Assessing the evolution of tree cover in Cantabria,1957-2020
A quantitative approach based on Artificial Intelligence
In the inner circles of forest specialists is a well-known secret that forests are expanding across Europe. The question is "To what degree?" Using AI, and particularly Deep Learning techniques, we try to shed some light about what is really happening in Cantabria, something that could also be the case in similar parts of Europe.
From long ago, in the inner circles of forest specialists is a well-known secret that forests are expanding -at least in some parts of Europe- in the last few decades, an idea to set against the general misbelief spilling over many parts of the Spanish and European society who actually think it is shrinking.
Societal changes such as the abandonment of traditional livestock and farming practices, along with the introduction of active reforestation policies, the increase of wood production for different industries and also ecological disturbances such as change in the temperature and precipitation patterns, can be responsible for a potential skyward shift.
We want to know and empirically prove the rate of change, where it happens across our region, and share our insights with other specialists and the whole society in order to better understand the processes happening in the territory.
So let's get started with kind of unpacking a little bit of what has been done! We will go fast and straightaway to what we've seen.
Tree cover in 2020
Deep Learning models have been trained to map the tree coverage using high definition orthorectified imagery captured back in 2020:
Tree cover in 1957
It is also possible to apply the same methods and techniques for imagery captured in different timeframes. Particularly, in Spain orthoimagery is available from the decade of 1940s -and with more quality in the 50s- thanks to the aerial surveys conducted in those days by the U.S. Army Map Service. That information has been used to train similar models of computer vision capable of automatically delineate the tree cover, with a similar result:
Side by side comparison
Both sets of imagery are orthorectified and technically have all the properties of a traditional map in terms of constant scale and controlled deformations according to the projection and reference system used. All the geospatial information showed uses EPSG:25830 as coordinate system, composed of ETRS89 as Reference System (based on the adoption of ellipsoid GRS80 among other parameters), and UTM in zone 30 north as projection system. So we can compare one with each other:
Gains vs losses: who wins?
Now, the losses are going to be plotted within the 2020 map (in red) and the gains are going to be also included the 1957 map (in light green). The result is astonishing: there are gains all over the place. The tree cover is expanding to a degree totally unexpected:
If we do the math, the tree cover has more than doubled. It has multiplied by a factor of 2,19x actually.
The surface covered by trees has more than doubled in 63 years. More precisely, it has multiplied by a factor of 2,19x. Please bear in mind that the surfaces and data of this work can differ with the official numbers derived from the National Forest Map of Spain, for a number of reasons:
- First, let us stress that we are not taking into account some forestry plantations oriented to the industrial use (typically stands of Eucalyptus globulus) that in the 2020 image appear as recently cut, due to its productive cycle. These spaces have not lost its ultimate forest use, but our models do not detect them as such (intentionally, because our target is to know the net areas effectively covered by trees in two different points in time).
- Different scale of detail. This study tries to detect every single tree and small stands, while the National Forest Map works with another scale of detail, and applies an adequate tessellation to segment the territory according to that scale.
Nonetheless, the data revealed by this study point in the exact same direction that the records derived from National Forest Map of Spain, but understanding that both are different things, created with different methodologies, is an important idea to come across.
For this reasons, we rather not to talk about specific quantities of areas covered in this study and use multiplication factors and graphical comparisons instead.
Spatial patterns
So, it has been demonstrated now that the tree cover is expanding all over Cantabria, but let's try to find out whether or not hidden spatial trends and patterns remain to be revealed. In order to do this, the data has been simplified and generalized using an hexagonal mesh. The total amount of areas of gains and losses falling in each of the hexagons have been added. The result let us see the data from another perspective and dimension:
Some interesting ideas can be derived from the image above:
- There is a notable increase in tree cover in the north east sector of the region, where two factors converge:
- First, the proliferation of multiple plantations of Eucaliptus globulus and other subspecies in Guriezo and Castro-Urdiales.
- Second, this north-east sector is also the domain of the Holly oak (Quercus ilex) also known as Evergreen or Holm oak, which has experienced a wide expansion, probably because the constant extractions of firewood stalled and started to decrease at some point in time, and perhaps also due to the warmer conditions and the increase of soil dryness due to a certain decrease in rainfall in the last few decades.
- Another important idea is a critical advance in the south of the region (mainly in Valderredible). There we see an huge expansion of woodlands of Pyrenean oak (Quercus pyrenaica) and the progression of Scotch pine (Pinus sylvestris) frequently in plantations, some of them abandoned nowadays but in pretty good health. This south part of Cantabria is pretty close to the boundary of the Mediterranean Biogeographical Region, in the transition zone, while the rest of the region belongs to the Athlantic European, Eurosiberian region.
- In Liébana (west extreme of the region) a lot of changes can also be seen, especially in the lower part of the valleys, where the temperatures are mild during most of the year. It is also worth noting again that low -and accessible- areas suffered a sustained extraction of firewood during the 40s and 50s, something that can be also applied, up to a point, to the higher parts of the woodlands throughout the entire region.
- In general, the lower parts of all the valleys present some kind of increase, probably because in this places the abandonment of the traditional livestock and farming practices has tanken place in a more intense way than in any other part, which creates better conditions for the expansion of the forest.
- Along the coast, plantations of Eucaliptus globulus and other subspecies are scattered some kilometers inland. This has been one of the main drivers of the tree expansion in Cantabria, especially since the Sniace Corporation, dedicated to the production of cellulose and rayon, established in Torrelavega back in 1946.
The losses are focused on the mid part of the region and almost all of them are related with recently cut plantations of Eucaliptus globulus for industrial production. As been said before, this areas have not been identified by the Deep Learning models intentionally.
Some final thoughts about AI
This has been one small example of how AI, and specifically Deep Learning, can help us to progress in many fields of knowledge. In the middle of all the noise and marketing of this crazy world of ours, AI is probably not fully delivering on its promise for solving the complex problems of society, but that being said we do firmly believe that this technology will change the world in the years to come. It’s just a baby, that needs to grow up. And mature...
Somebody once said in that great novel and then movie: “This was just a first step. In time, you'll take another (…) Small moves, Ellie, small moves”.