Spectral Responses to Larval and Artificial Defoliation in Eucalyptus dunnii: Implications for UAV-Based Detection of Gonipterus Damage

Highlights

What are the main findings

• Larval feeding by Gonipterus sp. n. 2 and artificial defoliation significantly altered multispectral reflectance and pigment-related vegetation indices in Eucalyptus dunnii, reducing visible and NIR reflectance and chlorophyll, anthocyanin, and carotenoid indices due to exposure of older leaves.
• Artificial defoliation produced stronger and earlier spectral responses than insect feeding but followed similar temporal patterns, indicating that canopy defoliation dominates the spectral signal and that artificial defoliation is a suitable proxy for insect-induced damage.

What are the implications of the main findings?

• UAV-based multispectral imagery can reliably detect defoliation related canopy and physiological changes, supporting its application for finescale, timely monitoring of forest health and pest impacts.
• Similar spectral responses from insect feeding and artificial defoliation indicate limited capacity of current multispectral methods to distinguish causes of damage, highlighting the value of artificial defoliation as an experimental proxy and the need for improved approaches for cause-specific pest detection.

Abstract

Remote sensing advancements have enhanced defoliation monitoring in forests, but distinguishing insect-specific damage from general canopy stress remains challenging due to overlapping spectral signatures. This study addresses this gap by analyzing multispectral reflectance changes in Eucalyptus dunnii caused by Gonipterus sp. n. 2 larval feeding and artificial defoliation (AD). A randomized complete block design with five replicates tested four treatments: No Damage, Medium (100 larvae/tree) and High (200 larvae/tree) larval inoculation, and AD (80% leaf removal). Twenty potted E. dunnii trees were monitored over 16 days using UAV-based multispectral 10-band imagery. Five multispectral flights were conducted during the experiment. The reduction in visible and near-infrared (NIR) reflectance likely reflects structural changes in canopy composition, namely an increased proportion of mature foliage. Both larval feeding and AD treatments decreased reflectance in these spectral regions, probably due to the removal of young leaves and exposure of older, darker leaves. This explanation is inferred from morphological observations; further biochemical measurements would be required to confirm the underlying mechanisms. Larval feeding and AD reduced chlorophyll-related vegetation indices (CVI, NDRE), decreased anthocyanin-related vegetation indices (mARI, ARI), and also caused a drop in relative carotene content (MTVI, CTRI/RE). The effects were strongest in the AD and peaked soon after the treatment, indicating that these pigment effects can mostly and also be attributed to the older leaves becoming more exposed. Statistically significant interactions between date and treatment were found for the pigment-sensitive indices, the Anthocyanin Reflectance Index (ARI) and the Chlorophyll Vegetation Index (CVI). They displayed opposite reflectance trends—CVI increased while ARI decreased—but followed a consistent pattern aligned with insect feeding. EVI values also exhibited a distinguishable pattern that matched this trend. Due to the inherent difficulty of studying insect feeding in natural settings, AD trials may serve as a practical proxy for assessing the impact of pest-induced damage on vegetation reflectance and physiological indices.

1. Introduction

Invasive insect pests pose a serious threat to global forest ecosystems and commercial plantations, reducing productivity, biodiversity, and long-term sustainability [1,2,3]. Among these, the Eucalyptus Snout Beetle (ESB; Gonipterus spp., Coleoptera: Curculionidae) Gyllenhal, native to southeastern Australia and Tasmania, has emerged as an important invasive pest in Eucalyptus plantations worldwide [4]. ESB refers to three species: Gonipterus platensis, G. pulverulentus and an undescribed Gonipterus sp. n. 2, which all belong to the G. scutellatus species complex [4,5]. Since its first detection in South Africa in 1916, ESB has spread globally, with Gonipterus sp. n. 2 remaining a key invasive pest in South Africa [5,6]. This species preferentially infests Eucalyptus globulus, E. viminalis, and E. smithii, with recent evidence highlighting its growing threat to E. dunnii, a species essential to South Africa’s pulp and paper industry [7,8,9].

Both larval and adult life stages of Gonipterus sp. n. 2 are leaf-feeding and contribute to defoliation and canopy damage. Early instar larvae feed on leaf epidermis, while later instars consume entire young leaves. Adult beetles primarily target the edges of mature leaves, creating a distinctive scalloped appearance [4]. Although classical biological control using the egg parasitoid Anaphes nitens Girault (Hymenoptera: Mymaridae) has historically suppressed outbreaks, climate mismatches and variable host resistance continue to limit parasitoid efficacy [4,10,11]. Approximately 17.5–86% of tree volume growth can be lost due to Gonipterus sp. n. 2 feeding after severe infections [6,12,13]. These challenges highlight the need for improved monitoring tools to detect infestations before irreversible damage occurs.

