| Abstract |
Accelerating global biodiversity loss, along with the deterioration of critical ecosystems and the release of greenhouse gases, together demand a radical reconsideration of existing responses. We need to better measure the impact of human activities on biodiversity and ecosystems globally, while also tracking shifts in the spatial distribution of habitats and organisms as a result of climate change. Existing systems are not up to the task: they aggregate fragmented and misaligned data derived from expert opinions and field inventories, and most digitisation efforts cannot model the hugely complex dynamics involved nor address these problems at scale. Hence, policymakers and the private sector cannot accurately identify the potential impact of critical decisions such as where and how to focus conservation activities, while still supplying food, wood, energy and water to people. We propose a planetary insights system for policymakers, business and scientists that can radically improve decision-making for achieving biodiversity conservation and restoration goals alongside meeting human resource needs. We will draw from computer, ecological, conservation and remote-sensing science to combine extensive earth observation data (satellites, drones) with terrestrial data (sensor networks, citizen science, habitat maps) to create a new "Terra" model of the world using self-supervised AI training. Terra will provide predictive properties for very many terrestrial plant and animal species with sparse data and enable monitoring of biodiversity metrics at fine spatial and temporal scales and at much higher resolution and accuracy than is currently possible. We will use Terra to build a suite of critically important predictions about life across our planet: - Deep species distribution modelling will classify changes in global land use and the plants and animals that live there, providing much-needed insights for policymakers to factor into their decision making. We plan to use Terra to create high-resolution, regularly updated databases that accurately classify habitats for wild species and agriculture crop and pasture types, providing a unified classification and change map. - Linking data on long-running as well as contemporary changes in species' area of suitable habitat together with information on their habitat preferences will allow us to model how human actions impacts extinction risks, and crucially where and how these might be most efficiently mitigated. These new datasets exist for ~30K species of terrestrial vertebrates; we plan to also apply these methods, for the first time, to first ~10k and then ~150k data-deficient plant species. These datasets will be ultimately used to analyse the impacts of human food consumption on nature with much greater accuracy and breadth than currently possible. Our system will combine Terra with additional information about the world, such as supply chain datasets about how food travels from field to fork, to identify the spatial distribution of the threat to biodiversity from a unit of consumption of each of a set of agricultural commodities. This will, for the first time, allow decision makers to accurately analyse the impact of their food production, procurement distribution and processing decisions on biodiversity worldwide, and do so on an evidence-driven basis from planetary scale observations that are transparent and reproducible. |
