The pattern, change and driven factors of vegetation cover in the Qin Mountains region

Vegetation is a key indicator of ecosystem function, biodiversity, and land–atmosphere exchanges. Under global change, vegetation cover has shifted notably across regions, motivating long-term monitoring. Remote sensing enables frequent, accurate assessment of vegetation dynamics, with NDVI widely used as a proxy for vegetation cover. Prior studies using GIMMS NDVI (1982–2012) and MODIS NDVI (post-2000) reported broad greening trends and highlighted the influence of China’s Grain for Green Project (GFGP) on vegetation dynamics. However, coarse resolution (GIMMS) can miss regional spatial details, while MODIS lacks pre-2000 coverage. The Qin Mountains, a major climatic boundary in China with complex topography and strong human influence, require high-resolution, long-term analysis. This study uses Landsat-based NDVI on Google Earth Engine to examine: (1) when and where vegetation changes occurred over the last three decades in the Qin Mountains, and (2) what major factors drove NDVI changes over time and elevation.

Multiple regional studies document increasing NDVI across China and attribute changes to both climate variability and human activities. GIMMS-based analyses (e.g., Xinjiang, Loess Plateau) found widespread greening with seasonal patterns, and MODIS-based studies linked NDVI increases to the GFGP, showing correlations between afforestation area and NDVI. Combined GIMMS–MODIS efforts revealed upward NDVI trends before the mid/late 1990s under hotter, drier conditions and the effects of drought. In the Qin Mountains (narrow definition within Shaanxi), MODIS-based work (2000–2015) showed significant greening and ranked precipitation > temperature > potential evapotranspiration as influences. Many studies, however, neglected complex topography and were limited by spatial or temporal resolution (GIMMS coarse resolution; MODIS post-2000), motivating Landsat-based, high-resolution, long-term analyses via GEE to better resolve spatial heterogeneity and topographic effects.

Study area: The Qin Mountains in southern Shaanxi, China (104°30′–110°05′E, 32°40′–34°35′N), elevation 95–4591 m, form a climatic boundary between North and South China and host diverse vegetation types that vary with altitude. Datasets: (1) Landsat surface reflectance imagery (30 m): Landsat 5 TM (1987–2011; LEDAPS), Landsat 7 ETM+ (2012; LEDAPS, gap-filled), Landsat 8 OLI/TIRS (2013–2018; LaSRC). Cloud, shadow, water, snow masks via CFMASK. (2) DEM: SRTM V3 (1 arc-second, ~30 m). (3) Land cover: GlobeLand30 (2010, ~30 m), reclassified to cultivated land, forest, grassland, shrubland, water bodies (incl. wetland), artificial surfaces, bare land. (4) Climate: TerraClimate monthly fields (precipitation, temperature, reference evapotranspiration PET, ASCE Penman–Monteith) at ~5 km, 1987–2018. Image processing: Using Google Earth Engine, images were filtered by pixel quality to remove clouds and shadows. NDVI computed as (NIR–RED)/(NIR+RED) using appropriate Landsat bands (NIR: band 4 for L5/7, band 5 for L8; RED: band 3 for L5/7, band 4 for L8). For each year, the maximum NDVI during the growing season (March–November) was extracted per pixel to represent optimal vegetation conditions. Elevation zoning: Based on NDVI–elevation relationships, the area was divided into four zones: I <1300 m, II 1300–1800 m, III 1800–3300 m, IV >3300 m. Trend and change-point analysis: Theil–Sen trend estimation (with significance via p-values) assessed pixel-wise and regional NDVI trends; significant greening defined as slope > 0 and p < 0.05, significant degradation as slope < 0 and p < 0.05, and insignificant otherwise. Accumulated variance (residual mass curve) analysis identified abrupt change points in the temporal NDVI series; the whole period was partitioned into 1987–1999, 1999–2006, and 2006–2018 (noting GFGP inception in 1999 and detected breakpoint). Causality analysis: Pearson correlation was computed between annual growing-season NDVI and climate variables (precipitation, temperature, PET) across the region and stratified by elevation, with significance assessed (p < 0.05). Spatial analyses and mapping were performed in GEE and ArcGIS 10.2. Additional comparisons contrasted Landsat-derived NDVI changes with MODIS-derived results to illustrate differences in delineating degradation features (e.g., mining, roads, urban areas).

