Hyperspectral Imaging Reveals Chlorophyll Temporal Dynamics in Masson Pine Under Pine Wood Nematode and Abiotic Stresses

Highlights

What are the main findings?

• Identification of stress-specific spectral fingerprints: This study identified unique hyperspectral signatures for Masson pine under different lethal stressors. Waterlogging stress was characterized by sensitivity in the green band (534–536 nm), while pine wood nematode infection triggered a response in the blue to red edge region (450–760 nm). A “linear-dominant, nonlinearly-mixed” link between spectral indices and needle chlorophyll content was also revealed for the first time.
• Superior performance of optimized narrow-band indices and machine learning models: Optimized narrow-band spectral indices, particularly NDSI (689, 907) and continuum-removed spectra at 534–536 nm, strongly correlated with needle chlorophyll content (R² ≈ 0.84), surpassing previous studies. RF and XGBoost models showed the best accuracy and robustness for needle chlorophyll content inversion in multi-stress conditions.

What are the implications of the main findings?

• Provides a reliable methodology for early stress diagnosis: The stress-specific spectral imagings and the high-accuracy inversion models enable the early detection and differentiation of pine wilt disease, drought, and waterlogging stress before visible symptoms appear, which is crucial for proactive forest management.
• Offers practical tools for large-scale forest health monitoring: The findings demonstrate the high efficiency of combining hyperspectral imaging (especially UAV-based) with machine learning models. This provides a viable technical solution for non-destructive, real-time monitoring of Masson pine health status over large areas, supporting precision forestry and pest control strategies.

Abstract

Masson Pine (Pinus massoniana), an important afforestation species in southern China, is severely threatened by pine wilt disease caused by pine wood nematode (Bursaphelenchus xylophilus, PWN). To differentiate mortality induced by B. xylophilus from that caused by abiotic environmental factors, hyperspectral imaging and needle chlorophyll content were measured and analyzed for the early detection physiological changes in Masson pine seedlings under various environmental stressors. Four-year-old Masson pine seedlings were subjected to PWN inoculation, mechanical injury, drought, and waterlogging treatments. Hyperspectral reflectance and needle chlorophyll content of Masson pine were measured concurrently at 7-day intervals. The results showed that hyperspectral imaging responses varied among the stressors. Both PWN and waterlogging stress induced rapid mortality, with spectral changes observed as early as the 3rd week and reaching statistical significance by the 5th week. Under PWN infection, hyperspectral reflectance increased markedly in the 405–580 nm range, accompanied by a pronounced blue-shift of the red edge position (680–750 nm), while needle chlorophyll content declined sharply from approximately 0.8 mg g⁻¹ to 0.48 mg g⁻¹. Waterlogging stress produced a uniform increase in reflectance within the 500–580 nm range, with the hyperspectral curve gradually flattening, and needle chlorophyll content decreasing from 0.75 mg g⁻¹ to 0.6 mg g⁻¹. Conversely, drought-stressed seedlings exhibited only minor hyperspectral changes and maintained relatively stable chlorophyll levels, demonstrating the inherent drought tolerance of Masson pine. The RF and XGBoost models performed best in fitting the entire process of pine wood nematode infection and waterlogging stress, with all R² values greater than 0.69. The distinct hyperspectral imaging patterns under nematode infection and water-related stresses provide a reliable basis for early diagnosis and monitoring pine wilt disease in Masson pine stands.

1. Introduction

Pine wilt disease, caused by Bursaphelenchus xylophilus, the pine wood nematode (PWN), is one of the most damaging emerging pest problems in forests around the world [1,2,3,4,5]. Since the introduction of PWD into China, its affected area continued to expand. By 2025, PWD has been detected in 455 county-level administrative regions across 20 provinces in China [6]. Masson Pine (Pinus massoniana) is a major coniferous species in southern China [7], valued for its rapid growth, high productivity, and economic value, and is extensively cultivated in the region. Although it exhibits strong environmental adaptability and is distributed in regions with plentiful annual rainfall, the species confronts three major threats in summer: (1) pine wilt disease caused by B. xylophilus [8], (2) recurrent high-temperature extremes [9], and (3) excessive rainfall during certain subtropical periods [10].

Currently, PWD caused by the pine wood nematode remains the most severe threat to Masson pine survival [11]. The occurrence and prevalence of PWD are strongly linked to climatic factors, such as high temperatures [12,13], which intensify the damage caused by nematode infestation in heat-stressed Masson pine. Under global warming, the impacts of high-temperature stress on Masson pine are expected to intensify, leading to decreased photosynthetic capacity and increased mortality [14,15,16]. Beyond high-temperature stress, water-related abiotic stresses have also become a major threat to Masson pine in tropical and subtropical regions [17,18]. Frequent typhoons and severe convective weather during summer in these areas often lead to surges in rainfall, causing temporary water accumulation in low-lying forest areas and subsequent waterlogging stress [19]. Early monitoring of Masson pine health is therefore critical for distinguishing and addressing the combined threats posed by PWD and abiotic stresses.

In recent years, hyperspectral remote sensing has become a key tool for the efficient monitoring of forest health and PWD [20]. This technique captures spectral data across very narrow wavelength intervals, enabling the precise extraction of spectral signatures from ground objects, and has been widely applied in monitoring plant health and pest infestation [21]. For example, Zhang et al. combined hyperspectral data with field measurements of sugar and moisture content in Masson pine, and identified the 455–677 nm spectral region as diagnostic for the early detection of PWN infestation [22]. Zhang also reported that drought stress reduces the photosynthetic rate in Masson pine [23], further leading to a decline in chlorophyll content [24]. Similarly, Huang’s hyperspectral analysis of black pine needles revealed a significant linear relationship between needle chlorophyll content and the progression of PWD [25]. These studies indicate that plant physiological indicators can be effectively used to evaluate plant physiological status [26,27,28].

