Assessing the trophic status of a tropical microtidal estuary applying TRIX and Random Forest - A combined approach

Abstract

The present study assessed the trophic status of a medium-sized microtidal estuary, Rushikulya, India using a combination of mutimetric trophic indices (TRIX, TRBIX) and a machine learning approach (Random Forest). A total of 108 samples were considered to build a predictive model for chlorophyll a (Chl a) and 17 response water variables by observing two annual periods (2021-2023) at six sampling sites. Mean values of TRIX (5.04 ± 0.72) and TRBIX (0.17 ± 0.08) reflected that the estuary has a moderate degree of eutrophication with 'good' water quality and 'biomass saturated'. However, the threshold of TRIX represents a transition state from 'moderate' to 'high' eutrophic. Random Forest model reflected that no apparent association between Chl a and water turbidity above 30 NTU and eutrophication in the estuary fluctuated mainly due to PO₄^3--P along with electrical conductivity. Linear statistical correlations showed high correlation between Chl a and conductivity and a negative correlation between Chl a and dissolved oxygen, unlike PO₄^3--P.

Keywords: Ecosystem health; Machine learning; Multimetric indices; Water variables.