Climatic and Human Drivers of Forest Vegetation Index Changes in Mainland Southeast Asia: Insights from Protected and Non-Protected Areas

Abstract

Forests’ dynamics have become increasingly complex under climate change and human activities. Mainland Southeast Asia (MSEA), with extensive forest cover and a mosaic of protected and non-protected areas, is an ideal region for examining forest responses to climate and anthropogenic activities. To ensure robust long-term monitoring, we retrieved EVI2 from daily AVHRR and MODIS surface reflectance data and performed pixel-wise assimilation of the two datasets, substantially reducing systematic bias and constructing a consistent 1982–2024 annual EVI2 dataset. Using this harmonized dataset, along with land use, climate, and protected area data, we analyzed over four decades of forest greenness trends, variability, and drivers in protected and non-protected zones. Results show that forests in MSEA maintained high coverage (mean EVI2 = 0.6253) and exhibited a clear greening trend (+0.014 per decade). Temperature was the dominant driver, contributing over 50% of the variation, followed by human activities (>30%), while precipitation played a smaller and mixed role. Forests in protected areas were 1.3 times more stable than those outside (CV = 4.1% vs. 5.2%), highlighting the buffering role of protection. These findings provide a reliable long-term benchmark for forest monitoring and inform targeted conservation and sustainable management strategies in MSEA. The constructed assimilated long-term EVI2 dataset is available to support future research on vegetation dynamics, climate impacts, and ecosystem sustainability.

1. Introduction

Forests are the largest terrestrial ecosystem on Earth, absorbing about 7 gigatons of carbon dioxide annually—approximately 25% of anthropogenic emissions—and playing a crucial role in biodiversity protection. They are therefore essential for maintaining global ecological balance and supporting sustainable development [1,2,3]. However, according to the Global Forest Resources Assessment, global forest area declined continuously between 1990 and 2020, with losses particularly pronounced in tropical regions [4,5]. Yet in Mainland Southeast Asia (MSEA), one of the world’s tropical biodiversity hotspots, the mechanisms driving long-term forest change remain poorly understood [6,7,8]. Existing global analyses often overlook this region’s complex interactions between climate variability and human disturbance, highlighting the need for region-specific, long-term assessments. The Food and Agriculture Organization of the United Nations (FAO) further emphasizes that national development planning should integrate forest conservation, sustainable utilization, and ecological restoration to balance economic growth with habitat quality.

Extensive research, largely supported by satellite remote sensing, has been conducted on long-term and systematic monitoring of forest cover, revealing the drivers and mechanisms of forest change across regions and time scales [9,10,11,12,13]. Remote sensing provides consistent, large-scale, and repeatable observations that enable detection of long-term vegetation dynamics, forest degradation, and recovery processes under both climatic and anthropogenic influences [14]. For example, in the Three Gorges Reservoir area, Tian et al. found that temperature had a stronger influence on vegetation indices than precipitation, with pronounced seasonal and spatial differences [15]. Forzieri et al. further reported that the resilience of tropical, arid, and temperate forests has declined significantly due to increasing water limitations, whereas boreal forests have shown enhanced resilience primarily as a result of warming [1]. Meanwhile, with rapid global economic development, the direct impacts of human activities on forests have become increasingly evident. Curtis et al. estimated that about 27% of global forest loss can be attributed to logging and land-use conversion, and that commodity-driven deforestation has continued to rise since 2001 [16].

Establishing protected areas is a common and effective approach to forest conservation, but their effectiveness varies across regions [17,18]. Previous studies have mainly focused on short-term or site-specific assessments, often constrained by inconsistent data sources and limited spatial comparability. For example, multi-country analyses showed that protected areas can reduce deforestation but are influenced by socio-economic and governance conditions [19]; other studies revealed that local reserves may still experience vegetation decline under external pressures [20,21]. These findings indicate that protection effectiveness cannot be generalized without accounting for regional heterogeneity and climatic influences, highlighting the need for consistent, long-term evaluation frameworks that integrate climate and human drivers at broader scales.

