Advanced statistical and mathematical modelling of a constructed wetland, treating abattoir waste water: A new modelling method for resilience analysis

Abstract

The meat processing industry produces high strength toxic waste from the slaughtering processes. Conventional methods were to ship the blood waste products to other companies for treatment, leading to high costs in storage, transport and licensing. A constructed wetland (CW) is an artificial treatment system consisting of soil, plants and water used to passively treat contaminated wastewater. The use of constructed wetlands may provide a cost effective passive treatment for the high strength toxic waste by-products, with the potential to implement effective waste treatment and lowering the overall costs of storage and transport. The output treated wastewaters are expected to show an overall reduction in toxicity. Typical wastewater generated by the abattoir industry has a high organic content (both soluble and particulate) with high nutrient and microbial loads. There are limited studies on the application of constructed wetlands treating high strength wastewaters especially in the meat industry and within the scope of the Irish landscape even fewer. The main objective of this research was the application of statstical and mathemathical modelling methodologies to provide an in-depth review of the constructed wetland treating abattoir wastewater. The research is based on a free water constructed wetland system, with twelve interconntected ponds treating abattoir wastewater. Samples were retrived from six locations. The dissolved air flotation (DAF) plant, pond 1, pond 6, pond 9, pond 12 and the local stream. This work reviews the input (DAF) and output (stream) variables of the wetland, but sampling of specific internal locations within the wetland system; the aim was to understand what was happening inside the wetland system. This work delves into the internal processes i.e. pond 1, pond 6, pond 9 and pond 12. The methods employed used advanced statistical methods such as Canonical variate analysis (CVA), Discriminant Functional Analysis (DFA) and Principal Component Analysis (PCA). Soft-computing methods such as self-organising maps (SOMs), artificial neural networks (ANNs), fuzzy logic and Bayesian belief networks were utilised. Canonical variate analysis (CVA) potentially revealed a second source of E.coil entering the wetland. Principal component analysis (PCA) highlights the dominance of indicator bacteria within the wetland system. The use of indicator bacteria as a viable tracer to deterimine the hydraulic retention time (HRT), with modelled values between 55 days to 128 days depending on the season. The use of Bayesian networks and sensitivity analysis on the wetland soils and sludge reveals that the wetlands soil porosity may be an issue in particular from pond 9 to pond 12. Self-organised criticality (SOC) analysis was employed, indicating the potential that indicator bacteria and nutrients are not being treated correctly within the CW and water depth was highlighted as an issue. The use of self-organising maps (SOMs) and artificial neural networks (ANNs) were employed to understand complex interactions within the CW. The SOM models can provide what-if scenarios. The use of coefficient of reliability (COR) and Cronbachs’ Alpha which measure reliability of the CW. Indicating a questionable reliability of the CW and potentially show that indicator bacteria provide a good indicator as to the overall performance of the CW. Combining soft-computing methods such fuzzy logic and Bayesian network analysis into hybrid models to understand the CWs resilience. The models indicated issues in pond 9 to pond 12, and that the overall resilience of the CW was an undesirable system to treat the abattoir wastewater.