Process-Based Spatiotemporal Simulation and Prediction of Remote Sensing Phenology

Xiaoqiu Chen

DOI: https://doi.org/10.1007/978-3-662-49839-2_8

2017-01-01

Abstract:To predict vegetation phenology under global climate change scenarios on continuous and large geographical ranges, satellite-derived start date of season (SOS) retrieved from AVHRR third generation NDVI data from 1982 to 2006 were fitted using UniForc and UniChill models at all pixels in the deciduous broadleaf forest region of northern China. The pixel-specific optimum models were then validated using daily mean air temperature and SOS data from 2007 to 2011. Further, the pixel-specific optimum models were employed to reconstruct SOS time series from 1950 to 2005 and predict SOS time series from 2006 to 2100 across the research region based on the calibrated historical daily mean air temperature data (1950-2005) and daily mean air temperature data under scenarios of the Representative Concentration Pathway (RCP) 4.5 and 8.5 (2006-2100) from outputs of the Regional Climate Model HadGEM3-RA. Because the UniForc model performs better in simulation efficiency than the UniChill model at 96.4 % of pixels, it was determined as the optimum model at these pixels. The temporal validation of the pixel-specific UniForc models during 2007-2011 confirmed the feasibility of the UniForc model in predicting SOS date in most parts of the research region. The predicted regional mean SOS dates show a significantly advancing trend during 2006-2100 at rates of 0.65 days (under RCP 4.5) and 1.79 days (under RCP 8.5) per decade. Taking the mean SOS date during 19611990 as the reference value, an overall advancement of decadal mean SOS date was detected in each decade during 2006 and 2100. The advancement amplitude and the spatial standard deviation (under RCP 8.5) of decadal mean SOS dates would increase with time. At the end of 21st century, the predicted regional mean SOS date under the RCP 4.5 and RCP 8.5 scenarios would be 11.8 and 20 days earlier than the regional mean SOS date during 1961-1990, respectively. Meanwhile, the spatial patterns of spring vegetation phenology under global warming would likely turn into more uncertain than at present. Furthermore, SOS dates at each pixel show also an advancement in each decade during 2011-2100 under both RCP 4.5 and RCP 8.5 scenarios. The advancement amplitude of decadal mean SOS dates under the RCP 8.5 scenario would be larger in southern parts of the deciduous broadleaf forest region than in northern parts. Thus, an enhanced sensitivity of spring phenology response to air temperature would be expected under future climate warming, especially in colder areas or higher latitudes where the greatest climate warming is predicted.