Exploring the Roles of Ancient Trees in Disturbance and Recovery Processes Using Monthly Landsat Time Series Analysis

Highlights

What are the main findings?

• Natural ancient tree patches in natural forests showed more stable NDVI change signatures than nearby neighborhoods.
• Planted ancient tree patches in planted forests showed stronger recovery-related NDVI change signatures than nearby neighborhoods. A total of 86% of changes occurred before 2010, and 60% were short-lived (≤5 years).

What are the implications of the main findings?

• Ancient trees’ locations provide spatial cues for targeted monitoring; patterns likely reflect habitat context and stewardship as well as tree presence.

Abstract

Quantifying forest patch dynamics is essential for understanding how forest patch characteristics vary in relation to ancient tree locations. This study developed a satellite-based framework to analyze the differences among forest patches associated with natural and planted ancient trees across the Sichuan–Chongqing region, China. Using monthly LandTrendr on Google Earth Engine, we analyzed long-term (1990–2024) and high-frequency observations of forest dynamics at a 180 m × 180 m (6 × 6 pixels) spatial scale. Disturbance and recovery events were characterized by their magnitude, rate, timing, and duration. Patches were classified into six categories based on ancient tree type and proximity and further subdivided by land use type. The results show that in natural forests, patches with natural ancient trees are associated with more stable change signatures, whereas in planted forests, patches containing planted ancient trees are associated with stronger recovery-related change patterns. Over 60% of detected changes were short-lived (≤5 years), indicating that most disturbances and recovery processes were transient rather than persistent. These findings show that the presence and spatial context of ancient trees are associated with differences in patch change patterns. The proposed workflow provides a scalable approach for integrating multi-temporal remote sensing into large-scale monitoring and management of ancient trees and their associated forest patches.

1. Introduction

Ancient trees are defined as individual trees older than 100 years occurring in human-dominated landscapes, following the national technical guideline ‘Technical Guidelines for Document Establishment of General Survey of National Old-Famous Trees’. Ancient trees are recognized as keystone structures that reinstate the functionality of perturbed forest patches and maintain their ecological stability [1,2]. They support forest patch recovery by boosting biodiversity and biomass [3,4], facilitating nutrient cycling [5], maintaining habitat connectivity [6], and promoting long-term carbon storage [7,8]. In addition to their ecological functions, ancient trees are closely associated with utilitarian practices and cultural-spiritual value [9], which shape human protection behaviors and influence surrounding landscape patterns and processes. Ancient trees supply multiple material resources, such as food provision, medicinal resources, and material supply, which promote their retention and management in human-dominated landscapes [10,11]. They also hold strong cultural and spiritual significance. In China, such values are commonly embedded in culturally managed landscapes, such as temples [12] and fengshui forests [13], where long-term human stewardship has allowed them to persist for centuries [14]. These protection mechanisms, driven by practical uses and cultural-spiritual values, enhance the longevity of ancient trees in human-dominated landscapes. They also create localized ecological refugia that support key ecological processes in surrounding landscapes. Recent studies have shown that ancient trees are found not only in natural forests but also widely distributed across various human-modified landscapes [15,16]. A nationwide inventory in China recorded nearly 1.8 million century-old trees scattered throughout human-dominated landscapes [11]. The origins of these ancient trees differ substantially; some are relics of long-term natural forest succession and regeneration [17], whereas many others result from deliberate planting or cultivation by humans driven by cultural, ecological, or economic motivations [15,18]. These distinct origins imply that ancient trees are embedded in different landscape contexts and management histories, making it necessary to distinguish between natural and planted ancient trees when interpreting observed change patterns.

Conventional field-based or plot-scale approaches often fail to capture and quantify the broader spatial difference in ancient trees on surrounding forest patches, limiting the understanding of their ecological functions at the landscape level. Therefore, using a remote sensing-driven change detection approach [19] to quantify the association of ancient trees on patch stability, disturbance buffering, and recovery across large spatial scales offers valuable insights into long-term vegetation dynamics [20] and supports more informed landscape-scale conservation planning. Satellite remote sensing has become indispensable for ecological monitoring, providing extensive spatial coverage and continuous temporal records that enable comprehensive analysis of both current status and historical evolution of forest patches [21,22]. Spanning almost five decades, the Landsat data archive provides abundant earth observation resources for ecological and environmental studies [23].

