sUAS-Based High-Resolution Mapping for the Habitat Quality Assessment of the Endangered Hoolock tianxing Gibbon

Abstract

The endangered Gaoligong hoolock gibbon (Hoolock tianxing) faces significant threats from habitat degradation and loss, making accurate habitat assessment crucial for effective conservation. This study explored the effectiveness of high-resolution small unoccupied aerial system (sUAS) imagery for evaluating habitat quality, comparing its performance against Sentinel-2 satellite data. Focusing on the critically fragmented habitat of this primate in Yingjiang County, China, we aimed to (1) assess habitat quality at the patch level using a sUAS; (2) apply the InVEST Habitat Quality (IHQ) model; and (3) compare the effectiveness of sUAS and Sentinel-2 imagery, across different resolutions, for habitat quality evaluation. We utilized sUAS imagery (0.05 m resolution) obtained from a DJI Mavic 3 drone and Sentinel-2 data (10 m resolution) for a comparative analysis. The InVEST IHQ model was then used to analyze nine habitat patches, examining how data resolution impacts habitat quality assessments. Our results showed that habitat quality varied considerably across space, with lower quality observed near villages due to agricultural activity and infrastructure development. The sUAS imagery proved superior at capturing detailed landscape features and delineating small, fragmented patches compared to Sentinel-2. Furthermore, the sUAS achieved higher classification accuracy. Although both data sources indicated generally high habitat quality, Sentinel-2 tended to overestimate both habitat quality and degradation compared to the sUAS. High-resolution sUAS imagery therefore provides a clear advantage for detailed habitat quality assessment and targeted conservation planning, especially in fragmented landscapes. Integrating sUAS data with other remote sensing methods is essential to improve the protection of endangered primate habitats. This research emphasizes the value of sUAS for fine-scale habitat analysis, providing a strong scientific basis for developing targeted habitat restoration strategies and guiding conservation management.

1. Introduction

Habitat fragmentation is a pervasive global environmental issue that significantly threatens biodiversity, particularly for species with specialized habitat needs and limited dispersal abilities. Primates are especially vulnerable to this fragmentation because it disrupts essential habitat connectivity, leading to population isolation and an increased risk of extinction. Worryingly, current assessments indicate that 60% of primate species worldwide are classified as critically endangered, endangered, or vulnerable. The situation is even more critical in Asia, where 73% of primate species and 95% of populations are declining [1]. The gibbon family, native to the tropical and subtropical evergreen broadleaved forests of Southeast Asia, is crucial for biodiversity conservation. Gibbons are recognized as the most diverse group among great apes, encompassing 4 genera and 20 species [2,3,4]. Within China, gibbons are represented by three genera and seven species [5,6,7]. Unfortunately, over the past four centuries, the distribution and abundance of gibbons in China have drastically decreased [8,9,10,11,12].

The Gaoligong hoolock gibbon (Hoolock tianxing) is classified as endangered (EN) on the International Union for Conservation of Nature (IUCN) Red List, with its range spanning northern Myanmar and southwestern Yunnan, China [6]. Historically, in the 1980s, the Gaoligong hoolock gibbon was widespread across nine counties in southwestern Yunnan [13,14,15]. However, by 1992, the species had disappeared from many areas, with the total population plummeting to fewer than 300 individual gibbons [16]. By 2009, a further decline left fewer than 200 individual gibbons inhabiting fragmented habitats within the Gaoligong Mountain National Nature Reserve, Tengchong Houqiao, and near the Tongbiguan Provincial Nature Reserves [7]. Recent surveys estimate the Gaoligong hoolock gibbon population at approximately 170 gibbons, dispersed across four subpopulations. Each subpopulation is further divided into 1–4 small groups, geographically isolated by human-made barriers like roads and agricultural land [12]. Notably, some habitats are located less than 100 m from human activities, highlighting that habitat fragmentation and isolation are major threats to this species (Figure 1) [12,13,14,15,16,17,18]. The Gaoligong hoolock gibbon is highly dependent on mid-mountain moist evergreen broadleaved forests for its survival. These habitats are characterized by high plant species diversity, including families such as Lauraceae, Theaceae, Fagaceae, and Rosaceae, which provide the complex canopy structure essential for their arboreal lifestyle [19]. Research indicates that Gaoligong gibbons travel approximately 1 km daily, with a home range not exceeding 100 hectares [20], emphasizing the critical need to maintain habitat integrity and connectivity. Therefore, the effective protection of the Gaoligong hoolock gibbon requires scientifically rigorous habitat assessment and conservation planning to ensure the preservation of the tall canopy and interconnected habitat networks vital for their long-term survival.

Landscape analysis is a valuable and commonly used technique in animal habitat conservation planning [21,22], and it has been gaining increased application recently. However, it is important to recognize that landscape analysis results are inherently influenced by scale-dependent factors, especially the spatial extent and resolution of the input data [23,24]. Studies have shown that using low-spatial-resolution data (≥100 m) can introduce errors in landscape analysis and subsequent habitat assessments [25]. Although considerable research has explored the impacts of habitat fragmentation on non-human primates, the Gaoligong hoolock gibbon has received relatively limited attention, particularly regarding the use of high-resolution remote sensing for habitat assessment. Existing studies have largely relied on lower-resolution data, which may not accurately capture the fine-scale structure and patch quality of the habitat. This limitation can lead to landscape analysis outputs that lack the precision needed to effectively guide targeted conservation actions [26]. Moreover, there is a notable gap in systematic investigations into how high-resolution data can optimize and refine conservation strategies [27].

Driven by advances in analytical methods and tools, the increasing availability of higher-resolution imagery has greatly enhanced the capacity of landscape analysis to produce more detailed and accurate results [28,29,30]. High-spatial-resolution data offer several key advantages in landscape analysis. First, they can represent more complex landscape geometric features and minimize the loss of spatial information [31,32,33]. Second, they enable the accurate identification of critical habitats and offer valuable guidance for habitat restoration [34]. Third, they facilitate the monitoring and analysis of large-scale ecological processes [35]. Finally, they advance the field of landscape ecology, shifting from traditional two-dimensional to more sophisticated three-dimensional spatial analysis [36].

