Remote Sensing-Enhanced Structural Equation Modeling for Evaluating the Health of Ancient Juglans regia L. in Tibetan Traditional Villages

Abstract

Ancient walnut trees (Juglans regia L.), revered as “cultural heritage in motion,” have coexisted harmoniously with dense clusters of Tibetan traditional villages for centuries. However, accelerating climate change and expanding human activities along the middle reaches of the Yarlung Tsangpo River have increasingly threatened their survival. To quantitatively evaluate the health of these ancient trees and identify the underlying driving mechanisms, this study developed a remote sensing-enhanced Structural Equation Model (SEM) that integrated satellite-derived ecological indices, land-use intensity, and field-measured morphological and physiological indicators. A total of 135 ancient walnut trees from villages such as Gamai in Jiacha County, Tibet, were examined. Key findings: (1) The SEM demonstrated an excellent model–data fit (Minimum Discrepancy Divided by Degrees of Freedom (CMIN/DF) = 1.372, Root Mean Square Error of Approximation (RMSEA) = 0.053, Tucker–Lewis Index (TLI) = 0.956, and Comparative Fit Index (CFI) = 0.962), confirming its robustness. (2) Among the latent variables, overall condition exerted the strongest influence (weight = 0.360), whereas foliage condition contributed least (0.289). (3) Approximately 35.56% of trees were healthy or sub-healthy, while 61.48% showed varying levels of decline. (4) Tree health was jointly shaped by intrinsic and extrinsic factors, with intrinsic drivers exhibiting stronger explanatory power. Externally, human disturbance negatively affected health, whereas ecological quality was positively associated. These results highlight the effectiveness of integrating remote sensing and SEM for ancient tree assessment and underscore the urgent need for long-term monitoring and adaptive conservation strategies to enhance ecological resilience.

1. Introduction

Ancient trees (commonly >100 years old [1]) represent irreplaceable ecological, cultural, and scientific heritage, sustaining biodiversity, moderating microclimatic conditions, and preserving long-term ecological information [2,3,4,5]. Their exceptional structural integrity and longevity make them unique biological archives capable of recording climatic fluctuations and ecological processes across historical timeframes [6]. Within traditional settlements along the middle reaches of the Yarlung Tsangpo River, ancient walnut trees (Juglans regia L.) are particularly prominent. Walnut cultivation has a long history in this region, and ancient walnut individuals are extensively distributed within and around village spaces. Functionally, they serve as dominant native tree species, forming essential components of the local ecological substrate and landscape structure [7,8,9]. Embedded in these culturally shaped environments, ancient trees mediate human–environment relationships and contribute to the ecological stability of village ecosystems [7].

Despite their immense significance, the health of ancient trees is increasingly under threat. Climate change-induced extreme events, such as droughts, floods, and lightning, combined with pest infestations, have accelerated physiological decline and structural deterioration. Age-related senescence further reduces resilience [10]. Concurrently, rural urbanization and infrastructure expansion since the early 21st century have intensified human disturbance, surpassing natural processes as the primary force shaping landscapes. The continuous loss of ecological space [11] has degraded habitats, compromised the survival of ancient trees, and triggered broader ecological consequences, including vegetation loss and soil erosion. Although legal instruments such as the Ancient and Famous Trees Protection Regulations and technological advancements such as LiDAR monitoring have strengthened conservation efforts, the ecological and cultural functions of ancient trees remain fragile under the combined impacts of natural and anthropogenic stressors [12].

The evaluation of ancient tree health has traditionally relied on indicator-weighting methods, in which morphological, physiological, and environmental indicators are classified and assigned specific weights to construct assessment systems. These approaches generally fall into three categories: subjective, objective, and integrated weighting methods [13,14]. Subjective methods, such as the analytic hierarchy process (AHP), depend heavily on expert judgment and prior knowledge but often overlook correlations among indicators [15]. For instance, Xie et al. [16] applied AHP to evaluate 365 ancient trees in Beijing based on trunk damage and inclination. Objective approaches, such as principal component analysis (PCA), derive weights directly from statistical properties but may underestimate the intrinsic importance of certain indicators, thereby introducing bias [17]. Integrated approaches, including AHP-fuzzy comprehensive evaluation, partially address qualitative aspects yet fail to eliminate redundancy among correlated indicators.

Structural equation modeling (SEM) represents a methodological advancement by integrating path analysis and factor analysis. SEM explicitly defines hypothesized causal relationships while identifying latent variable structures, effectively overcoming issues such as multicollinearity and neglected inter-indicator associations. This approach has been widely applied in studies on ecosystem functionality, ecological security, and forest health [18,19]. Because ancient tree health is influenced by morphological traits, habitat quality, environmental stressors, and human disturbance [20,21], SEM offers a robust and comprehensive framework for developing more reliable evaluation systems. Although previous SEM-based studies have made valuable progress in evaluating ecological conditions and tree health, many of them prioritize variance explanation and indicator associations, and thus do not explicitly disentangle the mechanistic contributions of ecological and anthropogenic drivers. In addition, external drivers such as human disturbance, land-use change, and habitat alteration, while recognized conceptually, have seldom been explicitly modeled as causal pathways affecting ancient tree health. For example, proximity to buildings may induce abnormal growth in ancient walnut trees [22], whereas complex ecosystems enhance resilience and create favorable microhabitats [2,23]. These patterns underscore the need to explicitly assess the interplay between anthropogenic drivers and tree health.

Remote sensing products enable village-scale measurement of vegetation cover, soil moisture, humidity, and Land Surface Temperature (LST) [11]. The remote sensing ecological index (RSEI), introduced by Xu [24], has been widely validated for integrated ecosystem quality monitoring across heterogeneous landscapes [25,26].

Human Activity Intensity (HAI) quantifies anthropogenic pressure and has been widely used in ecological evaluations [27,28,29]. Yet most studies operate at administrative scales (county, watershed), producing coarse resolution unsuitable for village-level ancient tree assessment [30,31]. Fine-scale habitat–disturbance interactions around individual-tree locations are rarely analyzed.

Although SEM has been increasingly adopted in ecological and forest-health studies, most applications remain indicator-driven and primarily emphasize explaining statistical variance rather than elucidating ecological mechanisms. In particular, external drivers such as land-use change, microhabitat modification, and anthropogenic disturbance are frequently simplified as contextual variables instead of being rigorously modeled as causal pathways [32,33]. As a consequence, existing SEM-based evaluations are limited in their capacity to reveal how intrinsic tree attributes interact with environmental stressors, thereby restricting their suitability for diagnosing stress mechanisms and informing practical management strategies for ancient trees.

Meanwhile, current approaches to quantifying Human Activity Intensity (HAI) predominantly operate at coarse administrative or regional scales, resulting in ecological assessments that lack precision in fine-grained environments such as traditional villages [25,34]. This scale mismatch obscures localized microenvironmental pressures surrounding individual tree locations, constraining the relevance of existing findings to conservation applications. To overcome these limitations, the present study advances a village-scale, second-order SEM framework by simultaneously integrating physiological and morphological indicators of ancient walnut trees with remote sensing-derived ecological indicators (RSEI) and GIS-based HAI quantification [35,36]. This unified framework enables explicit causal inference among intrinsic health traits, external ecological conditions, and anthropogenic disturbance, thereby addressing both the mechanism-related and scale-related gaps and providing a more actionable scientific basis for ancient-tree conservation in traditional cultural landscapes.

