This happened to me too, but after playing around with both I realised that it's because samtools mpileup counted in properly mapped only, while bedtools used all the mappings.

Solution: filter the .bam file first using samtools view -b -f 2 to get only mappings with properly mapped reads, then pass that output to bedtools. You will get exactly the same results for both of the methods.

Based on my comments above, I looked more closely into this. I ran the exact commands you used on a test file I have. It is never the case that the samtools depth is reported to be higher than bedtools. When discrepant, bedtools is always higher.

Here is an example. In this case, bedtools reports a coverage of 5, whereas samtools reports coverage of 1. Every other case I check out has the same pattern. I actually doubt there is a bug in the samtools logic for counting depth. I suspect that instead, the parameters you are using aren't behaving exactly the way you expect, and as such, some alignments are just being filtered out before being counted.

A guess: I know that samtools outputs only positions with depth > 0. Might it be that genomeCoverageBed instead outputs a line for every position in the reference? If your reference is 1,027,173 bases long then this is likely.

There might be a couple of things going on here. The first is that I think you are using an outdated version of BEDTools - I can't find genomeCoverageBed anywhere in the current git repository, and I think it has been superseded by coverage. In any case, you should take a look at the -split option in the documentation for coverage as I suspect that your reads have insertions and deletions, and that BEDTools is not correctly splitting the CIGAR string of these reads to produce coverage in the way that samtools does. This could result in a different number of bases in your reads "covering" the bases in the alignment, and a different number of lines in your output.

Edit: Looks like lh3 has the answer.

hi:)

i have a question regarding the depth calculated by samtools, when there is a deletion in the read.

scaffold79    978    N    5    g$gggg    HGHGH
scaffold79    979    N    4    aaaa    HHGH
scaffold79    980    N    4    g-1ng-1ng-1ng-1n    HH?H
scaffold79    981    N    4    ****    HHEH
scaffold79    982    N    4    tttt    HHEH
scaffold79    983    N    4    tttt    HHGH
scaffold79    984    N    4    tttt    HHGD

at position 980 it is indicated, that here is a deletion in 4 reads at the following position. So these 4 reads do not really cover the next position 981. Should i check for these positions and correct the depth to gain the real coverage, or is this the accepted way to calculate the coverage?

Many thanks in advance!:)