The choice of distance metric is mainly decided on how you define similarity, which I discuss in this video: ru-vid.com/video/%D0%B2%D0%B8%D0%B4%D0%B5%D0%BE-uWf__KIKzPQ.html To select linkage function I have previously used some sort of bootstrapping, where you can select the linkage function that generates the most robust clusters, see for example this paper: pubmed.ncbi.nlm.nih.gov/29079427/