2. Materials and Methods

2.1. Overview of the Study Area and Data Processing

2.1.1. Study Area

Jiacha County is located in the central section of the middle reaches of the Yarlung Tsangpo River Valley, a mid-latitude, high-altitude region characterized by a plateau semi-temperate and semi-humid climate. The terrain slopes gradually from west to east, with an average elevation of approximately 4000 m. Within the valley, altitudes range from 3100 to 3500 m, where fertile soils, synchronized rainfall and heat, and abundant water and thermal resources create highly favorable conditions for agriculture. Consequently, this region has long been recognized as one of the major agricultural centers of the Tibet Autonomous Region [37,38]. The valley supports a variety of economically valuable tree species, particularly walnut (Juglans regia L.) and wild peach (Amygdalus mira Koehne), with walnut being especially dominant. Historically, walnuts harvested from this region were selected as tributes to successive Dalai Lamas, leading to Jiacha being recognized as the “Economic Forest Base of Southeastern Tibet” [39]. In addition to their cultural prominence, ancient walnut trees (Juglans regia L.) perform essential ecological functions within mountainous village ecosystems. As long-lived native broadleaf trees with deep and well-developed rooting systems, they improve soil structure and enhance slope stability, thereby reducing erosion risks on steep Himalayan terrain [40,41]. Their large and persistent canopy architectures moderate near-surface temperature and humidity, creating thermally buffered microhabitats that benefit understory vegetation and associated fauna [42]. Moreover, Juglans-dominated agroforestry configurations increase landscape heterogeneity and support biodiversity maintenance in traditional Tibetan villages [43,44]. The synergistic integration of cultural heritage, ecological stability, and habitat provisioning highlights walnut trees as uniquely appropriate focal species for evaluating ancient-tree health and environmental conditions in traditional settlement contexts.

This study focused on four representative villages, Gamai, Gaji, Gadu, and Liegang (Figure 1). Agriculture remains the primary livelihood in these communities, and ancient walnut trees are widely distributed throughout the area. Notably, the region is home to a 2000-year-old “King Walnut Tree,” symbolizing the enduring coexistence between local residents and walnut cultivation. In recognition of its ecological and cultural significance, Gamai Village was included in the sixth batch of China’s National List of Traditional Villages in 2023.

2.1.2. Data Source and Processing

Information on ancient walnut trees was obtained from a joint census conducted by the Forestry and Grassland Bureau of Jiacha County and the Walnut Industry Research Institute of the Tibet Plateau at Yangtze University. In October 2024, 135 registered ancient walnut trees were surveyed within the study area (Figure 1d). To reduce subjective variability, all observers participated in a pre-survey calibration session that standardized indicator definitions, scoring thresholds, and representative examples. During field assessment, each tree was evaluated independently by two trained investigators, and any discrepancies were resolved through reassessment and consensus-based discussion; additionally, a subset of trees was rescored at a later date to verify consistency of interpretation. Indicator selection was initially informed by previous studies [45,46], and indicators exhibiting redundancy, insufficient discriminative power, or poor feasibility under field conditions were excluded during calibration. The final set of 20 indicators was retained to ensure comprehensive representation of the major structural dimensions of tree health—specifically crown architecture, foliage condition, and trunk integrity—while also incorporating remotely sensed metrics that capture habitat stressors relevant to the high-altitude village environment.

Geospatial vector data were obtained from the National Platform for Common Geospatial Information Services (Tianditu), provided by the Ministry of Natural Resources of China. Sentinel-2A Level-2A imagery acquired on 2 April 2025 was downloaded from the ESA Copernicus Data Hub (https://www.copernicus.eu/en/access-data/conventional-data-access-hubs, accessed on 23 July 2025). For image compositing, three of the ten Sentinel-2A spectral bands available at 10 m spatial resolution (B2–blue, B3–green, and B4–red) were selected because they provide superior separability among vegetation, bare soil, and built-up surfaces, while avoiding redundancy introduced by the remaining bands [47,48]. Image preprocessing (radiometric correction, geometric rectification, projection transformation, and clipping) was carried out in ArcGIS 10.7. Land-use classification was then performed using supervised Maximum Likelihood Classification (MLC), which assigns pixels to the most probable class based on class-specific mean vectors and covariance matrices [49,50]. Cross-validation was applied to minimize overfitting, and the resulting classified raster was used to calculate the area of each land-use category (Figure 1c).

For the RSEI and the Normalized Difference Vegetation Index (NDVI), Landsat 8 OLI/TIRS imagery acquired on 18 June 2025, with less than 5% cloud cover, was downloaded from the USGS Earth Explorer platform (https://earthexplorer.usgs.gov, accessed on 29 July 2025). Radiometric calibration, Flaash atmospheric correction, image mosaicking, and clipping were performed in ENVI (v5.6; NV5 Geospatial, Boulder, CO, USA), followed by raster calculations in ArcGIS (v10.7; Esri Inc., Redlands, CA, USA) to generate the final datasets (

Table 1. Details of satellite imagery datasets.

Data Type	Data Source & Platform	Satellite (Product)	Image ID/Name	Spatial Resolution (m)	Preprocessing	Bands	Date
Land-use data	ESA Copernicus SciHub (Sentinel Open Access Hub) (https://dataspace.copernicus.eu/, accessed on 23 July 2025)	Sentinel-2A L2a	S2A_MSIL1C_20250402T043221_N0511_R133_T46RDT_20250402T081109	10 m × 10 m	Geometric correction; coordinate & projection transform; supervised classification	2, 3, 4	2 April 2025
Ecological Index	USGS Earth Explorer (https://earthexplorer.usgs.gov, accessed on 29 July 2025)	Landsat 8 OLI TIRS	LC08_L2SP_136040_20250609_20250618_02_T1	30 m × 30 m	Radiometric calibration; FLAASH atmospheric correction; mosaicking; clipping	2, 3, 4, 5, 6, 7, 10	18 June 2025

). Among these, OLI refers to the Operational Land Imager, a multispectral sensor onboard the Landsat-8 platform, which was utilized in this study to derive RSEI and NDVI.

2.2. Development of the Health Evaluation Model for Ancient Walnut Trees

2.2.1. Selection of Health Evaluation Indicators

Tree health is influenced by a combination of intrinsic attributes and extrinsic drivers that collectively reflect natural processes and anthropogenic pressures. To establish a comprehensive and scientifically robust evaluation framework, multiple dimensions were integrated. Based on structural and morphological characteristics, three primary indicators—Overall Condition (OC), Leaf Health (LH) and Trunk Health (TH)—were defined to provide a holistic assessment of tree vitality. For the OC dimension, four secondary indicators were selected: Tree Vigor (TV), Tree Inclination (TI), Biotic Damage (BD), and Environmental Stress (ES). Leaf Health (LH) was characterized using Crown Structure (CS), Leaf Color (LC), Major Branch Damage (MBD), and Apical Shoot Dieback (ASD). Trunk Health (TH) was assessed using Trunk Cavities (TC), Trunk Damage (TD), Bark Necrosis (BN), and Hollow Trunk (HT). To capture external pressures, four Anthropogenic Disturbance (AD) indicators—Human Activity Intensity (HAI), Land Surface Imperviousness (LSI), Land-use Type (LUT), and Distance to Roads/Buildings (DRB)—were incorporated. External Habitat Condition (EHC) was represented by four environmental indices: NDVI, RSEI, light availability (LA), and neighboring vegetation competition (NVC). Together, these indicators enabled an integrated evaluation of internal physiological status and external ecological constraints affecting tree health.

