New high-resolution land use and land cover maps are expected to improve climate models

What is the total area of trees, crops, or buildings on Earth's surface? This knowledge is necessary for estimating the greenhouse gases that are released into the air and predicting how the planet's climate will change in the coming years. However, the spatial resolution of maps containing this information is often too low, which does not allow accurate tracking of deforestation or small-scale agricultural and urban growth and leads to significant errors. CREAF researchers Lluís Pesquer and Cristina Domingo have been involved in a project lead by the European Space Agency (ESA) that has released new land use and land cover maps with a resolution ten times higher than current ones. They are expected to improve the accuracy of climate predictions.

The new land use and land cover maps have a resolution ten times higher than the previous maps.

The project has focused on three territories particularly relevant to predict how climate changes, such as the Amazon rainforest, the Sahel drylands and the transition zone between taiga and tundra in Siberia, respectively related to deforestation, desertification and melting ice. The new maps enhance their spatial resolution from 300 to 30 metres and display land cover changes in these areas every 5 years, starting from 1990 to 2019. In addition, a 10-metre static mapping of 2019 has been produced over a wider area across 3 territories.This level of detail makes it possible to classify into very precise categories the types of land cover on plots of up to 100 square metres, revealing roads, buildings, streams and deforested areas hitherto undetectable by the 300-metre maps. Based on the results of these areas and the needs of the climate modellers, new potential study areas in Asia, the Philippines or Malaysia are being evaluated.

Interactive map showing the three zones mapped over the Earth's surface. Source: ESA Climate

These maps automatically classify whether an area is covered by trees, bushes, crops or buildings and, more importantly, allow these areas to be quantified. In addition, they make it possible to analyse historical changes in the same area and compare the current situation with that of 5, 10 or 20 years ago and establish patterns.

The high-resolution land cover maps are one of the products of ESA's Climate Change Initiative, which is developing global data records on the 26 Essential Climate Variables. These variables are the physical, chemical or biological factors that are essential to characterise the Earth's climate, such as surface temperature, ozone or sea level. Each is a necessary ingredient for making large-scale climate forecasts and informing the international community.

Maps of the 3 most influential territories on climate

Deforestation in the Amazon is causing the rapid replacement of tree cover by crops and grasslands, resulting in soil erosion, decreased photosynthesis, and decreased carbon stocks. For instance, thanks to these new high-resolution maps it is possible to distinguish the presence of new roads and observe how in the years following their construction, the forest disappears in favour of crops.

“What happens in the Amazon not only affects those who live there, which would be significant, but it influences the entire planet. It is a key area for predicting climate changes through atmospheric circulation models."

LLUÍS PESQUER, CREAF researcher in the GRUMETS group, specialized in remote sensing.

The case of the Sahel, a transitional territory between the Sahara desert and the savannah, is particularly relevant for climate models, as it affects the dynamics of the Monsoons. Recurrent droughts have been observed to displace vegetation with limited adaptability, resulting in the conversion of scarce scattered forests into scrublands. Likewise, some areas of scattered vegetation end up being used for crops and then urbanised. "It's a dog chasing its tail, since extreme weather conditions are the main trigger for changes in land cover which, in turn, impact the climate again," reports Lluís Pesquer, who believes this process echoes the aridification issue we encounter in southern Spain.

In Siberia, on the other hand, it is key to detect how the increase in temperatures associated with climate change causes the loss of permafrost, which realises large amounts of carbon into the atmosphere and contributes very negatively to the greenhouse effect and, again, on climate change. The northward movement of this ice is accompanied by the movement of forest and scrub in the same direction. "This area is especially problematic due to the lack of optical images from satellites due to the high cloud cover and few hours of daylight during half of the year, which we try to compensate with radar images," says Cristina Domingo, also a CREAF researcher and member of the GRUMETS group.

Clouds, the great enemy

The high resolution improvement comes at a great computational and economic cost. Artificial intelligence is used to generate these maps, by classifying a huge bank of remote sensing data. "The system is firstly trained manually indicating which areas are crops, trees or buildings, so that the classifier ends up making an autonomous decision. The more data you provide, the better," Domingo points out. Not having enough images means the AI doesn't have the information it needs, resulting in less accurate classification. The more you want to go back in time, the more problems arise due to the limitations in technology at that moment, making it difficult to obtain maps prior to 2015. Still, knowing what the situation was like, for example, in 1990 is very essential for studying global change.

“The system is firstly trained manually indicating which areas are crops, trees or buildings, so that the classifier ends up making an autonomous decision. The more data you provide, the better.”

CRISTINA DOMINGO, CREAF researcher in the GRUMETS group.

Occasionally clouds block the view of optical satellites, making it difficult for researchers to see.

On top of that, clouds are the great enemy that hinders the researchers' task. Optical satellites orbit the Earth above the clouds, which sometimes block their view. The current satellites of the Sentinel fleet that have sufficient spatial resolution acquire images every 4 days and only since 2016, when they were put into orbit. "A sensor that takes images every day, even when there are clouds, ends up finding a small window of time to look through. If you get a picture every two weeks, as was the case before 2016, and it's cloudy, you'll be left with nothing until the next opportunity. There may be periods of more than half a year without images", Pesquer explains.

Comparison of the change in resolution between the old land use and land cover maps (right, 300 meters) and the new ones (left, 10 meters). Images adapted from ESA.

To solve the clouds issue, images from these optical satellites are often complemented by radar satellite measurements. Although they can see through the clouds, they only produce a texture map instead of an image. This may not be sufficient on its own, but it can help to differentiate between a forest or a plantation, for example. Moreover, when examining past periods, pictures from other kinds of optical satellites can sometimes be combined to create high-resolution maps.

Simulating changes in climate

At the moment, the researchers aim to disseminate the results of this first phase of the project and receive feedback from climate modellers. Some simulations have already been carried out by project partners using the ORCHIDEE model. Preliminary evidence suggests that these new maps enhance the accuracy of water and energy balance calculations. High-resolution mapping of other parts of the Earth will be continued based on the needs of these modellers.