Wasserman et al. [25] conducted a comparative study evaluating the performance of Landsat 8 (resolution 30 m), Sentinel-2 (resolution 10 m), and LiDAR (resolution 1 m) in landscape analysis. Their findings demonstrated that higher-resolution data significantly improved the accuracy of depicting the spatial structure of forests. Furthermore, the emergence of small unoccupied aerial system (sUAS) photography has enabled the acquisition of even finer-resolution data (resolution 0.05 m), providing robust technical support for achieving higher classification accuracy [37] and effectively identifying structural variations in forest restoration sites in tropical regions [38]. Lazaro et al. [39] further confirmed the advantages of sUAS imagery in habitat assessment, showing its superior ability to display more detailed spatial information of habitat categories compared to Sentinel-2 imagery. In animal monitoring, Li et al. [40] highlighted the crucial contribution of sUAS infrared and visible light image fusion technology in providing important references for the observation and conservation of the Hainan gibbon (Nomascus hainanus). Beyond gibbon conservation, sUASs have found diverse applications in wildlife management. For example, in Sri Lanka, sUASs have been used to mitigate human–elephant conflicts, track poachers, detect illegal plantations, and conduct wildlife surveys [41]. Linchant et al. [42] explored the potential of innovative rose-shaped flight patterns for improving wildlife monitoring effectiveness. Yan et al. [43] used a sUAS to monitor Asian elephants in Xishuangbanna, successfully identifying key conservation areas and clarifying the impact of different land use types on hotspot distribution. Francis et al. [44] demonstrated that both sUASs and ground surveys provide unique and complementary insights into waterbird nesting dynamics, advocating for the integration of both methods for comprehensive waterbird monitoring and conservation.

Given the critically fragmented habitat of the Gaoligong hoolock gibbon in Yingjiang County, where small, isolated subpopulations are confined to habitat patches within a human-dominated landscape, this study aims to (1) assess the habitat quality of individual patches using sUAS photography; (2) apply the InVEST Habitat Quality (IHQ) model to quantitatively evaluate habitat quality; and (3) compare the performance of sUAS imagery against 10 m resolution Sentinel satellite data at the patch scale, based on varying image resolutions. By pursuing these objectives, this research intends to emphasize the significant value of high-resolution sUAS data in guiding targeted conservation efforts for this endangered primate species facing complex ecological and human-induced challenges. Specifically, we integrate human visual interpretation with the IHQ model to quantitatively demonstrate the advantages of sUAS imagery in habitat quality assessment, thereby providing a strong scientific foundation for developing and implementing targeted habitat restoration measures.

2. Methods

2.1. Study Site

The study site is situated within Sudian Lisu Autonomous Township and Zhina Township, Yingjiang County, Yunnan Province, China (97.75°–98.19° E, 25.00°–25.34° N), bordering Myanmar. The elevation of the area varies considerably, ranging from 624 m to 3404 m above sea level. This region experiences a monsoonal climate characterized by a southwestern prevailing wind and indistinct seasonal variation. Annual rainfall averages approximately 1402.7 mm, with a pronounced wet season from June to November and a dry season from December to May [45]. The rivers in this area are tributaries of the Irrawaddy River, as illustrated in (Figure 2).

2.2. sUAS Data Acquisition and Processing

2.2.1. Data Acquisition

In a prior population survey utilizing acoustic monitoring, Guan et al. [12] investigated the distribution of Gaoligong hoolock gibbons from October 2022 to February 2023, the local winter season. Building upon the distribution data from this survey [12], we established the locations of gibbon groups in Yingjiang as centers and defined gibbon habitat extent by delineating a 0.75 km radius, intended to encompass the home range of each group. High-resolution imagery of these gibbon habitats was acquired using a DJI Mavic 3 multispectral (Shenzhen, China) sUAS system. These sUAS images were captured between 19 and 24 July 2023. This period coincided with the rainy season in Yingjiang, corresponding to a time of lush vegetation. Image acquisition covered seven habitat patches in Sudian Township (lmh1, lmh2, lmh3, lmh4, lmh5, lmh6, and pw01) and two in Zhina Township (xb03 and xby), Yingjiang County. Each patch was inhabited by a single, geographically isolated Gaoligong hoolock gibbon group (

Table 1. The correspondence between villages and patches.

Village Name	Patch Name
lamahe	lmh1
	lmh2
	lmh3
	lmh4
	lmh5
	lmh6
pawa	pw01
xiangbai	xb03
xinbaiyan	xby

).

Flight missions were conducted on days with clear weather conditions. To account for terrain variations, a terrain-following flight mode was employed to maintain consistent spatial resolution of the imagery. The sUAS flight altitude was maintained at 180 m, yielding a spatial resolution of approximately 0.05 m, with a forward overlap of 80% and a side overlap of 70%. The ISO setting was set to automatic, and Real-Time Kinematic (RTK) positioning was enabled throughout all flights.

For the Lama River area, flight missions across six of the seven patches in Sudian Township were executed between 11:33 and 17:32 on 19 July 2023. The remaining Pawa patch in Sudian Township was surveyed from 10:44 to 11:25 on 24 July 2023. In Zhina Township, flight missions for the Xiangbai and Xinbaiyan patches were carried out from 17:16 to 17:42 on 20 July 2023 and from 9:57 to 10:28 on 23 July 2023, respectively.

2.2.2. sUAS Data Preprocessing

The acquired images were imported into Pix4D software 4.3 [46], with the geographic coordinate system set to WGS 1984. The processing workflow in Pix4D comprised the following key stages: (1) Initial Processing: This involved calibration, the determination of internal and external camera parameters, utilizing these parameters to ascertain the position and orientation of each image, matching common features between images, tie point creation, and point cloud generation. (2) Point Cloud and Mesh Generation: Spurious or outlier points were removed from the point cloud, followed by point classification into categories such as ground, vegetation, and buildings, based on their spatial position and spectral characteristics. (3) DSM, Orthomosaic, and Index Calculation: Images were projected onto the Digital Surface Model (DSM) to generate an orthomosaic, and relevant indices were calculated. Following these procedures, an orthomosaic image was generated.