TV reflects the integrated physiological performance of the crown, leaves, and trunk. TI denotes trunk leaning caused by external forces such as wind, solar asymmetry, or anthropogenic cutting. BD incidence indicates pathogenic or insect infestation, with severity serving as a direct measure of health decline [51]. ES accounts for shading from neighboring trees or buildings and spatial constraints imposed by roads or debris, thereby reflecting the degree of external disturbance [52]. CS describes canopy fullness [53], while LC reflects photosynthetic activity and efficiency. MBD and ASD represent structural deterioration caused by natural senescence or human interference. TC disrupts nutrient transport between above- and below-ground compartments, weakening structural stability. TD, BN and HT further impair physiological function, collectively serving as strong indicators of aging and vitality loss [54].

AD was quantified using LUT and LSI, while DRB provided a spatial measure of construction-related impacts. NDVI and RSEI were employed to represent vegetation vigor and overall ecosystem quality, whereas LC and NVC reflected habitat pressures.

2.2.2. Tree Health Evaluation Indicators

Evaluation criteria were established with reference to previous studies and standardized guidelines, and subsequently adapted to the walnut tree health assessment framework [55,56,57,58]. Seventeen key indicators were classified into five ordinal levels and scored on a scale from 0 to 4, where a score of 4 indicated optimal health and a score of 0 represented the poorest condition (

Table 2. Evaluation criteria for health indicators of ancient walnut trees.

Primary Indicators	Secondary Indicators	Description	Tree Vigor
			4	3	2	1	0
Overall Condition	Tree Vigor	Overall growth condition	Growth is vigorous with no adverse signs	Good growth with minor local impacts	Noticeable decline in vigor	Very poor growth, extremely weak	Entire tree desiccated, near death
	Tree Inclination	Degree of deviation from vertical	≤5%	≤10%	≤20%	≤30%	>30%
	Biotic Damage	Degree of damage caused by insects, fungi, etc.	<5% affected	5–25% affected	25–50% affected	50–75% affected	>75% affected
	Environmental Stress	Extent of external environmental restrictions on growth	No shading/obstruction, sufficient space	Minor shading or debris	Considerable shading, limited space	Severe shading, highly restricted	Completely shaded or occupied, no space
Leaf Condition	Crown Structure	Integrity of crown form	Round and complete	Nearly natural	≤20% crown loss	≤40% crown loss	>40% crown loss
	Leaf Color	Percentage of discolored foliage	All leaves green	No obvious change (<10%)	Slight discoloration (<25%)	Moderate discoloration (25–60%)	Severe discoloration (>60%)
	Major Branch Damage	Extent of damage to large branches	None	Very minor	Noticeable	Severe, incl. breakage	Majority damaged
	Apical Shoot Dieback	Extent of dieback at crown top	<5% dieback	5–25% dieback	25–50% dieback	50–75% dieback	>75% dieback
Trunk Condition	Trunk Cavities	Proportion of trunk area affected by cavities/decay	No decay/cavity	<1/8 trunk area	1/8–1/4 trunk area	1/4–1/2 trunk area	>1/2 trunk area
	Trunk Damage	Proportion of bark circumference damaged	No damage	<1/3 circumference	1/3–1/2 circumference	1/2–2/3 circumference	>2/3 circumference
	Bark Necrosis	Proportion of bark area dead or detached	None	<1/8 trunk area	1/8–1/4 trunk area	1/4–1/2 trunk area	>1/2 trunk area
	Hollow Trunk	Extent of internal decay	<5%	5–15%	15–30%	30–50%	>50%
Anthropogenic Disturbance	Land Surface Imperviousness	Degree of surrounding land imperviousness	<5%	5–25%	25–50%	50–75%	>75%
	Land-use Type	Type of land-use unit where tree grows	Forest land	Farmland	Grassland	Bare land	Impervious surface
	Distance to roads/buildings	Proximity to built structures	>30 m	20–30 m	10–20 m	5–10 m	0–5 m
External Habitat Condition	Light Conditions	Shading by surrounding vegetation or buildings	<5% shaded	5–15% shaded	15–30% shaded	30–50% shaded	>50% shaded
	Neighboring Vegetation Competition	Extent of competition with neighboring plants	No competition	Minor competition	Moderate competition	Considerable competition	Very strong competition

).

2.2.3. Determination of Indicator Weights

The structural equation model (SEM) employed in this study comprises two components: the measurement model [Equations (1) and (2)] and the structural model [Equation (3)]. The measurement model delineates the associations between latent constructs and their observed indicators, whereas the structural model specifies the directional relationships among the latent constructs themselves. The general formulations of these two components are presented as follows:

X = ΛXξ + δ

(1)

Y = ΛYη + ε

(2)

η = +Гξ + ζ

(3)

where X and Y denote the vectors of exogenous and endogenous observed variables, respectively; ΛX and ΛY represent the factor loading matrices; δ and ε are the measurement errors; ξ and η denote the exogenous and endogenous latent variables, respectively; Γ represents the effects of exogenous latent variables on endogenous latent variables; and ζ is the disturbance term. In this study, no directed paths were specified among endogenous latent variables; therefore, the path matrix B was fixed to zero to avoid model misspecification.

The SEM was statistically identified under standard confirmatory factor modeling principles. Each latent construct was measured by at least three observed indicators, ensuring local identification, and one factor loading per latent variable was fixed to unity to establish the measurement scale. Model identification was further confirmed by the fact that the number of known sample moments exceeded the number of freely estimated parameters, yielding positive degrees of freedom and thereby rendering the model over-identified and statistically testable. This identification strategy follows established SEM methodological guidelines [59,60,61].

SEM analysis was conducted in five stages: (i) theoretical model construction, (ii) hypothesis formulation, (iii) assessment of construct reliability and validity, (iv) evaluation of model fit, and (v) model respecification when necessary. Model estimation was performed using the maximum likelihood method in AMOS 26.0 (IBM Corp., Armonk, NY, USA).

• Theoretical model construction

Tree Health was defined as the sole endogenous latent construct in the structural model. To maintain both conceptual coherence and statistical rigor, tree health was specified as a second-order reflective latent variable, represented by three first-order latent factors: Overall Condition (OC), Leaf Health (LH), and Trunk Health (TH). Each first-order construct was measured through multiple observed indicators. Specifically, OC was assessed using indicators such as tree vigor and crown morphology; LH was represented by foliage coloration and branch dieback; and TH was characterized by trunk damage, cavity presence, and bark condition. Two exogenous latent variables, Anthropogenic Disturbance (AD) and External Habitat Condition (EHC), were hypothesized to exert direct influences on tree health, with AD expected to have a negative effect and EHC a positive one. This model specification ensured consistency between theoretical constructs and statistical estimation, thereby minimizing risks of theoretical–statistical misalignment. The conceptual framework is illustrated in Figure 2.

• Model hypotheses

Based on previous studies of ancient tree health and the ecological characteristics of walnut trees, the following hypotheses were formulated (

Table 3. Hypotheses for the health evaluation model of old Juglans regia L.