• Regional NDVI trend (1987–2018): Mean NDVI across the Qin Mountains was relatively high (mean ≈ 0.68) and increased significantly at 0.0053 per year over 32 years (R² = 0.8361, p < 0.01). NDVI rose from approximately 0.624 to 0.776. - Change-point: An abrupt point occurred in 2006. Post-2006 NDVI increased faster at 0.0094 per year (R² = 0.8159, p < 0.01) during 2006–2018, exceeding rates in 1987–1999 (slope 0.0029, R² = 0.2169, p < 0.01) and 1999–2006 (slope 0.0025, R² = 0.3415, p < 0.01). - Spatial distribution: About 92.2% of the area had NDVI > 0.5; highest values (≈0.8) occurred in central Shaanxi and Henan. Areas with initially lower NDVI showed faster increases (slopes > 0.2 in 1987–2018), while high-NDVI regions increased more slowly (slopes < 0.2). - Elevation patterns: NDVI increased with elevation up to ~1300–1800 m (peak), then remained ~0.7 between 1800–3300 m, and decreased above 3300 m. All elevation zones showed increases from 1987 to 2018: Zone I (<1300 m) mean 0.67, total increase (slope over period) 0.18; Zone II (1300–1800 m) mean 0.71, increase 0.15; Zone III (1800–3300 m) mean 0.69, increase 0.13; Zone IV (>3300 m) mean 0.65, increase 0.12, with higher interannual variability (Std 0.0785). - Vegetation types by elevation (before vs after 2006): In Zones I–III (<3300 m), cropland and forest dominate. Forest NDVI was high and increased (e.g., Zone II: 0.74→0.82; Zone III: 0.71→0.82). Cropland NDVI notably increased in low elevations (Zone I: 0.55→0.66; +21.31%). Grassland, though less extensive in Zones I–III, also increased (e.g., Zone I: 0.56→0.67). In Zone IV (>3300 m), forest (35.76%) and grassland (62.68%) dominate; grassland showed relatively higher NDVI and increase (+11.80%) compared to forest (+10.59%). - Climate–NDVI relationships (time scale): Correlation coefficients between NDVI and climate factors strengthened over time. R(NDVI, precipitation) increased from 0.06 to 0.21; R(NDVI, temperature) from 0.10 to 0.31; R(NDVI, PET) peaked around 0.29 by 1987–2016 and was 0.24 for 1987–2018. - Climate–NDVI by elevation: Before 2006, precipitation showed weak or negative correlations with NDVI overall, with relatively higher correlations at 2300–2700 m; above 3000 m correlations decreased. Temperature correlated more strongly with NDVI at 0–1300 m before 2006, shifting to >3000 m after 2006. PET–NDVI correlations were higher at 800–1200 m and 2400–3300 m before 2006, and mainly >2600 m after 2006. - Drivers: Below 3300 m, NDVI was mainly driven by thermal factors (temperature, PET) before 2006; after afforestation and other positive human interventions, human activities became important drivers in low-elevation areas. Above 3300 m, hydro-climatic factors (especially temperature and PET) remained dominant throughout. - Sensor comparison: Landsat-based NDVI captured degradation features (mining, roads, urban expansion) with clearer boundaries and more internal detail than MODIS-derived NDVI, highlighting the value of high-resolution, long-term Landsat data via GEE.

The study demonstrates significant and widespread greening in the Qin Mountains over the last three decades, with an accelerated increase after the 2006 breakpoint. This temporal pattern aligns with delayed vegetation response following large-scale afforestation under the Grain for Green Project, indicating that human activities increasingly influenced vegetation in low-elevation, agriculture-dominated zones after 2006. Elevation modulated both NDVI levels and drivers: low to mid-elevations exhibited strong greening and a shift from climate to human-dominated drivers, while high elevations (>3300 m) remained primarily climate-driven, with temperature and PET exerting consistent influence. Spatial heterogeneity was substantial, with initially low-NDVI areas experiencing faster improvement. The high-resolution Landsat record via GEE enabled finer delineation of degradation hotspots (e.g., mining, infrastructure, urban areas), improving attribution and management implications compared to coarser sensors. These findings answer where and when changes occurred (broad greening, especially post-2006; strongest gains in low-NDVI and lower-elevation areas) and why (combined effects of afforestation and climate, with topographic controls). The results underscore the importance of integrating topography in vegetation–climate analyses and demonstrate the utility of cloud-based processing of dense Landsat time series for regional ecosystem monitoring and policy evaluation.

Using high-resolution Landsat NDVI time series on GEE, this study quantified vegetation cover patterns, trends, and drivers across the Qin Mountains from 1987 to 2018. NDVI increased significantly overall, with an abrupt change around 2006 and a tripled growth rate thereafter compared to earlier periods. Over 92% of the area had NDVI > 0.5, with greening particularly pronounced below 3300 m in cropland and grassland. Above 3300 m, where forest and grassland dominate, NDVI values were lower and more tightly linked to climatic controls. Before 2006, temperature and PET primarily drove NDVI below 3300 m; after 2006, human activities (notably afforestation) increasingly influenced low elevations, while hydro-climatic factors continued to dominate high elevations. Elevation-conditioned analyses improved understanding of vegetation dynamics and can guide targeted policies to mitigate degradation. Future work should incorporate additional climatic and hydrologic variables (e.g., soil moisture, humidity, aridity index), assess seasonal and lagged responses, and continue leveraging high-resolution time series for fine-scale monitoring of land cover change and ecosystem responses.

• Climate drivers were limited to precipitation, temperature, and PET; other influential variables (e.g., soil moisture, humidity, aridity index) were not included and may exhibit lagged effects relative to NDVI. - Correlation analysis does not establish causality; disentangling human vs climate influences would benefit from methods separating trends (e.g., RESTREND, attribution models) and incorporating land management data. - Land cover was derived from GlobeLand30 (2010), which may not fully represent land cover changes across the entire 1987–2018 period. - TerraClimate’s ~5 km resolution may smooth local climate variability in complex terrain. - Landsat 7 ETM+ data required gap filling (2012), which can introduce uncertainties, and cloud contamination can reduce data density in some years/areas.