Computational protocol: Idiosyncratic Responses of High Arctic Plants to Changing Snow Regimes

Similar protocols

Protocol publication

[…] Since the data was collected in a hierarchically organized experimental set up, we used linear mixed-effects models for analysis. We assumed a unimodal relationship between time and plant size since plants usually do not grow cumulatively throughout the vegetation period but with a peak during early to mid-season, followed by a decline of live plant tissue due to senescence and leaf drop. In the analysis, we hence fitted a second order polynomial of each of the nine different temperature variables to the size measurements () of each species separately, with random effects for fence area, fence (i.e. plot), subplot and individual. We selected the best among the nine temperature models per species based on the Akaike Information Criterion (AIC), and the one with the lowest AIC was chosen. We then sequentially removed each term of the selected full model and compared which of the reduced models was the best fitting one (i.e. with the lowest AIC) and then used that to predict temperature sum needed until (and magnitude of) maximum plant size for each species and snow regime by determining the peaks of the hump-shaped functions.Treatment effects on (1) melt-out dates, (2) the different temperature variables, and (3) the average size of the species throughout the season were also evaluated by means of mixed effects models with the same group structure as defined above. Potential heteroscedasticity was considered for as much grouping levels as possible (i.e. parameter estimation algorithms converted). With respect to plant size, these analyses were conducted for every species separately as well as for all species together to test for generic trends. In the latter case, we scaled all species-specific size measurement to a common range between 0 and 1 and added growth form and habitat association as predictor variables, as well as species identity as an additional (highest) group variable. Each model was then sequentially reduced and the one with the lowest AIC selected. All analyses were conducted with R using the packages nlme and lattice . […]

Pipeline specifications

Software tools lme4, nlme
Application Mathematical modeling
Diseases Inert Gas Narcosis