As the core region of Southeast Asia, MSEA is characterized by extensive forest cover and high ecological significance, and its forest conservation status is closely linked to regional and even global carbon cycling and biodiversity [22,23]. However, long-term and systematic analyses of forest disturbances in this region remain limited. In particular, the relative contributions of temperature, precipitation, and human activities to long-term forest change have not been clearly quantified, and comprehensive assessments comparing protected and non-protected areas are still lacking. To fill this gap, this study developed a harmonized long-term EVI2 dataset (1982–2024) by first resampling AVHRR and MODIS surface reflectance data to a common spatial resolution and deriving vegetation index (EVI2). The two datasets were then assimilated through pixel-level calibration over their overlapping period (2000–2013) using scaling factors, which effectively reduced Average Deviation (AD) and Symmetric Mean Absolute Percentage Error (SMAPE). This process minimized cross-sensor spectral and atmospheric inconsistencies and provided a consistent basis for long-term forest monitoring. Using this dataset together with land use, climate, and protected area data, we aimed to (1) quantify the relative influence of climatic and anthropogenic factors on forest EVI2 variability, and (2) compare the temporal stability of forest greenness trends between protected and non-protected areas. Using this dataset together with land use, climate, and protected area data, we analyzed forest greenness trends, variability, and the relative roles of climate and human drivers, with a focus on differences between protected and non-protected areas.

The results offer new insights into forest dynamics in MSEA and support evidence-based conservation and sustainable management. The overarching goal is to improve understanding of the mechanisms underlying forest dynamics in MSEA and to provide scientific evidence for regional forest conservation and sustainable development.

2. Materials and Methods

2.1. Study Area

MSEA spans approximately 92.0° E–109.0° E and 5.5° N–28.5° N, with a total area of about 2.065 million km² (Figure 1a). MSEA has a typical tropical monsoon climate, with mean annual temperatures ranging from 24 to 29 °C. Precipitation shows strong seasonality: abundant rainfall occurs from June to October under the influence of the southwest monsoon, while from November to May the northeast monsoon brings dry conditions with little rainfall. Topography generally slopes from north to south and is dominated by plateaus and mountains. Forests are the primary land cover type [24], accounting for about 35% of the total area based on the 2020 ESA CCI Land Cover dataset (https://cds.climate.copernicus.eu/, accessed on 12 June 2025).

In recent years, the economy of MSEA has developed rapidly. To promote ecological balance and regional sustainability, phased forest conservation programs have been gradually implemented since the beginning of the 21st century [25]. The establishment of nature reserves has been one of the key measures. At present, the study area contains 562 protected areas, covering a total of approximately 383,000 km², mainly distributed in the southeastern and central parts of the region (Figure 1b).

2.2. Data Sources

The data used in this study fall into four main categories: (1) remote sensing surface reflectance (SR) data, used to derive the EVI2 index for analyzing forest vegetation changes and trend characteristics; (2) climate data, including temperature and precipitation, used for attribution analysis of forest changes; (3) land use/land cover data, used to extract forest patches; and (4) protected area data, used to distinguish between protected and non-protected zones.

The first category consists of two complementary surface reflectance datasets used to derive a consistent long-term EVI2 time series. (1) AVHRR surface reflectance data were obtained from the NOAA CDR AVHRR dataset (Version 5) (https://www.ncei.noaa.gov/, accessed on 5 June 2025), which has provided daily surface reflectance since 24 June 1981, at a spatial resolution of approximately 0.05° (~5 km). (2) MODIS surface reflectance data were obtained from the MCD43A4.061 dataset provided by the National Aeronautics and Space Administration (NASA, https://ladsweb.modaps.eosdis.nasa.gov/, accessed on 5 June 2025), with a spatial resolution of 500 m and a daily temporal resolution, available since 24 February 2000.

For the AVHRR and MODIS surface reflectance datasets, strict quality control was applied to ensure data reliability before EVI2 retrieval. For AVHRR, quality assurance (QA) flags were used to remove pixels affected by clouds, cloud shadows, and nighttime observations. Specifically, only pixels with valid channel information (CH1–CH2) and clear-sky conditions were retained, while those with high solar zenith angles (>80°) or invalid reflectance values (outside 0–1) were excluded. For MODIS, only observations with valid red and near-infrared reflectance (>0) were used. To further enhance reliability, additional filtering based on the MCD43A2 BRDF/Albedo QA dataset was performed: pixels with full BRDF inversion quality (mandatory quality = 0) in both Band 1 (red) and Band 2 (NIR) were retained. This effectively minimized the influence of cloud contamination, atmospheric noise, and low-quality BRDF retrievals. These filtering steps ensured that only high-quality surface reflectance data were used for EVI2 calculation, improving the temporal consistency and cross-sensor comparability of the assimilated dataset.

