### Mapping global mountain vegetated landscapes
We first mapped MVL in 2000 as a base map from which too estimate changes. To do so, we combined a global land-cover dataset in 2000 with a global mountain dataset. For land cover, some moderate spatial resolution products exist, such as the MODIS Land Cover Type from 2001 to 2016 (500 m), the ESA Climate Change initiative (ESA-CCI) from 1992 to 2015 (300 m), Copernicus Global Land Cover Layers (CGLS-LC100) from 2015 to 2019 (100 m), Finer Resolution Observation and Monitoring of Global Land Cover product (FROM-GLC) in 2010, 2015 and 2017 (30 m/10 m), Globeland30 in 2000, 2010 and 2020 (30 m), global 30 m land cover classification with a fine classification system (GLC_FCS30D) in 2000 and 2020 (30 m) and ESA’s WorldCover in 2020 and 2021 (10 m)78,79,80,81. However, we chose GLC_FCS30D in our study mainly due to its high spatial resolution, suitable temporal resolution, fine classification system and more stable accuracy on a global scale compared to other 30-m global land-cover products82.The GLC_FCS30D product was produced by combining time series of Landsat imageries and high-quality training data from the Global Spatial Temporal Spectra Library on the google Earth Engine cloud computing platform and presented at an approximately 30-m resolution80. The accuracy of GLC_FCS30D has been validated using numerous validation samples, with an overall accuracy of more than 81 %83,84 (available from Zenodo:
Methods: Quantifying MVL Loss
Table of Contents
We quantified global MVL loss driven by human expansion and natural disasters using a combination of remote sensing data, spatial analysis, and change detection methods.
(1) Human Expansion-Driven MVL Loss
We utilized expansion of impervious surface,cropland,and mining areas as indicators of human impact. Two global mining datasets were leveraged, encompassing large-scale and artisanal/small-scale mining activities. These datasets comprised polygons covering all ground features related to mining (open cuts, tailing dams, waste rock dumps, water ponds, processing infrastructure) derived from high-resolution satellite imagery (≤10m resolution in Google Earth and Sentinel-2) around 2020 22,23. accuracy assessments yielded overall accuracies of 88.3% (Maus et al.22; https://doi.org/10.1594/PANGAEA.942325) and ~92% (Tang et al.23; https://zenodo.org/records/7894216).
The two datasets were merged using union analysis in ArcGIS 10.8, resulting in a combined dataset of 81,962 polygons covering 120,413 km2 (Supplementary Fig. 9).This was then converted to a 30m resolution grid image.Spatial overlay analysis was employed to address potential pixel overlap between impervious surfaces,cropland,and mining areas,prioritizing mining data due to its superior spatial resolution and accuracy.
MVL loss was quantified according to the following:
(i) Human settlement growth: Expansion of impervious surface between 2000 and 2020 was identified using a pixel-level change detection method. MVL loss driven by settlements represents MVL pixels in 2000 converted to impervious surface by 2020 (fig.1a).
(ii) Agriculture expansion: Cropland expansion between 2000 and 2020 was identified using a change detection method. MVL loss driven by agriculture represents MVL pixels in 2000 converted to cropland by 2020 (Fig. 1a).
(iii) Mining: Spatial overlay analysis identified intersections between mining areas and MVL. MVL loss driven by mining represents MVL pixels in 2000 converted to mining surface by 2020 (Fig. 1a).
(2) Natural Disasters-Driven MVL Loss
Multi-source datasets of wildfires, floods, landslides, and drought events occurring between 2000 and 2020 were collected to track MVL loss due to natural disasters. For wildfires…
## Methods
We used the Normalized difference vegetation Index (NDVI) to identify areas experiencing Mountain Vegetation Loss (MVL). NDVI is calculated from the red and near-infrared reflectance of land surfaces, with values closer to 1 indicate dense vegetation coverage, while values closer to 0 suggest very little vegetation, and values less than 0 indicate no vegetation101,102. We collected all available global Landsat images (TM/ETM + /OLI, 30-m resolution) in 1999 (the year prior to our focus) and 2021(the year after our focus) from the United States Geological Survey (USGS) (
We quantified MVL losses driven by four natural disasters: wildfires, floods, landslides, and droughts. to do so, we used a consistent spatial overlay and NDVI thresholding approach. We first delineated MVL in 2000 and intersected them with respective disaster footprints (burned/flooded/landslide/drought areas). If the NDVI at the end of the study period in a given pixel is less than or equal to 0.1, while it was greater than 0.1 at the start of the period, these pixels are labelled as MVL loss104,105(Fig. 1a).### The MVL loss in separate ten world regions
To examine the spatial pattern of MVL in different geographical zones, we divided the world into ten separate regions by merging national administrative boundaries (i.e, Sub-Saharan Africa, Southeast Asia, Middle East, South Asia, Central Asia, East Asia, North America, Latin America, Europe & Russia, and Oceania) (Supplementary Fig. 2). We then quantified the MVL loss (including total MVL loss and MVL loss types, Fig. 2) and the contribution of drivers (Fig. 2) in these ten regions through spatial analysis (zonal statistics tool of ArcGIS) between global MVL loss layer and the regions (Supplementary Table 2).
