This repository contains code to create an experimental design for the Bromeliad Working Group (BWG). Bromeliads are flowering plants native to the Neotropics. These plants collect rainwater in their leaves; this rainwater forms the habitat for a diverse community of invertebrates. This tiny ecological community depends entirely on that rainwater. But rain will change with the world's changing climate: becoming more or less intense, and/or falling in heavier storms than normal. How will these changes affect the structure and function of the bromeliad food web?
The BWG conducted a field experiment manipulating both the amount and distribution of rainfall reaching bromeliads. We replicated this experiment in seven countries, each with its own community of invertebrates and its own rainfall regiem. In order to manipulate the amount and distribution of rainfall independently, we wrote code which
- fits a statistical distribution to observed rainfall data
- modifies the parameters of this distribution in a factorial combination
- approximates these distributions within the 60-day time limit of the experiment, giving the number of dry days and rainy days (and the amount of rain on each)
- finally, arranges these rainy and dry days into a "plausible" natural sequence.
The following links contain an illustrated detailed description of this algorithim as applied to each experimental site. Below is a guide to the resulting schedules themselves (intended for use by the different fieldsite teams).
For a discussion of the challenges in fitting extreme values of parameters, see here
This is a quick guide to the precipitation experiment schedule file. Files will be named FIELDSITEschedule.csv, where FIELDSITE will be replaced with the name of your country or site (e.g. Cardososchedule.csv, FrenchGuianaschedule.csv) The description follows every column name:
trt.name = treatment name, following the naming convention suggested by Diane. Treatments are named after the two parameters of the negative binomial distribution (mu and k). Names are in the form “muXkY”, where X and Y are numbers which multiply mu and k, respectively, in different treatments.
mu = the actual parameter values used in this field site. The “mu1k1” treatment was calculated from the original data, the others were derived from mu1k1 by multiplying those values of mu and k by the range of factors agreed upon by the working group (mu: 0.1, 0.2, 0.4, 0.6, 0.8, 1, 1.5, 2, 2.5, 3 and k: 0.5, 1, 2)
k = as above, but for k
temporal.block = after intended.mu and intended.k were calculated, the treatments were divided into three random groups; these groups begin on different days (the 1st, 2nd and 3rd days of the experiment). The grouping was tested with an ANOVA to confirm that there is no significant difference in intended.mu or intended.k.
day.1-day.68 = these columns represent “days from beginning of experiment”. They do not represent calendar dates. Each column gives the amount of water in mm that should be added to each plant. Remember to multiply these by the catchment area of your bromeliad species, and by the correction factor for the canopy, as described by Diane and her team in Costa Rica. Therefore, treatments are applied to bromeliads over a period of 68 days, not including the preparation of bromeliads beforehand, the calculation of correction factors, and the final destructive sampling at the end. Plants receive no water on days marked both NA and 0. (NAs represent days added by the experimental design, for example the temporal blocks, in contrast to 0s which are predicted by our model of rainfall). Additionally, plants are not watered on days marked “sample” or “insects”. These represent days set aside for the collection of protist samples (sample) and destructive sampling for macroinvertebrates (insects).
Looking along each row (rather than down each column), the pattern of treatments for each bromeliad is as follows:
-
0–2 days delay, depending on the temporal.block (marked NA)*
-
1 day median water of the mu1k1 treatment
-
sample the protist community (no water on this day)
-
60 days of treatment watering
-
1 day treatment median
-
sample the protist community again (no water on this day)
-
one NA day with no watering or sampling
-
destructively sample the bromeliad for all macroinvertebrates (insects)
-
0–2 days delay, depending on temporal.block*
*The number of these days always sums to 2 (i.e. they are either at the beginning or end of the experiment, depending on temporal.block)
Thus the full treatment period takes 68 days.
Note that the NA days at the end (i.e. on days 67 or 68) are simply included to keep the schedule looking “square”.
We release the code in this repository under the MIT software license. see text of license here