NB: A New Dissimilarity Measure with Robust Clustering Evidence from Ecological Community Data

Distance and dissimilarity measures are central to ecological clustering, particularly for sparse, high-dimensional, and heavy-tailed community data, where classical measures may become unstable as cluster separability decreases. Motivated by these limitations, we propose a normalized and bounded (NB) dissimilarity measure that combines weighted absolute differences with quadratic normalization to promote scale stability and reduce sensitivity to extreme values. NB was evaluated under complete-linkage and K-means clustering using a new robustness-aware multi-metric soft ranking framework that jointly considers average performance, worst-case stability, performance degradation, and trade-off behavior. In simulations based on ecological community models, NB exceeded the classical benchmark average in five of six complete-linkage scenarios and all six K-means scenarios, with significant paired improvements (mean differences = 0.226 and 0.463; exact permutation p = 0.031 and 0.016, respectively). In real ecological datasets, NB produced stable and high-quality clustering outcomes, whereas traditional measures showed dataset-dependent sensitivity to sparsity and distributional irregularities. However, NB should be viewed as a robustness-oriented alternative for ecological community data rather than a universal measure for now; its performance may decline when rare species signals dominate, extreme sparsity saturates pair-wise contrasts, or metric properties are required.