Google Earth Engine (GEE) is a cloud-based geospatial processing platform, offering comprehensive user-friendly APIs [24]. Users can analyze freely available satellite imagery, generate statistics and maps, and visualize their findings through parallel computing [25]. The LandTrendr (LT) algorithm uses trajectory and spectral-temporal segmentation methods to detect annual land disturbance and recovery [19,20]. LT-GEE is an implementation of the LT algorithm on the GEE platform, which enables the detection of land cover changes [26,27]. By combining trajectory analysis and spectral-temporal segmentation, LT-GEE enables the detection of annual land disturbances and recovery, offering valuable insights into land dynamics [20]. The algorithm considers factors like magnitude (the total amount of spectral change), rate (slope of change), year of disturbance (start time), and duration (time span of change) [20,27]. However, the LT algorithm is limited in accurately capturing intra-annual changes [20,28]. To effectively capture within-year patch dynamics and accurately describe the complex changes, the monthly LT is essential for quantifying patch dynamics.

Ancient trees are increasingly documented across diverse land use contexts, yet there remains a limited understanding of whether and how natural and planted ancient trees differ in their associations with surrounding landscape change patterns. It is also unclear how such differences vary across land use types and with spatial proximity to ancient trees. To address these gaps, this study first distinguishes ancient trees by origin (natural versus planted), then defines patch units based on spatial proximity to ancient trees, including ancient tree patches and adjacent patches, and finally categorizes patches by land use types into natural forest, planted forest, and cropland. Patch-scale change characteristics were quantified using temporal segmentation with the LandTrendr algorithm applied to monthly time series data. This study comparatively examines change patterns associated with natural and planted ancient trees across different land use contexts and spatial distances. The analysis provides a basis for interpreting how ancient tree origin, landscape setting, and spatial proximity relate to observed patch dynamics.

2. Materials and Method

2.1. Study Area

The Sichuan–Chongqing region (97°21′E–110°11′E, 26°03′N–32°13′N), located in southwestern China, is a key economic and cultural hub of the upper Yangtze basin, including Sichuan Province and Chongqing Municipality. According to the Seventh National Population Census (https://www.stats.gov.cn/ (Accessed on 27 May 2025)), in 2020, the permanent population of Sichuan Province exceeded 80 million, with its capital Chengdu having a population of over 20 million; Chongqing Municipality had a permanent population of over 30 million. According to the National Bureau of Statistics of China (NBS) (https://www.stats.gov.cn/sj/ndsj/2024/indexch.htm (Accessed on 27 May 2025)), in 2022, the urbanization rate of Sichuan Province was 59.49%, while that of Chongqing Municipality was 71.67%. 94% of Sichuan’s total area consists of mountains, whereas in Chongqing, mountains account for 76% and hills for 22% of the land (Figure 1a). This region has a subtropical monsoon humid climate, with mild winters and hot summers. Annual rainfall is between 1000 and 1400 mm, concentrated primarily between May and September. The complex topography-climate interactions in this region lead to persistent cloud cover and frequent fog formation. Although the species richness in the Sichuan–Chongqing region is high, only a few of these areas are included in protected areas (Figure 1b) [29].

2.2. Data Source

This study used four datasets. Surface reflectance data from Landsat 5, 7, and 8 (Collection 2, Tier 1, Level 2) obtained from GEE [24] were used to detect the changes within human-modified areas of the Sichuan–Chongqing region. Ancient tree data shared by [11] were used to create ancient tree, adjacent, and second-order adjacent patches. The 30 m Global Natural and Planted Forest Map dataset (https://zenodo.org/records/10701417 (Accessed on 24 December 2024)) [30] and 30 m Annual Land Cover datasets (https://zenodo.org/records/12779975 (Accessed on 24 December 2024)) [31] provided the basis for pixel-level land use classification.

2.2.1. Landsat Time Series

Using Landsat time series imagery, we analyzed forest and cropland patch dynamics in anthropogenically modified areas of the Sichuan–Chongqing region during 1990–2024. Landsat satellite image data, with over 40 years of historical archives and a 30 m spatial resolution, makes it suitable for capturing long-term dynamic changes in large-scale forest patches [32]. The GEE platform provides Collection 2 Level-2 datasets acquired by TM, ETM+, and OLI sensors, all subjected to thorough noise removal procedures. The noise reduction procedure comprised three steps: elimination of low-quality and cloud-contaminated pixels, outlier removal through 1st–99th percentile filtering, and Savitzky–Golay smoothing for phenological noise and residual cloud effects [33]. Monthly time series composites were constructed using all available Landsat image data (including Landsat 5 TM, 7 ETM+, and 8 OLI) across the Sichuan–Chongqing region.

