The reason for that is to get rid of the Nextera and transposase adapter sequences. In the experimental workflow of ATAC-seq, you use a transposase that integrates Illumina adapters into regions of open chromatin while simultaneously fragmentating these regions. The result are fragments from these open regions, being flanked with adapters that can be targeted by suitable PCR primers to make these regions ready for sequencing. The workflow creates fragment of different lengths, see the original paper Figures 1 and 2.
If you then sequence your samples, with lets say Nextseq 75bp cycles, but your actual sequence (sequence between the adapters) is only 60bp long, then you also pick up 15bp of the adapters, which will lead to false alignment results.
=> Therefore, you must get rid of any adapter content prior to aligning for fastq files. The actual Nextera adapter sequence (if using standard Illumina Nextera Sample Prep Kit) is CTGTCTCTTATA on both strands. You can use tools,such as Cutadapt (TrimGalore) or Skewer.
I personally use Skewer with these options to trim the Nextera transposase sequences:
skewer -x CTGTCTCTTATA -y CTGTCTCTTATA -m pe -q 20
I think they used 30bp because 30 is probably the minimal fragment length that the assay created in their hands, so it is kind of the length that does for sure not contain adapter contaminations. Still, I would use adapter trimming software instead of fixed trimming sizes (which pretty much all publications did so far).