Currently, remote estimation of physiological parameters using hyperspectral data is commonly based on the construction of spectral indices. Compared with broadband indices derived from multispectral sensors, narrowband indices take full advantage of the high spectral resolution of hyperspectral data to capture subtle biochemical and physiological variations associated with plant stress [29]. For early stress detection, narrowband indices provide superior sensitivity, as they can precisely target the characteristic absorption features of key pigments and detect minor shifts in the red edge position (REP)—signals that are often obscured or smoothed by the broadband channels of conventional sensors [29]. The construction of spectral indices, including both traditional vegetation indices and narrow-band spectral indices, typically relies on combinations of two or more spectral bands, often enhanced by mathematical transformations. Regression-based modeling enables the accurate identification of sensitive spectral intervals associated with plant physiological parameters, thereby enhancing the predictive performance of estimation models [30,31].

With the rapid advancement of deep learning, advanced models based on attention mechanisms including Transformers and their variants have achieved breakthrough performance in remote sensing image processing, especially in semantic segmentation and fine-grained classification of hyperspectral images. For example, Liu et al. reported that the SegHSI model, which employs clustered attention, enables efficient and accurate segmentation with limited labeled samples, addressing the challenges of high labeling costs and complex spatial contexts in hyperspectral data [32]. Meanwhile, Multi-Scale Multi-Attention Networks that integrate self-attention and cross-attention modules have been widely used to fully exploit multi-scale and multi-modal information in remote sensing data, greatly improving the ability to identify complex scenes and subtle stress characteristics [33]. These cutting-edge algorithms provide effective tools for the automated extraction of highly discriminative stress-related features from high-dimensional complex spectral data.

In this study, we weekly measured the chlorophyll content and hyperspectral characteristics of needles on Masson pine under PWN, drought, and waterlogging stresses, to explore the hyperspectral responses of Masson pine to different stressors and the relationship between needle chlorophyll content and hyperspectral features. We identified the characteristic spectral bands under each stress condition and developed four machine learning models. This research provides a scientific basis for the remote sensing monitoring of PWD and the health status of Masson pine forests.

2. Materials and Methods

2.1. Experimental Materials

Pine wood nematodes: The nematodes (B. xylophilus) used in this study were isolated from a Masson pine tree with PWD in the Washan Forest Farm, Chuzhou City, Anhui Province, China. The species of the nematodes was confirmed as B. xylophilus by morphological and molecular identification and subsequently preserved at 4 °C in the Anhui Provincial Key Laboratory of Biological Control, Anhui Agricultural University.

Masson pine seedlings: Four-year-old Masson pine seedlings, approximately 1.2–1.4 m in height, were purchased from Anhui Siji Forestry Co., Ltd. (Huangshan, Anhui, China). The seedlings were acclimatized for one month at the Hefei Base of the Anhui Academy of Forestry Sciences before the experiments.

Hefei City (31°49′14″N, 117°13′38″E) and Chuzhou City (32°15′22″N, 118°19′59″E) in Anhui Province share a subtropical monsoon climate characterized by distinct seasons, with hot and humid summers and cold, damp winters. The annual average temperature ranges from 15 to 16 °C, and the annual precipitation falls between 900 and 1000 mm. The terrain of Hefei is predominantly plains and hills, while Chuzhou is mainly composed of hilly and mountainous areas, with Masson pine being the predominant tree species.

2.2. Experimental Methods

2.2.1. Cultivation and Harvest of Pine Wood Nematodes

Artificial cultivation: Botrytis cinerea was inoculated onto Potato Dextrose Agar (PDA; 200 g potato, 20 g glucose, 20 g agar, 1000 mL distilled water) and incubated at 25–28 °C. After the fungus completely covered the medium (approximately 5 days), a liquid of B. xylophilus was introduced onto the PDA and further incubated at 25–28 °C to propagate the nematodes.

Nematode harvest: The Baermann funnel method was used to harvest the well-cultivated nematodes [31]. A sterilized glass funnel (12 cm in diameter) was placed on a stand. Sterile water was added to the funnel, and the PDA medium with B. xylophilus well cultured was wrapped in tissue paper and submerged in the water for 12 h. Then the pinch clamp was released, and 10 mL of the nematode liquid was collected in a centrifuge tube. The liquid was centrifuged at 3500 rpm for 3 min at room temperature. The supernatant was discarded, and the nematode sediment was washed 2–3 times with sterile water before storage at 4 °C.

2.2.2. Stress Treatments and Sample Collection

The experiments were conducted at the Hefei Base of the Anhui Academy of Forestry Sciences, located more than 1 km from any known PWD outbreak area. A 15 m × 15 m experimental plot was set up under a shading shelter (2.5 m height, 50% shading rate) enclosed with 40 mesh insect-proof netting on all sides to ensure adequate ventilation and stable temperature and humidity. A total of 250 Masson pine seedlings were selected. The seedlings were transplanted with a spacing of 1.5 m × 1.5 m and allowed to grow for one month under standard watering practices. Five treatments were established, each with five replicates of 10 seedlings per replicate: (1) Pine wood nematode inoculation group (A): A wedge-shaped incision, not exceeding one-third of the stem diameter, was made into the xylem at the midpoint of the first and second branches at the bottom using a sterile scalpel. A sterile cotton ball soaked in B. xylophilus liquid (5000 nematodes per seedling) was inserted into the incision, which was then sealed with sealing film and plastic wrap. (2) Mechanical injury group (B): The same procedure as group A, but 20 mL of sterile water instead of the nematode liquid to assess the effects of the mechanical injury. (3) Drought stress group (C): Watering was completely halted. Soil water content was monitored until it dropped below 10% and this level was maintained. No nematode inoculation was performed in this group. (4) Waterlogging stress group (D): Seedlings were transferred to a tank containing field soil, and the water level was maintained 5 cm above the root zone throughout the experiment. No nematode inoculation was performed in this group. (5) Control group (E): No treatment; plants were allowed to grow naturally. Hyperspectral data and needle samples from Masson pine seedlings were collected at 3 h after the treatments and periodically measured and sampled at 7-day intervals. The collected needles were sealed in plastic bags, stored in an insulated box with ice packs, and immediately transported to the laboratory for chlorophyll determination.