Temperature and precipitation data were obtained from the ERA5-Land Monthly Aggregated dataset provided by the European Centre for Medium-Range Weather Forecasts (ECMWF, https://www.ecmwf.int, accessed on 5 June 2025). The dataset covers the period from February 1950 to the present, with a spatial resolution of 0.1°. Monthly mean temperature and precipitation data for 1982–2024 were retrieved and resampled to a 1 km grid using the nearest-neighbor method to maintain the original statistical integrity of the ERA5-Land fields. To evaluate potential scale effects, we additionally conducted sensitivity experiments using bilinear and cubic convolution resampling. The results showed negligible differences in regional trend estimates and attribution outcomes across the three methods, confirming that the chosen resampling approach does not materially affect the analysis accuracy.

Land use/land cover data were derived from the global dataset provided by the European Space Agency (ESA) under the Climate Change Initiative (CCI, https://cds.climate.copernicus.eu/, accessed on 12 June 2025). To identify forest patches, we performed a binary classification of land cover types into forest and non-forest categories. Pixels labeled as forest in any year of the available time series were aggregated to define the maximum forest extent during the study period, ensuring that all areas with forest presence at least once were included in the final forest mask. This approach captures the complete spatial distribution of forested land while avoiding potential omission of intermittently forested areas.

Protected area data were obtained from the World Database on Protected Areas (WDPA, https://www.protectedplanet.net/, accessed on 5 June 2025). In MSEA, there are 562 protected areas with a total area of approximately 383,000 km².

2.3. Construction of the Assimilated EVI2 Dataset

We constructed a consistent annual EVI2 dataset for 1982–2024 by integrating AVHRR and MODIS observations. EVI2 was first derived from AVHRR and MODIS surface reflectance data (R and NIR bands), using the standard two-band EVI2 formulation [26].

$$EVI2=2.5\times {\displaystyle \frac{NIR-R}{NIR+2.4·R+1}}$$

(1)

For both EVI2 datasets, we generated annual maximum value composites by first removing anomalous or contaminated observations (e.g., due to clouds) and then selecting the maximum value for each year. Because the native resolutions of AVHRR and MODIS differ substantially, all data were first standardized to a 1 km spatial grid before assimilation. The resampling was based on the reflectance information within each coarse AVHRR pixel rather than surrounding pixels, in order to preserve the original radiometric integrity and avoid introducing artificial spatial smoothing. Although vegetation indices are continuous variables, nearest-neighbor interpolation was chosen to maintain the discrete spectral characteristics of the AVHRR pixel during spatial alignment. Subsequent pixel-wise calibration with MODIS data ensured that the final assimilated dataset retained continuous and spatially coherent vegetation index values across sensors.

To remove inter-sensor biases and produce a temporally homogeneous series, AVHRR and MODIS EVI2 data were assimilated at the pixel level over their overlap period (2000–2013) [8]. A scaling factor (SF) was computed for each pixel as the ratio of its multi-year mean MODIS EVI2 to AVHRR EVI2, and this factor was applied to adjust all pre-2000 AVHRR values. This assimilation minimized systematic discrepancies and improved temporal continuity, as further validated by the reduced AD and SMAPE after scaling. The resulting dataset provides an annual, 1 km resolution EVI2 record.

$${sf}_{(x,y)}={\displaystyle \frac{\sum _t{MODIS}_{(x,y,t)}}{\sum _t{AVHRR}_{(x,y,t)}}}$$

(2)

$$\overline{{AVHRR}_{(x,y)}}={sf}_{(x,y)}\times {AVHRR}_{(x,y)}$$

(3)

Here, ${sf}_{(x,y)}$ is the SF for pixel, $\overline{{AVHRR}_{(x,y)}}$ is the assimilated AVHRR EVI2 value for pixel $(x,y)$, and $t$ represents the overlap years between MODIS and AVHRR, which in this study are 2000–2013.

To evaluate the effectiveness of the assimilation approach, we compared the AVHRR EVI2 data before and after assimilation with MODIS EVI2 for the overlapping period (2000–2013). Figure 2 illustrates the spatial differences between AVHRR and MODIS multi-year mean EVI2 values before (a) and after (b) assimilation. Before assimilation, large negative biases were widespread, with AVHRR systematically underestimating EVI2 across most of the region. After assimilation, the spatial differences were greatly reduced, and most pixels showed near-zero deviations, indicating improved consistency.