Conventional forest health monitoring relies on labour-intensive field surveys [14]. Remote sensing offers a scalable alternative by detecting physiological stress through foliar pigment and leaf biomass dynamics, including in less accessible stands. Most importantly, it is so much faster than conventional forest monitoring. Advances in Unmanned Aerial Vehicle (UAV) remote sensing enable high-resolution imagery compared to freely available satellite data and allow scalable detection of stress signals at the individual tree level [14,15,16]. However, linking a spectral change or a change in vegetation indices (VIs) to a specific biotic or abiotic stress remains challenging due to a lack of understanding of the spectral impact of the specific insect feeding, as well as to confounding factors such as leaf age, phenology, and abiotic stressors [17]. Insect feeding reduces leaf area, noticeable through reductions in the near-infrared spectrum which can be detected using VIs focusing on biomass. Insect feeding can also alter pigment content and consequently, a reduction in chlorophyll is expected, which should be detectable with specific VIs sensitive to the red-edge region. As a response to lower chlorophyll, trees can increase carotenoid and/or anthocyanin levels as a photoprotection mechanism, which can also be detected with targeted indices [18,19].

Specifically for Gonipterus sp. n. 2 defoliation, Nzuza et al. [20] have leveraged UAV-derived VIs and machine learning to predict Gonipterus sp. n. 2 defoliation, identifying the Chlorophyll Index (CI), Photochemical Reflectance Index (PRI), and Anthocyanin Reflectance Index (ARI) as key predictors [20]. Similarly, Lottering et al. [21] used high-resolution satellite imagery to assess canopy damage by Gonipterus sp. n. 2, with chlorophyll-sensitive indices (e.g., Modified Chlorophyll Absorption Reflectance Index, MCARI) and stress-related indices like ARI and Normalized Difference Vegetation Index (NDVI) ranking high in feature importance. While these studies establish which VIs correlate with Gonipterus sp. n. 2 damage, they leave a fundamental mechanistic question unresolved: do these indices primarily detect structural canopy loss (defoliation), insect-induced biochemical stress (pigment alteration), or a combination of both? This distinction is critical because physical defoliation (leaf area reduction) and herbivore-induced biochemical responses arise from different physiological processes, may follow different temporal trajectories, and carry different implications for forest management and yield forecasting.

A second unresolved issue concerns the widespread use of artificial defoliation (AD) as a proxy for insect herbivory. AD provides a controlled framework for isolating the structural effects of canopy loss [22]; for example, Barry et al. [23] used it to demonstrate compensatory photosynthetic responses in Eucalyptus following defoliation. However, AD does not replicate the full complexity of natural herbivory. Real insect feeding involves not only tissue loss but also the introduction of herbivore-associated molecular patterns (HAMPs) such as salivary effectors that activate distinct phytohormonal signalling cascades [24]. These biotic cues can systematically alter plant physiology, including the regulation of photosynthetic pigments, in ways that physical wounding alone does not. Consequently, while AD effectively mimics structural canopy damage, it may fail to elicit the same biochemical stress responses that influence spectral signatures in the visible spectrum. Critically, no study has directly compared AD with natural Gonipterus sp. n. 2 larval feeding to determine whether their spectral responses are equivalent. Without such validation, it remains unclear whether conclusions drawn from AD experiments accurately represent natural infestation dynamics.

To date, no study has experimentally separated these mechanisms, nor isolated the specific spectral effects of the larval stage—the primary defoliator. The first innovation of this study is therefore to experimentally isolate and characterize the spectral signature of larval feeding dynamics, determining whether previously identified VIs detect structural loss, biochemical stress responses, or both.

The second innovation of this study is to provide the first direct experimental comparison between AD and natural larval herbivory, testing whether AD is a valid proxy for infestation or whether insect-specific interactions generate distinct spectral signatures.

We set up a controlled open-field trial experiment in which potted trees were infested with Gonipterus sp. n. 2 larvae, and UAV multispectral imagery was collected over 17 days with five flights. We hypothesize that

-

H1 (Structural Impact): Both larval feeding and AD will cause significant decreases in VIs primarily sensitive to canopy structure and biomass (EVI, NDVI). This prediction is based on the direct removal of photosynthetic tissue by both treatments, which reduces overall leaf area and consequently lowers near-infrared reflectance [22,23].