2.2.3. Determination of Needle Chlorophyll Content

Freshly conifer needles from 10 Masson pine seedlings were randomly sampled for chlorophyll content extraction and measurement. Fresh needle leaves from each treatment were uniformly mixed. A 0.25 g sample of needles was randomly selected and cut into 0.2 cm segments. These segments were ground into a homogenate in a mortar with 80% acetone, 2–3 g CaCO₃, and quartz sand. The homogenate was filtered into a 25 mL volumetric flask. The mortar, pestle, and glass rod were rinsed several times with 80% acetone, and the rinses were also filtered into the same volumetric flask. The final volume was adjusted to the 25 mL mark with 80% acetone. The absorbance of the solution was measured at 663, 646, and 470 nm using a spectrophotometer. Chlorophyll and carotenoid contents were determined using standard formulas [34]:

$${\mathrm{C}}_{\mathrm{a}} = (12.21{\mathrm{A}}_{663}-2.81{\mathrm{A}}_{646}) \times \mathrm{V}/1000 \mathrm{m}$$

(1)

$${\mathrm{C}}_{\mathrm{b}}=(20.13{\mathrm{A}}_{646}-5.03{\mathrm{A}}_{663}) \times \mathrm{V}/1000 \mathrm{m}$$

(2)

$${\mathrm{C}}_{\mathrm{total}}={\mathrm{C}}_{\mathrm{a}}+{ \mathrm{C}}_{\mathrm{b}}$$

(3)

In the equations, ${\mathrm{C}}_{\mathrm{a}},{ \mathrm{C}}_{\mathrm{b}}$ and ${\mathrm{C}}_{\mathrm{total}}$ represent the contents (mg g⁻¹) of chlorophyll a, chlorophyll b, and total chlorophyll (sum of chlorophyll a and b), respectively. ${\mathrm{A}}_{663}$ and ${\mathrm{A}}_{646}$ denote the absorbance of the chloroplast pigment extract at 663 nm and 646 nm, respectively.

2.3. Hyperspectral Data Acquisition and Preprocessing

2.3.1. ASD Spectroradiometer Measurements

From each Masson pine seedling, 3–5 clusters of needles were randomly sampled from the middle-upper part of the crown as measurement targets, thus ensuring the spectral data reflected the natural progression of plant physiological status. Hyperspectral reflectance of Masson pine needles was measured using a portable spectroradiometer (ASD FieldSpec4 Wide-Res, Malvern Panalytical Ltd., Malvern, Worcestershire, UK), which has a mean spectral resolution of 1 nm and a spectral range of 350–2500 nm. All measurements were conducted around noon. The instrument was placed in a well-ventilated, dry location and allowed to warm up for at least 30 min prior to use to ensure stable operation. The needle samples cover an area of approximately 2 cm × 3 cm, fully encompassing the probe’s field of view, preventing spectral interference from the background. A standard Spectralon whiteboard was used for reference calibration before each measurement to eliminate errors caused by fluctuations in natural light intensity and instrument drift.

2.3.2. Data Preprocessing

Raw hyperspectral data were exported using MATLAB R2022b, then organized and standardized in Microsoft Excel 2021. Preprocessing was performed to minimize environmental noise, instrument baseline drift, and spectral scattering. The aim was to improve the signal-to-noise ratio and feature discriminability. Several preprocessing methods were evaluated, including multiplicative scatter correction, normalization, moving window smoothing, first and second-order differentiation, and Savitzky–Golay smoothing. Comparative assessments indicated Savitzky–Golay smoothing was optimal for analysis. The core parameters for the Savitzky–Golay smoothing and derivative transformations were set as follows: a window width of 7 data points and a polynomial order of 3. First-derivative spectra (derivative order = 1) were subsequently calculated based on the smoothed spectra for further analysis. The preprocessed hyperspectral data were used to construct a comprehensive library of narrow-band spectral indices, including ratio, normalized difference, and difference types. All data computations and processing were performed in the Python 3.9 environment, by utilizing scientific computing libraries such as Pandas and NumPy. Dynamic variations in measured needle chlorophyll content and needle spectral reflectance were analyzed across different stressing stages. Correlations between each spectral index and needle chlorophyll content were calculated to identify key narrow-band combinations most sensitive to different stressors. Specifically, Pearson correlation analysis was adopted to calculate the correlation coefficient (r) between each constructed narrow-band spectral index (NDSI, DSI, RSI) and the measured needle chlorophyll content of Masson pine under different stressors. The absolute value of r was used to evaluate the strength of the correlation. The narrow-band combinations with |r| > 0.6 (p < 0.01) were defined as highly sensitive combinations to chlorophyll content variation. These sensitive combinations were then ranked by the absolute value of the correlation coefficient, and the top-ranked combinations were selected as the key narrow-band combinations for each stress type. All correlation analyses and significance tests were performed using SPSS 26.0 software, with a 95% confidence interval.

