However, only 50% of
pixels simultaneously exhibited significant trends (p-value < 0.01)
for EVI_mean, MAP, and MAT, so we eventually decided to
include all Sen’s slopes in the LMM analysis regardless of their
significance. Both tests were run using the ManneKendall package
available through the MATLAB Central file exchange (http://www.
mathworks.com, accessed September 2009). Second, to build the
LMMwe considered the slope of EVI_mean trends as the dependent
variable. Independent variables were the trends in MAP, the trends
in MAT, vegetation type, lithological substrate, and their first order
interactions as fixed effects; and meteorological stations as random
effects.