Table 1. Validation of EVI2 data assimilation.

Year	Before Assimilation		After Assimilation
	AD	SMAPE	AD	SMAPE
2000	−0.126	0.308	0.038	0.150
2001	−0.125	0.283	0.043	0.115
2002	−0.137	0.300	0.031	0.097
2003	−0.158	0.331	0.010	0.089
2004	−0.119	0.344	0.039	0.125
2005	−0.139	0.327	0.026	0.113
2006	−0.176	0.374	−0.014	0.094
2007	−0.173	0.373	−0.014	0.096
2008	−0.163	0.354	−0.003	0.097
2009	−0.173	0.368	−0.012	0.098
2010	−0.176	0.375	−0.017	0.093
2011	−0.185	0.389	−0.027	0.096
2012	−0.171	0.362	−0.011	0.088
2013	−0.176	0.363	−0.012	0.087
Multi-year average	−0.157	0.346	0.005	0.103

provides a quantitative assessment. Both the AD and the SMAPE decreased substantially after assimilation. These metrics were calculated using all pixels across the study region, ensuring a comprehensive evaluation of assimilation performance. The mean AD improved from −0.157 to 0.005 (a 96.8% reduction), while SMAPE dropped from 0.346 to 0.103 (a 70.2% reduction). These results confirm that the assimilation procedure effectively removed systematic biases and enhanced the temporal consistency of the long-term EVI2 dataset.

2.4. Analysis of Vegetation Index Dynamics

To systematically evaluate the dynamic characteristics of the forest EVI2, this study applied the pixel-wise Mann–Kendall (MK) test to annual EVI2 time series. This method identifies the direction, magnitude, and significance of EVI2 changes, thereby providing a comprehensive characterization of regional vegetation dynamics [27]. The MK test is a classical non-parametric method widely used in meteorology, hydrology, and ecology. It can effectively detect long-term trends in time series without relying on assumptions about data distribution.

Building on the trend test, this study further applied Sen’s slope estimator to quantify the magnitude of vegetation index trends. This method provides a robust estimate of the trend magnitude in a time series without being constrained by assumptions about data distribution.

2.5. Attribution Analysis of Vegetation Index Changes

To quantitatively assess the relative contributions of climatic factors and human activities to forest vegetation changes, this study employed multiple linear regression combined with residual analysis. Specifically, the annual EVI2 series was used as the dependent variable, while climate factors were treated as independent variables to construct a multiple linear regression model at the pixel level. After model fitting, the residuals represent the portion of EVI2 variation not explained by climatic factors, which is interpreted as the approximate effect of human activities.

$$y=\sum _i{a}_i\times x_i+d$$

(4)

$$∆y_i=a_i\times ∆x_i$$

(5)

$$∆y_h=∆y-\sum _i∆y_i$$

(6)

Here, $y$ is the dependent variable, $x_i$ is the $i$-th independent variable, $a_i$ is the regression coefficient of $x_i$, and $d$ is intercept. $∆x_i$ represents the change in $x_i$ over the study period, calculated as the mean value for 2004–2024 minus the mean value for 1982–2003; $∆y_i$ denotes the contribution of $∆x_i$ to the change in the dependent variable; and $∆y_h$ represents the contribution of human activities to the change in the dependent variable, determined from the regression residuals. In this study, $y$ corresponds to EVI2, $x_1$ to temperature, and $x_2$ to precipitation.

Subsequently, the contribution ratio of each influencing factor to EVI2 change was calculated using the following formula, and a threshold of 0.5 was applied to identify the dominant factor. This threshold is commonly used to determine the primary driver in multi-factor attribution studies. In cases where no single factor exceeded 0.5, it indicates that two or more variables contributed comparably to EVI2 variation, reflecting the joint influence of climatic and anthropogenic processes on forest dynamics.

$${CR}_i={\displaystyle \frac{\left|∆y_i\right|}{\sum _{i=1}^3\left|∆y_i\right|}}$$

(7)

Here, ${CR}_i$ denotes the contribution ratio of the $i$-th factor, and $∆Y_i$ is the contribution of the $i$-th factor.