These sensitive bands were used as specific spectral markers to distinguish between the different stressors. Given the challenges in acquiring the stringent input parameters required by fully physical radiative transfer models (RTMs) under complex field conditions, which would introduce significant uncertainties. We opted for a hybrid approach combining empirical spectral indices with machine learning algorithms. To develop high accuracy chlorophyll content inversion models and enable early diagnosis of stress types, four widely used machine learning regression models, Random Forest Regression (RFR), Partial Least Squares Regression (PLSR), Support Vector Regression (SVR), and Extreme Gradient Boosting (XGBoost), were developed and compared. All models were implemented using the Scikit-learn and XGBoost libraries in Python. The datasets were divided into training and testing subsets. Model performance was evaluated using the coefficient of determination (${\mathrm{R}}^2$) and the root mean square error (RMSE) to identify the model with the highest inversion accuracy under different lethal stress conditions (see Figure 1). This methodology provides a core algorithmic foundation for real-time, non-destructive monitoring of Masson pine growth status.

2.4. Construction of Narrow Band Spectral Indices from Needle Hyperspectral Data

By pairwise combination of any two bands within the hyperspectral range, narrow band spectral indices can be constructed. Three index formulations, including the Normalized Difference Spectral Index (NDSI), the Difference Spectral Index (DSI), and the Ratio Spectral Index (RVI), were used to investigate the relationship between narrow band spectral indices derived from Masson pine needle hyperspectral data across the 350–1050 nm range and needle chlorophyll content. Specifically, NDSI, DSI, and RVI were systematically constructed for the original reflectance spectra, first derivative spectra, and continuum-removed spectra, based on all possible two-band combinations. The formulas are given in Equations (4)–(6):

$$\mathrm{N}\mathrm{D}\mathrm{S}\mathrm{I}={\displaystyle \frac{{\mathrm{R}}_{\mathrm{i}}-{\mathrm{R}}_{\mathrm{j}}}{{\mathrm{R}}_{\mathrm{j}}+{\mathrm{R}}_{\mathrm{i}}}}$$

(4)

$$\mathrm{R}\mathrm{S}\mathrm{I}={\displaystyle \frac{{\mathrm{R}}_{\mathrm{i}}}{{\mathrm{R}}_{\mathrm{j}}}}$$

(5)

$$\mathrm{D}\mathrm{S}\mathrm{I}={\mathrm{R}}_{\mathrm{i}}-{\mathrm{R}}_{\mathrm{j}}$$

(6)

In the equations, i and j represent the hyperspectral wavelengths, and ${\mathrm{R}}_{\mathrm{i}}$ and ${\mathrm{R}}_{\mathrm{j}}$ denote the hyperspectral reflectance values at wavelengths i and j, respectively. This study takes the original spectrum, the first derivative spectrum, and the continuum-removed spectrum separately.

3. Results and Analysis

3.1. Effects of Different Stressors on Needle Chlorophyll Content of Masson Pine

Continuous morphological observations revealed different temporal changes in needle symptoms under different stressors. Temporal variation in needle chlorophyll contents for different treatments exhibited a high degree of distinctiveness. A summary of these findings is presented in

Table 1. Temporal changes in symptoms of Masson pine under different treatments.

Stress Treatment	D1–D8	D9–D15	D16–D22	D22–D36
A	Green	Needles turned yellowish-brown	Needles turned yellowish-brown	Needles turned yellowish-brown
B	Green	Green	Green	Green
C	Green	Needles appeared dark green	Needles appeared dark green	Needles appeared dark green
D	Green	Yellow spots appeared at leaf tips	Yellow spots appeared at leaf tips	Yellow spots appeared at leaf tips
E	Green	Green	Green	Green

Notes: D1–D8 represents the first week, D9–D15 the second week, D16–D22 the third week, and D22–D36 the fourth and fifth weeks.

and Figure 2.

Needle chlorophyll content of Masson pine for all treatments remained almost stable in the first two weeks (Figure 2). No significant difference in needle chlorophyll content between different treatments and the control, except for Group C (Drought stress) on the eighth day. A transient but obvious drop in chlorophyll content was observed in Group C on day 8. This likely reflects a short-term shock response to sudden watering cessation, rather than sustained drought stress. This initial fluctuation did not form a trend, and chlorophyll levels subsequently recovered and stabilized.

After one additional week, needle chlorophyll contents in Group A (Nematode stress) and Group D (Waterlogging stress) notably declined to 0.55 mg g⁻¹ and 0.68 mg g⁻¹, respectively. These values were significantly lower than that of the control, indicating obvious physiological deterioration. In the 5th week, chlorophyll content in Groups A and D reached lower levels of less than 0.50 mg g⁻¹ and 0.55 mg g⁻¹, significantly lower than that of the control (0.85 mg g⁻¹). Needle chlorophyll content in Group C showed fluctuations over time, indicating that the effect of drought stress on Masson pine needs a longer time. No significant difference appeared between Group B (Mechanical injury) and Group E (Control), verifying that the mechanical injury had a minor effect on needle chlorophyll content.

According to morphological observations, there was one week of potential before morphological abnormalities appeared on needles in Groups A and D. Meanwhile, needle chlorophyll content in the two groups did not change much until the 3rd week. This represents a key time point at which chlorophyll metabolism in Masson pine underwent significant alteration, serving as a distinguishing marker for the health detection of Masson pine.

3.2. Spectral Characteristics of Masson Pine Needles

The spectral curves of all treatment groups in the first two weeks maintained relative stability, while some early features appeared in Groups C (drought stress) and D (Waterlogging stress) (Figure 3). A typical spectral curve of “two valleys one peak” representing absorption valleys in the 450–550 nm and 600–750 nm ranges and a reflectance peak near 550 nm in Groups B and E exhibited spectral features of healthy Masson pine. The spectral curve in Group A (nematode infection) did not show any obvious change in the visible and near-infrared regions, while reflectance in the visible region increased slightly in Group C. Similarly, there was a mild rise in reflectance in the green light region in Group D.

