Periodic Reporting for period 2 - TOFDRY (Trees outside forests in global drylands)
Periodo di rendicontazione: 2022-05-01 al 2023-10-31
Drylands cover approximately 65 million km² of the Earth’s land surface but their tree and shrub cover is a major unknown in terrestrial research. This is because a large proportion of dryland trees grow isolated without canopy closure and most scientific and non-scientific interest is devoted to forests, while the density and size of trees outside of forests is not well documented. However, these non-forest trees play a crucial role for biodiversity and provide ecosystem services such as carbon storage, food resources, livelihoods and shelter for humans and animals. The limited attention devoted to the quantification of dryland trees leads to an underrepresentation of non-forest trees in development strategies and climate\vegetation models, and the economic and ecological importance of non-forest trees is largely unknown at large scale.
This project will work towards a wall-to-wall identification of trees in global drylands, and study their ecological services and socio-environmental determinants. The breakthrough is that trees are not assessed as canopy fraction of an area, but as individuals, allowing to identify not only their coverage but also their density, crown size, and key ecological services. The project applies a new generation of satellite imagery at sub-meter resolution and extensive field data in conjunction with fully convolutional neural networks, a deep learning technique being able to identify objects within imagery at an unprecedented accuracy. In doing so, the project lays the groundwork for new insights into the contribution of human agency and climate change to the distribution of dryland trees and their role in mitigating degradation, climate change and poverty.
This project will work towards a wall-to-wall identification of trees in global drylands, and study their ecological services and socio-environmental determinants. The breakthrough is that trees are not assessed as canopy fraction of an area, but as individuals, allowing to identify not only their coverage but also their density, crown size, and key ecological services. The project applies a new generation of satellite imagery at sub-meter resolution and extensive field data in conjunction with fully convolutional neural networks, a deep learning technique being able to identify objects within imagery at an unprecedented accuracy. In doing so, the project lays the groundwork for new insights into the contribution of human agency and climate change to the distribution of dryland trees and their role in mitigating degradation, climate change and poverty.
- Several deep learning based frameworks to detect single trees and estimate their height and biomass have been developed. Those can be used with sub-meter satellite or aerial imagery, and with PlanetScope/RapidEye imagery. The frameworks have been made publicly available and the methods are documented in several publications.
- We have set up storage and data processing systems to deal with the large amounts of data.
- PlanetScope images have been purchased, downloaded, quality filtered and mosaiced at a global scale. The year 2019 has been completed, more years are in preparation.
- Vast amounts of publicly available aerial and satellite images as well as lidra point clouds have been collected and processed (about 500 tb). The data cover all major continents.
- Tree detections based on PlanetScope data have been run and validated for Europe, North America, China, Africa and India. This was published in Nature Communications.
- In depth tree biomass analyses based on sub-meter images have been conducted for the Sahel, Rwanda, South Africa, Finland and Denmark. For the Sahel, we have mapped the biomass for 15 billion individual trees. This was published in Nature and Nature Climate Change.
- All baobab trees in the Sahel have been mapped over an area of 1.5 million km2 using sub-meter satellite images. The study has been submitted.
- Field work has been conducted in Rwanda, Africa.
- Socio-environmental variables determining distribution and properties of trees have been studied for Africa. The study has been submitted.
- A framework has been developed to study temporal dynamics at the tree level, this work is ongoing.
- Urban trees have been studied at the tree-level for all Chinese cities from 2010-2019 using RapidEye and PlanetScope. The study has been submitted.
- The direct outcomes of TOFDRY have been published in Nature, Nature Climate Change, Nature Communications (2x), PNAS Nexus, and Science Advances (accepted). All these publications are gold open access. We also made codes, data, and frameworks available.
- We have set up storage and data processing systems to deal with the large amounts of data.
- PlanetScope images have been purchased, downloaded, quality filtered and mosaiced at a global scale. The year 2019 has been completed, more years are in preparation.
- Vast amounts of publicly available aerial and satellite images as well as lidra point clouds have been collected and processed (about 500 tb). The data cover all major continents.
- Tree detections based on PlanetScope data have been run and validated for Europe, North America, China, Africa and India. This was published in Nature Communications.
- In depth tree biomass analyses based on sub-meter images have been conducted for the Sahel, Rwanda, South Africa, Finland and Denmark. For the Sahel, we have mapped the biomass for 15 billion individual trees. This was published in Nature and Nature Climate Change.
- All baobab trees in the Sahel have been mapped over an area of 1.5 million km2 using sub-meter satellite images. The study has been submitted.
- Field work has been conducted in Rwanda, Africa.
- Socio-environmental variables determining distribution and properties of trees have been studied for Africa. The study has been submitted.
- A framework has been developed to study temporal dynamics at the tree level, this work is ongoing.
- Urban trees have been studied at the tree-level for all Chinese cities from 2010-2019 using RapidEye and PlanetScope. The study has been submitted.
- The direct outcomes of TOFDRY have been published in Nature, Nature Climate Change, Nature Communications (2x), PNAS Nexus, and Science Advances (accepted). All these publications are gold open access. We also made codes, data, and frameworks available.
- The detection of single trees and their biomass at continental scale breaks new ground. The maps set a new standard on tree cover and biomass.
- The developed methods go beyond previously published techniques and advance the field from a computer scientific perspective.
- We have used tree crown allometry to estimate the biomass of billions of trees at national scales.
- In the coming years, a new heatmap based detection method will replace the previously used segmentation method.
- We will develop improved biomass estimation techniques that do not need allometric models.
- Results will be expanded to a global scale. A temporal depth will be added and changes will be monitored at the tree level and continental scale. Observed changes will be linked with human management and climate change.
- Information on tree/forest types will be added to understand economic and ecological values of trees.
- The spatial distribution of tree cover will be further studied to better understand transition zones.
- The developed methods go beyond previously published techniques and advance the field from a computer scientific perspective.
- We have used tree crown allometry to estimate the biomass of billions of trees at national scales.
- In the coming years, a new heatmap based detection method will replace the previously used segmentation method.
- We will develop improved biomass estimation techniques that do not need allometric models.
- Results will be expanded to a global scale. A temporal depth will be added and changes will be monitored at the tree level and continental scale. Observed changes will be linked with human management and climate change.
- Information on tree/forest types will be added to understand economic and ecological values of trees.
- The spatial distribution of tree cover will be further studied to better understand transition zones.