-

H2 (Chlorophyll Dynamics): Larval feeding, involving sustained tissue removal and the introduction of herbivore-associated molecular patterns, will trigger a significant decrease in chlorophyll content, detectable through chlorophyll-sensitive indices (NDRE, CI). We expect that AD will have a weaker or delayed effect on these indices, as the biochemical signaling cascades leading to chlorophyll degradation are less strongly activated by mechanical wounding alone [24].

-

H3 (Photoprotective Pigment Induction): Larval feeding will increase photoprotective pigment concentrations, detectable as a rise in anthocyanin-sensitive indices (ARI, mARI) and indices related to the carotenoid: chlorophyll ratio. This response is expected as a downstream consequence of herbivore-induced chlorophyll degradation and sustained biotic stress, which triggers the upregulation of these pigments to dissipate excess energy and mitigate oxidative damage [18,19]. We expect that AD will not elicit a comparable increase in these indices, as the systemic biochemical signals required for their accumulation are largely absent.

-

H4 (Comparative Spectral Pathways): Collectively, we expect that larval damage will result in a compound spectral response: increased reflectance in the visible spectrum (400–700 nm) due to reduced chlorophyll absorption, coupled with a decreased reflectance in the near-infrared (NIR) due to reduced biomass. We hypothesize that the spectral impact of AD will be dominated by the NIR decrease from tissue loss, with minimal accompanying changes in the visible spectrum. A significant divergence in the visible-spectrum response between the two treatments would indicate that Gonipterus sp. n. 2 damage cannot be attributed to defoliation alone, and that herbivore-specific biochemical processes are a key driver of stress-induced spectral changes.

2. Materials and Methods

2.1. Study Area

This study was conducted at the University of Pretoria’s experimental farm in Pretoria, Gauteng, South Africa (25°44′57.4″ S, 28°15′38.6″ E), from 2 to 18 September 2024 (Figure 1). Pretoria has a humid subtropical climate with summer rains, an average annual rainfall of 706 mm [25], an average summer maximum temperature of 26.7 °C and minimum temperature of 13.9 °C (https://en.climate-data.org/africa/south-africa/gauteng/pretoria-154/#climate-graph, accessed on 25 March 2026).

2.2. Experimental Design

The study investigated the effects of Gonipterus sp. n. 2 larval feeding and AD on the physiological recovery of Eucalyptus dunnii. A randomized block design with five replications was implemented. Twenty uniform potted E. dunnii trees (height: 1.2–1.5 m) were obtained from the Forestry and Agricultural Biotechnology Institute (FABI) nursery. Each tree was planted in a 10-litre pot filled with two bags of compost and one shovel of soil, supplemented with 10 g of a balanced N-P-K (15-15-15) fertilizer to ensure healthy growth. Trees were spaced 1 m apart in an open field and watered three times per week throughout the experimental period to maintain consistent soil moisture. The experiment included four treatments, randomized within blocks: (1) a control group with undamaged trees maintained under natural environmental conditions (No Damage); (2) a medium inoculation, where 100 Gonipterus sp. n. 2 larvae were placed on each tree (Medium); (3) a high inoculation, where 200 Gonipterus sp. n. 2 larvae were placed on each tree (High); and (4) AD to simulate severe defoliation through manual removal of new flush leaves. On each branch, 80% of the new flush leaves were removed using secateurs to simulate defoliation in a standardized way. Leaves were distinguished between old and new flush, with older leaves appearing darker.

Larvae were collected in stands in Melmoth, KwaZulu-Natal, and transported to the Forestry and Agricultural Biotechnology Institute (FABI), University of Pretoria the week before the trial started. During transportation, the larvae were kept in a brown paper bag with Eucalyptus leaves inside a cooler box. Upon arrival, the larvae were transferred to a container with fresh Eucalyptus leaves and incubated for two days in a Memmert IN110 incubator (Memmert GmbH + Co. KG Schwabach, Germany). Trees were infested on 2 September 2024. The larvae were put in a cooler box and were then carefully placed onto fresh leaves using a 6.4 mm paint brush to minimize disturbance, allowing them time to attach before proceeding. A mixture of different instar larvae were used to replicate natural feeding behaviour, and infested trees were monitored daily to confirm larval activity. Larvae remained on the trees until they naturally dropped off. All treatments, including larval inoculation and AD, were applied once at the start of the experiment.

2.3. UAV Flights

The flights were carried out between 2 and 18 September 2024 (

Table 1. UAV flight information.