Building upon the Pix4D processing outputs, we developed a manual interpretation key to classify seven land cover types. The high spatial resolution (0.05 m) of the sUAS imagery, coupled with distinct canopy structure variations, facilitated the differentiation between broadleaf evergreen forests and coniferous forests. Consequently, seven categories were defined: forest, pine forest, grass, farm, built-up, bare land, and water. Subsequently, the visual interpretation of the orthomosaic image was conducted based on the established interpretation key (Figure 3) to produce a land use/land cover (LULC) map of the study area.

2.3. Collection and Processing of Sentinel-2 Satellite Images

2.3.1. Data Collection

To investigate the influence of varying spatial resolutions on landscape analysis within the gibbon habitat, Sentinel-2 satellite imagery, acquired temporally proximal to the sUAS flights, was utilized. Sentinel-2 images (10 m spatial resolution) of the habitat were downloaded for the period spanning June to October 2021, according to the delineated study area. These images were obtained from the Copernicus Open Access Hub (https://scihub.copernicus.eu/dhus/#/home accessed on 1 March 2023). Images with a cloud cover of less than 5% were selected, satisfying the criteria for this investigation. It should be noted that June and November are both rainy seasons in the region, and significant changes in vegetation are not anticipated between these months [45].

2.3.2. Satellite Image Preprocessing

For the Sentinel-2 imagery, atmospheric correction was performed using the Sen2Cor v2.12 plugin [47]. Geometric correction was executed utilizing the RPC file inherent in the original Sentinel-2 data. Subsequently, the corrected images were resampled to a 10 m spatial resolution.

Supervised classification was conducted using the maximum likelihood method. This approach establishes a probabilistic model for each pixel class based on statistical parameters, such as standard deviation and mean, thereby achieving the classification objective [48]. The images were classified into six land cover classes: forest, farm, built-up, water, bare land, and grass. Land use/land cover (LULC) maps with a 10 m spatial resolution were generated. Finally, the images were cropped to the extent of the drone flight area.

To evaluate the spatiotemporal consistency of the classification results, using the sUAS-derived land cover map as a reference, accuracy assessment metrics derived from the confusion matrix were calculated. These metrics included the overall classification accuracy, the user’s accuracy, the producer’s accuracy, and the Kappa coefficient. The formulas for LULC classification accuracy evaluation are as follows:

Overall accuracy (OA):

$$\mathit{OA}={\displaystyle \frac{{\sum }_{i=1}^k{M}_{ii}}{N}}$$

(1)

$M_{ii}$: the values on the diagonal of the confusion matrix (correctly classified pixels for each class); N: the total number of pixels; k: the number of classes.

User’s accuracy (UA):

$$U{A}_i={\displaystyle \frac{M_{ii}}{{\sum }_{j=1}^k{M}_{ij}}}$$

(2)

$M_{ij}$: the value in the confusion matrix where class iii is classified as class j.

Producer’s accuracy (PA):

$$P{A}_i={\displaystyle \frac{M_{ii}}{{\sum }_{j=1}^k{M}_{ji}}}$$

(3)

$M_{ji}$: the value in the confusion matrix where the true class is iii but it is classified as j.

Kappa coefficient (κ):

$$\kappa ={\displaystyle \frac{OA-P_e}{1-P_e}}$$

(4)

where $P_e$ is the expected agreement by chance:

$$P_e={\displaystyle \frac{{\sum }_{i=1}^k({\sum }_{j=1}^k{M}_{ij})({\sum }_{j=1}^k{M}_{ji})}{N^2}}$$

(5)

2.4. Habitat Quality Assessment Based on the InVEST Model

The InVEST (Integrated Valuation of Ecosystem Services and Tradeoffs) model is a widely recognized tool for assessing habitat quality. This model evaluates alterations in habitat quality across various land use scenarios by quantifying habitat ecological functions and biodiversity levels. Specifically, the “Habitat Quality” (HQ) module within InVEST utilizes land use data, integrating factors such as the maximum impact distance and relative weight of threat sources, habitat suitability for each land use type, and sensitivity to these threats for a comprehensive habitat quality assessment. By considering the relative impact of each threat source, the sensitivity of each habitat type, and the spatial proximity between habitats and threats, the HQ module determines the spatial distribution of habitat quality within the study area. The habitat quality assessment is as follows:

$$Q_{xj}=H_j(1-({\displaystyle \frac{D_{xj}^z}{D_{xj}^z+k^z}}))$$

(6)

where $Q_{xj}$ is the habitat quality of grid cell x within the habitat type; $H_j$ is ecological suitability; z is the normalization constant; k is the half-packet saturation constant; and $D_{xj}$ represents the degree of habitat degradation of grid x within habitat type, and its calculation formula is as follows:

$$D_{xj}={\sum }_{r=1}^R{\sum }_{y=1}^{y^r}({\displaystyle \frac{{\omega }_r}{{\sum }_{r=1}^R{\omega }_r}})r_y{i}_{rxy}{\beta }_x{S}_{jr}$$

(7)

where the variables ${\omega }_r$ and $i_{rxy}$ represent the weight of each danger element r, the influence of danger r in raster y on habitat x, and the accessibility level of grid x, respectively; ${\beta }_x$ represents the accessibility level of grid x, and values closer to 1 indicate higher accessibility; and $S_{jr}$ is the sensitivity of land cover type j to threat factor r, with values closer to 1 indicating higher sensitivity.