The spectral curves exhibited marked divergence under different stressors between the 3rd and 5th weeks. In Group A, reflectance increased significantly in the 405–505 nm range, the absorption valley disappeared, the reflectance peak in the 505–600 nm range diminished, and the red edge feature became indistinct after 5 weeks of nematode infection. This is because stressed plants produce less chlorophyll, leading to reduced light absorption associated with chlorophyll decreases [35]. Consequently, reflectance in the red and green regions increases, resulting in leaf chlorosis and yellowing. In Group C, reflectance gradually decreased in the 520–680 nm range and beyond 680 nm. Because the contents of soluble protein and soluble sugar showed no significant changes under moderate and severe drought stress, the seedlings exhibited lower oxidative damage and enhanced drought resistance [7,36]. Therefore, the early-stage spectra of Masson pine in Group C may undergo slight changes, but due to its inherent drought resistance, it does not exhibit disease symptoms. In Group D, reflectance increased uniformly across the 500–600 nm range, leading to an overall flattening of the spectral curve. According to Chen Wei’s research, the main reason for the prolonged waterlogging of Masson pine is that the fallen debris of coniferous trees contains a large amount of tannins, resins, lignin, etc. These substances will produce acidic substances after decomposition, and the acidic soil will increase with the length of waterlogging time, resulting in an increase in soil organic matter content [37]. The quantity decreases, leading to the death of Masson pine. This effect was more pronounced in the 5th week. The spectral curves of the 3rd week and 5th week almost completely overlapped in Groups B (Mechanical injury) and E (Control), indicating a consistently healthy spectral profile. These distinct temporal changes in spectral curves provide a reliable method to distinguish between different stressors.

Significant differences in needle chlorophyll content and narrow band spectral indices were observed before and after three weeks (D22) of treatment. The observation period was therefore divided into pre-D22 (potential) and post-D22 (symptom) stages. Separate spectral inversion models were developed for each stage.

3.3. Correlation Analysis Between Narrow Band Spectral Indices and Chlorophyll Content in Masson Pine

Based on needle spectral reflectance data collected with an ASD spectroradiometer, dual band spectral indices derived from three types of transformed spectra (original, first derivative, and continuum-removed spectra) were constructed in Python 3.9. The correlation between narrow-band spectral indices and needle chlorophyll content under different stressors was investigated through correlation isoplots (Figure 4).

In the pre-D22 stage, the original pine needle spectra of Group A were negatively correlated with needle chlorophyll content in the 450–700 nm and 750–1000 nm ranges. The maximum correlation coefficients between the nine narrow band spectral indices and needle chlorophyll content ranged from 0.5416 to 0.6628. The correlation between first derivative spectra and needle chlorophyll content was variable. A positive correlation was observed in the 750–950 nm range, but the correlation index was lower than that of the original spectra. Continuum-removed spectra exhibited a negative correlation in the 650–750 nm range, with a maximum correlation coefficient of 0.6296 at 663 nm. In Group C, original needle spectra exhibited a strong correlation with needle chlorophyll content, forming concentrated and continuous high-correlation zones in the combined regions of 700–800 nm and 1000–1050 nm. The absolute correlation coefficients for NDSI and RSI exceeded 0.7500. The index derived from first-derivative spectra showed a maximum correlation of 0.668 (DSI), with multiple isolated strong correlations observed, typically in the 550–750 nm range and above 1000 nm. Continuum-removed spectra displayed negative correlations in the 500–600 nm and 650–720 nm ranges, with correlation coefficients exceeding 0.7200 near 550 nm. In Group D, the original spectrum-based NDSI (689,907) exhibited a very high positive correlation with chlorophyll (r = 0.8372). The optimal RSI derived from first-derivative spectra showed a correlation above 0.6. Continuum-removed spectra in Group D showed negative correlations in the 350–450 nm range; the optimal indices were RSI (352,444) and DSI (359,435), with correlation coefficients of −0.7412 and −0.7392, respectively, presenting a shift of the most sensitive bands to the blue-violet region. In Group E, continuum-removed spectra directly reflected chlorophyll absorption features at the sensitive wavelengths of 560 nm and 690 nm.

In the post-D22 stage, the NDSI and RSI indices of Group A were concentrated in the bands (664,748) and (664,749), with correlation coefficients exceeding 0.66. The inversion performance in Group C was poor, with correlations only above 0.4. Conversely, the original spectrum-based NDSI (471,1045) in Group D achieved a correlation coefficient of –0.8433 with chlorophyll, and the continuum-removed spectra in the extremely narrow 534–535 nm ranges reached correlations of 0.8297 and 0.8081. First-derivative spectra consistently demonstrated high sensitivity to stressors across all groups, with the NDSI (474,478) band expressing a very strong correlation of –0.8441, and continuum-removed spectra near 534 nm provided the best inversion capability in Group D. Notably, a weakened response in spectral band information appeared in Group C (Figure 5).

When performing data inversion on the chlorophyll content of Masson pine, the transformed spectra from first derivative and continuum-removed performed better than the original spectra, yielding higher and more stable correlation coefficients. First derivative processing effectively highlighted subtle variations in spectral curves, responding more rapidly to changes in chlorophyll and showing extremely high sensitivity in specific narrow bands. At the same time, continuum-removed spectra, with appropriate band combinations, also proved to be an effective method for spectral transformation in chlorophyll inversion. The optimal band combinations for narrow band spectral indices of chlorophyll under the stressors were selected for both the pre-D22 and post-D22 stages, according to their maximum correlation coefficient (

Table 2. Optimal band combinations of narrow band spectral indices in the pre-D22 stage.

