Modeling nutrients, algal blooms and fish mortality in a large reservoir with extreme hydrological variability using mechanistic, machine learning and hybrid approaches

ABSTRACT Tropical semiarid reservoirs exhibit strong interannual and seasonal hydrological variability that greatly affects eutrophication. This study analyzed 15 years of monthly data from a large semiarid reservoir in Brazil, encompassing high-flow (2008-2011), drought (2012-2016), and intermediate (2017-2022) periods associated with massive algal blooms and fish mortality. Variables included inflow, outflow, volume, fish-cage production, total phosphorus (TP), total nitrogen (TN), chlorophyll-a (Chla), and fish mortality. A transient complete-mixing mechanistic model and four machine learning (ML) algorithms, linear regression (LR), Ridge regression (RR), decision tree (DT), and k-nearest neighbors (KNN), were calibrated using hydrological and aquaculture inputs to predict TP or TN. The mechanistic TP model achieved R2 ≈ 0.30 and PBIAS ≈ ±11%, while KNN performed best (R2 ≈ 0.53, PBIAS ≈ ±9%). For TN, the mechanistic and DT models outperformed the others (R2 ≈ 0.28, PBIAS = ±10%). Combining mechanistic or best ML (KNN, DT) models with an empirical power-law for Chla prediction (as a function of TP, TN, volume and rainfall) yielded high accuracy (R2 > 0.70, PBIAS < 3%). Using only ML for Chla prediction produced poorer results (R2 = 0.22, PBIAS = 17%). Fish mortality correlated with TN (R2 = 0.37), and a hybrid mechanistic-DT model predicted mortality with high accuracy (R2 = 0.90, PBIAS = –8.47%). Overall, model results revealed that water quality improves under higher storage volumes, lower inflow rates and higher outflow rates. Despite data limitations, the study offers robust and generalizable models for managing water quality and assessing ecological risks in reservoirs worldwide.