When the distance attenuation effect of the threat factor r on the grid x is expressed as a linear function, the formula is as follows:

$$i_{rxy}=1-({\displaystyle \frac{d_{xy}}{d_{r max}}})$$

(8)

When the distance attenuation effect of the threat factor r on the grid x is expressed as an exponential function, the formula is as follows:

$$i_{rxy}=\mathrm{e}\mathrm{x}\mathrm{p}(-({\displaystyle \frac{2.99}{d_{r max}}})d_{xy})$$

(9)

$d_{xy}$ is the distance between grid x and grid y; $d_{r max}$ is the maximum scope of influence of the threat factor.

In this study, model parameters were defined based on a synthesis of existing literature, model guidelines, expert consultation, and the known ecology and habitat preferences of Gaoligong hoolock gibbons [12,49,50]. Six threat factors were selected: coniferous forest, cropland, sod, waterbody, bare ground, and residential buildings. The maximum impact distance and relative weight for each of these factors (

Table 2. Threat parameters used in this study.

Threat	Max_Dist	Weight	Decay
residential buildings	1	1	exponential
bare ground	0.6	0.5	linear
cropland	0.4	0.4	linear
sod	0.3	0.3	linear
waterbody	0.2	0.2	linear
coniferous	0.1	0.2	linear

), along with habitat suitability and sensitivity values for each land use type (

Table 3. Habitat sensitivity parameters for land use/land cover (LULC) categories, and potential impacts of threat factors on respective LULC classes.

LULC	Habitat Suitability	Residential Buildings	Bare Ground	Cropland	Sod	Waterbody	Coniferous
no data	0	0	0	0	0	0	0
grass	0.4	0.4	0.5	0.3	1	0.5	0.6
pine forest	0.7	0.6	0.7	0.6	0.6	0.5	1
farm	0.2	0.3	0.4	1	0.5	0.4	0.6
forest	1	0.9	0.8	0.8	0.7	0.6	0.3
water	0.5	0.7	0.6	0.5	0.6	1	0.5
bare	0	0.4	1	0.4	0.4	0.4	0.7
built-up	0	1	0.4	0.4	0.6	0.2	0.8

), were determined. Subsequently, using ArcGIS 10.8 [51], each threat land use type was extracted. These data, along with Table 2 and Table 3, and the land use type map, were input into the HQ module to execute the model, generating a habitat quality map with HQ values ranging from 0 to 1. Values closer to 1 indicate higher habitat quality, while values closer to 0 represent lower quality [52,53]. For analytical purposes, the habitat quality map was categorized into five levels: very low [0, 0.2], low [0.2, 0.4], medium [0.4, 0.6], high [0.6, 0.8], and very high [0.8, 1]. The area proportions for each habitat quality level were then quantified. The technical workflow is illustrated in Figure 4.

3. Results

3.1. sUAS Survey in Gaoligong Hoolock Gibbon Habitat in Yingjiang

A sUAS survey was conducted to acquire imagery of the habitat of nine Gaoligong hoolock gibbon groups within Sudian and Zina townships, encompassing a total area of 735.86 hectares. Habitat patch sizes exhibited considerable variation, ranging from 43.68 hectares for the smallest patch (lmh1) to 132.19 hectares for the largest (lmh4). The average patch size was calculated to be 81.76 hectares (Figure 2).

Analysis of the sUAS imagery revealed significant variations in land use structure across the nine habitat patches (Figure 5). The xby patch exhibited the highest proportion of evergreen broadleaved forest, constituting 91.53% of its total area. The pw01 patch also presented a high forest cover, exceeding 90%. Conversely, the lmh6, lmh2, and lmh4 patches demonstrated comparatively lower proportions of evergreen broadleaved forest, representing a reduced fraction of their respective landscapes. Specifically, in the lmh6 patch, agricultural land comprised 12.15%, grassland comprised 19.76%, and evergreen broadleaf forest accounted for only 62.68% of the total area. The lmh2 patch showed a forest cover of merely 67.77%, while the lmh4 patch contained approximately 9% bare land (

Table 4. Percentage of each land use type at sUAS resolution.

	Forest	Grass	Farm	Bare	Pine Forest	Built-Up	Water
xby	91.53%	0.94%	5.12%	0.33%	2.08%	0.01%	0%
pw01	87.92%	6.25%	2.16%	0%	2.20%	0.86%	0.61%
xb03	84.87%	3.55%	5.88%	1.51%	1.09%	0%	0%
lmh5	84.51%	12.50%	0.57%	1.00%	1.42%	0%	0%
lmh1	79.22%	4.73%	4.69%	3.78%	5.00%	2.21%	0.36%
lmh3	77.21%	6.98%	5.37%	0.72%	6.06%	1.94%	1.71%
lmh4	70.70%	15.57%	3.99%	8.96%	0.77%	0%	0%
lmh2	67.77%	21.98%	2.22%	0.05%	5.43%	0.89%	1.67%
lmh6	62.68%	19.76%	12.15%	0.74%	3.88%	0.54%	0.25%

). Collectively, the data indicated a general trend of decreasing evergreen broadleaf forest cover coupled with an increasing proportion of agricultural land and other land uses among the surveyed patches (Figure 5).

3.2. Classification Accuracy Assessment

To evaluate the accuracy of the 10 m resolution satellite image classification, a manually interpreted land use/land cover map was utilized as the reference dataset (Figure 6). The accuracy assessment revealed an overall classification accuracy of 60.77% and a Kappa coefficient of 0.5023 for the 10 m resolution data (

Table 5. Evaluation results of classification accuracy of different land use/cover types for 10 m resolution data.

	Forest	Grass	Farm	Built-Up	Bare	Water	Producer’s Accuracy/%
Forest	215	0	0	0	21	5	100
Grass	0	70	26	8	4	0	93.33
Farm	0	4	21	9	8	11	17.36
Built-up	0	1	43	59	0	73	77.63
Bare	0	0	31	0	2	0	5.71
Water	0	0	0	0	0	11	11
User’s Accuracy/%	89.21	64.81	39.62	33.52	6.06	100
Overall accuracy		60.77%
Kappa coefficient		0.5023

). Among different land cover categories, the forest category exhibited the highest producer’s accuracy (100%) and user’s accuracy (89.31%). Grassland followed, with producer’s and user’s accuracies of 93.33% and 64.81%, respectively. However, it is noteworthy that these accuracies for farmland and bare land remained considerably lower.

