Phase 1 (2025) – Spatial distribution and drivers of greening
Act. 1.1 – Uncovering regional-level distribution of greening trends
We analyzed Landsat satellite data from 1984 to 2023 using Google Earth Engine to track long-term vegetation trends. We used Tier 1 surface reflectance images from Landsat 5, 7, and 8, selecting only those with less than 80% cloud cover. Cloud- and snow-affected pixels were removed using the CFmask algorithm. To correct for differences in lighting and viewing angles, we applied BRDF normalization, standardizing reflectance values across time and space. We also used cross-sensor calibration to align reflectance data between the three Landsat missions. For each year, we calculated the maximum NDVI value (NDVImax) during the growing season, based on BRDF-corrected and calibrated red and near-infrared bands. To avoid bias from increasing image frequency over time, we adjusted NDVImax using a pixel-based phenology model. NDVI time series were reconstructed with the HANTS method, and trend analysis was done using an autoregressive model, which accounts for temporal autocorrelation and provides more reliable slope estimates than traditional methods.
Key findings:
-
~40–44% of pixels above 1500 m show significant greening.
-
Greening hotspots concentrate on north-facing slopes between 1800–2200 m.
-
Areas with strongest greening include: Făgăraș, Parâng, Rodna, Călimani, Retezat, Bucegi.
Act. 1.2 – Assessing greening trends variability across land cover types
Using Sentinel-2 SR imagery (2018–2021), we built a detailed land cover map distinguishing six key vegetation types: screes, grasslands, Ericaceae shrublands, Juniperus shrublands, Pinus mugo shrublands, and Picea forests.
Over 1500 training points were manually digitized from high-resolution aerial imagery and Google Earth.
A Random Forest classifier using spectral indices (EVI, MNDWI, NARI, NCRI, CARI) produced a high-resolution (10 m) land cover map, later resampled to 30 m for integration with Landsat.
To quantify how greenness trends relate to land cover, we used projection pursuit regression with compositional data constraints. Individual Conditional Expectation (ICE) plots indicated:
-
Ericaceae shrublands show the strongest, most consistent greening.
-
Juniperus shrublands also exhibit positive trends.
-
Picea forests and screes show weak or neutral greening.
-
Grasslands show variable responses depending on elevation and topoclimate.
Act. 1.3 – Exploring the effect of regional climate change
Using ERA5-Land surface temperature (1984–2024), we extracted March–October mean temperatures for all study sites. By pairing NDVImax and temperature series, we examined the synchrony between climatic warming and vegetation productivity.
Main outcomes:
-
Two major greening waves were identified (~1993–1996 and ~2016–2019), with an intermediate weaker wave (early 2000s).
-
The first wave is tightly linked to strong warming episodes.
-
The second wave occurs despite reduced or locally negative warming trends, indicating that greening may be increasingly influenced by non-thermal factors (moisture availability, snow dynamics, land-use).
-
Site-level NDVI–temperature regressions show strong positive relationships in some Ericaceae and grassland sites, weak responses elsewhere, and occasional negative responses, reflecting spatial climatic heterogeneity.
These findings refine our understanding of long-term regional climatic controls on mountain greening.
Act. 1.4 – Investigating the role of microclimate
We integrated temperature logger data with ERA5-Land and Landsat NDVImax to examine how microclimatic variability influences greening.
Key results:
-
Microclimate (GDDm from loggers) explains NDVImax variation far better than regional ERA5 data.
-
Responses differ strongly between sites:
-
Some sites show negative NDVI–temperature relationships (thermal or moisture stress).
-
Others show positive relationships (longer growing seasons).
-
Some sites show no clear thermal control, indicating the dominance of water balance, snow persistence, or vegetation structure.
-
-
ERA5 often underestimates or misses local temperature fluctuations important for vegetation.
This highlights the critical role of microclimate buffering/amplification in shaping greening patterns.
Act. 1.5 – Evaluating the consequences of land use changes
A case study in Argeș county (southern Făgăraș Mountains) compared NDVImax from grazed vs. ungrazed alpine pastures with 30+ years of livestock (ovine) statistics.
Key insights:
-
Grazed pastures generally show higher NDVImax than ungrazed ones.
-
Sharp post-1990 decline in sheep numbers does not cause an immediate greening of ungrazed pastures.
-
Instead, NDVImax declines in both grazed and ungrazed areas, likely due to:
-
early shrub colonization,
-
changes in species composition,
-
degradation patches.
-
-
Moderate grazing appears to sustain stable, productive vegetation, whereas both overgrazing and abandonmentcan reduce or destabilize NDVI.
Land-use change is therefore a non-linear driver of greening, interacting with climate and shrub expansion.
Summary of 2025 Achievements
-
100% completion of all planned Phase 1 activities.
-
A robust multi-scale analytical framework integrating Landsat, Sentinel-2, ERA5, and logger data.
-
Identification of major greening hotspots and the environmental & land-use mechanisms shaping them.
-
Establishment of complete land cover databases, climate–NDVI relationships, and first evidence linking pastoral decline to greening patterns.
-
A Q1 publication, conference presentations, media exposure, and a fully functional project website and social media presence
Phase 2 (2026) – Ecological processes of greening and consequences on mountain biodiversity
Act. 2.1 – Investigating shrub encroachment
Act. 2.2 – Evaluating the increase in shrubland density
Act. 2.3 – Studying the impact on plot-scale species diversity
Act. 2.4 – Assessing the consequences on landscape heterogeneity