Group	Spectral Transformation	Band Combination	Correlation (r)
A		NDSI (664,748)	0.6610
	Original spectrum	RSI (664,749)	0.6628
		DSI (900,902)	0.5630
		NDSI (377,900)	−0.5238
	First derivative spectrum	RSI (401,899)	0.5416
		DSI (900,1045)	−0.5441
		NDSI (655,706)	−0.6272
	Continuum Removal	RSI (701,653)	−0.5932
		DSI (663,710)	−0.6296
B		NDSI (690,959)	0.7496
	Original spectrum	RSI (690,958)	−0.7496
		DSI (699,902)	0.6467
		NDSI (783,798)	0.6308
	First derivative spectrum	RSI (765,798)	0.6288
		DSI (622,753)	0.6464
		NDSI (696,698)	0.6025
	Continuum Removal	RSI (696,698)	0.6225
		DSI (663,706)	0.6024
C		NDSI (689,769)	0.7547
	Original spectrum	RSI (689,780)	−0.7547
		DSI (699,1045)	0.626
		NDSI (379,1034)	−0.5928
	First derivative spectrum	RSI (378,750)	−0.5812
		DSI (633,757)	−0.668
		NDSI (563,566)	−0.7265
	Continuum Removal	RSI (549,577)	−0.7269
		DSI (651,704)	−0.6338
D		NDSI (687,1000)	0.7858
	Original spectrum	RSI (506,786)	−0.7851
		DSI (752,754)	−0.6181
		NDSI (774,798)	−0.5973
	First derivative spectrum	RSI (774,798)	−0.6026
		DSI (634,752)	−0.5973
		NDSI (359,376)	0.7335
	Continuum Removal	RSI (352,444)	−0.7412
		DSI (359,435)	−0.7392
E		NDSI (689,907)	0.8368
	Original spectrum	RSI (689,907)	0.8372
		DSI (751,752)	0.6846
		NDSI (370,1045)	0.6806
	First derivative spectrum	RSI (370,751)	0.7298
		DSI (557,764)	0.7225
		NDSI (697,698)	0.7971
	Continuum Removal	RSI (697,698)	0.7975

Notes: NDSI is the Normalized Difference Spectral Index, RSI is the Ratio Spectral Index, and DSI is the Difference Spectral Index. The same as below.

and

Table 3. Optimal band combinations of narrow band spectral indices in the post-D22 stage.

Group	Spectral Transformation	Band Combination	Correlation (r)
A		NDSI (368,673)	−0.6749
	Original spectrum	RSI (685,769)	0.7144
		DSI (663,748)	−0.7224
		NDSI (469,526)	−0.7609
	First derivative spectrum	RSI (470,527)	−0.7612
		DSI (984,986)	0.7142
		NDSI (662,695)	0.6841
	Continuum Removal	RSI (702,703)	0.6692
		DSI (662,695)	0.6841
C		NDSI (371,377)	−0.4123
	Original spectrum	RSI (371,377)	0.4143
		DSI (372,377)	0.4381
		NDSI (359,754)	−0.4139
	First derivative spectrum	RSI (393,733)	−0.4548
		DSI (375,417)	0.4430
		NDSI (371,377)	0.4123
	Continuum Removal	RSI (371,377)	0.4143
		DSI (371,377)	0.6024
D		NDSI (370,579)	−0.5924
	Original spectrum	RSI (689,780)	−0.7547
		DSI (392,571)	−0.5570
		NDSI (471,1045)	−0.8433
	First derivative spectrum	RSI (474,478)	−0.8441
		DSI (679,765)	0.7046
		NDSI (534,536)	0.8297
	Continuum Removal	RSI (534,535)	0.8297
		DSI (534,535)	0.8081

Notes: NDSI is the Normalized Difference Spectral Index, RSI is the Ratio Spectral Index, and DSI is the Difference Spectral Index.

).

3.4. Regression Models Based on Narrow Band Spectral Indices

Linear regression models were developed with needle chlorophyll content as the dependent variable. The independent variables included the optimal band combinations of the Difference Spectral Index, Ratio Spectral Index, and Normalized Spectral Index derived from original, first-derivative, and continuum-removed spectra. The best univariate prediction model for each spectral transformation was selected, as shown in Figure 6. The results indicated that the optimal narrow-band spectral indices exhibited a linear relationship with needle chlorophyll contents in all cases. The coefficient of determination (${\mathrm{R}}^2$) exceeded 0.50 for all stress groups, except for the drought-treated group in the post-D22 stage under the three spectral transformations. Among the four constructed machine learning regression models, the RFR and XGBoost models performed best in predicting needle chlorophyll content of Masson pine under multi-stress conditions. Specifically, RFR and XGBoost consistently achieved the highest coefficient of determination and the lowest root mean square error across all stress groups, both in the pre-D22 (potential) stage and post-D22 (symptom) stage. Conversely, PLSR and SVR models presented relatively lower prediction accuracy, particularly in the post-D22 stage under drought stress.

4. Discussion

4.1. Needle Chlorophyll Content Response to Pine Nematodes and Water Stresses

A rapid and accurate determination of needle chlorophyll content enables a precise assessment of the growth status and photosynthetic capacity of Masson pine [36]. Changes in needle chlorophyll content reflect the physiological and survival status of the trees. When Masson pine is inoculated with pine wood nematodes, the transport of water and nutrients is hindered, chlorophyll synthesis is suppressed, and needle chlorophyll content decreases [38]. Chen et al. verified that chlorophyll a and b decreased as the infection severity of PWD increased [39].

Various stressors induce distinct temporal changes in needle chlorophyll content. For instance, following inoculation with pine wood nematodes, needle chlorophyll content in Masson pine declines sharply within a short period [40]. In contrast, needle chlorophyll content remains elevated to maintain normal photosynthetic function during prolonged drought stress [41]. Under waterlogging stress, fine roots can recover within one week after drainage if the flooding persists for less than 17 days. However, waterlogging lasting 17–32 days causes irreversible damage, including decreased photosynthetic rate and reduced needle chlorophyll content [42]. Consistent with these findings, the present study observed a significant reduction in needle chlorophyll content within 3 weeks under PWN and waterlogging stresses, whereas no such decrease was detected under drought and mechanical injury. Notably, although external symptoms were not obvious in the early stages under PWN and waterlogging stress, changes in needle chlorophyll content were already evident.