2.2.2. Ancient Tree Data

Ancient trees in this study were defined according to the authoritative document, Technical Guidelines for Documenting the General Survey of National Old-Famous Trees. Specifically, ancient trees are those aged over 100 years and located in human-inhabited areas, excluding individuals found within protected areas and forestry plantations. The information on ancient trees was first collected with help from local villagers and subsequently verified through on-site investigations by forestry administration personnel. The classification of planted ancient trees was based on species-level evidence of historical cultivation and traditional human use. In the published dataset, an ancient tree was classified as a planted ancient tree if its species was documented as having a clear cultivation history or traditional human use. Information on traditional uses was mainly obtained from the following sources: (1) Flora of China, which documents the traditional uses of most plant species in China. (2) Published books, including Economic Flora of China, Chinese Tree Culture, and Pharmacopoeia of the People’s Republic of China. (3) Published literature, including journal articles, review papers, and academic theses on Chinese ethnobotany, was systematically searched. Information on cultivation history and geographical spread was derived mainly from Chinese Cultivated Plants and Flora of China. Ancient trees that did not meet these criteria for historical cultivation or documented traditional use were classified as natural ancient trees. The resulting data, including geographical coordinates and ancient tree type information, were published in Nature Plants and underwent rigorous validation to ensure accuracy [11]. The database documented a total of 101,947 ancient trees, comprising 16,183 natural ancient trees and 85,764 planted ancient trees. High-density clusters were predominantly distributed in central and eastern Sichuan–Chongqing, with gradually decreasing densities toward the western, southern, and northern regions.

2.2.3. Land Use Types

Changes in land use and land cover markedly affect the ecological functions of ancient trees [34]. Therefore, the land use types of ancient tree, adjacent, and second-order adjacent patches were classified using the 30 m annual land cover datasets (https://zenodo.org/records/12779975 (Accessed on 24 December 2024)) developed by [31]. This dataset was produced from all available Landsat data on GEE, achieving an overall accuracy of 80%. For analytical purposes, land use was categorized into forest and cropland, as these two classes represent the dominant land cover types in which ancient trees are distributed.

2.2.4. Natural and Planted Forests Map

Significant differences exist in the ecosystem services between natural and planted forests [35]. Natural forests, characterized by their structural complexity and high species diversity, create favorable microhabitats that support the stability and regeneration of ancient trees. In contrast, planted forests provide limited ecological facilitation, exerting weaker positive effects on adjacent ancient tree populations [36,37]. For example, natural forests enhance pollinator abundance and seed dispersal networks that benefit ancient tree recruitment, whereas planted forests minimally contribute to such critical ecological processes [38]. Based on land use classification, forests were further categorized into natural and planted forests based on the 30 m Global Natural and Planted Forests Map dataset (https://zenodo.org/records/10701417 (Accessed on 24 December 2024)) provided by [30]. Derived from dense Landsat time-series data processed through a random forest classifier, this dataset attained 85% overall accuracy upon validation with reference data.

2.3. Method

We proposed a framework to analyze the impact of ancient trees on forest patch changes based on the monthly LandTrendr algorithm (Figure 2). Firstly, ancient trees were classified into natural and planted types. Three patch types were then constructed: (1) ancient tree patches, (2) adjacent patches, and (3) second-order adjacent patches (Figure 2a). Secondly, three land use categories (natural forest, planted forest, and cropland) were defined at the pixel level to ensure comparability in the subsequent analyses. (Figure 2b). Thirdly, a monthly time series was developed based on the LandTrendr algorithm to enhance the frequency of patch dynamics monitoring in Sichuan and Chongqing from 1990 to 2024 (Figure 2c). Finally, changes in magnitude, rate, timing, and duration were compared across the three patch types: ancient tree, adjacent, and second-order adjacent (Figure 2d).

2.3.1. Extracting the Ancient Tree, Adjacent, and Second-Order Adjacent Patches

In this study, an ancient tree patch was defined as a patch containing an ancient tree. An adjacent patch was defined as a patch sharing a boundary with the focal ancient tree patch. A second-order adjacent patch was defined as a patch separated from the focal ancient tree patch by one intervening patch. Based on the previous definition of ancient trees, patches were classified according to the type of their focal ancient tree. Accordingly, we distinguished six patch categories: natural/planted ancient tree patches, patches adjacent to natural/planted ancient tree patches, and second-order adjacent patches to natural/planted ancient tree patches.

To establish contrastive patches, four adjacent patches were generated per ancient tree patch by 180 m centroid offsets along cardinal directions (east, south, west, north), with corresponding 360 m offsets producing second-order adjacent patches (Figure 2a). The 180 m distance was chosen to match the 30 m spatial resolution of Landsat imagery, ensuring that the scale is an integer multiple of the pixel size. This alignment ensures coherence between the spatial resolution of the imagery and the analysis scale, minimizing potential spatial mismatches. Initially, we used the 90 m scale, but no significant differences were observed between ancient tree patches and adjacent patches (Figure 3). We then extended the analysis to 150 m and 180 m scales and found that the results for both scales were similar. To explore a broader spatial extent, we chose the 180 m scale for further analysis. A total of 16,183 natural and 85,764 planted ancient tree patches were generated. Based on the four-cardinal-direction spatial relationship, these comprised 64,732 adjacent/second-order adjacent patches for natural ancient trees and 343,056 for planted ancient trees. This processing step was implemented in ArcGIS Pro 3.1.5.

