Abstract:
This study investigated the Tororo Wastewater Treatment Plant’s stabilization ponds under varying
climatic conditions and increasing influent loads, aiming to guide sustainable water management
in alignment with Uganda's Vision 2040 and SDGs 6, 11, and 13. Methodologically, historical data
(2014–2024) covering influent rates, wastewater characteristics, and key climate variables were
analyzed using descriptive statistics, Principal Component Analysis (PCA), Cluster Analysis (CA),
and regression analyses to characterize pond performance and identify dominant influencing
factors. Subsequently, three data-driven predictive models Multivariate Regression, Artificial
Neural Networks (ANN), and a Hybrid GRU-LSTM Recurrent Neural Network (RNN) were
developed and tested using normalized and partitioned datasets. An LSTM-based forecasting
model was also developed to project future influent parameters.
Key results showed that Facultative Ponds 1 and 2 frequently experienced negative TSS removal
efficiency and moderate BOD/COD reductions (19–26%). In contrast, the Maturation Pond
demonstrated more stable TSS reduction (around 26.9%) and slightly higher BOD removal (up to
37.4%), but struggled with ammonia removal efficiency. All ponds achieved consistently high fecal
coliform attenuation (≥ 89%). The Hybrid GRU-LSTM model delivered superior predictive
accuracy for effluent parameters (R² up to 0.97 for BOD, COD, TSS), outperforming other models
by effectively capturing temporal dependencies. Feeding forecasted influent parameters into this
Hybrid GRU-LSTM model achieved robust accuracy for future effluent predictions (overall R² ≈
0.81). These findings highlight that data-driven forecasting, coupled with targeted operational
interventions (e.g., optimized sludge management, improved pond hydraulics), can enhance
stabilization pond resilience. This approach ensures more reliable wastewater treatment and
supports sustainable development goals in rapidly urbanizing settings.