Date	Days After Treatment	Time	Flight Conditions
2 September 2024	0	11:00	Sunny
6 September 2024	4	11:30	Sunny
10 September 2024	8	12:30	Cloudy
13 September 2024	11	11:10	Sunny
18 September 2024	16	11:30	Sunny

). The flight on 2 September was executed immediately before the inoculation and AD treatment, and can be considered as a baseline. All timepoints are hereafter referred to as days after treatment (DAT). The UAV flights were conducted using a DJI M300 UAV with a dual gimbal connector (DJI, Shenzhen, Guangdong, China) and equipped with a multispectral camera (MicaSense RedEdge Dual MX Camera, AgEagle, Wichita, KS, USA) (not on a gimbal) and a high-resolution RGB camera (DJI Zenmuse P1). Nine permanent ground control points (GCPs) were placed around the experimental area, and their exact location was retrieved with an RTK-GNSS receiver (S10 GNSS Receiver, Stonex, Paderno Dugnano, Milan, Italy).

DJI Pilot 2 was used for the waypoint flight planning and execution. The automated flight missions were conducted at a flight altitude of 12 m (Table 1). Overlap of 80% was maintained horizontally and vertically, with the flight speed set to 5 m/s [26]. The RedEdge Dual MX camera is a multispectral camera with 10 spectral bands in the coastal blue (444 ± 28 nm), blue (475 ± 32 nm), green (531 ± 14 nm & 560 ± 27 nm), red (650 ± 16 nm & 668 ± 14 nm), red-edge (705 ± 10 nm, 717 ± 12 nm & 740 ± 38 nm), and near-infrared (842 ± 57 nm) spectrum. A single grey reference panel measurement was taken by holding the UAV about 1 m above a standard single panel (Calibrated Reflectance Panel, MicaSense), avoiding shadows on the reference panel and on the downwelling light sensor [27]. This sensor configuration is particularly significant because it includes three relatively narrow red-edge bands (centered at approximately 700 nm, 717 nm, and 740 nm). The red-edge region (680–750 nm) is highly sensitive to changes in vegetation chlorophyll content while being less sensitive to leaf area influences—hence, VIs using red edge can be used to single out chlorophyll effects from leaf biomass/area effects. The availability of these specific bands is fundamental to this study, as it enables the accurate calculation of stress-sensitive indices such as NDRE and CTRI/RE. By leveraging these dedicated red-edge bands, our analysis can effectively capture the subtle spectral signatures of vegetation stress, ensuring that the derived indices are both reliable and physically meaningful for interpretation. The DJI Zenmuse 45-megapixel P1 RGB camera is capable of acquiring images with a resolution of 8192 × 5460 pixels and had a ground sampling distance (GSD) of 0.15 cm/pixel.

Aerotriangulation was performed during image alignment in Metashape Professional v1.4.3 (Agisoft LLC, St. Petersburg, Russia), to optimize camera exterior orientation parameters through bundle adjustment. The process yielded a mean reprojection error of 0.32 pixels, indicating high-quality image matching and model stability. Georeferencing using nine well-distributed ground control points (GCPs) resulted in a horizontal RMSE of 4.2 mm and a vertical RMSE of 5.6 mm, demonstrating sufficient spatial accuracy for detecting treatment effects at the individual pot scale.

2.4. Image Processing, Vegetation Index Retrieval, Statistical Analysis and Leaf Counts

Orthomosaics for multispectral and RGB imagery were generated in Agisoft Metashape Professional v1.4.3 (Agisoft LLC, St. Petersburg, Russia). For the multispectral imagery, radiometric calibration was performed within Agisoft Metashape by applying grey reference panel corrections and converting raw digital numbers (DN) to surface reflectance values across all flights. The information from the downwelling light sensor (DLS) was used for the flight in cloudy conditions (10 September 2024, DAT8). The GCPs were used for georeferencing. The orthomosaics were exported as TIF files for subsequent processing.

The Soil-Adjusted Vegetation Index (SAVI), calculated as SAVI = ((NIR − Red)/(NIR + Red + L)) × (1 + L), was used to refine vegetation classification by improving the separation of vegetative areas from the soil background. To further minimize non-foliar background effects, vegetation masking was enhanced using the SAVI, ensuring accurate spectral analysis. Individual tree crowns were delineated using a semi-automated workflow. Initial crown boundaries were generated through threshold-based segmentation and grouping of contiguous vegetation pixels, followed by manual refinement to separate adjacent crowns and correct boundary inaccuracies. For each finalized crown polygon, the mean spectral reflectance and VIs were extracted using zonal statistics in Python 3.8. In total, eight VIs were calculated (see Table 2), targeting vegetation biomass (leaf area), chlorophyll levels, anthocyanin levels, and the carotene-to-chlorophyll ratio.