2.3.2. Classifying Patch Land Use Types

At the pixel level, patch land use types were initially categorized as forest or cropland based on a land use classification product [31]. Patches whose land use type was classified as forest were further subdivided into natural forest and planted forest based on a forest classification dataset [30] (Figure 2b). This two-step approach reduces inconsistencies arising from mixed land use types within individual patches. Cropland was included as a patch land use type because many ancient trees are distributed within agricultural areas. This allows assessment of whether the association of ancient trees on patch dynamics differs between forested and agricultural contexts, which is critical for understanding ancient-tree-related patterns in human-dominated landscapes. All data processing and analysis for this component were performed using ArcGIS Pro.

2.3.3. Quantifying Patch Change Dynamics Based on the Monthly LandTrendr Algorithm

Few studies use remote sensing technology to quantify the relationship between ancient trees and surrounding patch dynamics in human-modified landscapes. In this study, we addressed this gap by analyzing patch dynamics around ancient trees using long-term Landsat observations. The LandTrendr (LT) algorithm was applied to Landsat image stacks (L5-TM, L7-ETM+, L8-OLI) for detecting patch dynamics across the Sichuan–Chongqing region using the Google Earth Engine platform (Figure 2c). Monthly NDVI time series from 1990 to 2024 were constructed to enhance the monitoring frequency of patch dynamics. To ensure that the detected changes were not confounded by seasonal fluctuations, we applied a temporal encoding strategy that compensates for phenological noise, as proposed by [28]. The deseasonalized NDVI residuals were used as inputs to the LT algorithm, which allows us to capture both abrupt and gradual changes in vegetation, improving the detection of non-seasonal dynamics and increasing sensitivity to disturbance and recovery events. By isolating non-seasonal changes, the method improves the accuracy of detecting patch dynamics related to ancient trees.

The LT algorithm is designed for annual time series. To enable the use of monthly observations, we adopted the temporal encoding strategy proposed by [28]. For each pixel, cloud-free monthly NDVI composites were first generated. Each monthly observation was then assigned a unique pseudo-year index, allowing the LT algorithm to treat the monthly series as an evenly spaced annual sequence. The new time coordinate $t$ for the month in the year was defined as follows:

$$t=Y_0+\left(y-Y_0\right)\times 12+m$$

(1)

where $Y_0$ is the starting calendar year of the time series (1990), $y$ is the calendar year of observation, and $m$ is the calendar month (1–12). For example, January 1990 was mapped to 1991, February 1990 to 1992, and so forth. This encoding preserves the original monthly temporal resolution while allowing LandTrendr to operate on the data. The encoded monthly NDVI series was then segmented using LT, and segments with the maximum change magnitude were used to characterize patch dynamics. We further extended the approach of [28] by incorporating phenological compensation to reduce seasonal noise (code available at https://code.earthengine.google.com/f2cfc1a65c25ea57547ace0b220f6512 (Accessed on August 2024)).

Normalized Difference Vegetation Index (NDVI)was used to characterize vegetation dynamics, calculated as follows:

$$\mathrm{N}\mathrm{D}\mathrm{V}\mathrm{I}=(\mathrm{N}\mathrm{I}\mathrm{R}-\mathrm{R}\mathrm{E}\mathrm{D})/(\mathrm{N}\mathrm{I}\mathrm{R}+\mathrm{R}\mathrm{E}\mathrm{D})$$

(2)

where $\mathrm{N}\mathrm{I}\mathrm{R}$ and $\mathrm{R}\mathrm{E}\mathrm{D}$ represent surface reflectance in near-infrared and red bands, respectively.

To reduce the influence of seasonal vegetation cycles, a deseasonalized procedure was implemented. A multi-year monthly NDVI was constructed using all available Landsat observations from 1990 to 2024. For each pixel, the seasonal baseline for a given calendar month was calculated as follows:

$${Baseline}_{month} = 1/N_{years}\sum _{y=1990}^{2024}{NDVI}_{month,y}$$

(3)

where ${NDVI}_{month,y}$ represents the NDVI value for a given pixel in calendar month (January-December) and year, and $N_{years}=34$ is the length of the study period. This procedure generated 12 monthly baseline images representing the mean phenological cycle. These baselines were used to remove seasonal signal from the monthly NDVI time series.

The seasonal component was removed from each monthly NDVI observation as follows:

$${Residual}_{month,y} = {NDVI}_{obs}-{Baseline}_{month}$$

(4)

where ${NDVI}_{obs}$ denotes the observed NDVI value for a given pixel in calendar month of year, and ${Residual}_{month,y}$ is the deseasonalised NDVI residual series.