Hypothesis	Statement
H1	Overall Condition (OC) is a significant reflective dimension of Tree Health.
H2	Leaf Health (LH) is a significant reflective dimension of Tree Health.
H3	Trunk Health (TH) is a significant reflective dimension of Tree Health.
H4	Anthropogenic Disturbance (AD) has a significant negative effect on Tree Health.
H5	External Habitat Condition (EHC) has a significant positive effect on Tree Health.

).

• Reliability and validity testing

Internal consistency was evaluated using Cronbach’s alpha (α > 0.7 indicating satisfactory reliability) [62]. Sampling adequacy was examined through the Kaiser–Meyer–Olkin (KMO) statistic, where values above 0.5 were considered acceptable, 0.5–0.7 moderate, 0.7–0.8 good, 0.8–0.9 very good, and greater than 0.9 excellent [63]. All statistical analyses were performed using IBM SPSS Statistics 22.0 (IBM Corp., Somers, NY, USA).

• Model fit evaluation

Confirmatory factor analysis was conducted to assess the overall goodness of fit of the model. Four key indices were used: the chi-square to degrees of freedom ratio (χ²/df, denoted as NC), root mean square error of approximation (RMSEA), Tucker–Lewis Index (TLI), and comparative fit index (CFI). An NC value between 1 and 3 indicated an excellent fit, whereas values between 3 and 5 were considered acceptable [60]. An RMSEA below 0.05 suggested a close fit, while values between 0.05 and 0.08 indicated a reasonable fit [64]. TLI and CFI values greater than 0.9 reflected strong model adequacy, and values above 0.8 were deemed acceptable [65].

$$R_{\mathrm{MSEA}}=\sqrt{{\displaystyle \frac{F_0}{df}}}=\sqrt{max\left({\displaystyle \frac{F_{\mathit{ML}}}{df}}-{\displaystyle \frac{1}{N-1}},\hspace{0.17em}0\right)}$$

(4)

$$TLI={\displaystyle \frac{\left(\frac{{\mathsf{\chi }}_N^2}{d{f}_N}-\frac{{\mathsf{\chi }}_T^2}{d{f}_T}\right)}{\frac{{\mathsf{\chi }}_N^2}{d{f}_N}-1}}$$

(5)

$$CFI=1-{\displaystyle \frac{\mathrm{m}\mathrm{a}\mathrm{x}\left({\mathsf{\chi }}_T^2-d{f}_T,\hspace{0.17em}0\right)}{\mathrm{m}\mathrm{a}\mathrm{x}\left({\mathsf{\chi }}_N^2-d{f}_N,\hspace{0.17em}0\right)}}$$

(6)

where F₀ denotes the population discrepancy function value; df represents the model’s degrees of freedom; F_ML is the value of the maximum likelihood estimation function; N refers to the sample size; ${\mathsf{\chi }}_N^2$ is the chi-square value of the null model; ${\mathsf{\chi }}_T^2$ is the chi-square value of the hypothesized model; $d{f}_N$ represents the degrees of freedom of the null model; and $d{f}_T$ denotes the degrees of freedom of the hypothesized model.

• Model modification

If the hypothesized model did not adequately fit the observed data, adjustments were performed based on the modification indices (covariance, variance, and regression weights) provided by AMOS. All modifications were theoretically justified to ensure conceptual validity. Each revised model was subsequently re-evaluated through an iterative process until the final model achieved both statistical robustness and theoretical coherence.

2.2.4. Comprehensive Health Scoring

The comprehensive health index (S) for each tree was computed as follows:

$$S=\sum _{i=1}^m{W}_i{A}_i$$

(7)

where W_i denotes the weight of the i-th indicator, A_i represents its assigned score, and m is the total number of indicators.

To categorize tree health status, K-means clustering was applied to the 135 surveyed ancient walnut trees. The reliability of the clustering outcomes was subsequently validated using Fisher’s linear discriminant analysis, ensuring robust classification consistency and accuracy.

2.3. Model Indicator Extraction and Composite Index Construction

2.3.1. RSEI and NDVI

The RSEI integrates multiple ecological dimensions, including vegetation vigor, soil moisture, surface dryness, impervious surface extent, and LST. It has been widely employed to assess regional ecological quality [10,23]. In this study, the NDVI was used as an indicator of greenness; the normalized difference built-up and soil index (NDBSI), derived from the Soil Index (SI) and the index-based built-up index (IBI), represented surface dryness; the tasseled cap wetness component (WET) quantified soil and vegetation moisture; and LST reflected the thermal condition [66,67]. Accordingly:

$$\mathrm{RSEI}=f\left(\mathrm{Greenness},\hspace{0.17em}\mathrm{Wetness},\hspace{0.17em}\mathrm{Heat},\hspace{0.17em}\mathrm{Dryness}\right)$$

(8)

where greenness denotes vegetation vigor and canopy density, wetness represents surface moisture content, heat indicates land surface temperature, and dryness characterizes the degree of surface aridity or bare soil exposure. In the context of remote sensing, the RSEI can therefore be expressed as:

$$\mathrm{RSEI}=f\left(\mathrm{NDVI},\mathrm{WET},\mathrm{LST},\mathrm{NDBSI}\right)$$

(9)

where NDVI, Wet, LST, and NDBSI denote the vegetation index, the surface moisture component, the land surface temperature, and the normalized difference built-up and soil index (NDBSI), respectively.

• Greenness (NDVI) reflects biomass, leaf area index, and canopy cover, calculated as:

$$\mathrm{NDVI}={\displaystyle \frac{{\rho }_{\mathrm{NIR}}-{\rho }_{\mathrm{Red}}}{{\rho }_{\mathrm{NIR}}+{\rho }_{\mathrm{Red}}}}$$

(10)

where p_NIR and p_Red denote the reflectance in the near-infrared and red spectral bands, respectively, The NDVI ranges from −1 to 1,

• Wetness (WET) was derived from the tasseled cap transformation for Landsat-8 OLI surface reflectance, using the sensor-specific coefficients proposed by Baig et al. [68]:

$$\mathrm{Wet}=0.1511\hspace{0.17em}{p}_{\mathrm{Blue}}+0.1973\hspace{0.17em}{p}_{\mathrm{Green}}+0.3283\hspace{0.17em}{p}_{\mathrm{Red}}+0.3407\hspace{0.17em}{p}_{\mathrm{NIR}}-0.7117\hspace{0.17em}{p}_{\mathrm{SWIR}1}-0.4559\hspace{0.17em}{p}_{\mathrm{SWIR}2}$$

(11)

where p_Blue, p_Green, p_Red, p_NIR, p_SWIR1 and p_SWIR2 denote the reflectance of the blue, green, red, near-infrared, and the first and second shortwave infrared bands, respectively.

• The heat index (LST) was represented by LST. Numerous studies worldwide have retrieved LST from satellite imagery [69,70,71]. In particular, Vineesha Singh (2017) proposed an automated algorithm for LST mapping using Landsat 8 data [72]. The data processing steps were as follows:

1.

First step: Conversion to TOA Radiance

Thermal infrared sensor (TIRS) data were converted to spectral radiance (L_λ) using the radiance rescaling factors provided in the metadata file, as expressed by the following equation:

$$L_{\lambda }=M_L\hspace{0.17em}{Q}_{\mathrm{cal}}+A_L$$

(12)

where L_λ = TOA spectral radiance (Watts/(m² × sr × μm)), M_L = Band-specific multiplicative rescaling factor from the metadata (RADIANCE_MULT_BAND_x, where x is the band number), A_L = Band-specific additive rescaling factor from the metadata (RADIANCE_ADD_BAND_x, where x is the band number), and Q_cal = quantized and calibrated standard product pixel values (DN).

