Random Forest model and TRIX used in combination to assess and diagnose the trophic status of Bizerte Lagoon, southern Mediterranean

The observed TRIX values vary from 5.18 to 6.12, reflecting waters ranging from moderate to poor quality with a high trophic level. The results of the RF model show that the most important predictor of chlorophyll a variation appears to be dissolved oxygen, followed by nitrate and temperature, suggesting a reduced model for this variable. The link between chlorophyll a and observed variables was also studied using numerical models, analyzing the linear statistical correlations and a recently proposed non linear model, the Random Forest. Both methods highlight a high correlation between chlorophyll a and dissolved oxygen as opposed to chlorophyll a and nitrate.