4.2. Hyperspectral Response to Different Environmental Stressors

Remote estimation of vegetation biophysical and biochemical parameters using hyperspectral imagery has been widely validated [43,44]. For example, Kim et al. analyzed hyperspectral data of needles infected with PWD and observed changes in red-band reflectance in adult Masson pine within two months of infection [45]. Selection of the most sensitive wavelength bands to specific stressors is crucial for reducing data redundancy and avoiding model overfitting in hyperspectral analysis [44]. In the present study, hyperspectral changes revealed that PWN infection mainly affected the blue-green region (405–580 nm) and the red edge region (680–750 nm), whereas waterlogging stress primarily influenced the green reflectance peak (505–580 nm). These findings are generally consistent with some previous research [46,47]. Li et al. reported that the red edge band at 714 nm alone could detect early symptoms of nematode infection in Masson pine. Similarly [48], Kathrin et al. found that the red edge band at 721 nm could identify diseased spruce trees [49]. These studies suggest that combining hyperspectral measurements—especially narrow-band data—with physiological indices such as needle chlorophyll content can significantly improve the monitoring precision of health status of Masson pine.

4.3. Comparison of Different Spectral Preprocessing Methods, Underlying Mechanisms, and Model Construction

4.3.1. Sensitivity of Spectral Preprocessing Methods

In the present study, the first-derivative transformation was more sensitive in monitoring spectral changes in leaf needles during the potential period, particularly in detecting chlorophyll responses to waterlogging and nematode infection. Meanwhile, continuum removal transformation showed a clear advantage in extracting absorption features directly associated with needle chlorophyll content, especially in key bands such as 534–536 nm, 663 nm, and 697–698 nm. These results align with previous studies by Pan et al. [50,51], Gong et al. [52], and Zhang et al. [53], verifying that transforming original spectra for physiological parameter inversion in Masson pine is feasible.

This study demonstrated that first-derivative and continuum-removal transformations significantly enhance the sensitivity of spectral data to variations in needle chlorophyll content—a result consistent with earlier findings by Ta et al. [54] and Tsuchiya et al. [55]. Building on analyses of spectral transformations and chlorophyll-sensitive bands, Masson pine exhibited distinct spectral responses under different stress conditions: it was particularly sensitive to the green light region (534–536 nm) under waterlogging stress, whereas sensitivity shifted to the blue–red edge range (450–760 nm) during nematode infection. These stress-specific spectral signatures align with those reported by Zhu et al. [56] for waterlogged winter wheat and Pan et al. [38] for nematode-infested Masson pine, collectively supporting the notion that physiological disturbances in plants are reliably encoded in characteristic spectral features—a pattern that appears to hold across species and stress types.

4.3.2. Physiological Mechanisms Underlying Stress-Type-Specific Spectral Sensitivities

The stress-specific spectral sensitive regions identified in this study can be traced to different physiological and pathological disturbance processes. The significant sensitivity to waterlogging stress in the green light region (534–536 nm) is primarily due to the cascade of physiological responses induced by root hypoxia. Waterlogging stress significantly inhibits the net photosynthetic rate and transpiration rate of Masson pine [57], which not only directly suppresses chlorophyll synthesis but also accelerates chlorophyll degradation due to the accumulation of reactive oxygen species [58]. In addition, waterlogging-induced changes in leaf cell structure (such as increased intercellular spaces in spongy tissue) further alter the light scattering pathways within the leaves [59]. Therefore, the high sensitivity of indices such as NDSI (534,535) results from the combined effects of pigment changes driven by chlorophyll degradation and alterations in cell structure, leading to changes in optical properties [60]. The primary factor driving the increase in reflectance within the blue-green region following PWN infection is the severe loss of chlorophyll, particularly chlorophyll a. This phenomenon fundamentally results from the water stress and physiological collapse induced by PWN-related vascular damage, which ultimately leads to the depletion of essential leaf pigments responsible for light energy capture [61].

4.3.3. Performance of Optimized Narrow-Band Indices and the Hybrid Response Pattern

Notably, the narrow-band spectral indices optimized in this study achieved markedly higher correlations with needle than previously reported. Specifically, NDSI (689,907) and the continuum-removed reflectance in the 534–536 nm band yielded R² values of 0.84 and 0.83, respectively—surpassing the performance of spectral indices of Masson pine evaluated by AlDwairi et al. [62]. In this study, the proposed algorithm achieved an approximately 8.7% improvement in explanatory power (${\mathrm{R}}^2$) for chlorophyll content estimation, increasing from 0.70 to 0.84. This indicates that the selected band combinations more effectively capture chlorophyll-related variability, thereby strengthening the explanatory power of linear models.

Beyond confirming the widely reported linear relationships between optimized narrow-band spectral indices and chlorophyll content, this study is the first to identify a hybrid spectral–chlorophyll response pattern in Masson pine under multiple stresses, characterized by a predominantly linear relationship with localized nonlinear deviations. For example, in the post-D22 stage of the waterlogging group, the first-derivative spectrum at 474–478 nm exhibited peak correlation (r = 0.8441), yet data points showed obvious nonlinear dispersion. With decreasing chlorophyll content, some observations diverged from the linear trend and instead followed an exponential decay trajectory. Similarly, the continuum-removed spectra in the blue region revealed multiple local extrema. These findings go beyond the strictly linear associations reported by Zhang et al. [63], providing a more refined and comprehensive understanding of the mechanistic relationship between spectral dynamics and chlorophyll status.

