Quantitative diatom analysis for drowning-site determination: Evaluation of accuracy and a practical decision guide

Thirty-two aquatic-lung pairs from confirmed drowning cases were compared using quantitative exploration of diatom data gathered by the lossless HNO3-PCF extraction method. Permutational analysis of variance (PERMANOVA) demonstrated that aquatic-lung pairs share a highly significant, pair-specific diatom signal (R² = 0.86, p = 0.001), while differences between aquatic and lung samples remained insignificant, establishing lung samples as highly representative of the site-specific aquatic community. Beta regression analysis proved that the quality of geographical aquatic-lung matching (site spatial tightness) was the sole significant factor affecting pairing accuracy (pseudo-R² = 0.50), confirming the strong site-specific character of these pairs. In real forensic scenarios where one lung was compared to several aquatic samples, the mutual nearest neighbour (MNN) criterion revealed that pairing accuracy reached 95% for two aquatic samples, maintaining an accuracy of 79% when the candidate pool increased to eleven. To control false positives, the nearest neighbour distance ratio (NNDR) was introduced as the decision parameter. Empirical NNDR 95% thresholds were calibrated to limit the false-positive rate to 5% within the resampling framework (with corresponding true-positive sensitivity reported for the three best-performing quantitative distance metrics: Bray-Curtis, quantitative Jaccard, and Hellinger). Based on these findings, a practical decision guide was developed to enable both the selection of the most probable drowning site and the assessment of match distinctiveness relative to the candidate pool. Our results suggest that quantitative analysis based on sample distance metrics provides a robust, statistically quantifiable framework for drowning-site determination in cases where the true drowning site is present in the candidate pool. However, rejecting false pairings when the true site is absent remains beyond the sensitivity of the method.