Understanding the linkages between between climatic and surface properties that influence water uptake and loss by vegetation is essential for understanding the impact of drought on dryland regions. The Normalized Difference Vegetation Index (NDVI) is a common metric used to identify vegetation condition across LULC types. Here we employ empirical dynamic modeling (EDM) to forecast NDVI changes for savannas, grasslands, and croplands across East Africa at a dekadal (10-day) time scale using satellite-derived environmental forcing variables. The model relies on state space reconstruction with lagged coordinate embedding of multiple time series observations to recover the dynamic environmental system that links vegetation dynamics to environmental forcing. We apply convergent cross mapping based on Takens’ Theorem to detect the impact of landcover on directional causal interactions and time delays between driving (e.g. LST, rainfall) and response variables (NDVI). The model brings to light that certain regions are highly consistent in their trajectories and therefore easier to project while other regions are more dispersive and thus more difficult to determine anomalies. In terms of land cover, we are able to make projections with high accuracy for grasslands out to 6 months ahead while croplands and savannas show reduced forecast skill overall and prove less useful after 3 months. The use of historical seasonal NDVI patterns to diagnose the manner by which landcover and land use determine climate-land surface couplings provides a means for defining critical areas of inquiry related to the impacts of future change, particularly the expansion of agricultural areas. In addition, the EDM approach provides a robust means for creating short term vegetation forecasts across LULC types in East Africa. These predictions can assist relief organizations in advising drought management, declaring food security classifications and providing early response to famine.

Access to groundwater leaves riparian plants in drylands resistant to atmospheric drought but vulnerable to changes in climate or water use that reduce streamflow and groundwater tables. Despite the vulnerability of riparian vegetation to water balance changes few extensible methods have been developed to automatically map riparian plants at the scale of individual stands or stream reaches, to assess their response to changes in moisture due to drought and climate change, and to contrast those responses across plant functional types. We used LiDAR and a sub-annual timeseries of NDVI to map vegetation and then assessed drought response by comparing a drought index to variation in a remotely sensed metric of plant health. First, a random forest model was built to classify vegetation communities based on phenological changes in Sentinel-2 NDVI. This model produced community classes with an overall accuracy of 97.9%; accuracy for the riparian vegetation class was 98.9%. Following this initial classification, LiDAR measurements of vegetation height were used to split the riparian class into structural subclasses. Multiple Endmember Spectral Mixture Analysis was applied to a timeseries of Landsat imagery from 1984 to 2018, producing annual sub-pixel fractions of green vegetation, non-photosynthetic vegetation, and soil. Relationships were assessed within structural subclasses between mid-summer green vegetation fraction (GV) and the Standardized Precipitation-Evapotranspiration Index (SPEI), a measure of soil moisture drought. Among riparian vegetation subclasses, all groups showed significant positive correlations between SPEI and GV, indicating an increase in healthy plant material during wetter years. However, the relationship was strongest for herbaceous plants (R^2=0.509, m=0.0278), intermediate for shrubs (R^2=0.339, m=0.0262), and weakest for the largest trees (R^2=0.1373, m=0.0145). This implies decoupling of larger riparian plants from the impacts of atmospheric drought due to subsidies provided by groundwater resources. Our method was extended successfully to multiple climatically-dissimilar dryland systems in the American Southwest, and the results provide a basis for ongoing studies on the fine-scale drought response and climatic vulnerability of riparian woodlands.