Vegetation dynamics in a climate change hotspot: trend analysis in a Spanish dehesa

      Introduction The Mediterranean region has been identified as a hotspot of climate change characterized by a large tree mortality. Extended drought periods, shifts in rainfall patterns, and increasing water stress are probably the main drivers. Especially holm (Quercus ilex L.) and cork oak trees (Quercus suber L.) in high-value and nature-based agroforestry systems (in Spain known as dehesa) have multiple positive effects on the microclimate, carbon storage, erosion prevention, increase of soil water content, and soil nutrient concentration, for example. With their positive effect on wind velocity, they are also considered the last natural barrier protecting the Iberian Peninsula and Central Europe from desertification processes advancing from North Africa.
     Objective We assume that wrong management, biotic causes like pests and diseases, and especially water stress are responsible for a decreased resilience of oak trees. Our goal was to analyse the vegetation dynamics with the help of the Normalized Difference Vegetation Index (NDVI) time series as an indicator for greenness and vitality. In particular, we focused on the trend of NDVI over about two decades.
    Material and Methods We have selected eight plots (250 m x 250 m) with different topographical conditions and analysed an 18 years long NDVI time series (2003 - 2020) from MODIS (MYD13Q1). To extract the trend, we decomposed the time series into trend, seasonal component, and the high-frequency remainder. Subsequently, we did the Mann-Kendall test on the trend component to determine whether the trend is significant. Since environmental time series are rarely linear or stationary, many statistical decomposition methods are not suitable to produce physically meaningful results. Therefore we used the data-driven method Complete Ensemble Empirical Mode Decomposition with adaptive Noise (CEEMDAN) by Torres et al. 2011.
     Results Depending on the topographical conditions of the plot, we were able to extract different NDVI trend signals from the time series. The NDVI values on the north-facing plots were larger than on the south-facing plots. The extracted trends were positive and significant (p <0.01). The seasonal component corresponded to the expected annual cycle.
      Conclusion In order to assess vegetation dynamics, NDVI time series can be regarded as a good starting point, although one indicator alone does not allow to make final conclusions about vegetation changes. The purely data-driven decomposition method CEEMDAN avoids strong assumptions about the shape of the trend.