The impact of inter-annual variability in remote sensing time series on modeling tree species distributions

Predictions of species occurrence as indicators of ecosystem integrity are of high relevance for decision-makers in conservation biology, invasive species' management, and climate change research. Remote sensing data can serve as valuable input for Species Distribution Models (SDMs) since they provide information on current habitat conditions and disturbance factors besides bioclimatic suitability which is commonly derived from climatic data. However, little is known about the usefulness of multi-temporal remote sensing data in general for modeling species distributions and the related effects of inter-annual variability on the extent and accuracy of modeled distribution ranges. This study investigates the above-mentioned questions for two tropical tree species, Brosimum alicastrum and Liquidambar macrophylla, in Mexico. From the MODIS 16-day vegetation index product (MOD13A2), 18 annual phenological metrics (time-related, NPP-related and seasonality-related) were computed for the period from 2001 to 2009 and combined to a set of multi-year average values (covering 3, 5, 7, and 9 years). The results show that inter-annual variability has a significant impact on model predictions and that models based on longer composite periods show less deviance from observed species presence-absence field data.