3.3. Habitat Quality Assessment Based on sUAS Images

Habitat quality was evaluated across nine habitat patches, revealing an average habitat quality index exceeding 0.9 and an average degradation level of 0.68 (

Table 6. Average habitat quality and degradation levels of the two resolution datasets.

	Habitat Quality		Current Level of Degradation
	sUAS (0.05 m)	Sentinel-2 (10 m)	sUAS (0.05 m)	Sentinel-2 (10 m)
average	0.92	0.94	0.68	0.76

—sUAS). Nevertheless, spatial analysis revealed significant spatial heterogeneity in habitat quality. Patches proximal to the village (lmh4, lmh5, lmh6, xb03, and xby) demonstrated significantly reduced proportions of high-quality habitat area and diminished overall habitat quality compared to patches located further from the village. Specifically, the average proportion of high-quality habitat area in these village-proximal patches was notably low at only 12%, ranging from a maximum of 29.25% to a minimum of 0%. Notwithstanding these findings, evergreen broadleaved forest cover in these patches remained consistently above 60%, suggesting high forest cover, yet their ecological function appeared constrained by the observed decline in habitat quality (Figure 7—sUAS and A1a).

3.4. Significant Impact of Image Resolution on Habitat Quality and Degradation Level Assessment and Difference Analysis

Analyzing habitat quality derived from land cover data with differing spatial resolutions demonstrably revealed the significant impact of resolution on habitat quality assessment (Figure 7). Overall, the nine study patches exhibited high habitat quality. The average habitat quality, calculated using sUAS and Sentinel images, exceeded 0.9, averaging 0.92 and 0.94, respectively. However, regarding habitat degradation levels, the results from different image resolutions revealed notable discrepancies.

The average degradation level derived from sUAS images was 0.68, whereas it was 0.76 for Sentinel images, a significantly higher value. This discrepancy might stem from the differential sensitivity of image resolution to land cover characteristics; lower-resolution Sentinel images could potentially overestimate degradation levels in localized areas. Habitat quality and degradation levels exhibited significant variations across patches. Specifically, habitat quality ranged from 0.769 to 0.983 for sUAS images and from 0.746 to 0.999 for Sentinel images. Concurrently, degradation levels varied from 0.63 to 0.72 and from 0.7 to 0.81 for sUAS and Sentinel images, respectively (Table 6).

Considering the distribution across five habitat quality levels, irrespective of whether sUAS or Sentinel images were used, the four patches (lmh4, lmh5, lmh6, and xb03) proximal to the village exhibited smaller proportions of high-quality habitat compared to patches farther away. This observation indicates that habitat quality near villages is significantly affected by human activities and land use change. For the xby patch, also located near the village, data analysis from the two image resolutions revealed substantial discrepancies. In the sUAS imagery, the high-quality habitat area was zero, whereas in the Sentinel imagery, it accounted for 25.71%. The primary reason for this difference lies in the Sentinel imagery’s ineffective identification of smaller fragmented patches under forest canopy, leading to an overestimation of high-quality habitat area. In contrast, the higher resolution of sUAS imagery more accurately captures these finer habitat units, resulting in a more conservative habitat quality assessment. For patches lmh1 and lmh2, which are also farther from the village, contrasting trends were observed. Classification results and habitat quality maps (Figure A1) highlight limitations in the classification accuracy of Sentinel-2 images. Specifically, in patch lmh1, roads and smaller fragmented understory patches were not fully identified, leading to high-quality habitat overestimation. In patch lmh2, rivers and evergreen broadleaf forests were misclassified as roads and grasslands, resulting in habitat quality underestimation. Such classification inaccuracies directly compromise the reliability of habitat quality analysis results. Conversely, the consistent habitat quality assessments for the remaining patches across both image sets underscore the necessity of high-resolution imagery for reliable habitat quality assessment in heterogeneous landscapes (Figure 7).

Overall, the five patches proximal to the village (lmh4, lmh5, lmh6, xb03, and xby) exhibit a low proportion of high-quality habitat, which is likely associated with disturbances from human activities. Patch lmh1, more significantly impacted by road proximity, also displays a low proportion of high-quality habitat. Conversely, the three patches distant from the village (lmh2, lmh3, and pw01) show a high proportion of high-quality habitat, suggesting that distance from human activity centers may contribute to maintaining habitat quality integrity.

4. Discussion

4.1. sUAS Imaging Technique in Analysis of Gaoligong Hoolock Gibbon Habitat Patch Quality

This study leveraged high-resolution sUAS imagery to perform a detailed assessment of habitat quality across nine distinct patches crucial for the Gaoligong hoolock gibbon in Yingjiang County. Moving beyond the limitations of coarse satellite data, this approach allowed for a detailed analysis of habitat characteristics. Our analysis revealed considerable variation in forest canopy cover among patches, a key component of gibbon habitat, ranging from 62.68% in patch lmh6 to 91.53% in xby (Table 4). Importantly, the sUAS data highlighted a nuanced relationship between forest cover and forest type. Patches with greater forest cover, such as xby and pw01, were characterized by a higher prevalence of mature evergreen broadleaf forests—the preferred habitat for gibbons. Conversely, patches with lower forest cover, including lmh6 and lmh2, exhibited a different ecological profile. These areas showed clear impacts from human activities, such as understory plantations, clearings, and roads, indicating a landscape increasingly shaped by human influence. Furthermore, the fine spatial resolution of sUAS imagery facilitated the detection of subtle, yet ecologically important, variations in land use across patches, revealing a mix of human-made and natural elements. This detailed understanding of land use heterogeneity is crucial, as it directly affects habitat suitability for the sensitive Gaoligong hoolock gibbon and significantly influences the survival of its fragmented populations [54].

