Aims: Unequal sampling is a common issue in field-based community ecology. Choosing α-diversity metrics that remain robust when sample sizes vary among plots is critical for reliable biodiversity assessment. This study evaluated the performance of nine diversity indices, including five “observed” indices calculated directly from the data: (1) species richness, (2) Shannon index, (3) Simpson index, (4) Hurlbert’s rarefied richness, and (5) Fisher’s α; and four “richness-estimator” indices: (1) Chao1 index, (2) abundance-based coverage estimator (ACE), (3) the extrapolated value of iNEXT (interpolation/extrapolation), and (4) total expected species (TES).

Methods: Using simulation, the performance of each index was evaluated under a gradient of minimum-sample thresholds, and for each case the accuracy and precision of between-sites variance (linear regression R²) was recorded. The simulation built up 20 sites in which “true” species richness (S) was linearly correlated with an environmental gradient (x) with a theoretical coefficient of determination R² = 0.80. Four unequal-sampling scenarios were then generated by imposing different minimum sample sizes per site. For each scenario, linear models were fitted between every diversity index and x, recording the corresponding R².

Results: The results indicate that sample size (the number of individuals recorded at a sampling site, as well as the equivalent sampling completeness) is the primary factor determining index performance. As sample size increased, model R² of all diversity metrics significantly improved. Under extremely low sampling (minimum < 20 individuals; sampling coverage < 20 %), rarefied richness had a higher R² than other indices. When the minimum sample size reached 100 individuals, the estimator indices group outperformed the observed indices. This study further clarified the minimum sample size and the corresponding sampling completeness required for each index to recover the predetermined R².

Conclusion: Overall, rarefied richness is recommended for highly unequal, sample size-poor scenarios. In practice, rarefaction threshold should be set at a relatively high level (e.g., > 40 individuals) to enhance the overall comparability among sampling sites, even if it results in the exclusion of extremely under sampled sites. Once sampling completeness is adequate, richness estimators are preferable, as they can generate extrapolated richness that are close to the true gradient.