Forest maps are an essential tool for forest management. They help in understanding the distribution, expansion and health of forests and they give spatial and temporal context to the drivers of forest degradation and potential nature-based solutions for restoration. However, at the global and regional levels, the existing sources of forest cartography present several limitations. The definition of forests is generally too generic (broad classes of forest cover, or distinction only on coniferous, deciduous and mixed forests) and the scale is too broad. In addition, the map accuracy might me insufficient depending on the methodology used and the availability of field truthing data. Mediterranean forests are very diverse in terms of tree species, forest types and tree density. They are generally composed of more broadleaf trees and mixed stands, often with a lower tree density than in other temperate and boreal forests. Moreover, the Mediterranean region is highly affected by human impacts and climate change. In this context, a map of Mediterranean forest types, based on dominant species, with a high temporal and spatial resolution and high map accuracy is needed to support forest management at the local scale. At a regional scale, this map supports conservation and restoration policies, such as the Convention on Biological Diversity, the UN Framework Convention on Climate Change, the European Green Deal, the Sustainable Development Goals, and the EU Forest and Biodiversity Strategies for 2030. In the current frame of open-source Earth Observation satellite Big Data and the development of Artificial Intelligence (AI) for massive data storage and analysis, it is feasible to generate forest maps for the entire Mediterranean region on a yearly basis and at a resolution of few meters. To achieve a good map accuracy today, the bottleneck is the forest data to feed the models. There is a need of harmonizing forest databases to achieve maps at the global and regional levels with a homogeneous accuracy along with the mapped territory. In addition, the data must reach remote sensing requirements. The map is based on Sentinel-2 multispectral imagery, NASA/JAXA ASTER Digital Elevation Model and derived thematic layers. More than 80.000 forest samples were gathered and curated for feeding the models, including approximately 100 tree species into 30 forest classes. Spectral separability analysis was used to confirm the suitability of the ecological description of forest types into remote sensing classification. This data comes from forest databases of several sources, which needed an extensive work on harmonization, as they come in different formats, collect different variables and handle different forest type’s definitions. Several National Forest Inventories (Spain, Tunisia and Lebanon) were used, together with some databases from the European Vegetation Archive (EVA). Therefore, we acknowledge the potential benefit in creating standards at European and regional levels for National Forest Inventories, for their use in remote sensing applications.

This work was developed in the framework of the EnBIC2Lab project, funded by the EU LifeWatch ERIC program, in collaboration with FAO Silva Mediterranea, Medforval network, the European Environmental Agency (EEA) and the EEA-Eionet.