Due to adverse weather conditions, the flight on DAT8 (10 September 2024) was executed in cloudy conditions (Table 1). This affected the reflectance. We applied a temporal normalization procedure to account for measurement variability across dates. For each spectral band (i), the reference reflectance of the No Damage trees was defined as the average of the measurements obtained on DAT4 (6 September 2024) and DAT11 (13 September 2024), under the assumption that the reflectance on DAT8 should be intermediate between these dates. The original reflectance of each band was then corrected across the entire image:

R_{corr_DAT,i} = R_{orig_DAT8,i} × (R_{ND_DAT4,11}/R_{ND_DAT8})

(1)

with R_{corr_DAT8,i} and R_{orig_DAT8,i} the corrected and uncorrected reflectance values for band i on DAT8, respectively, and R_{ND_DAT8} the average uncorrected reflectance of the No Damage trees on DAT8. R_{ND_DAT4,11} is the mean reflectance of the No Damage trees on DAT4 and DAT11. R_{ND_DAT8} and R_{ND_DAT4,11} were both obtained after masking non-vegetative pixels, as described above.

The number of new flush leaves was manually counted from high-resolution Zenmuse P1 orthomosaics by zooming into each tree canopy. This was done for each date and each tree separately.

Table 2 lists the VIs calculated from the reflectance data, focusing on biomass (EVI, NDVI), chlorophyll concentration (NDRE, CVI), anthocyanin concentration (ARI, mARI), and the ratio of carotene over chlorophyll concentration (MTVI, CTRI/RE). Notice that we focused on the ratio rather than the absolute carotene concentration, as carotene is typically correlated with chlorophyll and is highest in healthy, unstressed leaves. Under stress, carotene may increase while chlorophyll decreases, making their ratio a more reliable indicator of biotic or abiotic stress [28].

Table 2. Overview of the different vegetation indices calculated in this study.

Target Property	Vegetation Index	Formula	References
Biomass	Enhanced Vegetation Index (EVI)	$2.5\mathrm{X}{\displaystyle \frac{{NIR}_{842}-R_{668}}{{NIR}_{842}+{6xR}_{668}-7.5\mathrm{x}{\mathrm{B}}_{475}+1}}$	[29]
Biomass	Normalized Difference Vegetation Index (NDVI)	$\mathrm{N}\mathrm{D}\mathrm{V}\mathrm{I}={\displaystyle \frac{{NIR}_{842}-R_{668}}{{NIR}_{842}+R_{668}}}$	[30]
Chlorophyll	Normalized Difference Red Edge (NDRE)	$\mathrm{N}\mathrm{D}\mathrm{R}\mathrm{E}={\displaystyle \frac{{NIR}_{842}-{RE}_{717}}{{NIR}_{842}+{RE}_{717}}}$	[31]
Chlorophyll	Chlorophyll Vegetation Index (CVI)	$\mathrm{C}\mathrm{V}\mathrm{I}={\displaystyle \frac{{NIR}_{842}\ast R_{668}}{G_{560}}}-1$	[32]
Anthocyanin	Anthocyanin Reflectance Index (ARI)	$\mathrm{A}\mathrm{R}\mathrm{I}=\left({\displaystyle \frac{1}{G_{560}}}\right)-\left({\displaystyle \frac{1}{{\mathrm{R}\mathrm{E}}_{705}}}\right)$	[33]
Anthocyanin	Modified Anthocyanin Reflectance Index (MARI)	$\mathrm{M}\mathrm{A}\mathrm{R}\mathrm{I}=\left(\left({\displaystyle \frac{1}{G_{560}}}\right)-\left({\displaystyle \frac{1}{{\mathrm{R}\mathrm{E}}_{705}}}\right)\right)\ast {NIR}_{842}$	[34]
Carotene/Chlorophyll	Modified Transformed Vegetation Index (MTVI)	$\mathrm{T}\mathrm{V}\mathrm{I}=1.2\ast \left[1.2\ast \left({NIR}_{842}-G_{560}\right)-2.5\ast \left(R_{668}-R_{560} \right)\right]$	[35]
Carotene/Chlorophyll	Carotenoid Triangular Ratio Index/Chlorophyll Red-Edge Index (CTRI/RE)	$\mathrm{C}\mathrm{T}\mathrm{R}\mathrm{I}/\mathrm{R}\mathrm{E}=1.2\ast [1.2\ast ({\mathrm{N}\mathrm{I}\mathrm{R}}_{842}-{\mathrm{G}}_{560})-2.5\ast ({\mathrm{R}}_{668}-{\mathrm{G}}_{560}\left)\right]]/{\mathrm{G}}_{531}/[({\mathrm{R}\mathrm{E}}_{750}-{\mathrm{R}\mathrm{E}}_{705})/{\mathrm{R}\mathrm{E}}_{705} ]$	[28]

Table 2. Overview of the different vegetation indices calculated in this study.

