Near-Real-Time Forecasting and Change Detection for a Fire-Prone Shrubland Ecosystem
Ecosystems provide essential goods and services, and maintaining their health is of crucial importance to meet the Sustainable Development Goals (SDGs). Shrubland ecosystems, where trees may be present but not dominant, make up >40% of the global total ecosystem organic carbon and contain a substantial proportion of the world’s biological diversity. Advances in Earth Observation, in situ measurements, and ecological models allow the development of near real-time monitoring tools that report on the state and changes in vegetation, the ecosystem functions, processes and services they drive, and the pressures they face. Unfortunately, to the best of our knowledge, no operational tools of this nature exist for dynamic shrubland ecosystems that are prone to fire and other natural disturbances, short-term (event-driven) variability, and long-term trends.
This project proposes to develop an operational system and tools for monitoring the vegetation state of a fire-prone shrubland ecosystem. Our study region is the Cape Floristic Region of South Africa, which contains 20% of Africa’s plant diversity and is a Global Biodiversity Hotspot and UNESCO World Heritage Site. Our main end-user organization is the South African Environmental Observation Network (SAEON), which will further facilitate the adoption of our developed tools by other government departments and conservation organizations in the region. The indigenous vegetation in the study region is threatened by climate change, human-induced habitat loss, and invasion by alien species. Due to a lack of readily available, reliable, and up-to-date information on the state of the ecosystem, our end users face challenging decisions about where to allocate extremely limited human and other resources for the management and conservation of protected areas. This project will combine Earth observations (e.g., NDVI from Landsat and MODIS), in situ observations, and ecological forecasting models to characterize the spatial and temporal variation of the vegetation state (including structure, productivity, natural disturbance dynamics, and seasonal phenology) for near real-time monitoring and change detection in the study region.
We will produce an operational system and tools which leverage iterative ecological forecasting and deep learning to predict natural land surface processes, evaluate near-real-time changes in the vegetation state, and classify the factors causing abnormal vegetation changes. The produced system and tools can support the decisions of our end users by providing various types of near real-time information, such as unmapped fires, locations where vegetation has been damaged or cleared, areas where alien plant species are invading indigenous vegetation, and the locations where high plant mortality has occurred. The system and tools will be deployed on SAEON’s computational infrastructure for operational forecasts and monitoring. Our developed system and tools are readily extensible to similar ecosystems (e.g. the Californian Chaparral, Australian Kwongan, and parts of the Mediterranean Basin), and can be adapted to ecosystems with different dynamics. This project focuses on subsection 2.1 of the ROSES call on measuring and monitoring protected area outcomes.
Major components of the project workflow. Our primary end-user is SAEON, but they will serve as a liaison to other provincial, national, and international organizations.
The map below illustrates how this approach can identify anomalous recovery trajectories on the Cape of Good Hope, South Africa.