Scalgo Live land cover map expands to France
- New releases
France is next in our steady roll-out of new nationwide land cover maps. With a 25 cm resolution and 11 distinct land cover classes, this is the most detailed map of France available.
In early 2024, we released the first land cover map for Germany, following on from Denmark, Sweden, Finland and Norway. The recent release of the France land cover map shows new advancements in the Scalgo AI model, which we have been developing since 2019, when we made the first map in collaboration with Aarhus University.
“Our methodology is very powerful because the model keeps learning and getting better every time we apply it,” Jonas Tranberg, data scientist at Scalgo, says. “And the first results we got for France, using the AI model that had been trained on the Nordic countries and Germany, were very accurate. The new thing we trained the model to identify was dry vegetation, since the vegetation in the south of France dries out in summer to a degree that is uncommon in northern Europe.”
Land cover has immense impact on the amount of runoff from a surface, and the map’s high resolution provides valuable information for local as well as regional analysis to ultimately enable improved planning of surface water flows.
Powerful AI methodology
The new map distinguishes between 11 land cover classes in 25cm resolution (we are, in fact mapping the whole world in 25cm resolution, one country at a time).
Artificial intelligence (AI) was used to develop the map, and we are using a variation of the UNET model which has been used to solve many segmentation problems, including tumour detection in MRI scans and wildfires from satellite images.
As Jonas explains, “We create the AI output by alternating between training and evaluation phases, starting from a model trained on orthophotos outside of France to identify areas where performance is insufficient. Next, we add new training data from these areas and re-train. We repeat the process until the model is sufficiently accurate, which entails at least 95% accuracy on randomly sampled locations.”
Finally, the AI output is combined with the BD TOPO data from IGN describing buildings, roads, railroads and water, and RPG data describing agricultural fields, to further refine the map.
Using the map for improved outcomes
Over the past few decades, the degree of surface artificialisation (land that humans have ended up transforming) has increased, although many policies have been aimed at halting or reversing this trend. In the France land cover map, we have combined the land cover map with the official administrative boundary maps for France to provide detailed land cover statistics.
As a result, you can see the current degree of artificialisation in your arrondissement, commune, department or region. What commune has the highest forest coverage? Where would you find the greatest density of buildings? How are the most environmentally efficient regions planned? This is all invaluable data you can access via the map’s analyses and tools.
In addition, we calculated the land cover statistics within all watersheds, large or small. So now when you use the interactive Watershed tool and query the watershed at any point along any flow path in France, the info box includes a breakdown of the land cover within the watershed.
With this new land cover integration in Scalgo Live, you can also easily assign a runoff function to each of the land cover classes or to groups of classes using the infiltration modules in the flash flood map. It’s also easy to add your own land cover classes for special cases such as, for example, a green roof or a pervious parking lot.
