Inferring River Condition from eDNA Using Supervised Machine Learning

ID: 1711

Presenting Author: Laura Steel Pascual

Session: 516 - Asian S3EA: Strategic, Spatial and Sustainable EA with effective information

Status: pending


Summary Statement

eDNA and machine learning identify species-level indicators of river disturbance across spatial scales, supporting scalable biomonitoring and conservation in Pacific island ecosystems.


Abstract

Environmental DNA (eDNA) metabarcoding provides a sensitive, non-invasive approach for monitoring aquatic ecosystems. This study integrates eDNA (collected during the Asian Development Bank Technical Assistance project TA-10135-SOL, in collaboration with the Solomon Islands Government) with machine learning (ML) to classify impacted vs reference river sites across Guadalcanal. Key questions include: (1) Can eDNA profiles distinguish site condition? (2) Which taxa are most predictive? (3) How does riparian buffer width affect classification?
Water samples from 47 sites across 8 rivers were sequenced, detecting 861 species. Supervised classifiers – Decision Tree, Logistic Regression, Random Forest, and eXtreme Gradient Boosting – were benchmarked using species count profiles across buffer widths (100-1000m) and land-use intensification thresholds (0.03125; 0.0625). Stratified 5-fold cross-validation and permutation testing ensured robust evaluation.
Random Forest performed best, achieving AUCs up to 0.94 ± 0.04 (p<0.001) and accuracy of 0.83 ± 0.05 at the 1000m buffer. Wider buffers improved classification by capturing broader anthropogenic signals. SHAP and feature importance analyses identified key taxa: vascular plants (kava, pua keni keni), protozoans, fishes (fringelip mullet, gobies), and invasives (cane toad, Giant African snail).
This work shows how eDNA and ML can enable scalable, data-rich impact assessments, highlights spatial scale’s role in interpreting ecological signals, and identifies candidate taxa for future Pacific island monitoring frameworks.


Author Bio

Laura Steel Pascual is an Environmental and Social Policy Consultant at DDA International Consulting, specialising in impact assessments, due diligence, vulnerability, and machine learning.


Coauthor 1: Mara Fehling

Coauthor 2: Darko Annandale

Coauthor 3: David Annandale

← Back to Submitted Abstracts