2.

Second step: Conversion to At-Satellite Brightness Temperature

The TOA spectral radiance (L_λ) was converted to brightness temperature (T_B) using the thermal constants (K₁ AND K₂) provided in the metadata file, as expressed by the following equation:

$$T_B={\displaystyle \frac{K_2}{In\left(\frac{K_1}{L_{\lambda }}+1\right)}}-273.15$$

(13)

where T_B = At-satellite brightness temperature (K), L_λ = Top of Atmospheric (TOA) spectral radiance (Watts/(m² × sr × μm)), K₁ = Band-specific thermal conversion constant from the metadata (K₁_CONSTANT_BAND_x, where x is the thermal band number), and K₂ = Band-specific thermal conversion constant from the metadata (K₂_CONSTANT_BAND_x, where x is the thermal band number)

3.

Third step: Calculation of NDVI value

The NDVI was estimated from the OLI sensor’s optical bands after stacking Band 4 and Band 5, using the algorithm presented in Equation (10):

4.

Fourth step: Calculation of Pv

Proportion Vegetation: Pv was estimated using NDVI Threshold method.

$$P_v={\left({\displaystyle \frac{\mathrm{NDVI}-\mathrm{N}\mathrm{D}\mathrm{V}{\mathrm{I}}_{\min }}{\mathrm{N}\mathrm{D}\mathrm{V}{\mathrm{I}}_{\max }-\mathrm{N}\mathrm{D}\mathrm{V}{\mathrm{I}}_{\min }}}\right)}^2$$

(14)

5.

Fifth step: Calculation of Emissivity value step

Deriving Land surface emissivity (LSE):

$$\epsilon =0.004\hspace{0.17em}{P}_v+0.996$$

(15)

6.

Sixth step: Estimation of LST

LST: To derive LST, it is first necessary to calculate the LSE of the area.

$$\mathit{LST}={\displaystyle \frac{\mathit{BT}}{1+\left(\frac{w\hspace{0.17em}\mathit{BT}}{\rho }\right)\ln \left(\epsilon \right)}}$$

(16)

where BT is the satellite brightness temperature (K); λ is the wavelength of emitted radiance (m); ρ = h·c/σ (1.438 × 10⁻² m·K), where h is the Planck constant (6.626 × 10⁻³⁴ J·s), k is the Boltzmann constant (1.38 × 10⁻²³ J·K⁻¹) and c is the speed of light (2.998 × 10⁸ m·s⁻¹).

• Dryness (NDBSI):
• Calculated as the mean of SI and IBI:NDBSI = (IBI + SI)/2,

$$\mathrm{IBI}={\displaystyle \frac{2{\mathsf{\rho }}_{\mathrm{S}\mathrm{W}\mathrm{I}\mathrm{R}1}-{\mathsf{\rho }}_{\mathrm{NIR}}-{\mathsf{\rho }}_{\mathrm{Green}}}{2{\mathsf{\rho }}_{\mathrm{S}\mathrm{W}\mathrm{I}\mathrm{R}1}+{\mathsf{\rho }}_{\mathrm{NIR}}-{\mathsf{\rho }}_{\mathrm{Green}}}}$$

(17)

$$\mathrm{SI}={\displaystyle \frac{\left({\rho }_{\mathrm{SWIR}1}+{\rho }_{\mathrm{Red}}\right)-\left({\rho }_{\mathrm{NIR}}+{\rho }_{\mathrm{Blue}}\right)}{\left({\rho }_{\mathrm{SWIR}1}+{\rho }_{\mathrm{Red}}\right)+\left({\rho }_{\mathrm{NIR}}+{\rho }_{\mathrm{Blue}}\right)}}$$

(18)

where p_Blue, p_Green, p_Red, p_NIR and p_SWIR1 denote the reflectance of the blue, green, red, near-infrared, and the first shortwave infrared bands, respectively.

PCA was applied to integrate the four standardized indicators, NDVI, WET, LST, and NDBSI. Prior normalization eliminated dimensional inconsistencies and minimized the influence of outliers. The initial ecological index (RSEI₀) was derived from the first principal component and subsequently rescaled to a range of 0–1:

$$N{I}_i={\displaystyle \frac{I_i-I_{min}}{I_{max}-I_{min}}}$$

(19)

where $N{I}_i$ denotes the normalized value of the indicator, ranging from 0 to 1. $I_i$ represents the value of the indicator at pixel i, while $I_{max}$ and $I_{min}$ present the maximum and minimum values of the indicator, respectively.

PCA was conducted using the normalized indicators, and the resulting principal components were subsequently normalized to ensure comparability across variables.

$$\mathrm{R}\mathrm{S}\mathrm{E}{\mathrm{I}}_0=1-\mathrm{P}{\mathrm{C}}_1\left(f\left(\mathrm{Wet},\hspace{0.17em}\mathrm{NDVI},\hspace{0.17em}\mathrm{LST},\hspace{0.17em}\mathrm{NDBSI}\right)\right)$$

(20)

$$\mathrm{R}\mathrm{S}\mathrm{E}\mathrm{I}={\displaystyle \frac{\mathrm{R}\mathrm{S}\mathrm{E}{\mathrm{I}}_0-\mathrm{R}\mathrm{S}\mathrm{E}{\mathrm{I}}_{0,\min }}{\mathrm{R}\mathrm{S}\mathrm{E}{\mathrm{I}}_{0,\max }-\mathrm{R}\mathrm{S}\mathrm{E}{\mathrm{I}}_{0,\min }}}$$

(21)

where PCA denotes the Principal Component Analysis; RSEI₀ represents the initial ecological index; and RSEI presents the constructed Remote Sensing Ecological Index, with values ranging from 0 to 1. Higher RSEI values indicate better overall ecological quality.

2.3.2. HAI

HAI was quantified from a landscape ecology perspective, wherein anthropogenic processes reduce the natural attributes of different land-cover types [26,27]. The disturbance coefficients were interpreted as representing relative gradients of anthropogenic modification rather than precise numerical estimates. Accordingly, the HAI values serve comparative rather than absolute purposes, and a formal sensitivity or uncertainty analysis was not required within the current exploratory framework. On this basis, HAI was defined as follows:

$$\mathrm{HAI}=\sum _{i=1}^n{\displaystyle \frac{A_i{P}_i}{T_{\mathrm{A}}}}$$

(22)

where A_i denotes the area of landscape type i, T_A represents the total area, and P_i is the AD coefficient. Higher HAI values indicate more intense human disturbance and greater ecological stress.

2.4. Correlation Analysis

To investigate the relationships between tree health status and its influencing factors, correlation analysis was performed using IBM SPSS Statistics 22.0 (IBM SPSS, Somers, NY, USA) [73]. External drivers (e.g., AD and EHC) and intrinsic tree attributes (e.g., OC, LH, and TH) were analyzed in relation to the health assessment scores. As the sampled trees are spatially distributed, spatial dependence among observations may occur; thus, the correlations are interpreted as exploratory associations rather than strict spatially independent inferences. Nevertheless, the analysis offers meaningful diagnostic insight into the principal drivers of tree health and supports the development of targeted conservation and management strategies.

3. Results

3.1. Comprehensive Evaluation of RSEI

Table 4. PCA result of RSEI.

