Analyzing savannah vegetation phenology with remotely sensed data, lagged time-series models and phenopictures

It is predicted that savannah regions will see changes in precipitation patterns due to current climate change pro-jections. The change will most likely affect leaf phenologywhich controls net primary production. It is thereforeimportant to; 1) study those changes and its drivers, 2) to be able to correctly model the changes to vegetationphenology due to climate change. To our knowledge there is no existing global savannah phenologymodel thatcan capture both the phenologicalevents and the vegetation state between the events. We therefore, investigate howday length, mean annual precipitation and soil moisture affects and controls the vegetation phenologyof savannahs(using MODIS NDVI as a proxy for phenologicalstate) with a lagged time series model for global application. Wefurthermore use phenologicalpictures (phenopictures) to investigate savannah tree and grass phenology. Phenopic-tures are pictures taken with a digital time-lapse camera with the purpose of recording and studying phenologicalevents. We used climate data from 15 flux towers sites located in 4 continents together with normalized differencevegetation index from MODIS for the model development. Two of the sites located in Africa were further ana-lyzed using phenopictures. The developed model identified all three considered variables as usable for modellingof savannah leaf phenologybut showed some inconsistent result for some of the sites indicating the difficultiesin creating a simple common model that works equally well across sites. We attribute some of these difficultiesto site specific differences (e.g. grazing or tree and grass ratio) that the simplified model did not consider. Butwe expect it to on average give the cross-validated result (r2= 0.6, RMSE = 0.1) when applied to other savannahareas. The preliminary analysis of the phenologicalpictures with respect to tree and grass to some extent supportthis by showing differences in the start of the leaves development in the beginning of the season. However, thisdiffered between the two studied sites which further highlights the difficulties in creating a common model thatworks equally well for individual sites.