Control for spatial autocorrelation
In common garden studies the spatial autocorrelation, or the probability
that individuals growing closer together are more similar, of samples
must be taken into account (Stopher et al., 2012). To account for
spatial patterns within each of our gardens (Clatskanie and Corvallis),
we used a thin-plate spline method (Blumstein et al., 2020; Evans et
al., 2014) via the fields (9.6) (Nychka, Furrer, Paige, & Sain,
2017) package in R . This method fits an interpolated surface to
the garden, which uncovers regions of each site that significantly
differ from the mean. To correct these patterns of spatial concordance,
we take the residuals from the thin plate spline and add them back to
the model intercept, thus removing spatial trends and placing sample
values back on a biologically meaningful scale. We did this for each of
our metrics independently; sugar concentration, starch concentration,
total nonstructural carbohydrate (TNC) concentration, the proportion of
starch (starch / TNC), and diameter at breast height (DBH).