Target Property	Vegetation Index	Formula	References
Biomass	Enhanced Vegetation Index (EVI)	$2.5\mathrm{X}{\displaystyle \frac{{NIR}_{842}-R_{668}}{{NIR}_{842}+{6xR}_{668}-7.5\mathrm{x}{\mathrm{B}}_{475}+1}}$	[29]
Biomass	Normalized Difference Vegetation Index (NDVI)	$\mathrm{N}\mathrm{D}\mathrm{V}\mathrm{I}={\displaystyle \frac{{NIR}_{842}-R_{668}}{{NIR}_{842}+R_{668}}}$	[30]
Chlorophyll	Normalized Difference Red Edge (NDRE)	$\mathrm{N}\mathrm{D}\mathrm{R}\mathrm{E}={\displaystyle \frac{{NIR}_{842}-{RE}_{717}}{{NIR}_{842}+{RE}_{717}}}$	[31]
Chlorophyll	Chlorophyll Vegetation Index (CVI)	$\mathrm{C}\mathrm{V}\mathrm{I}={\displaystyle \frac{{NIR}_{842}\ast R_{668}}{G_{560}}}-1$	[32]
Anthocyanin	Anthocyanin Reflectance Index (ARI)	$\mathrm{A}\mathrm{R}\mathrm{I}=\left({\displaystyle \frac{1}{G_{560}}}\right)-\left({\displaystyle \frac{1}{{\mathrm{R}\mathrm{E}}_{705}}}\right)$	[33]
Anthocyanin	Modified Anthocyanin Reflectance Index (MARI)	$\mathrm{M}\mathrm{A}\mathrm{R}\mathrm{I}=\left(\left({\displaystyle \frac{1}{G_{560}}}\right)-\left({\displaystyle \frac{1}{{\mathrm{R}\mathrm{E}}_{705}}}\right)\right)\ast {NIR}_{842}$	[34]
Carotene/Chlorophyll	Modified Transformed Vegetation Index (MTVI)	$\mathrm{T}\mathrm{V}\mathrm{I}=1.2\ast \left[1.2\ast \left({NIR}_{842}-G_{560}\right)-2.5\ast \left(R_{668}-R_{560} \right)\right]$	[35]
Carotene/Chlorophyll	Carotenoid Triangular Ratio Index/Chlorophyll Red-Edge Index (CTRI/RE)	$\mathrm{C}\mathrm{T}\mathrm{R}\mathrm{I}/\mathrm{R}\mathrm{E}=1.2\ast [1.2\ast ({\mathrm{N}\mathrm{I}\mathrm{R}}_{842}-{\mathrm{G}}_{560})-2.5\ast ({\mathrm{R}}_{668}-{\mathrm{G}}_{560}\left)\right]]/{\mathrm{G}}_{531}/[({\mathrm{R}\mathrm{E}}_{750}-{\mathrm{R}\mathrm{E}}_{705})/{\mathrm{R}\mathrm{E}}_{705} ]$	[28]

2.5. Statistical Processing

Repeated Measures Analysis of Variance (Repeated Measures ANOVA) was conducted to evaluate the interaction between treatment and date on VIs. Prior to the ANOVA, data were tested for normality and homoscedasticity using Shapiro–Wilk and Levene’s tests, respectively. Both tests indicated positive outcomes, with the Shapiro–Wilk test confirming that residuals met the assumption of normality (p > 0.05) and Levene’s test showing no evidence of heteroscedasticity (p > 0.05). Following the ANOVA, Tukey’s Honest Significant Difference (HSD) test was performed as a post hoc analysis, with a Bonferroni correction applied to adjust for multiple comparisons. All statistical analyses were conducted in Python version 3.8.

3. Results

3.1. Number of New Flush Leaves

In both the medium- and high-damage treatments, based on visual observation, feeding activity left distinct tracks on the leaves, which gradually turned reddish-brown in colour. This discolouration became more apparent by DAT6, when the effects were significant and continued to progress thereafter. The damaged leaves also showed a noticeable reduction in leaf area, particularly under the high-damage treatment, where the symptoms appeared more severe.