Furthermore, this study distinguishes between protected and non-protected areas in order to compare their differences, and the analyses were conducted separately for the two categories.

3. Results

3.1. Forest Distribution and Long-Term Mean Vegetation Index

As shown in Figure 3a, forests in MSEA are mainly distributed in the northern mountainous regions and the eastern highlands, while appearing as fragmented patches across the central and southern lowlands. The total forest area in the region was approximately 730 × 10³ km², accounting for 35.4% of the land area. Among these, protected areas contained 218 × 10³ km² of forest (29.9% of the total forest area), whereas non-protected areas contained 512 × 10³ km² (70.1%). The overall forest coverage in the region was 35.4%.

The spatial distribution of mean EVI2 for forests in MSEA during 1982–2024 is shown in Figure 3b. The regional mean EVI2 was 0.6253, with 97% of the area exceeding 0.5, indicating generally high forest cover. Areas with EVI2 values of 0.5–0.6, 0.6–0.7, 0.7–0.8, and >0.8 accounted for 25%, 63%, 8%, and 1% of the region, respectively, mainly in mountainous and plateau areas. In contrast, regions with EVI2 below 0.5 comprised only 3%, scattered across the southeast and northwest.

3.2. Changes and Trends of Forest Vegetation Index

Based on the pixel-wise MK trend test results (Figure 4a), most forest areas in MSEA showed an increasing EVI2 trend, with 63% of the region experiencing significant increases and 24% showing non-significant increases; only 2% exhibited significant decreases. This indicates that forest conditions in the region have improved substantially over the past four decades. The spatial distribution of EVI2 change (Figure 4b), further reveals that the most pronounced increases (>0.04) occurred in northern mountainous regions and parts of the eastern highlands, together covering more than 50% of the study area. Areas of decline were limited (8%), suggesting a general trend of forest recovery and enhanced growth. It should be noted that potential temporal and spatial autocorrelation may influence local significance estimates, which is further discussed in Section 4.

Interannual variations show that forests in both non-protected (Figure 4c) and protected areas (Figure 4d) experienced significant upward trends (p < 0.01), with nearly identical linear rates of 0.014 year⁻¹, corresponding to an increase of about 0.014 per decade. Although the mean EVI2 levels were similar in the two types of areas, non-protected zones exhibited 1.27 times higher interannual variability (CV = 5.2%) than protected areas (CV = 4.1%), highlighting their greater sensitivity to anthropogenic disturbance.

3.3. Attribution of Forest EVI2 Variations

In non-protected areas, temperature exerted predominantly positive effects on EVI2 changes, with 67% of the area showing a positive response (Figure 5). Its mean contribution was 0.0080, accounting for 51.9% of the total, and nearly 45% of the area exhibited contributions greater than 0.01, indicating that temperature variation was the primary driver of EVI2 increases in non-protected forests. Precipitation had a mean contribution of 0.0016 (10.6%), exerting positive effects in 56% of the area but negative effects in 44%, suggesting a relatively limited overall influence. Human activities contributed an average of 0.0058 (37.5%), with 60% of the area showing positive impacts, implying that in non-protected areas, human interventions may both cause forest degradation and promote EVI2 increases through afforestation or vegetation restoration projects.

In protected areas, the mean contribution of temperature was 0.0088 (55.6%), with 68% of the area showing positive effects; nearly 48% of the area exhibited contribution values greater than 0.01, confirming that temperature was likewise the dominant factor driving EVI2 changes within reserves. Precipitation contributed an average of 0.0018 (11.5%), promoting growth in 55% of the area but exerting negative effects in 45%, indicating a relatively weak overall influence. Human activities contributed an average of 0.0052 (32.9%), with 65% of the area showing positive impacts—a higher proportion than in non-protected areas—suggesting that protective measures have, to some extent, mitigated the adverse effects of human disturbance.

Overall, temperature was the primary factor influencing forest EVI2 changes, followed by human activities, while precipitation made the smallest overall contribution. The influence patterns of temperature and precipitation were largely consistent across the two categories, whereas the positive effects of human activities were more pronounced within protected areas. This indicates that protected areas not only effectively buffer the negative impacts of human activities but also play an active role in maintaining forest vegetation stability.