Title 1	P1	P2	P3	P4
NDVI	0.563	−0.652	−0.469	−0.190
WET	0.264	0.057	0.557	−0.786
NDBSI	−0501	0.177	−0.610	−0.587
LST	−0.601	−0.734	0.313	−0.033
Eigenvalue	0.028	0.010	0.006	0.001
Variance contribution (%)	62.847	22.901	13.684	0.568

presents the PCA results of the RSEI for the new study area in 2025. In PC1, the greenness indicator (NDVI) and the WET exhibit positive loadings, whereas the dryness indicator (NDBSI) and the heat indicator (LST) show negative loadings. This pattern aligns with the ecological principle that greenness and wetness contribute positively to environmental quality, while dryness and heat exert negative effects. Given that the eigenvalue contribution rate of PC1 is approximately 62%, insufficient to comprehensively capture the variance across all indicators, and considering the unique characteristics of the plateau environment, the eigenvalue contributions of PC1 through PC3 were integrated to improve accuracy and minimize error. Using ArcGIS 10.7, the three principal components (PC1, PC2, and PC3) were weighted by coefficients of 0.63, 0.23, and 0.14, respectively, for raster-based computation. This integrated approach provided a more scientifically robust and reasonable representation of regional ecological quality. The resulting spatial distribution map of ecological environmental quality was subsequently generated (Figure 3). Furthermore, using ArcGIS 10.7, ancient-tree point data were overlaid with the RSEI and NDVI layers to quantitatively extract the corresponding RSEI and NDVI values for each ancient tree.

3.2. HAI and Its Spatial Distribution

Given the relatively small spatial scale of the village-level study area, the determination of human impact intensity coefficients for various land-use types was based on previous studies. Land use was classified into three categories: (1) unused layer, comprising unused land; (2) low-use layer, including grassland, forest, and water bodies; and (3) transformed/developed layer, encompassing cropland, bare land, and construction land. A 10-point scale was adopted to assign human impact intensity coefficients to each land-use type, with higher values indicating stronger human activity or development intensity. The corresponding coefficients are presented in

Table 5. Description of human impact intensity factor for landscape types.

Intensity Level	Land-Use Type	Landscape Description	Coefficient Interpretation	Coefficient
Unused layer Low-use layer	Forest land	Dominated by evergreen broadleaf forest, coniferous forest, and mixed conifer–broadleaf forest, with small areas of plantations	Natural surface cover remains unchanged; very limited utilization	2
	Grassland	Natural and artificial grasslands	Partial alteration and utilization of surface cover	3
	Water bodies	Natural lakes, rivers, artificial reservoirs, and canals	Partial alteration and utilization of surface cover	4
	Farmland	Paddy fields, dryland, and irrigated cropland	Surface cover altered; used for short-term or perennial crops	6
Transformed/developed layer	Bare construction land	Abandoned bare land after construction activities	Surface cover largely altered, with artificial isolation layers affecting energy exchange	9
	Construction land	Residential, transportation, and facility land	Surface cover completely altered, with artificial isolation layers affecting energy exchange	10

. Considering that the study area contains a substantial proportion of bare land subjected to intensive construction and development activities, the bare land category was assigned a coefficient value of 9.

For the quantitative assessment of HAI, land-use data of the study area were first utilized to calculate the overall characteristics of human disturbance based on the formula and coefficients presented in Table 5. Considering the spatial resolution of satellite imagery, the scale of the study area, the size of landscape patches, and the feasibility of analysis units, the Fishnet tool in ArcGIS 10.7 was employed to divide the study area into a uniform vector grid of 50 m × 50 m. Using spatial analysis techniques, the average HAI value was computed for each grid cell, thereby deriving the spatial distribution characteristics of baseline human disturbance at the grid scale. Subsequently, the calculated HAI values were classified into five levels using an equal-interval classification scheme (2-unit intervals), which is a widely applied method in ecological and spatial analyses [74]. The five disturbance categories—low (0 < HAI ≤ 2), relatively low (2 < HAI ≤ 4), moderate (4 < HAI ≤ 6), relatively high (6 < HAI ≤ 8), and high (8 < HAI ≤ 10)—describe a relative gradient of anthropogenic influence rather than absolute statistical thresholds. This equal-interval classification scheme enables clear visualization of spatial variation in disturbance intensity and supports spatially explicit mapping of AD patterns, thereby highlighting the heterogeneity of anthropogenic pressures across the study area (Figure 4).

As illustrated in Figure 4, the study area is generally subject to strong anthropogenic influence, though notable spatial heterogeneity exists across subregions. (1) Central village core: This area features relatively flat terrain and favorable hydrothermal conditions, supporting a broad distribution of walnut trees. Land-use Types are dominated by forest, farmland, and grassland, with only scattered construction land. Consequently, land-use coefficients are low, and human disturbance intensity remains at a relatively low level. (2) Southern Yarlung Tsangpo River bank: Due to hydropower development and the construction of major infrastructure such as the Lalin Railway, original land-use types have been extensively altered and converted into bare or construction land. The associated disturbance coefficients are high, resulting in strong human disturbance. (3) Northern sector: Characterized by natural mountainous terrain primarily covered by grassland and forest, this region exhibits low land-use coefficients and well-preserved landscape patches. Overall, disturbance intensity here is the lowest among the three sectors.

3.3. Health Assessment Framework for Ancient Walnut Trees

3.3.1. Model Construction

• Reliability and validity of indicator data

Reliability and factor analyses (

Table 6. Evaluation index data reliability and validity test statistic.

Latent Variable	Observed Variable	Cronbach’s Alpha	Kaiser–Meyer–Olkin (KMO) Test
Overall Condition (OC)	4	0.884	0.823
Leaf Health (LH)	4	0.856	0.723
Trunk Health (TH)	4	0.855	0.810
Anthropogenic Disturbance (AD)	4	0.872	0.825
External Habitat Condition (EHC)	4	0.819	0.791

) indicated strong internal consistency and sampling adequacy across the latent constructs. All three latent variables exhibited Cronbach’s alpha values exceeding 0.8, confirming high reliability, while KMO values for the seven observed variables were above 0.7, indicating satisfactory sampling adequacy. Collectively, these results demonstrate the dataset’s robust reliability and construct validity, validating its suitability for modeling the health status of ancient walnut trees.

• Model fit

As shown in

Table 7. Goodness-of-fit indices of the initial (theoretical) SEM and the modified SEM.

Model Fit Indices	NC (CMIN/DF)	RMSEA	GFI	AGFI	TLI	CFI
Theoretical Model	2.777	0.115	0.772	0.708	0.841	0.839
Modified Model	1.372	0.053	0.868	0.829	0.956	0.962

, the initial theoretical model exhibited a poor fit, with an RMSEA value of 0.115, exceeding the acceptable threshold of 0.08. Following model refinement, the goodness-of-fit indices improved substantially, yielding NC = 1.372, RMSEA = 0.053 (<0.08), TLI = 0.956, and CFI = 0.962. These values indicate that the revised model achieved an excellent overall fit to the observed data, confirming its statistical robustness and theoretical adequacy [67].

3.3.2. Indicator Weight Determination

The revised structural path diagram (Figure 5) demonstrates that the latent variables OC, LH and TH, exhibited strong and positive loadings on tree health, with standardized coefficients of 0.86, 0.69, and 0.84, respectively. These results confirm the hypothesized positive causal relationships among the constructs.