The residual series captures deviations from the mean seasonal signal. Positive residuals indicate increased vegetation greenness relative to the long-term monthly mean, whereas negative residuals indicate reduced greenness. Residual values close to zero indicate that vegetation conditions are close to the long-term mean for that calendar month, with no substantial departure from typical seasonal conditions. Removing phenological fluctuations reduces the risk that regular seasonal cycles are misidentified as disturbance or recovery events during subsequent temporal segmentation and improves sensitivity to non-seasonal vegetation changes while preserving the monthly temporal resolution. The deseasonalized series was subsequently used as input to the LandTrendr algorithm for temporal segmentation of patch dynamics.

We used deseasonalized monthly NDVI time series (1990–2024) as a remote sensing proxy for patch-level vegetation greenness. LandTrendr fits piecewise linear segments to each pixel’s NDVI trajectory and derives four change metrics from the main change segment: change timing (time), defined as the onset of the main change segment; change magnitude (mag), defined as the difference between the end and start NDVI values of this segment; change duration (dur), defined as the number of monthly time steps spanned by the segment; and change rate (rate), calculated as change magnitude divided by the duration. These metrics describe the timing and intensity of NDVI change signatures rather than ecological functions or mechanisms.

2.3.4. Examining the Effects of Ancient Trees on Patch Dynamics

This study makes descriptive and comparative inferences about whether the presence of natural versus planted ancient trees is associated with differences in patch-level change detection metrics (magnitude, rate, timing, and duration) and whether these differences show a spatial gradient across ancient tree, adjacent, and second-order adjacent patches. The analysis is correlational and does not test ecological mechanisms or causality. Each pixel was classified into one of three change status classes based on its main LandTrendr change segment (i.e., the segment with the largest absolute NDVI change). Pixels with a substantial positive NDVI change in the main segment were labeled as recovered, those with a substantial negative NDVI change were labeled as disturbed, and pixels without any pronounced positive or negative change in the main segment were labeled as unchanged. The association between the presence of ancient trees and patch dynamics was analyzed by integrating multi-temporal change detection results (change magnitude, change rate, change timing, and change duration) with land use classified patches using overlay analysis (Figure 2d). Given the correlational design and NDVI-based change metrics, we cannot disentangle the direction of this association and therefore interpret the results as an association between the presence of ancient trees and changes in surrounding vegetation, rather than as direct causal effects or a stabilizing effect. Metrics were summarized using descriptive statistics (frequency, median, mean, and the 25th and 75th percentiles) and compared among patch categories and land use types. For visualization, variables were categorized using the natural breaks (Jenks) method.

Analyses were conducted for three representative land use types (natural forests, planted forests, and croplands). Pixel-level change states were first classified as recovered, disturbed, or unchanged, and proportional areas were calculated for each patch category. To examine spatial gradients, comparisons were conducted among ancient tree patches, adjacent patches, and second-order adjacent patches. This gradient-based analysis was applied separately for natural ancient trees and planted ancient trees. In addition, direct comparisons were conducted between natural and planted ancient tree patches under the same land use types.

3. Results

3.1. Recovery and Disturbance Proportions in Patches

Table 1. Overall proportions of unchanged, recovered, and disturbed areas around natural and planted ancient trees.

Patch Group	Land Use Type	Proportion (%)
		Unchanged	Recovery	Disturbance
Patches surrounding natural ancient trees	Natural forest	21.66%	16.74%	3.85%
	Planted forest	5.98%	4.14%	1.12%
	Cropland	20.79%	20.71%	5.01%
Total		48.43%	41.59%	9.98%
Patches surrounding planted ancient trees	Natural forest	12.17%	13.00%	2.79%
	Planted forest	2.93%	2.84%	0.61%
	Cropland	25.28%	33.06%	7.33%
Total		40.38%	48.89%	10.73%

presents the proportions of unchanged, recovered, and disturbed areas in patches surrounding natural and planted ancient trees, including ancient tree patches and adjacent and second-order adjacent patches across different land use types. Overall, the region is dominated by unchanged and recovered areas, while disturbance remains minimal. Specifically, natural ancient tree patches comprise 48.43% unchanged, 41.59% recovered, and 9.98% disturbed areas, while planted ancient tree patches include 40.38% unchanged, 48.89% recovered, and 10.73% disturbed areas.

Comparative analysis between patches surrounding natural and planted ancient trees, including adjacent and second-order adjacent patches, revealed that patches surrounding natural ancient trees maintained a higher proportion of unchanged areas, reflecting stable patch states, whereas patches surrounding planted ancient trees exhibited a higher proportion of recovered areas, indicating a higher prevalence of recovery-dominated change states.

