A global forest reference set with time series annual change information from 2000 to 2020

Abstract

ABSTRACT Mapping and monitoring forest with time-series remote sensing methodologies requires reference data constructed in a sampling framework. Few worldwide reference data sets with time-series information are currently available for large areas. We produced a global forest reference set at 30 m with time-series information from 2000 to 2020 based on a stratified random sampling scheme. All available Landsat, high-resolution Google Earth images and other relative land cover/land change products were used in a visual interpretation approach. This reference dataset contains 10339 sample units (6252 persisting forest sample units, 2049 change sample units and 2038 persisting non-forest sample units) attributed with annual forest/non-forest information from 2000 to 2020. The results of our analysis highlight many cases of undetected forest change due to definition and other factors, and suggest that change of forest canopy cover could be monitored instead of land cover types. The reference set can be potentially used for many large-area applications such as validation of forest mapping projects or as auxiliary data for forest monitoring analysis. And this reference set will be available online (https://doi.org/10.5281/zenodo.5524258).