Data from: Landsat-based greening trends in alpine ecosystems are inflated by multidecadal increases in summer observations
Data files
Aug 08, 2024 version files 3.19 GB
-
DISTRIB_CLUST.zip
-
LandsatObservations.zip
-
README.md
Abstract
Remote sensing is an invaluable tool for tracking decadal-scale changes in vegetation greenness in response to climate and land use changes. While the Landsat archive has been widely used to explore these trends and their spatial and temporal complexity, its inconsistent sampling frequency over time and space raises concerns about its ability to provide reliable estimates of annual vegetation indices such as the annual maximum NDVI, commonly used as a proxy of plant productivity. Here we demonstrate for seasonally snow-covered ecosystems, that greening trends derived from annual maximum NDVI can be significantly overestimated because the number of available Landsat observations increases over time, and mostly that the magnitude of the overestimation varies along environmental gradients. Typically, areas with a short growing season and few available observations experience the largest bias in greening trend estimation. We show these conditions are met in late snowmelting habitats in the European Alps, which are known to be particularly sensitive to temperature increases and present conservation challenges. In this critical context, almost 50% of the magnitude of estimated greening can be explained by this bias. Our study calls for greater caution when comparing greening trends magnitudes between habitats with different snow conditions and observations. At a minimum we recommend reporting information on the temporal sampling of the observations, including the number of observations per year, when long term studies with Landsat observations are undertaken.
README: Landsat-based greening trends in alpine ecosystems are inflated by multidecadal increases in summer observations
https://doi.org/10.5061/dryad.1rn8pk13f
Datasets described in this repository are obtained from public data only (Landsat Constellation or MODIS).
Description of the data and file structure
There are two datasets in this repository described below:
There is one TIF for vegetation clusters in the European Alps, and one TIF per year (1984 to 2021) of Landsat clear-sky observations (37 files). Years are indicated in the filename. All files have the same extent
(1) Raster of vegetation clusters in the European Alps
To characterize vegetation phenology, we used the 250-m resolution 8-day composite of MODIS MOD09QA/Terra collection 6 products over the entire European Alps (N = 1,076,872 pixels). We assembled tiles h18v4 and h19v04 to cover the entire mountain range and reprojected red (RED) and near-infrared (NIR) surface reflectance values for high quality pixels (according to the MOD09QA Quality Control flag) and calculated NDVI according to (NIR-RED)/(NIR+RED). We implemented a Partitioning Around Medoids (PAM) model for N = 30,000 NDVI time series using a prescribed number of K = 4 clusters. The model was then used to calculate the distances between medoids, and the entire dataset and each pixel of the European Alps was assigned to the closest medoid. We interpreted the three vegetation clusters as early, intermediate, and late snowmelt sites based on their elevation distribution and growing season length.
(2) Rasters of Landsat clear-sky observations
We extracted annual time series of the number of clear-sky and snow-free Landsat observations during the growing season (OBSGS) from GEE by applying the CFmask to remove snow, clouds, and cloud shadows from Landsat 5 TM, 7 ETM+ and 8 OLI images (Collection 2) between day of year 152 to 243 (Zhu and Woodcock, 2012) over the European Alps. This dataset include 37 rasters, one per year.
Methods
The data available in this repository corresponds to (1) vegetation cluster distribution in the European Alps used as an example for computation; (2) rasters of Landsat clear-sky observations use to build the null model over the European Alps.