As shown in Table 7, the measurement model met recognized criteria for reliability and construct validity. The majority of observed indicators exhibited high standardized loadings on their corresponding latent constructs, demonstrating that the selected variables adequately capture the conceptual dimensions of tree health.

Normalization of standardized loadings is commonly used to express the relative contributions of observed indicators under each latent variable in SEM applications. To enhance clarity, the normalized indicator weights are summarized in

Table 8. Calculation results of health index weight of old Juglans regia L.

Latent Variable	Weight	Observed Variable	Weight
Overall Condition	0.360	Tree Vigor	0.095
		Tree Inclination	0.081
		Biotic Damage	0.100
		Environmental Stress	0.084
Leaf Health	0.289	Crown Structure	0.083
		Leaf Color	0.071
		Major Branch Damage	0.064
		Apical Shoot Dieback	0.071
Trunk Health	0.351	Trunk Cavities	0.097
		Trunk Damage	0.075
		Bark Necrosis	0.087
		Hollow Trunk	0.092
Anthropogenic Disturbance		Remote-sensing-based Human Activity Intensity	0.193
		Land Surface Imperviousness	0.270
		Land-use Type	0.273
		Distance to roads/buildings	0.264
External Habitat Condition		Normalized Difference Vegetation Index	0.216
		Remote Sensing-based Ecological Index	0.219
		Light Conditions	0.267
		Neighboring Vegetation Competition	0.298

.

Briefly, among the three endogenous dimensions, Overall Condition and Trunk Health contributed more strongly to the latent construct of tree health than Leaf Health. At the indicator level, only a subset of variables accounted for most of the explanatory power. Indicators associated with structural integrity and vigor decline, such as biotic damage, trunk cavities, and tree vigor, exhibited comparatively higher weights, whereas leaf-related symptoms contributed less.

Among exogenous influences, land-use configuration and competitive interactions from surrounding vegetation were the strongest determinants, whereas generalized greenness indices (e.g., NDVI) had relatively minor contributions. Overall, these results indicate that ancient walnut health is primarily governed by structural deterioration and vigor impairment, with anthropogenic land-use pressure and neighborhood competition acting as key external controls.

3.3.3. Health Assessment Outcomes

K-means clustering categorized tree health into five distinct classes: healthy, sub-healthy, declining, severely declining, and near death. Fisher’s linear discriminant analysis (

Table 9. Health discrimination results of old Juglans regia L. based on K-means clustering.

Health Category	Fisher’s Linear Discriminant Analysis (FLDA)
	Healthy	Sub-Healthy	Declining	Severely Declining	Near Death	Total
Healthy	12	0	0	0	0	12
Sub-Healthy	0	36	0	0	0	36
Declining	0	0	54	0	0	54
Severely Declining	0	0	0	29	0	29
Near Death	0	0	0	0	4	4
Total	12	36	54	29	4	135
Percentage (%)	8.89	26.67	40.00	21.48	2.96	100

) effectively validated the robustness of this classification scheme. Among the 135 trees evaluated, 12 (8.89%) were classified as healthy, 36 (26.67%) as sub-healthy, 54 (40.00%) as declining, 29 (21.48%) as severely declining, and 4 (2.96%) as near death. Overall, approximately 30% of the trees were in relatively good condition (healthy + sub-healthy), while about 20% were in a state of critical decline.

3.4. Correlation Between Health and Influencing Factors

As shown in

Table 10. Description and statistical analysis of the correlation of data related to various factors affecting the health of old Juglans regia L.

Indicator	Overall Condition	Leaf Health	Trunk Health	Anthropogenic Disturbance	External Habitat Condition
Mean ± SD	2.770 ± 0.662	2.900 ± 0.617	2.869 ± 0.647	2.525 ± 0.959	2.858 ± 0.570
p	<0.001	<0.001	<0.001	<0.001	<0.001

, both intrinsic (OC, LH, and TH) and extrinsic (AD and EHC) factors were significantly associated with overall tree health (p < 0.001). Among intrinsic factors Overall Condition (OC), Leaf Health (LH), and Trunk Health (TH) were all positively correlated with tree health (r = 0.820, 0.755, and 0.854, respectively). Notably, TH exhibited the strongest association, highlighting its role as a central physiological determinant of ancient walnut tree vitality (

Table 11. Analysis of the Correlation between the Health Status of old Juglans regia L. and Their Internal Factors.

Indicator	Tree Health	Overall Condition	Leaf Health	Trunk Health	Anthropogenic Disturbance	External Habitat Condition
Tree Health	1.000	0.820 **	0.755 **	0.854 **	−0.466 **	0.603 **
Overall Condition	0.820 **	1.000	0.481 **	0.584 **	−0.457 **	0.540 **
Leaf Health	0.755 **	0.481 **	1.000	0.510 **	−0.218 *	0.470 **
Trunk Health	0.854 **	0.584 **	0.510 **	1.000	−0.485 **	0.433 **
Anthropogenic Disturbance	−0.466 **	−0.457 **	−0.218 *	−0.485 **	1.000	−0.420 **
External Habitat Condition	0.603 **	0.540 **	0.470 **	0.433 **	−0.420 **	1.000

Note: ** Correlation is significant at the 0.01 level (two-tailed). * Correlation is significant at the 0.05 level (two-tailed).

).

Among extrinsic factors, Anthropogenic Disturbance (AD) was negatively correlated with tree health (r = −0.466), whereas External Habitat Condition (EHC) was positively correlated (r = 0.603). These results align with previous studies. For instance, previous studies have reported that urban development often fragments ancient tree habitats and disrupts soils and root systems, thereby reducing longevity [74]. In contrast, favorable habitats characterized by intact vegetation, stable soil moisture, and buffered microclimates enhance tree resilience. Similarly, other studies have shown that ancient trees typically persist in relatively undisturbed, stable environments, such as parks, wooded pastures, or forest edges, further supporting the present findings [75].

4. Discussion

4.1. Performance of the SEM-Based Health Assessment

We evaluated 135 ancient walnut trees using an SEM. The goodness-of-fit indices (NC, RMSEA, TLI, and CFI) fell within acceptable ranges, indicating that the SEM adequately represents the observed data and satisfies the requirements for reliable health assessment. These results confirm the feasibility and scientific validity of the SEM-based framework as a robust alternative to conventional indicator-weighting methods, providing a transparent, theory-driven basis for evaluation and evidence-informed management.

Conventional weighting approaches often underutilize information contained in raw data and neglect inter-indicator dependencies. In contrast, SEM integrates prior knowledge with multivariate relationships, explicitly encodes hypothesized causal pathways, and produces a logically coherent path structure following model refinement. By coupling SEM with quantitative remote-sensing indicators (e.g., RSEI), the framework enhances interpretability and minimizes subjectivity in weight derivation. Overall, the revised model demonstrates strong fit to the data and is well-suited for operational health assessment of ancient walnut trees.

4.2. Interpretation of Weights and Key Drivers

As illustrated in Figure 6. Across the three health dimensions, Overall Condition (OC), Leaf Health (LH), and Trunk Health (TH) all showed significant positive correlations with tree health (r = 0.82, 0.76, and 0.85, respectively), with TH exerting the strongest influence. These correlations do not indicate causal dominance; rather, they reflect the tendency for structural integrity to co-vary with overall vitality. Trunk integrity is critical for hydraulic transport and mechanical stability; trees with few, well-healed cavities, continuous bark, and non-load-bearing internal hollows are able to maintain key ecosystem functions, including carbon storage, habitat provisioning, and aesthetic services [76]. In contrast, expanding cavities, advanced decay, and bark necrosis that extend into load-bearing zones disrupt nutrient transport, compromise structural stability, and precipitate sharp declines in overall health [77].

