Bridging Multiple Lines Of Evidence To Quantify Plant Phenology And Assess Links To Dryland Ecosystem Function

The clear and pressing need to reliably identify and predict shifts in plant phenology at landscape scales requires a critical link between mechanistic understanding of climate drivers and broad scale forecasts of plant responses to climate change. A multi-scale phenology study co-located with two eddy covariance towers was initiated on the Jornada Basin LTER in New Mexico in 2010 to bridge phenology patterns at the plant level with those representing aggregated signals at the landscape level. The study integrates phenology observations collected in the field along with those collected via remotely using imagery from phenocams, unmanned aerial vehicles (UAVs), and satellite sensors along with estimates of carbon flux. We applied the Breaks for Additive Seasonal and Trend (BFAST) time series algorithm to MODIS 250-m NDVI greenness index values to partition the NDVI signal into components representing the long-term trend, seasonal periodicity, and residuals and identified significant shifts in the NDVI signal (i.e., "breaks"). Previous work verified breaks representing significant deviations from the BFAST seasonal and trend models using field-estimated plant biomass collected between 2000 and 2014. We subsequently examine estimates of fractional cover by functional group derived from UAV images acquired 2010 through 2015. At a mixed grassland site, the BFAST algorithm detected four breaks in the trend model denoting significant increases in NDVI in May 2004, July 2006, and March 2010 and a significant decrease in May 2012. The 2004 and 2006 breaks corresponded to herbaceous vegetation responses to rainfall following prolonged periods of drought. The 2012 decrease in NDVI corresponded to the marked reduction of herbaceous biomass following an exceptionally dry period in late 2010-2011. Seasonal breaks representing changes in the timing and magnitude of NDVI identified in July 2006 and September 2008 coincide with rapid increases in production of annual species in 2006 and perennial grasses in 2008. Combining extensive spatially-explicit UAV depictions of land surface characteristics and geographically constrained long-term plot data yielded compelling evidence to support ecologically meaningful interpretations of BFAST break points from MODIS NDVI in this water-limited grassland ecosystem.