Remotely Sensed Resilience for the Prediction of Droughtinduced Forest Decline

Organised by Laboratory of Geo-information Science and Remote Sensing

Wed 14 November 2018 09:00 to 09:30

Venue Gaia, gebouwnummer 101
Room 1

By Sophie Stuhler (Germany)

Although rates of tree mortality and forest decline are rising globally, there is still a lack of understanding the drivers and insufficient ability to reliably model the occurrence. Previous work has identified climatic factors and used a recent decline event in Catalonia to study the effect of soil moisture as seen from space and the effect of species. The research at hand identified Early Warning Signals as resilience indicators from vegetation indices of satellite time series to assess the stability of the forest ecosystems prior to this event on a large scale. Three different vegetation indices were used for the extraction of resilience indicators, each sensitive to a different type of plant functionality: NDVI to leaf chlorophyll content, NDMI to leaf water content and EVI to chlorophyll content and canopy structure. Strong trends of decreased resilience were found in areas with high forest decline occurrence. The resilience indicators were added to the existing logistic regression model and their significance, model fit and performance were reported. Overall, the best model explained about 31% of the deviance in the data out of which an additional 4% was explained by the resilience indicators. The most explanatory power within the Early Warning Signals was found in indices describing temporal autocorrelation when extracted from NDVI. However, the single best performing resilience indicator was the trend in spatial variance extracted from NDMI which explained about 1.3% of the deviance in forest decline by itself.