3.2. Magnitude and Rate of Patch Changes

Across all three land use types (Figure 4a–c), the interquartile ranges of change magnitude were consistently above zero, indicating that most detected changes were positive values (interpreted as recovery under our sign convention) rather than negative values (interpreted as disturbance). In natural forest (Figure 4a), natural ancient tree patches had similar 25th percentiles, medians, and 75th percentiles of change magnitude to adjacent and second-order adjacent patches, but a lower mean, suggesting a smaller contribution of relatively high magnitude positive changes to the overall distribution. In the planted forest (Figure 4b), planted ancient tree patches exhibited similar 25th percentiles and medians of change magnitude to their adjacent and second-order adjacent patches, but a higher mean change magnitude and 75th percentile, showing a distribution with a relatively heavier upper tail of positive change values in patches containing planted ancient trees. In croplands (Figure 4c), change magnitude distributions associated with natural ancient trees were highly similar among ancient tree patches, adjacent patches, and second-order adjacent patches, indicating only minor differences within these patch groups. In contrast, planted ancient tree patches showed a higher mean change magnitude, while the 25th percentile, median, and 75th percentile were similar to those of adjacent and second-order adjacent patches, suggesting that large positive values occurred more often in planted ancient tree patches, whereas central tendency and interquartile spread remained comparable across the three categories.

Patch statuses are visualized in Figure 4d, where recovery dominates the spatial pattern across the region. Unchanged areas were relatively limited and primarily concentrated in suburban and mountainous regions. Disturbed patches were sparsely distributed but showed a clear clustering tendency around urban peripheries, particularly in the vicinity of the Chengdu metropolitan area. In contrast, patches in Chongqing were predominantly in a state of recovery and stability, even within urban areas.

In natural forests (Figure 5a), natural ancient tree patches showed a lower mean change rate than their adjacent and second-order patches, while the 25th percentile, median, and 75th percentile were nearly identical among the three categories. This pattern is characterized by change rate values that are more tightly centered around near-zero levels in natural ancient tree patches when compared with adjacent categories. In planted forests (Figure 5b), second-order adjacent patches associated with natural ancient trees exhibited similar median change rates but a lower mean and lower 25th and 75th percentiles than natural ancient tree patches and their adjacent patches, indicating a shift toward lower change rate values from ancient tree patches to second-order adjacent patches within the spatial gradient defined in this study. In cropland (Figure 5c), change rate distributions were almost indistinguishable among all patch categories.

Spatially (Figure 5d), low change rates (−0.01, 0.01] dominated the study area. Low to moderate change rates (−0.06 ≤ rate ≤ −0.01 and 0.01 < rate ≤ 0.05) mainly occurred in urban areas, while high change rates (0.05, 0.21] were infrequent and primarily concentrated in rural regions, mountainous areas, and the surroundings of protected zones (Figure 1b).

3.3. Timing and Persistence of Patch Changes

For both natural (Figure 6a,c) and planted (Figure 6b,d) ancient tree patches and their surrounding patches, most change events occurred between 1990 and 1992, peaking markedly from March to May 1992. Early changes occurred throughout the study area, with notable clusters in the northeastern and central areas. Following this peak, the change frequency declined significantly and remained relatively low until the early 2000s, when several minor peaks were observed in 2000, 2001, 2003, 2006, 2008, and 2010. Thereafter, the change frequency gradually stabilized. This pattern was more pronounced in planted ancient tree patches and their surrounding areas.

Spatially (Figure 6a,b), early changes (before 2000) predominantly occurred in less urbanized regions, while mid-term changes (2000–2009) were mainly concentrated in urban and peri-urban areas, particularly extending from eastern Chengdu into the central and northeastern parts of Chongqing Municipality. Post-2010 changes exhibited a scattered distribution, but notable clusters emerged around major urban centers and in newly developing or peripheral regions.

Across patches surrounding natural (Figure 7a,c) and planted (Figure 7b,d) ancient trees, including ancient tree patches and their adjacent and second-order adjacent patches, short-duration changes dominated all change types. More than half of the disturbance and recovery events occurred within two years (approximately 53%). When the threshold was extended to five years, over 60% of disturbance and recovery events were captured, indicating most detected changes were concentrated within short timeframes (Figure 7). Based on the natural breaks classification, change durations of 3–9, 10–20, and 21–34 years accounted for progressively smaller shares of change events. For each duration class longer than five years, the proportion of pixels did not exceed 2%. Spatially, longer change durations were mainly observed in mountainous regions and major urban areas, whereas short-duration changes were widespread across the study region (Figure 7a,b).

Although the 25th percentile and median values of change duration were similar between patches surrounding natural and planted ancient trees, the 75th percentile and mean values were considerably higher in patches surrounding natural ancient trees (Figure 7c,d). This demonstrates that while the majority of pixels in all patch types exhibited short-to-medium change durations, patches surrounding natural ancient trees contained a greater proportion of pixels with longer change durations. This skewed the distribution upward, resulting in higher mean and upper quartile values. Overall, all patches were characterized by predominantly rapid and short-lived changes. However, patches surrounding natural ancient trees showed a slightly higher proportion of long-duration changes compared to those surrounding planted ancient trees.

