Fitness landscapes map genotypes to their corresponding fitness under given environments
and allow explaining and predicting evolutionary trajectories. Of particular interest is
the landscape ruggedness or the unevenness of the landscape, because it impacts many
aspects of evolution such as the likelihood that a population is trapped in a local
fitness peak. Although the ruggedness has been inferred from a number of empirically
mapped fitness landscapes, it is unclear to what extent this inference is affected by
fitness estimation error, which is inevitable in the experimental determination of fitness
landscapes. Here we address this question by simulating fitness landscapes under various
theoretical models, with or without fitness estimation error. We find that all eight
examined measures of landscape ruggedness are overestimated due to imprecise fitness
quantification, but different measures are affected to different degrees. We devise a
method to use replicate fitness measures to correct this bias and show that our method
performs well under realistic conditions. We conclude that previously reported fitness
landscape ruggedness is likely upward biased owing to the negligence of fitness estimation
error and advise that future fitness landscape mapping should include at least three
biological replicates to permit an unbiased inference of the ruggedness.
