Data analysis:
We considered Bd to have been detected in a sample if ≥1 PCR technical replicated tested positive for frog skin swabs. For eDNA samples, we considered Bd to have been detected at the site if ≥1 technical replicate was positive in ≥1 eDNA sample collected at the site. We considered a technical replicate to be positive if an exponential increase occurred at any point during the qPCR cycles (as described in Goldberg et al., 2013, see also Ellison et al., 2006).
To model Bd DNA detection probability, we used a multi-scale occupancy models in in R (version 3.6.0; R Project for Statistical Computing, Vienna, Austria) and package eDNAoccupancy (as described in Sepulveda et al., 2018; see also Dorazio & Erickson, 2017). We compared a null model to models fitted with covariates that affected the occurrence of Bd DNA in the sample (θ; sample type [swab vs. eDNA]), and covariates that affected the detection of Bd DNA in the technical replicate or subsample (p; analysis approach [field vs. lab]). We assessed the models using posterior-predictive loss criterion (PPLC) and widely acceptable information criteria (WAIC). We calculated detection probability and their standard errors using a Markov chain containing 11,000 iterations (1,000 burn-in).