I have a species abundance dataset with many surveys (samples) containing a total of 2-0 counts. Because of the high proportion of zeros and no-share sample pairs I have created a zero-adjusted Bray Curtis resemblance matrix using a dummy value of 1. This has done an excellent job of spreading the datapoints on the NMDS plot so that all surveys can be properly visualised. However, this has increased the stress value from 0.05 (Bray Curtis) to 0.21 (zero-adjusted Bray Curtis) making the plot less reliable. Square-root, fourth-root and log(x+1) pre-transformations of the count data either didn't reduce the stress value or increased it to 0.26.

Are there any other ways I can try to reduce the stress value? Perhaps a different dummy value for the zero-adjusted Bray Curtis or simply report three dimensions of my NMDS plot to lower the stress value?

Many thanks in advance.