Building on the forest cover analysis, we used the spatial precision of sUAS imagery to investigate the distribution of habitat quality levels. A notable pattern emerged: patches near village settlements (lmh4, lmh5, lmh6, xb03, and xby) showed a clear reduction in high-quality habitat compared to patches further from human disturbance. Specifically, the proportion of high-quality habitat in village-adjacent patches was significantly lower [55]. This spatial trend strongly indicates the pervasive influence of human activity and land use changes on habitat integrity in village peripheries. The ecological landscape around villages is marked by greater land use diversity and increased human disturbance. This complex interaction of factors results in a general decline in habitat quality in these areas [56]. Further investigation identified agricultural expansion, particularly the spread of tsaoko plantations, and infrastructure development, especially road construction, as major drivers of habitat fragmentation. These human-induced landscape changes not only directly eliminate critical habitat but also gradually reduce the quality of remaining patches by increasing edge effects and promoting ecological isolation [57]. Habitat patches close to villages are therefore especially vulnerable to intensified human pressures, leading to faster fragmentation and significant habitat quality reduction.

These findings align with existing scholarly research on Gaoligong hoolock gibbon ecology and the broader field of endangered primate conservation. In a key study, Bai et al. [58] rigorously demonstrated the gibbon’s strong preference for habitat patches with dense evergreen broadleaf forest cover and minimal human disturbance, clearly establishing evergreen broadleaf forest as the most important factor in gibbon habitat selection. However, the relentless pressures of human population growth and economic needs are driving a concerning trend: the ongoing degradation and outright loss of valuable evergreen broadleaf forests in the Gaoligong Mountains. Fan et al. [59] meticulously documented land use changes in Gaoligong hoolock gibbon habitat between 2000 and 2014, revealing a massive forest area loss of 861,700 hectares—with only 3% under protected status. These severe landscape alterations have directly caused habitat loss and fragmentation, leading to a catastrophic decline in the Gaoligong hoolock gibbon population size and range. Compounding this already precarious situation, our study revealed that all gibbon habitat patches in Yingjiang County are now confined to village peripheries, essentially existing as habitat islands within a human-dominated landscape. Consequently, these remaining patches are constantly exposed to varying levels of fragmentation and human disturbance. In sharp contrast to the relative safety found within protected areas, these village-adjacent patches face considerably greater disturbance pressures. This pervasive fragmentation pattern intensifies the survival pressures on the already-vulnerable gibbon populations. Drawing on a substantial body of research, habitat patch fragmentation has been shown to reduce gibbon home range size and decrease foraging efficiency [18,60], while also increasing the risks of inbreeding and harmful genetic drift within isolated populations [61]. Given the critically endangered status and alarmingly low population numbers of gibbons in Yingjiang County, the continued habitat degradation threatens to cause further population declines, potentially resulting in local extinction events.

4.2. Effects of Spatial Resolution on Classification Accuracy and Habitat Quality Analysis

The results of this study highlight that spatial resolution is not just a technical detail but a crucial factor influencing both classification results and subsequent ecological analyses. This is evident when comparing high-resolution sUAS (0.05 m) and medium-resolution Sentinel-2 (10 m) imagery. As expected, sUAS imagery showed significantly higher classification accuracy than Sentinel-2 [37]. This is primarily because sUASs can resolve fine-scale landscape features that are indistinguishable at 10 m resolution. Sentinel-2’s inherent limitations in capturing localized surface variations lead to inaccuracies and potential biases in both land cover classification and habitat quality assessment, particularly in complex ecological environments. Indeed, in diverse landscapes, the choice of spatial resolution fundamentally shapes how habitat quality is perceived and interpreted, potentially leading to different ecological conclusions [62]. While differences in timing between datasets might have some influence, this discussion mainly focuses on the fundamental impact of spatial resolution as a key driver of variability in ecological assessments.

Looking beyond overall classification accuracy, it is clear that image resolution acts as a filter, selectively revealing or obscuring ecologically relevant information. While 10 m spatial resolution data, such as those from Sentinel-2, may be suitable for mapping broad forest cover in large-scale monitoring efforts [25], their effectiveness is significantly reduced when focusing on more detailed ecological distinctions. Specifically, precisely identifying non-forest patches within forests, ecologically important areas like critical microhabitats [63], or nuanced assessments of forest quality within seemingly uniform forested areas [64], requires the fine detail provided by high-spatial-resolution data [33]. Essentially, important ecological information is often hidden within details that are lost at coarser resolutions. This is further demonstrated in land cover classification within fragmented landscapes, where our analysis showed a notable decrease in classification accuracy for land cover types like cropland and built-up areas as spatial resolution decreases. This is mainly due to the increased presence of mixed pixels in lower-resolution imagery. These mixed pixels blur boundaries and obscure fine-scale spatial details, reducing the ability of the classifier to effectively differentiate between land cover types [25]. In contrast, high-resolution imagery, like sUAS data, provides a more comprehensive range of textural and spectral information. This additional information significantly improves classification accuracy, especially in ecologically complex and fragmented landscapes [39], where subtle differences in spectral signatures and textural patterns are essential for accurate feature identification. Furthermore, the combined use of sUAS and Sentinel-2 imagery provides a strong approach for ecological analysis, creating a multi-scale dataset that combines detailed textural information from sUAS with the broader spectral coverage of Sentinel-2. When strategically integrated with field surveys, this data framework has the potential to greatly enhance the performance of Sentinel-2 imagery, advancing classification accuracy and expanding the scope of landscape ecological analysis [65].

