Community assembly following disturbance is a key process in determining the composition and function of the future community.  However, replicated studies of community assembly at whole ecosystem scales are rare.  Here, we describe a series of whole-lake experiments in which the recovery of zooplankton communities was tracked following an ecosystem-scale disturbance, i.e., application of the piscicide, rotenone.  Using a BACI design, fourteen lakes in eastern Washington were studied: seven lakes were treated with rotenone, while seven lakes acted as reference systems.  Each lake was monitored up to six months before and one to two years after the rotenone treatments.  Zooplankton samples and environmental measurements were collected approximately monthly from each lake. Community responses following disturbance were assessed using metrics of abundance, diversity, and community composition, as well as taxonomic group abundance.  Zooplankton recovery was also assessed using species traits related to habitat, feeding mode, trophic level, body size, and life history.  In addition to patterns of recovery, potential mechanisms were explored relating to abiotic conditions, biotic interactions, and traits. There were steep declines in the abundance (average across years: 99%) and diversity (average across years: 75%) of the zooplankton community following rotenone treatment.  Although abundance had recovered by the second year of the study, community diversity had not fully recovered after two years.  Communities from rotenone lakes appeared to be compositionally recovered within about eight months following disturbance.  Cyclopoid copepods were typically the first group to recover, and remained dominant for a few months, whereas cladocerans recovered more slowly, typically within ~6-7 months following rotenone. Calanoid copepods were not fully recovered two years after rotenone treatment.  Traits related to body size and feeding mode were associated with the zooplankton communities following rotenone treatment. We failed to observe significant spatial synchrony in recovery patterns of zooplankton across lakes, though we did observe significant synchrony of zooplankton taxonomic groups within lakes.  These findings suggest that traits related to ecological function, and to a lesser extent, biotic and abiotic factors, as well as characteristics of the disturbance itself, may be important in helping to understand recovery processes. 

Data were collected between 2014 and 2016 in a set of 14 lakes in eastern Washington state.  Seven lakes served as references, while seven lakes were treated with rotenone in either fall 2014 or fall 2015.  Monthly samples were collected for water quality parameters (water temperature, dissolved oxygen, specific conductance, pH, Secchi depth; the first four parameters at 1-m intervals from the deepest spot in the lake) and crustacean zooplankton (vertical integrated tows from the bottom to the surface; from three locations). In July of each sampling year, an integrated water sample of the epilimnion was collected for nutrient analysis.  Crustacean zooplankton samples from each date-lake combination were combined, weighting by volume of net tows, and were identified and enumerated using microscopy.  See paper for additional details.  Fish stocking densities were obtained from the Washington Department of Fish and Wildlife.

Zooplankton were analyzed with several metrics, including abundance, taxonomic group abundance (calanoids, cyclopoids, cladocerans), Shannon-Wiener diversity, adult species density, and trait composition. Surface measurements of dissolved oxygen, specific conductance, and pH were used in analyses.

Lakes that were treated with rotenone in 2014 and 2015 were analyzed in separate models (referred to as 2014 rotenone and 2015 rotenone), in order to account for the differences in treatment years.  Two linear mixed effects (LME) models were run for 2014 rotenone lakes: year one and year two following treatment.  The 2014 rotenone year one model contrasted the four rotenone lakes with the seven reference lakes from June 2014 to September 2015; whereas the year two model again used pre-impact data from June to September 2014 contrasted with post-impact data from October 2015 to September 2016.  Samples from reference lakes could therefore be classified as both “after treatment” (2014 treatment, year 1) but also “before treatment” (2015 treatment, year 1).

When monthly samples were not taken, an empty line of NA values are included so that figures show gaps appropriately.

A README file is included that describes the R scripts and associated datasets.