The initial number of new flush leaves was similar in all treatments (AD, High, Medium, No Damage), ranging between 600 and 650 leaves per tree at DAT0 (Figure 2). In the no-damage treatment, the number of new flush leaves increased as the experimental trial continued. In contrast, the medium- and high-damage treatments had a lower number of new flush leaves, with a minimum for the high damage at DAT4, and for the medium damage at DAT8, after which fresh leaf number recovered slightly. The AD treatment had the largest drop in fresh leaf number, with the lowest value at DAT4, recovering slowly afterwards.

3.2. Spectral Response to Gonipterus Feeding

The reflectance spectra did not differ at DAT0, but were markedly different afterwards. On all days after treatment, reflectance in the visual, red-edge and near-infrared region was lowest for the AD treatment, followed by the high- and medium-larval-damage treatment and finally no-damage treatment (Figure 3). Despite the similar flight hours and similar conditions (apart from the flights on 10 September), there were relatively large differences in absolute values in reflectance between the different days.

3.3. Vegetation Index Response to Gonipterus Feeding

Figure 4 presents the correlation matrix among the number of new flush leaves and the evaluated VIs, revealing two main patterns. First, the number of new flush leaves shows weak correlations with all spectral indices, indicating no strong linear association with any individual index. Second, clear interrelationships are evident among the VIs. ARI and mARI are strongly and positively correlated and show negative correlations with chlorophyll-associated indices such as CVI. EVI and NDVI are positively correlated with CVI and MTVI and negatively correlated with ARI and mARI. MTVI and CTRI-RE are strongly positively correlated with each other and with chlorophyll-related indices. Overall, the matrix demonstrates a structured pattern in which anthocyanin-related indices and chlorophyll-associated indices form opposing correlation groups, whereas flush leaf number does not cluster strongly with either group.

All indices satisfied the assumptions of homoscedasticity and sphericity. The Shapiro–Wilk test indicated that residuals did not deviate from normality (W < 0.05, p < 0.05) for all indices. The Treatment and Date variables were significant for most VIs, and the Treatment × Date interaction was significant for some but not all indices (

Table 3. Statistical analysis of vegetation indices using repeated measures ANOVA across treatments, dates and their interactions.

Target Property	Vegetation Indices	Treatment	Date	Treatment × Date
Biomass	EVI	<0.001	<0.001	0.17
	NDVI	<0.05	<0.001	0.09
Chlorophyll	NDRE	<0.05	<0.05	0.11
	CVI	<0.001	<0.001	<0.05
Anthocyanin	ARI	<0.001	<0.001	<0.05
	mARI	<0.001	<0.001	0.23
Carotene/Chlorophyll	MTVI	<0.001	<0.001	0.07
	CTRI/CIRed-edge	<0.05	<0.001	0.46

). For all indices, no significant differences were noted on DAT0, as expected, yet all other dates showed significant differences. The effects were in general most explicit on DAT4 (Figure 3). For the biomass indices, the Treatment effect was larger on EVI than on NDVI. In line with the hypotheses, the AD treatment had the lowest values, the no-damage treatment the highest values, and the medium- and high-larval-damage treatment had intermediate damage, with high damage differing significantly from no damage at DAT4; by the end of the experiment, there was no longer a significant difference.

The Treatment effect on the chlorophyll indices (CVI and NDRE) was similar. No damage had the highest and AD the lowest values throughout, with medium- and high-damage leaves having intermediate values, which were not statistically significant from the no-damage treatment. Both ARI and mARI were significantly affected by the Treatment and Day, and Treatment × Day interaction effect was also significant on ARI (Figure 5c). The AD treatment had the highest levels, indicating higher anthocyanin content, followed by high damage, medium and finally no damage with the lowest levels. A very similar pattern was observed for the relative carotene indices (Figure 5d).

4. Discussion

The first hypothesis, that both larval damage and AD would decrease reflectance in NIR, and subsequently the biomass-related VIs NDVI and EVI, was confirmed. This supports earlier findings by Lottering et al. [21], who showed that Gonipterus sp. n. 2 damage causes a decrease in EVI. The severity of the spectral response depended on larval density and stage, with early instar larvae expected to cause substantial biomass loss when fresh foliage is available [6].

In the visual bands, the spectral response was opposite to what we expected (H4) (Figure 3). We hypothesized an increase in reflectance in the medium- and high-damage treatments, particularly in the red spectra, due to the anticipated reduction in chlorophyll levels. Instead, the reflectance was consistently lower for the high- and medium-damage and AD treatment in the visual bands compared to the no-damage treatment, with AD treatment showing the sharpest reduction. A very similar spectral signal was earlier observed by Nzuza et al. [20] for UAV multispectral data of E. dunnii trees affected by Gonipterus sp. n. 2 in field conditions. New flush leaves typically have a higher reflectance in the visual spectra compared to mature leaves [36], which was also observed in Eucalyptus species [37] and by us when doing the fieldwork. For all the mentioned treatments, younger (new flush) leaves are removed, causing the mature, older leaves to be more exposed. As such, we believe that the observed spectral shift towards lower reflectance is mainly due to the increased contribution of older leaves on the signal.