At the indicator level, Trunk Cavities, Hollow Trunk and Biotic Damage carried the highest weights, whereas Leaf Color (LC), Major Branch Damage (MBD) and Apical Shoot Dieback (ASD) were comparatively less influential. This pattern aligns with prior studies: trunk inclination significantly affected the health of 695 oaks [78]; trunk condition was pivotal in Brazilian urban heritage trees [79]; and dead branches, cavities, and mechanical damage were key indicators of decline in 3659 urban trees. Observed deviations among studies likely reflect differences in sample size, climate, and species morphology [80,81].

Among Anthropogenic Disturbance (AD) indicators, Land-use Type (LUT) carried the greatest weight, whereas Remote-sensing-based Human Activity Intensity (HAI) contributed the least. Land-use mosaics integrate multiple disturbance pathways, including surface sealing, construction encroachment, tourism, and agricultural expansion, which can alter root-zone structure, reduce soil water retention, and fragment habitats. Consequently, their explanatory power often surpasses that of single disturbance metrics [82]. Indeed, land-use change is a dominant driver of forest stability and tree survival, frequently exerting greater influence than individual pressure variables [83].

Within External Habitat Condition (EHC) indicators, Normalized Difference Vegetation Index (NDVI) had the highest weight, highlighting the critical role of vegetation cover and biomass as foundational support for ancient tree vitality. NDVI has long been recognized as a reliable proxy for habitat quality and primary productivity [84]. By contrast, light availability exhibited the lowest weight, likely because mature canopy trees efficiently capture sunlight, and marginal light effects are often outweighed by neighbor competition and below-ground resource limitations [85].

4.3. Management Implications for Ancient Walnuts

Discriminant analyses indicate that the population of ancient walnut trees is generally in poor health: only 8.89% of trees were classified as healthy and 26.67% as sub-healthy, whereas 40.00% and 21.48% fell into the declining and severely declining categories, with 2.96% near death. The population is thus dominated by compromised health classes, consistent with global evidence of tree degradation under intense human pressure and habitat fragmentation [86].

Field observations corroborate these findings. Healthier trees typically exhibit strong vigor, full crowns, dense foliage, and intact stems, with minimal pest or disease incidence and low levels of AD. In contrast, many trees, though not necessarily the oldest, display sparse foliage, widespread twig dieback, severe trunk hollows, and exposed roots, patterns closely associated with construction-related vegetation removal and habitat loss, which exacerbate drought and soil erosion stress [87,88].

Ancient tree health arises from the interaction of biotic factors (e.g., parasitism, competition, and pests) and abiotic factors (e.g., wind, drought, and flooding). Walnuts require specific edaphic conditions, a warm, moist climate and loose, fertile, well-drained, slightly acidic soils, so shifts in soil physicochemical properties can have pronounced effects on health [89]. Although rhizosphere metrics were not included in the present study, future research should integrate high-resolution soil monitoring to better capture soil–tree interactions. In addition to ecological conditions, socio-cultural dynamics may influence health outcomes. Observational studies in Tibetan and Himalayan contexts have reported that culturally valued trees may receive differential care, reduced disturbing activities, and preferential protection, which can indirectly buffer ecological stress [90,91]. Such effects have been documented in other regions where traditional practices lead to reduced disturbance around sacred trees or groves, thereby promoting longevity and structural stability [92,93]. Accordingly, active community participation and integration of local knowledge constitute promising pathways to enhance conservation effectiveness.

For the middle Yarlung Tsangpo region, our standardized framework can be directly operationalized. Field measurements, combined with model-derived weights, enable rapid scoring and risk classification, supporting precise and time-efficient interventions. We recommend the following strategies:

(1)

Habitat-centric management: Treat the tree and its immediate surroundings as an integrated micro-ecosystem and biodiversity micro-hotspot, addressing the current overemphasis on individual trees while neglecting site integrity.

(2)

Diagnostics focused on trunk integrity and foliage color: Employ non-destructive tools such as sonic tomography and ground-penetrating radar (GPR) to enhance objectivity and reliability.

(3)

Targeted restoration for low-scoring compartments: Apply structural pruning, eco-friendly cavity treatment, and root-zone aeration. Codify management standards, adopt multidisciplinary practices, legislate protective zones, and promote community awareness to ensure long-term conservation.

(4)

Limitations and future directions: Spatial scope, sample size, and indicator coverage were constrained, and some diagnostics retain subjective components. Advances in technology will improve measurement rigor. Future studies should refine class thresholds and integrate both ecological and cultural variables to design more precise, context-specific management strategies.

Overall, ancient walnuts in Tibetan traditional villages along the middle Yarlung Tsangpo are in a precarious state. Coordinated action across ecological, cultural, and technological dimensions is essential to achieve sustainable conservation and management.

5. Conclusions

Using Gamai and neighboring villages in Jiacha County (Tibet Autonomous Region) as case studies, this research integrated satellite-based remote sensing with a tree health assessment framework to evaluate ancient walnut trees at the fine scale of Tibetan traditional villages. By applying SEM, we quantified ecological conditions, HAI, and tree health, and analyzed the interactions between intrinsic and extrinsic drivers. The key conclusions are as follows:

(1)

We developed an SEM-based health risk assessment model for ancient walnut trees, with all fit indices meeting accepted statistical thresholds, demonstrating strong model reliability and suitability for operational evaluation. These results reflect patterns within the present dataset and spatial–temporal sampling frame.

(2)

Overall health status was moderate; however, trees located in construction-affected areas outside village cores exhibited substantially reduced vigor and pronounced structural decline. This pattern is consistent with the fitted model, in which disturbance-related indicators show higher weights and negative path coefficients, confirming that anthropogenic pressure is a significant determinant of declining tree health.

(3)

SEM analysis identified trunk condition as the most critical diagnostic indicator, reflecting its decisive role in structural integrity and physiological stability. This finding is consistent with theoretical expectations, though influenced by the inherent complexity of field-based structural assessment.

(4)

Tree Health is influenced by both intrinsic and extrinsic factors. Externally, health declined under increased Anthropogenic Disturbance (AD) but improved with better External Habitat Condition (EHC), with land-use change exerting both direct and indirect effects through habitat modification. Internally, indicators such as Hollow Trunk (HT) and Bark Necrosis (BN) showed strong correlations with health, underscoring tree vigor as a central integrative trait.

In summary, integrating SEM with modern remote-sensing indicators offers a rigorous, transparent, and scalable framework for evaluating ancient tree health. Implementing dynamic monitoring, prioritizing declining individuals, optimizing site conditions, and mitigating anthropogenic pressures are essential for effective conservation. Beyond walnut trees, the proposed framework provides valuable guidance for the assessment and sustainable management of other heritage trees and contributes to the cultural landscape conservation of Tibetan traditional villages. While both visual scoring and remotely sensed metrics carry inherent limitations, the consistent application of these methods preserves internal comparability. Future work incorporating multi-season imagery, expanded physiological measurements, and quantitative monitoring approaches would further enhance the generality of the findings and reduce methodological uncertainty.