4. Discussion

This study investigated whether patch change dynamics differ in relation to the presence of different types of ancient trees across various land use types in human-modified landscapes, using a monthly LT algorithm. Focusing on the human-modified Sichuan–Chongqing region, we examined patch dynamics over the past 34 years since 1990. The results reveal that, in natural forests, patches containing natural ancient trees were associated with a higher prevalence of stable change patterns, whereas patches containing planted ancient trees were associated with more recovery-dominated change signatures.

Focusing solely on small-scale or minimally disturbed areas may obscure the broader roles of ancient trees across different land use contexts. By integrating long-term satellite observations with the monthly LT algorithm, this study provides a feasible and robust approach to quantify the spatiotemporal dynamics of patches. Time series remote sensing data reveal both gradual and abrupt patch changes [27], offering a deeper understanding of how ancient trees are associated with patch dynamics against human-induced disturbances at broader spatial scales. This study transcends the constraints of conventional plot-based research by revealing novel perspectives on the large-scale significance of ancient trees in various land use contexts.

Recovery and stability dominated the study area, while disturbance was limited. Natural ancient tree patches and their adjacent and second-order adjacent patches showed high stability, with 48.43% unchanged and 41.59% recovered, while planted ancient tree patches and their surrounding patches exhibited a high recovery proportion (48.89% recovered). These patterns are consistent with the broad influence of conservation and restoration policies implemented in China (e.g., NFPP, the Grain for Green Project, and relevant protection regulations), which may have reduced disturbance and promoted vegetation recovery across the landscape. In natural forests, patches containing natural ancient trees were associated with higher stability than their adjacent and second-order adjacent patches, which may reflect the long-term co-occurrence of historically stable habitats and ancient trees [39] and the ecological roles of ancient trees [1,3,40]. However, this pattern may also reflect survivorship bias, whereby natural ancient trees persist in historically stable, low-disturbance environments, rather than allowing a clear separation between tree-related ecological effects and long-term habitat stability. In planted forests, patches containing planted ancient trees exhibited a higher recovery-related NDVI change than their adjacent and second-order patches. This is likely related to long-term human management practices (e.g., irrigation, fertilization, protection) [34,41], which are associated with the cultural significance and historical use of the ancient trees themselves. These stewardship practices, motivated by these trees’ unique values, are likely associated with the recovery of surrounding vegetation rather than the trees’ biological effects alone. The observed patch level contrasts and spatial gradients in change detection metrics provide pattern-based, correlational evidence. While our analysis does not test ecological mechanisms, the direction and spatial structure of these associations are consistent with the broader literature describing large old trees and remnant trees as ecological legacies and structurally important elements in modified landscapes [5,15]. Such evidence offers an independent ecological context for interpreting why ancient tree-associated patches may show distinctive recovery- and disturbance-related change signatures, while emphasizing that functional validation would require field-based or process-oriented indicators. Nevertheless, in both natural and planted forests, differences between adjacent and second-order adjacent patches were relatively weak, suggesting a diminishing role with increasing distance. Regarding the following pattern in planted forests, although it is not clearly evident in the visual representation, our analysis indicates higher recovery rate in planted ancient tree patches and their second-order adjacent patches, but not consistently in adjacent patches. This phenomenon may reflect the clustered spatial distributions of planted ancient trees, leading to partial overlaps between second-order adjacent patches and ancient tree patches, which may introduce uncertainty in recovery estimation. In croplands, although the distribution of change magnitude around the same type of ancient trees showed no significant differences among the three patch types, clear differences were observed between natural and planted ancient trees. In contrast, the distributions of change rate showed no apparent differences across either ancient tree types or patch categories. This discrepancy likely reflects strong and persistent anthropogenic disturbances in croplands, which may obscure spatial contrasts in recovery rate even when differences in change magnitude are evident. Moreover, fragmented vegetation structure and limited spatial continuity in croplands [42] may also restrict the expression of stabilizing patterns at the patch scale.