### The MVL loss within PAs and AHRTMS
to quantify MVL loss within protected areas (PAs) and areas with high richness of threatened mountain-occurring species (AHRTMS), we first mapped mountain PAs and AHRTMS using global PA distributions and the IUCN Red List of threatened species. We collected the global distribution of PAs from the WDPA (version 2023.11,
The AHRTMS were designated as critically important biodiversity habitats based on the number of mountain-occurring threatened species (mammals, amphibians, reptiles, birds and plants) they contain. Our AHRTMS was not fully consistent with the global database of IUCN’s key biodiversity areas106 because we only focus on mountain areas that represent areas of the highest richness of threatened mountain-occurring species. To identify AHRTMS, we used the IUCN Red List of threatened species.“`html
Mapping Human Settlement Growth and its Impact on Vegetation Loss
Human expansion,driven by factors like population growth and economic development,is a primary cause of vegetation loss globally.Accurately mapping this expansion and quantifying its impact on natural landscapes is crucial for effective environmental management and sustainable planning. Recent research leverages high-resolution satellite imagery and advanced labeling techniques to reliably track human settlement growth and its correlation with vegetation loss, offering a powerful tool for understanding and mitigating these changes.
Understanding the Methodology
The core of this research relies on identifying and classifying impervious surfaces – areas covered by structures like buildings,roads,and parking lots – as a proxy for human settlement growth. This approach utilizes satellite imagery from the years 2000 and 2020, allowing for a clear comparison of land cover changes over a two-decade period. The process involves several key steps:
Data Acquisition and Labeling
High-resolution satellite imagery is essential for accurately identifying impervious surfaces. Researchers meticulously labeled samples based on a combination of factors, including the proportion of impervious surface area (compared to 2000 and 2020) and the ratio of artificial versus natural surfaces within those areas. A sample was considered “correct” if it represented an impervious surface in the high-resolution imagery.Conversely, samples depicting other land types (forest, grassland, cropland, etc.) were labeled as “incorrect.”
Accuracy Assessment
To ensure the reliability of the methodology, an accuracy assessment was conducted. Uncertain samples, representing only a small percentage (2.0%) of the total, were excluded from the analysis. The results demonstrated a high degree of accuracy, with approximately 92.4% of identified impervious surfaces being correctly classified. Notably, the vast majority of these impervious surfaces (85.9%) were artificial, while the proportion of natural surfaces like bare rock or soil was relatively small. This reinforces the validity of using impervious surfaces as a reliable indicator of human settlement growth.
Assessing MVL Loss Driven by Human Expansion
Beyond simply identifying settlement growth, the research also focused on quantifying the associated vegetation loss, termed “Moving Vegetation Loss” (MVL). Samples were labeled as “correct” if they represented areas covered by vegetation (forest, shrubland, or grassland) in 2000 but were converted to impervious surfaces, cropland, or mining areas by 2020. Incorrectly labeled samples were those that remained vegetated or underwent other land cover changes. This allows researchers to directly link human expansion to specific instances of vegetation loss.
Why Impervious Surfaces are a Reliable Proxy
The choice of impervious surfaces as a proxy for human settlement growth is based on several key advantages:
- Clear Definition: Impervious surfaces are easily identifiable in satellite imagery due to their distinct spectral characteristics.
- Direct Correlation: The presence of impervious surfaces directly indicates human activity and development.
- Long-Term Stability: Once established, impervious surfaces are relatively stable over time, providing a reliable indicator of past and present settlement patterns.
Implications and Future Directions
this research provides a robust and accurate method for mapping human settlement growth and its impact on vegetation loss. The high accuracy rate (92.4%) of identifying impervious surfaces strengthens the reliability of this approach. The findings have important implications for:
- Environmental monitoring: Tracking changes in vegetation cover and identifying areas at risk of further loss.
- Urban Planning: Informing sustainable urban development strategies and minimizing the environmental impact of expansion.
- Conservation Efforts: Prioritizing conservation efforts in areas experiencing rapid human settlement growth.
Future research could focus on integrating this methodology with other data sources, such as socioeconomic data and climate models, to gain a more extensive understanding of the drivers and consequences of human-induced vegetation loss. Further refinement of labeling techniques and the exploration of machine learning algorithms could also enhance the accuracy and efficiency of this approach.
Key Takeaways
- High