Materials and Methods
Quantitative samples of benthic macroinvertebrates from NEON wadeable-stream sites from 2017-2019 were downloaded to determine size-spectra relationships (National Ecological Observatory Network 2020). NEON is a National Science Foundation (USA) funded program which collects standardized samples from 81 sites (24 streams) across a variety of terrestrial and aquatic ecosystems in North America. Repeat collections include automated instrument recordings and observational field sampling throughout the year, and data are available as open source data products (https://data.neonscience.org/home). The stream sites range across a broad environmental gradient (Figure 1), spanning from 18 to 68° N latitude (majority between 33 to 45° N), mean annual temperatures of -4 to 25° C, and mean annual precipitation of 331 to 2530 mm (Table 1). Additionally, sites are in varying biomes and terrestrial plant communities, and subject to different local geologies and flow regimes.
Sites were sampled 1-4 times each year across the local growing season. Most sites had at least two years with three samples, except for COMO which only had one year with three samples (2018), and two samples collected in both 2017 and 2019. Additionally, three sites only had data available from a single collection in a single year; REDB in 2018, WALK and WLOU both in 2019. Details of sample collection and processing protocols are available at the NEON website. Briefly, macroinvertebrate samples were collected from a known area using the sampling method most suited to a site. Laboratory processing included subsampling for taxonomic identification and size class measurement (nearest mm) and estimating the total count per sample. Estimated total counts were standardized to individuals per m2 by dividing by the area sampled (Chesney 2019). Macroinvertebrate size classes (mm) were converted to individual dry mass (M, in mg) using published length-weight regression coefficients. Most (96%) taxa had taxon-specific length-weight regressions and these were used for model development and assessment. Approximately 8% of the observations were flagged by NEON as being damaged, affecting their length measurements, and were removed from the data. The final data set included length measurements for >82,000 individuals and counts that totaled >17 million individuals.
Size spectra slopes have been estimated using a variety of methods, with binning methods being common in the published literature (White et al. 2008; Sprules & Barth 2015; Edwards et al. 2017). However, recent comparative studies have shown binning methods to be inaccurate, providing biased slope estimates, and recommend using maximum likelihood methods (Edwards et al. 2017). We fitted body size data (dry weight estimated from length-weight regressions) of macroinvertebrates from each collection to a bounded power law distribution with a probability density function:
\begin{equation} f\left(x\right)=\ \frac{\left(b+1\right)x^{b}}{x_{\max}^{b+1}-x_{\min}^{b+1}},\ \ b\ \neq 1\nonumber \\ \end{equation}\begin{equation} f\left(x\right)=\ \frac{1}{\log x_{\max}-logx_{\min}},\ \ b=1\nonumber \\ \end{equation}
Where x is body mass, b is the scaling exponent (size spectra slope), and the distribution is bounded by the minimum (xmin ) and maximum (xmax ) body sizes observed in a collection. Maximum likelihood methods were used to estimate the exponent, b , using code modified from Edwards et al. (2017). Using these methods, a more negative bexponent represents a steeper slope of the size spectra. Slopes can be steeper due to either a relative increase in the abundance of small body sizes, or a relative decrease in the abundance of large body sizes, or a combination of the two. Regardless, steeper slopes represent a relatively smaller proportion of large individuals compared to small individuals within a community.
In addition to the size spectra slopes, we also examined how total community biomass varied across the collections. First, we multiplied the estimated biomass of individuals by their estimated density to calculate mg of dry mass per m2 for each sample within a collection (n = 8 samples per collection). The 8 samples per collection were used as individual observations in the hierarchical Bayesian analyses (see Statistical Analyses ).
Macroinvertebrates were generally sampled three times within the growing season from each site. However, due to the wide geographical distribution of sites, there was significant variation in the sample dates. For example, samples in Alaska (latitude 65-68° N) were collected between May and August, whereas samples in Alabama (latitude 33° N) were collected between February and November. Because we are interested in the effects of temperature on size spectra, we used the mean annual temperature in degrees Celsius for each site.