Treating floodplain lakes of large rivers as study units for variables that vary within lakes; an evaluation using chlorophyll a and inorganic suspended solids data from floodplain lakes of the Upper Mississippi River

Abstract

Contiguous floodplain lakes ('lakes') have historically been used as study units for comparative studies of limnological variables that vary within lakes. The hierarchical nature of these studies implies that study variables may be correlated within lakes and that covariate associations may differ not only among lakes but also by spatial scale. We evaluated the utility of treating lakes as study units for limnological variables that vary within lakes based on the criteria of important levels of among-lake variation in study variables and the observation of covariate associations that vary among lakes. These concerns were selected, respectively, to ensure that lake signatures were distinguishable from within-lake variation and that lake-scale effects on covariate associations might provide inferences not available by ignoring those effects. Study data represented chlorophyll a (CHL) and inorganic suspended solids (ISS) data from lakes within three reaches of the Upper Mississippi River. Sampling occurred in summer from 1993 through 2005 (except 2003); numbers of lakes per reach varied from 7 to 19, and median lake area varied from 53 to 101 ha. CHL and ISS levels were modelled linearly, with lake, year and lake x year effects treated as random. For all reaches, the proportions of variation in CHL and ISS attributable to differences among lakes (including lake and lake x year effects) were substantial (range: 18%-73%). Finally, among-lake variation in CHL and ISS was strongly associated with covariates and covariate effects that varied by lakes or lake-years (including with vegetation levels and, for CHL, log(ISS)). These findings demonstrate the utility of treating floodplain lakes as study units for the study of limnological variables and the importance of addressing hierarchy within study designs when making inferences from data collected within floodplain lakes.