the context of stress detection. And we also explore index combinations in the VIS to NIR range up to 970 nm (VNIR1), which are easier to realize on a multispectral sensor. A pixel-based application of the developed crown-based model was also tested for a more fine-scale prediction of stress responses, especially in larger crowns.
The objectives of this study are to:
Identify the best band and index combinations to detect stress symptoms in kauri crowns for both the full spectral range (VIS–SWIR) and the VNIR1 spectral range. The selected band-combinations should not exceed six wavelengths, to be suitable for a multispectral platform.
Test the performance of a pre-defined band combination for stress detection, which was defined in Meiforth et al. (2019) [11
] to locate kauri trees.
Test the performance of the model and indices-selections that was developed on mean crown values in a pixel-based approach by calculating the model on indices raster.
To support objective 1, we analysed the inner- and intra-crown spectral variability and described the spectral characteristics of kauri crowns for different crown size classes and stress symptom stages from non-symptomatic to dead.
This study addresses symptoms of stress in the canopy that can be caused by PA, but they can also have a range of other causes, such as drought, insect damage or other diseases. Proof of a PA infection still requires systematic soil sampling and analysis in the laboratory [81