Global Forest Mask for 2023

This application provides the visualization of global forest mask corresponding to the year 2023 at a ground sampling distance of 10 m, covering a latitude range from 57° S to 67° N. The dataset is available for download from the Zenodo repository https://doi.org/10.5281/zenodo.15741437 as well as from a public AWS S3 bucket (further instructions on Zenodo).

This is a derived product based on the estimation of Canopy Height (CH) and Canopy Cover (CC) by our unified deep learning model as described in the publication

Weber, M.; Beneke, C.; Wheeler, C. Unified Deep Learning Model for Global Prediction of Aboveground Biomass, Canopy Height, and Cover from High-Resolution, Multi-Sensor Satellite Imagery. Remote Sens. 2025, 17, 1594. https://doi.org/10.3390/rs17091594

which provides the details on the model approach and extensive evaluation metrics.

We use the widely accepted definition of forest by the UN Food and Agriculture Organization (FAO), requiring CH > 5 m and CC > 10% [1]. The forest mask (FM) is therefore defined as the output of the operation

FM = [(CH - CH_sd) > 5 m] & [(CC - CC_sd) > 10%]

We subtract one standard deviation from the source variables, as estimated by the model, in order to compensate for the slight over-estimation at low values of CH and CC resulting in a more conservative classification of forest. Further details are given in the publication mentioned above.

[1] Food and Agriculture Organization. (2000, November 2). FRA 2000 on definitions of forest and forest change (FRA Working Paper No. 33). Forest Resources Assessment Programme. Rome. Retrieved from FAO website https://www.fao.org/4/ad665e/ad665e00.htm