Analysis of change timing demonstrates that early changes (before 2000), primarily observed in mountainous and remote areas, were likely associated with large-scale afforestation efforts aimed at ecological restoration. These findings align with national policies. Since the 1990s, China has actively advanced ecological restoration projects via afforestation, farmland reforestation, and related initiatives, while attaching great importance to ecological protection [43]. Between 2000 and 2009, changes concentrated in peri-urban areas and rapidly growing urban areas and were likely caused by urban expansion and land use changes. Several minor change peaks were observed during the early 2000s, but this trend was more evident in planted ancient tree patches and their surrounding patches. This divergence may reflect their typical occurrence in more intensively managed and disturbance-prone landscapes. Such environments typically exhibit lower ecological stability due to ongoing urbanization, expansion, and land use transformations [34]. Conversely, natural ancient trees exhibited fewer fluctuations, likely reflecting their occurrence in less disturbed settings and long-term environmental stability [16,39], while the relative contributions of tree-related ecological effects and background environmental conditions cannot be disentangled in this study. Beyond the timing of changes, differences also emerged in how long these changes persisted. Analysis of patch change durations revealed that across all patch types, approximately 53% of disturbance events persisted for only 0–2 years, with over 60% occurring within a 5-year period, indicating that most disturbances were short-term. This prevalence of brief disturbance durations likely reflects rapid land use shifts and episodic anthropogenic impacts [44,45]. Even in natural patches, such short durations could be attributed to intermittent environmental events or edge effects from nearby anthropogenic activities. Similarly, approximately 53% of recovery events lasted for only 0–2 years, with over 60% persisting no longer than five years, suggesting that recovery-related changes often occurred over relatively short timeframes. This pattern may reflect a certain degree of ecological recovery, potentially associated with favorable site conditions, the co-occurrence of ancient trees, or effective restoration efforts.

Interpreting the effects of ancient trees based on patch dynamics is strongly scale-dependent. In this study, a 6 × 6 pixel window (180 m × 180 m at 30 m resolution) was used to represent ancient tree patches. This scale aligns with the spatial resolution of Landsat imagery and allows patch-level vegetation dynamics to be reliably detected from time series data. At this scale, differences among most patch categories could be observed. However, the relatively weak contrast between adjacent and second-order adjacent patches suggests that the difference in patch dynamics may be subtle at this spatial extent and that finer-scale spatial variability may not be fully resolved. At the same time, it should also be acknowledged that, at the 180 m patch scale, vegetation signals from the surrounding forest are inevitably integrated, and a dilution effect from these background dynamics cannot be entirely eliminated. To minimize the impact of large-scale regional trends, we employed comparative neighborhood-based approaches, focusing on localized, relative differences in patch dynamics rather than overall background vegetation trends. Although comparative analyses were applied to reduce this effect, the potential dilution from surrounding vegetation may still influence the results, especially in highly homogeneous forest stands or landscapes where background vegetation dynamics dominate. These findings emphasize that spatial scale is essential when assessing the role of ancient trees on surrounding patches. Future studies should explore a wider range of patch sizes to better understand how observed differences vary across spatial scales for ancient tree patches and their surrounding patches. In addition, because the indicators used in this study are derived from NDVI trajectories, the analysis captures pattern-based, correlational signatures of vegetation condition change. It does not directly quantify functional ecological roles, such as habitat provision, biodiversity support, microclimate buffering, or soil processes, nor does it allow causal inference. Future study will be required to test mechanistic explanations by integrating field measurements and functional remote sensing variables, such as structure from LiDAR, canopy height, biomass, or microclimate proxies. Such efforts would support the development of scientifically grounded protection zones around ancient trees and support site-specific conservation and forest management strategies.

5. Conclusions

This study employs a monthly LandTrendr-GEE cloud computing framework to detect the roles of natural and planted ancient trees on forest patch changes in the Sichuan–Chongqing region of China from 1990 to 2024. Stable and recovered areas dominated across all patches, with disturbed areas comprising only about 10%. Our results indicate that patch-level change detection metrics differ across ancient tree patches and their surrounding patch categories in a manner that depends on land use type and ancient tree type. In natural forests, patches containing natural ancient trees show smaller mean change magnitudes and change rates while having broadly similar interquartile ranges compared with adjacent and second-order patches, corresponding to distributions more centered near zero. In a planted forest, patches containing planted ancient trees showed comparatively larger positive changes (higher mean change magnitudes and higher 75th percentile values) relative to adjacent and second-order adjacent patches. Along the ancient tree adjacent to the second-order gradient, some cases exhibit a shift in change rate distributions toward lower values at the second-order level, indicating a distance-structured association within the spatial scale examined. These conclusions are association- and pattern-based inferences grounded in remote sensing change metrics; they do not demonstrate ecological functions or causal mechanisms. Temporally, most detected events occurred before 2010 and were short-lived (≤5 years), while long-duration changes (>5 years) were rare. These findings suggest that short-term disturbances reflect concentrated land use changes and anthropogenic impacts, while rapid recovery suggests that the presence of ancient trees may be associated with a swift return to a stable state. This research highlights the critical role of both natural and planted ancient trees in promoting forest recovery, buffering forest disturbance, and maintaining forest stability. These insights underscore the value of ancient trees and provide a valuable foundation for their conservation, as well as for the development of management strategies aimed at preserving ancient trees and forest patches in human-dominated landscapes.