4. Discussion

This study shows that between 1982 and 2024, forests in MSEA (MSEA) maintained a generally high level of greenness, with a mean EVI2 of 0.6253 and a significant upward trend of approximately 0.014 per decade. Both protected and non-protected areas exhibited similar long-term trajectories, indicating widespread forest improvement across the region. However, non-protected areas showed more pronounced interannual fluctuations (CV = 5.2%), 1.27 times higher than in protected areas (CV = 4.1%), reflecting their greater vulnerability to anthropogenic disturbance. These results highlight the effectiveness of protected areas in maintaining forest stability over the past four decades.

Given the relatively high vegetation cover in MSEA, EVI2 was used as the primary indicator to better capture the response of forest dynamics to climatic factors [28]. The results show that temperature was the dominant driver, exerting significant positive effects, while the contribution of precipitation was smaller and varied in direction. During the past decades, the region experienced overall warming and a decline in precipitation [8]. Moderate warming can enhance photosynthesis and growth, whereas insufficient rainfall may exacerbate water stress under conditions of high temperature, drought, or high elevation, ultimately leading to biomass decline [29,30,31].

In general, human activities exert a dual influence on forests. Our results indicate that management measures in protected areas effectively limited forest degradation, leading to EVI2 improvement in 65% of the area. Nevertheless, pressures from illegal logging, edge development, and tourism remain significant [32,33,34]. By contrast, the absence of strict regulation in non-protected areas resulted in more volatile forest dynamics: expansion of commercial plantations and logging caused degradation, while ecological restoration policies promoted recovery. Although protected areas mitigated negative human impacts, they may also shift development pressures to surrounding regions [35]. Therefore, forest conservation in MSEA requires strengthened regional cooperative governance, integrating policy enforcement, resource allocation, and development restrictions to achieve long-term ecosystem stability [36,37,38].

This study integrated a homogenized EVI2 dataset (1982–2024) with multi-source land cover, climate, and protected area data to reveal the long-term patterns and drivers of forest change in MSEA. The results provide important insights into the relative roles of climate and human activities and highlight the stabilizing effect of protected areas. Despite these advances, several limitations remain. Although the assimilation of AVHRR and MODIS data greatly improved temporal consistency, residual sensor differences, calibration uncertainties, and spatial resolution mismatch may still influence early-period estimates. In addition, uncertainty in land-cover classification could affect the accuracy of forest boundary identification and trend estimation. The residual component in the attribution analysis may also include unmeasured environmental effects in addition to human activity. However, its spatial consistency with known disturbance hotspots supports its reliability in representing non-climatic influences. Forest dynamics also exhibit temporal autocorrelation, as vegetation growth is not fully independent between years. Although the 43-year annual averaging of EVI2 reduces short-term fluctuations, some serial dependence may remain, introducing limited uncertainty to trend and attribution estimates [39]. Even so, the spatial coherence and long-term stability of observed patterns demonstrate the robustness of our findings. Future studies should incorporate additional environmental variables and advanced time-series models to better capture non-linear, spatially heterogeneous, and temporally dependent forest responses in MSEA.

5. Conclusions

Based on multi-source remote sensing and environmental data, this study analyzed the spatiotemporal patterns and drivers of forest EVI2 in MSEA from 1982 to 2024, with a comparative focus on protected and non-protected areas. The main conclusions are as follows:

(1)

The total forest area in MSEA was approximately 730 × 10³ km², accounting for 35.4% of the region’s land area. Among these, forests in protected areas covered 218 × 10³ km² (29.9% of the total forest area). Overall, forest conditions were favorable, with a mean EVI2 of 0.6253 and an increasing trend of about 0.014 per decade, showing similar trajectories in both protected and non-protected areas.

(2)

Temperature was the dominant factor driving EVI2 changes, with contributions of 55.6% in protected areas and 51.9% in non-protected areas, significantly higher than those of precipitation (11.5% and 10.6%) and human activities (32.9% and 37.5%).

(3)

Forest EVI2 in non-protected zones fluctuated considerably, likely influenced by activities such as logging and plantation expansion, which are common in tropical forest regions. In contrast, forests in protected areas remained relatively stable, playing an important role in mitigating the potential negative effects of human disturbance.

Overall, rising temperatures were the core driver of forest growth in MSEA, while the establishment of protected areas contributed to maintaining stability and reducing human disturbance. These findings provide a scientific reference for monitoring regional forest dynamics.