Supporting forest conservation and restoration policies - A Mediterranean forest types map

The Mediterranean basin is the third-richest hotspot in the world in terms of plant biodiversity [1] and one of the greatest sources of endemic plants on Earth [2]. Its plant diversity accounts for 25 000 plant species [2,3], 60 percent of which are endemic [4], of which more than 100 tree species are recorded in Mediterranean forests [5]. It is estimated that the region has more than 25 million hectares of Mediterranean forests and about 50 million hectares of other Mediterranean wooded lands [1].  

Forest and wooded lands are, at the same time, one of the most threatened ecosystems in the Mediterranean because of the large population concentration (537 million people censed on 2015 [6]) and the high pressure from human activities, such as tourism, urbanization, agriculture, etc. [7]. Finally, the Mediterranean is one of the most affected regions by climate change, worldwide [8,9]. Due to the human alterations of the landscape along millennia, only 5 percent of natural vegetation in the Mediterranean basin remains, the lowest of any hotspot [10].

Many international agreements (i.e. the Convention on Biological Diversity [11] –CBD–; United Nations Framework Convention on Climate Change –UNFCCC [12]– or the European Green Deal [13]) point out the need of preserving and restoring forests to tackle the Sustainable development Goals [14] –SDG–, the EU forest strategy [15] and the EU Biodiversity Strategy for 2030 [16].

For the conservation of forests and the identification of potential priority areas of restoration, first, it is essential to quantify and delineate them precisely. Forest area and tree species composition are, indeed, two of the seven indicators for Sustainable forest Management [17] and an important aspect of Ecosystem Accounting [18]. Forest cartography can help in understanding the state and evolution of forests, while identifying the drivers and impacts on forests in a spatial context.

To date, we count with several maps of forests globally and at the regional scale of the Mediterranean Basin, at different spatial and temporal scales [19,20,21]. These maps, produced from remote sensing data, discern between open and dense forests, and generic forest types, such as evergreen, broadleaved, coniferous or mixed forest. In addition, national and supra-national space institutions have facilitated open access data and AI tools for environmental monitoring, such as the ESA Copernicus Sentinels.

Here we present the results of a methodological workflow that classifies Mediterranean forest types based on dominant species to better address the needs of many different stakeholders, such as decision makers, forest managers, land developers and researchers. The usefulness of this type of maps increases highly if produced yearly, to monitor changes in land cover, to track drivers of change in the landscape and to check effectiveness of measures taken about conservation and restoration. This map of Mediterranean forest types, based on the dominant species, was developed in the framework of the EnBIC2Lab-project [22], framed on a cooperation between remote sensing, big data and Mediterranean forest experts. The aim of the present map is to create a regional baseline with comparable information and similar accuracy across all countries in the Mediterranean basin, by using data and a methodology that covers the region equally, independently of the existence/accessibility or not of national inventories and databases.

In our presentation we tackle the challenges of mapping Mediterranean forests, because of the diversity and complexity of this landscape. Mixed forests and open forests are frequent in the Mediterranean area, which are difficult to map using reflectance only. For this reason, in this project we modelled the variables that define the Mediterranean forest characteristics using diverse remote sensing data. Two main data sources were used:  ESA Copernicus Sentinel-2 MSI [23] and the NASA’s ASTER Digital Elevation Model (DEM) [24]. The biophysical properties of different forest types, defined by the dominant tree species, were defined by the multispectral data of Sentinel-2 MSI. Imagery of different seasons helped in the discriminating deciduous from perennial species, but also in separating species based on different phenology traits, such as plant productivity or flowering. The use of a DEM and the derived slope and aspect layers assisted in identifying the bioclimatic vegetation stages and species with different slope and illumination requirements [25,26,27]. Other layers such as texture supported in differentiating forests from other classes, such as shrublands, tree crops and plantations. A layer of distance to rivers extracted from the Global River Classification –GloRiC– [28] aided on defining riparian forests.

The relevance of the different remote sensing layers were evaluated by means of regression analysis and principal component analysis, per forest type. Once these layers were identified, forests were classified with a combined Artificial Intelligence technique, involving Random Forests and Regression Analysis extrapolation.

One of the major challenges of the project was to feed the models with reliable training samples. Several databases were filtered and harmonized to produce a database of forest types based on dominant species, co-occurrence of species and tree coverage. Among them, data was retrieved from the European Vegetation Archive [29] –EVA–, national inventories (Spain [30], Italy [31] and Lebanon), and expert’s counselling for the North-African countries [32].

Future work involves the use of Sentinel-1 data to improve the classification of certain forest types and other land cover classes, such as shrublands, plantations and tree crops.