Despite indicating overall ecological suitability, the habitat quality assessment, regardless of image resolution, consistently suggested high ecological suitability in the study area, indicating a generally healthy ecosystem. However, the seemingly small differences in average habitat quality (0.92 for sUAS images and 0.94 for Sentinel images) conceal the underlying sensitivity of habitat quality assessment to spatial resolution. This minor numerical difference masks a crucial finding: degradation level estimates derived from Sentinel-2 imagery were consistently and significantly higher (0.76) compared to sUAS imagery (0.68). This consistent overestimation of degradation by Sentinel-2 is not merely a statistical quirk; it reflects a fundamental limitation in its ability to accurately represent fine-scale ecological features [66]. The lower spatial resolution of Sentinel-2 inherently limits its ability to resolve subtle yet ecologically significant surface characteristics and small habitat patches. Conversely, sUAS imagery, with its superior spatial precision, excels at accurately delineating small, distinct patches within larger landscapes. This ability is vital for habitat quality assessment and, importantly, for targeted conservation planning. By allowing the identification of key microhabitat areas and ecological refuges [26], high-resolution data facilitate a more precise and ecologically meaningful assessment of habitat quality and degradation. In contrast, the generalization inherent in Sentinel-2’s coarser resolution may lead to the misclassification of fragmented forest gaps or localized heterogeneous areas as uniformly degraded zones, resulting in a consistent overestimation of overall habitat degradation. This highlights a critical point: assessments based solely on medium-resolution data may inadvertently inflate degradation metrics, potentially leading to biased ecological interpretations and misdirected conservation efforts [25].

To further illustrate the impact of resolution, patch-level habitat quality assessments reveal the resolution effect with greater clarity and ecological relevance. The contrasting analyses of the lmh1 and lmh2 patches provide clear examples of this effect. In the lmh1 patch, Sentinel-2 imagery clearly overestimated the area of high-quality habitat. This overestimation is a direct result of its inability to resolve fine-scale linear features, like roads, and small-scale forest fragmentation patterns under the canopy. The smoothing effect of lower resolution effectively masked these ecologically important features, misrepresenting a fragmented landscape as more uniform and, consequently, of higher habitat quality than it actually was. Conversely, in the lmh2 patch, Sentinel-2 underestimated habitat quality. This underestimation stemmed from classification errors, specifically misclassifying rivers and ecologically valuable evergreen broadleaved forests as roads and grasslands. These misclassifications are, again, indicative of the spectral mixing and information loss inherent in coarser-resolution data. These inaccuracies are not merely technical errors but also directly lead to flawed assessments of habitat suitability and ecological integrity. Critically, these findings underscore the inherent limitations of using Sentinel-2 imagery in ecosystems characterized by high spatial diversity and fine-grained patch dynamics [66]. Interestingly, our observations also showed that in areas with low ecological diversity, Sentinel-2 imagery produced results that were generally consistent with sUAS imagery. This agreement suggests that Sentinel-2 can still provide reasonably reliable habitat quality assessments in landscapes with uniform land cover and minimal patch-level complexity. Therefore, ecological context is vital: the advantages of high-resolution imagery in habitat quality assessment are most evident and ecologically meaningful in complex and highly diverse ecological environments. Conversely, in areas of low diversity, lower-resolution imagery can achieve comparable assessment outcomes with significantly reduced costs and data processing. Consequently, carefully selecting image resolution, based on the specific ecological features of the study area, is a crucial strategic decision. This strategic approach is not just about technical optimization but also about achieving efficient resource use while ensuring the ecological validity and accuracy of habitat quality assessments [25]. The main challenge is to find the best balance between data detail, analytical rigor, and practical resource limitations, always guided by the ecological questions being asked and the specific landscape being studied.

4.3. Advantages of sUAS Imagery in Animal Habitat Analysis

Image resolution is not simply a technical specification; it fundamentally determines our ability to accurately perceive ecological realities in diverse landscapes. The accuracy of habitat quality assessments, particularly in complex ecosystems, is directly tied to image resolution [67]. Insufficient resolution not only blurs the boundaries between habitat types, leading to classification errors, but, more importantly, it systematically distorts our understanding of habitat fragmentation and degradation. This distortion has significant consequences for conservation decision-making, potentially leading to misinformed resource allocation and management strategies. In contrast, high-resolution imagery is a powerful tool, enhancing our ability to understand land use complexities, minimizing information loss, and thereby improving the scientific rigor and reliability of habitat assessments [68].

Traditional field surveys and conventional satellite remote sensing techniques have inherent limitations, primarily due to constraints in time, cost, and, crucially, spatial and temporal resolution. These limitations have historically hindered comprehensive studies of the intricate relationships between wildlife and their habitats [26]. Consequently, these methods often fail to capture the fine-grained spatial diversity that characterizes many ecosystems and influences species distributions and ecological processes. The emergence of sUAS technology offers a major shift in addressing these methodological limitations. sUASs provide an unprecedented ability to acquire habitat information at both high spatial resolution (centimeter-level precision) and flexible temporal scales, effectively bridging the resolution gap and enabling better ecological understanding and targeted conservation action [33]. Furthermore, given the alarming global decline of primate populations, where habitat fragmentation is the second most significant threat after outright habitat loss [1], the ability of sUASs to analyze habitat complexity is not just advantageous but essential. For critically endangered species like the Gaoligong hoolock gibbon, whose survival depends on navigating increasingly fragmented landscapes and maintaining crucial habitat connectivity [18], the fine-scale insights provided by sUAS imagery are vital for developing robust conservation strategies.

Unlike conventional satellite remote sensing, sUAS imagery’s centimeter-level spatial resolution provides a range of transformative advantages, representing a significant leap forward in understanding habitat structure and function. First, sUAS reduces the pervasive loss of landscape information inherent in coarser resolution data [32], thus preserving the ecological integrity of habitat representations. This is especially crucial for accurately delineating patch edges and inter-patch connecting areas [69], enabling the precise identification and characterization of even small fragmented patches within larger habitat areas [25]. Such detailed information is not just descriptive; it is ecologically important, reflecting how many species experience and use fragmented landscapes. Second, the enhanced resolution of sUASs significantly improves our ability to detect and analyze small-scale, yet ecologically vital, microhabitat elements. This helps preserve important spatial details, reducing boundary blurring and identification errors common in lower-resolution approaches [32]. For instance, subtle yet ecologically vital microhabitat features (treefall gaps and understory vegetation patches) are often undetectable in coarser imagery but are crucial for species survival and various ecological processes. Third, sUAS imagery significantly enhances the characterization of patch diversity, revealing a more nuanced range of spatial variation within habitat patches [31]. The ability to discern intra-patch complexity is crucial for understanding resource availability, habitat quality differences, and how species interact with their environment and use resources. It allows us to move beyond simple binary classifications of habitat quality and embrace the inherent spatial complexity of ecological systems and processes.

