Advances in the availability of satellite earth observation data and processing tools allow for near real-time monitoring and study of ecosystems.

Unfortunately, most research is focused on forests, with little attention given to “open” non-forest ecosystems such as the shrublands, woodlands and grasslands that dominate most of Sub-Saharan Africa and much of the world.

These ecosystems contain a substantial proportion of the world’s biodiversity and organic carbon, among other important variables. Near-real time tools can be a game-changer for the sustainable management and stewardship of these ecosystems and their benefits to society.

Detecting abnormal change in forests is usually straightforward because they are typically relatively stable over timescales from years to centuries. Open ecosystems are far more challenging, due to noisy natural dynamism such as strong seasonality, high sensitivity to rainfall and fire.

Detecting changes in the state of Fynbos ecosystems

A team of SAEON researchers recently published an innovative approach that overcomes these difficulties in Fynbos by using a model to generate forecasts of the expected vegetation signal observed by satellite under natural conditions. This allows us to detect changes in the state of Fynbos ecosystems by identifying areas where the observed vegetation signal has deviated from the modelled expectation for healthy natural ecosystems.

The core model is a mathematical function that allows one to estimate the shape of the expected post-fire recovery trajectory (over roughly 10 to 50 years) and seasonality for a given site (see Figures 1 & 3). The exact shape of the curve varies predictably among sites based on their climate, topography and soils and is estimated from the observed satellite record and the environmental data.

Knowing the shape of the curve allows us to predict the expected natural range in vegetation signal for any given site so long as we know the environmental parameters (climate, topography and soils), time of year and fire history. While this core model is specific to ecosystems with multi-year fire cycles, similar models can easily be developed for other open ecosystems.

Testing our approach on the Cape Peninsula found that it is useful for identifying a range of change agents such as fire, alien plant species invasions, drought, pathogen outbreaks and clearing of vegetation. We have developed a prototype interactive web application (Figure 2), but the next step is to fully automate the workflow and work with end users (conservation managers, environmental compliance officers, etc.) to develop this into a full production near-real time change-detection system.