The observed effect of Gonipterus sp. n.2 feeding on the VIs was largely expected. Indeed, feeding by Gonipterus sp. n. 2 larvae led to a decrease in the chlorophyll-related VIs (Hypothesis H2), as well as an increase in anthocyanin-related indices (H3), also as expected. Elevated anthocyanin levels in leaf tissues are a well-documented photoprotective response to oxidative stress caused by excess light [38]. Likewise, indices associated with the carotene/chlorophyll ratio suggested a relative increase in carotene levels is a general stress reaction [39] (H3). However, it remains unclear whether this pattern reflects an active physiological response to leaf feeding or simply the greater visibility of older leaves in the imaging.

A consistent pattern in spectral reflectance and VIs emerged throughout the experiment. This pattern provides a general understanding of how Gonipterus sp. n. 2 feeding influences leaf reflectance. Nevertheless, the study had several limitations. First, the imagery on DAT8 was collected on a cloudy day. Second, some spectral index dynamics were inconsistent over time, for instance, showing a drop in ARI (Figure 5c) over time for all treatments. These inconsistencies are likely due to calibration inaccuracies. We used a single grey reference panel for calibration, which can influence overall data quality [27]. Future flights should use multiple grey reference panels for empirical line corrections. Furthermore, segmenting soil pixels from individual trees can be challenging due to mixed pixels—a common issue with high-resolution imagery [40]. The consistent patterns observed across treatments and dates suggest that any potential influence on spectral reflectance, particularly in the red and green regions, where soil reflectance is high, did not affect the overall results. However, we observed consistent reflectance responses across the days and across the treatments; so, we believe the influence of mixed pixels is minor.

Although the variations observed in ARI, mARI, and carotenoid/chlorophyll-related indices are consistent with documented plant stress responses such as increased anthocyanin accumulation for photoprotection under excess light and oxidative stress [38], these interpretations must be considered cautiously. In the absence of direct biochemical validation (e.g., pigment extraction or reliable in situ spectroscopy), the spectral index dynamics cannot be conclusively attributed to active physiological adjustments. The absence of hyperspectral measurements further limited our ability to derive validated pigment-specific indices, thereby restricting mechanistic inference. Accordingly, the observed changes in ARI, mARI, and carotene/chlorophyll-related indices should be interpreted as associational rather than causal.

Alternative explanations remain plausible. In particular, the removal of younger flush leaves in the high- and medium-damage, and AD treatments increased the proportional contribution of mature leaves to the canopy signal. Because leaf age strongly influences pigment composition and optical properties, the spectral shifts may reflect changes in canopy composition rather than induced pigment synthesis. The fact that the strongest differences occurred in the AD treatment and peaked at DAT4, when defoliation was maximal and before new flushing occurred, further supports the predominance of structural and compositional effects. While this interpretation is plausible, it remains hypothetical in the absence of direct measurements of leaf area index, pigment concentrations, or canopy biomass, and alternative stress-driven physiological mechanisms cannot be excluded. Future studies integrating direct pigment assays, structural measurements (e.g., LiDAR-derived canopy metrics), or validated hyperspectral datasets will be necessary to disentangle physiological responses from canopy compositional changes and to establish mechanistic causality.

5. Conclusions

This study provides evidence that Gonipterus sp. n. 2 larval feeding induces distinct spectral responses in E. dunnii that can be detected using UAV-based multispectral imagery. These spectral responses align with established patterns associated with insect feeding regimes. Both larval feeding and AD reduced near-infrared reflectance, as expected. The changes in the reflectance of the visual bands as well as in the pigment-related indices suggest that the main effect of AD or larval damage is the exposure of older, darker leaves, with higher pigment concentration, although a response by increasing anthocyanin and relative carotene content cannot be excluded. Overall, the findings demonstrate that UAV-based multispectral imagery can effectively detect damage caused by Gonipterus sp. n. 2. To disentangle structural canopy effects from active physiological responses, future studies should integrate direct pigment assays, structural measurements such as leaf area index, and validated hyperspectral datasets. Such an approach would enable a more mechanistic understanding of the spectral signatures associated with pest infestation.