In our study of Gaoligong hoolock gibbon habitat quality, the practical advantages of sUAS imagery became strikingly clear. sUAS data significantly facilitated more accurate assessments of patch quality. Specifically, it precisely differentiated between ecologically distinct forest types, such as evergreen broadleaved forests and coniferous forests, and, crucially, it resolved smaller fragmented patches within the forest matrix [39]. In contrast, the analysis using Sentinel imagery revealed significant limitations. Specifically, in the lmh1 patch, Sentinel imagery failed to resolve roads and smaller fragmented patches beneath the forest canopy, resulting in a consequential overestimation of high-quality habitat area. Conversely, in the lmh2 patch, spectral confusion in Sentinel data led to the misclassification of rivers and evergreen broadleaved forests as roads and grasslands, paradoxically causing an underestimation of habitat quality. This pattern highlights a critical flaw in low-resolution assessments: they can simultaneously overestimate habitat integrity in some areas while underestimating it in others, resulting in a fundamentally flawed and inconsistent representation of habitat quality [37]. The higher overall degradation level inferred from Sentinel data, compared with sUAS-derived assessments, suggests a potential systematic overestimation of fragmentation by lower-resolution data. This overestimation poses tangible risks for conservation planning, potentially skewing important decisions about corridor design and the strategic placement of conservation efforts. Conversely, the ability of sUAS imagery to assess habitat quality at a finer scale directly contributes to reduced uncertainty in conservation decision-making, providing a more robust and reliable basis for action [70]. Furthermore, the ability of sUAS data to differentiate pine forests from evergreen broadleaved forests and to resolve small fragmented patches within complex vegetation structures, as shown in our study, highlights the inherent advantages of high-resolution data, especially in ecologically complex and diverse habitats. These advantages are not just technical refinements; they are particularly relevant when analyzing the habitat of the Gaoligong hoolock gibbon, a species highly sensitive to habitat nuances. Building on recent research identifying critical dispersal corridors for this species in northern Yingjiang [71], our findings further emphasize the crucial role of sUAS data in accurately characterizing habitat features at the patch level. This accuracy is essential for informing conservation actions aimed at maintaining and enhancing habitat connectivity, including strategically vital initiatives like corridor construction and targeted habitat restoration.

Nevertheless, it is important to acknowledge that sUAS remote sensing technology, while transformative, is not without limitations in species habitat evaluation. Inherent limitations on flight altitude and operational range restrict the acquisition of synoptic, large-scale landscape pattern information in a single deployment. Therefore, a practical and effective strategy requires the combined use of sUASs with complementary technologies, like satellite remote sensing, to capitalize on the strengths of each approach and mitigate individual limitations [65]. Moreover, the practical use of sUASs is inherently influenced by weather conditions. Montane forest environments, in particular, present challenges due to frequent cloud cover, fog, and strong winds, all of which can negatively affect data quality. Strategic flight planning, including carefully selecting optimal flight times and take-off locations, becomes vital to mitigate these environmental constraints [72]. The computational demands associated with processing and analyzing large high-resolution imagery datasets also pose significant logistical considerations. Effective data processing pipelines and analytical models, along with access to high-performance computing infrastructure, are essential to fully use the potential of sUAS-derived data. Furthermore, while sUAS data are excellent at providing high-precision vegetation structural information, validation through strong field survey data remains crucial [73].

Future research should adopt integrated approaches that combine field-based habitat assessments, incorporating gibbon population density, food resource availability, and other key ecological metrics, with analytical findings derived from sUAS imagery. Given the early stage of widespread sUAS application in ecological research, there is an urgent need to establish long-term, multi-temporal datasets to facilitate dynamic habitat monitoring and change analysis over ecologically relevant timescales. This is essential for understanding habitat trends, evaluating conservation effectiveness, and informing adaptive management strategies. Addressing the current lack of long-term data requires dedicated efforts to enhance the acquisition, archiving, and collaborative sharing of multi-temporal sUAS datasets within the scientific community. Moving forward, integrating multi-platform, multi-sensor remote sensing data, alongside strengthened interdisciplinary collaboration with ecology and conservation biology, will be crucial for developing a comprehensive, three-dimensional system for wildlife habitat monitoring. Such advancements will significantly enhance the scientific foundation for biodiversity conservation, particularly for endangered primates like the Gaoligong hoolock gibbon. By combining sUAS data with complementary remote sensing technologies and ecological field studies, future research can foster more holistic and effective conservation strategies.

5. Conclusions

This innovative study has demonstrated the effectiveness of high-resolution sUAS imagery for detailed habitat quality assessment across nine patches used by the Gaoligong hoolock gibbon. Notably, this is the first time that fine-scale sUAS technology has been used for the habitat evaluation of endangered primates. By identifying critical characteristics within fragmented gibbon habitat patches, this research provides crucial baseline data for developing targeted conservation strategies.

Our findings clearly show the accuracy of sUAS imagery in characterizing gibbon habitat quality, particularly its effectiveness in delineating fragmented landscape features such as agriculture and built-up areas adjacent to forests. Habitat analysis further confirmed sUAS imagery’s superior ability to detect fragmentation within habitats, thereby minimizing mixed pixels and generating more reliable data for informed conservation decisions in fragmented landscapes. Consequently, this research highlights the significant advantages of sUAS technology for fine-scale habitat mapping, quality assessment, and corridor planning for endangered species, due to its high spatiotemporal resolution, mobility, and cost-efficiency.