If these samples were ran through the same PCR run and sequencing run then you can denoise them together, otherwise it is recommended that you run each run through DADA2 separately with the same trimming parameters then merge afterward. The trimming parameters need to be exactly the same, because even a single nt difference in 2 features will lead to them being identified as different ASVs. You will see very clearly on an ordination plot that your samples will cluster more strongly based on sequencing run if your trimming parameters are not the same. The same goes for truncating parameters (cutting from 3’) if you were to use single-end reads, but since these are paired end and merge on 3’, the truncating is not an issue.
To identify batch effect is a bit tricky but the easiest way I go about doing this is as I mentioned just visually looking on your ordination plots to see you see some artificial clustering based on sequence runs (a column run you should add to your metadata file). As far as I’m aware there is one q2-plugin that deals with normalization of batch effects, q2-perc-norm, might be worth reading up on.