Hi @Seok_Jun_Kim,
This looks just like this inquiry we had recently which might be worth following as well. My initial thought is that you are truncating too much of your reads, thus not leaving enough overlap coverage for dada2 to properly merge your reads. See here. In addition you may be discarding some reads from the initial dip you see in your Forward read's 5' which you may want to trim as well. If you could share the stats-output from this run we take a closer look (you'll need to upgrade your q2 environment).
A few additional notes.
OTU picking and Denoising methods are very different from each other so don't expect the same results here as it would be comparing apples and bananas. 
. But you should count on the dada2 data to be truer to the truth than OTU picking. If you need to compare OTU methods between qiime1 and qiime2, you could try the vsearch method and even then you can still denoise your reads with dada2 before that. Generally, I would advise against OTU picking methods on their own unless you have a specific reason to do so.
Not sure what you mean by "as denoising phase" but just be aware that you shouldn't do any prior modifications or denoising before dada2 as it may interfere with its own internal denoising.
See my link above regarding trim/truncating guidelines.
This message is essentially telling you that at the position where your mouse is hovering at there are reads that are shorter than that position. This is likely not an issue as some shorter reads (true sequences or artifacts) are inevitable and commonly found in these runs. If a majority of your reads however are shorter than you expect then this would be worth looking more in detail. In the most recent version of q2 (2018.6) you can also retrieve information about read lengths, so I recommend upgrading your environment if you want to look at this more closely. In fact you should always keep up with the most updated version anyways 
Hope this helps!