In addition to comparing model accuracy, this study systematically evaluates model suitability within the complex context of temporal chlorophyll dynamics in Masson pine under concurrent abiotic and biotic stresses. Random forest and XGBoost consistently achieved the highest inversion accuracy, with overfitting effectively mitigated through rigorous hyperparameter optimization. However, in the post-D22 stage, Group C exhibited significantly reduced inversion performance, due to the physiological resilience of Masson pine under drought stress. Unlike nematodes or waterlogging stress that directly disrupts chlorophyll synthesis, drought stress typically causes a delayed response in chlorophyll degradation [64]. Despite these conditions, RF and XGBoost maintained superior performance, consistent with previous research [65], demonstrating that ensemble tree models, such as XGBoost, are effective at capturing the complex nonlinear relationships associated with drought-induced physiological changes. In contrast to the single-target modeling approach of Zhang et al. [66], which focused exclusively on chlorophyll mass fraction, the comparative framework employed here offers greater methodological breadth and practical relevance for real-world monitoring applications.

RF and XGBoost models demonstrated superior performance, providing a robust algorithmic foundation. Their inherent resilience to noise and overfitting, proven scalability for large-area applications, and successful track record in forest health surveillance collectively support their technical feasibility for transitioning from controlled experiments to operational UAV or satellite-based monitoring systems [67,68,69,70,71].

4.4. Implications for Compound Stress Scenarios: Interactive Mechanisms

Natural stresses in forest ecosystems rarely occur separately. The interaction between drought and PWD is one of the most concerning compound stresses in reality. Drought stress significantly accelerates the lethal process of pine trees by PWN. This is because drought weakens tree physiological vigor and disrupts their protective enzyme systems, thereby synergistically increasing susceptibility to nematode infection [72]. However, drought-induced stomatal closure may temporarily reduce transpiration pull, which in the short term can mitigate embolism caused by nematodes, reflecting a potential complex antagonistic relationship between stresses. At the molecular level, Masson pine activates complex defense metabolic pathways against biotic stresses [73], whereas abiotic stresses such as drought may disturb these signaling networks, resulting in an extremely sophisticated physiological stat. This study provided critical knowledge for future analysis and differentiation of dominant stress factors under compound stresses. The next step involves advanced algorithms such as spectral unmixing and deep learning to quantify the contribution rates of different stresses from the mixed spectral signals of trees under compound stressors and to develop an intelligent monitoring system.

4.5. Future Implications

The highly sensitive narrow-band spectral indices identified in this study, together with optimized and validated machine learning models, establish a robust algorithmic foundation for translating laboratory findings into real-time field monitoring tools. The most practical application involves hyperspectral remote sensing mounted on unmanned aerial vehicle (UAV) platforms. The sensitive spectral bands identified all fall within the effective detection range of mainstream commercial UAV-hyperspectral sensors [46,66]. Recent studies have demonstrated that stress-related characteristic bands from UAV-acquired hyperspectral imagery enable effective identification of early symptoms of pine wilt disease (PWD) and accurate delineation of infested areas [8,48]. As a result, forest management departments can deploy such UAV platforms to conduct targeted, high-frequency patrol surveys, facilitating the generation of high-resolution stress distribution maps at the stand level. Additionally, by leveraging the stress-specific spectral fingerprints revealed in this study, it is possible to directly differentiate and locate PWD infection hotspots, waterlogging stress areas, or drought-affected zones within the imagery. This approach supports early warning, precise localization, and classified management, thereby significantly improving the timeliness and accuracy of forest health monitoring.

5. Conclusions

Temporal variations in hyperspectral reflectance data and needle chlorophyll content were evaluated under pine wood nematode and water stress to assess the feasibility of early detection of healthy conditions in Masson pine. The main conclusions are as follows:

(1)

Needle chlorophyll content exhibited distinct temporal responses under different stressors. Pine wood nematode infection and waterlogging stress resulted in a continuous, steady decline in needle chlorophyll content, and a significant decline after 3 weeks. Chlorophyll changes were minor under the stressors of drought and mechanical injury within 5 weeks.

(2)

Each stressor caused diagnostic changes in specific spectral regions. The pine wood nematode caused a notable increase in reflectance in the 405–580 nm region and a distinct blue-shift of the red edge (680–750 nm), while waterlogging stress produced a uniform rise in reflectance of the green band (505–580 nm) and a flattened spectral curve. Furthermore, drought stress had only a weak effect on the visible and near-infrared spectral regions.

(3)

Continuum removal and first-derivative spectral transformations significantly improved the sensitivity and accuracy of stress identification. Continuum removal effectively enhanced chlorophyll absorption features in the 534–536 nm and 663 nm bands, while first-derivative transformation captured subtle spectral changes during early stress stages, such as in the 474–478 nm band. Combined, these methods provide high-dimensional spectral features for early-stage stress classification.

(4)

Machine learning models demonstrate strong performance in chlorophyll inversion and stress classification. RF and XGBoost consistently achieve the highest accuracy among spectral models for both pine wood nematode infestation and waterlogging stress. The R² values for groups A and D range from 0.68 to 0.82, with RMSE values between 0.034 and 0.044. In the post-D22 stage, for groups F and H, the ${\mathrm{R}}^2$ values remain high, all exceeding 0.73.

(5)

There was a potential period before obvious symptoms appeared under different stressors. Early differentiation and monitoring of pine wood nematode, drought, and waterlogging stress can be achieved by a combination analysis on temporal changes in spectral indices and needle chlorophyll content.

This study achieved for the first time an effective distinction between biological stress (PWN) and abiotic stress (drought, watering) experienced by Masson Pine. Its core innovations lie in, firstly, constructing a high-precision chlorophyll inversion method that integrates narrow-band spectral indices and machine learning models; secondly, identifying and clarifying specific spectral response characteristics that distinguish the three types of stress, thus providing a new methodological basis for early monitoring and early warning of Masson pine forests under multi-source stress environments.