2. Model selection
In the full model, interactions between filter pore size and water temperature and between target gene and water temperature were statistically significant (both P < 0.05), and effects of the filter pore size and interaction between fragment size and water source were marginally significant (both P < 0.1) (Table 2). All supported models resulting from model selection included the effects of filter pore size, target gene, and water source, whereas the effects of fragment size and temperature were uncertain, owing to their small coefficient and large SE. However, we focused on the effects of the interactions among variables; all supported models included interactions between filter pore size and temperature (Figure 1) and between target gene and temperature (Figure 2). In addition, 11 of the 13 models included the interaction between target gene and water source (Figure 3), and four models included the interaction between filter pore size and water source (Figure S1). Other interactions were included in less than three supported models, and the uncertainties of the corresponding coefficients were relatively large.
Although DNA fragment size was included in most supported models, its effect was relatively small due to its high variability (Table 2). Considering the smaller number of eDNA decay rate constants targeting longer DNA fragments as mentioned previously, we instead assessed the relationship between the eDNA decay rate and shorter DNA fragment size (<200 bp). Consequently, the fragment size was found to have a significantly positive effect on the decay rate (P < 0.01; Figure S2).