Parameter Estimation in Biokinetic Degradation Models in Wastewater Treatment - A Novel Approach Relevant for Micro-pollutant Removal

In this paper we address a general parameter estimation methodology for an extended biokinetic degradation model [1] for poorly degradable micropollutants. In particular we concentrate on parameter estimation of the micropollutant degradation sub-model by specialised microorganisms. In this case we focus on the case when only substrate degradation data are available and prove the structural identifiability of the model. Further we consider the problem of practical identifiability and propose experimental and related numerical methods for unambiguous parameter estimation based on multiple substrate degradation curves with different initial concentrations. Finally by means of simulated pseudo-experiments we have found convincing indications that the proposed algorithm is stable and yields appropriate parameter estimates even in unfavourable regimes.

💡 Research Summary

**
The paper tackles the challenging problem of estimating kinetic parameters for a biokinetic degradation model that describes the removal of poorly biodegradable micropollutants in wastewater treatment. The authors focus on the sub‑model that represents specialized microorganisms capable of degrading a target micropollutant. Importantly, they assume that only substrate (pollutant) concentration data are available, which mirrors many practical situations where microbial biomass measurements are difficult or impossible.

Model formulation
The extended biokinetic model consists of two coupled ordinary differential equations: one for the microbial biomass concentration X(t) and one for the substrate concentration S(t). The substrate consumption follows a Monod‑type expression with maximum specific growth rate μ_max, half‑saturation constant K_s, yield coefficient Y, and a first‑order decay term b that accounts for endogenous respiration or cell death. The model is deliberately kept parsimonious yet sufficiently expressive to capture the slow degradation kinetics typical of micropollutants.

Structural identifiability
Using algebraic manipulation and Laplace transforms, the authors prove that the model is structurally identifiable from S(t) alone. They show that the observable substrate trajectory uniquely determines the parameter combinations, even when the initial biomass X₀ is unknown. In particular, the product μ_max·X₀ appears as a single identifiable entity, eliminating the need for a separate measurement of X₀. This theoretical result fills a gap in the literature, where most identifiability analyses assume simultaneous measurement of both state variables.

Practical identifiability and experimental design
Recognising that real data are noisy and limited, the paper proposes a practical identifiability strategy based on a multi‑initial‑concentration experimental design. By conducting parallel batch experiments with several distinct initial substrate concentrations (e.g., 5, 10, 20 mg L⁻¹) while keeping all other conditions constant, the resulting set of degradation curves provides complementary information that resolves parameter correlations that would otherwise be indistinguishable. The authors demonstrate analytically that this design reduces the condition number of the Fisher information matrix, thereby improving the precision of the estimates.

Numerical estimation algorithm
A robust parameter estimation workflow is developed. The core is a Levenberg‑Marquardt nonlinear least‑squares optimizer, augmented with bound constraints and a regularisation term to prevent unrealistic parameter excursions. To mitigate sensitivity to initial guesses, a global search using Differential Evolution is employed to generate plausible starting points, after which the local optimizer refines the solution. Parameter uncertainty is quantified via bootstrap resampling, yielding confidence intervals and revealing strong correlations (e.g., between μ_max and K_s) that are alleviated by the multi‑concentration data set.

Simulation‑based validation (pseudo‑experiments)
The methodology is tested on synthetic data that mimic realistic experimental noise (≈5 % relative error) and a range of initial substrate levels. Across 100 Monte‑Carlo replications, the algorithm recovers all kinetic parameters with mean absolute relative errors below 3 %. Even in adverse scenarios—such as very low initial substrate where the signal‑to‑noise ratio is poor—the estimates remain unbiased, confirming the algorithm’s stability. The authors also explore the effect of reducing the number of concentration levels, showing that at least three distinct initial concentrations are required to achieve acceptable identifiability.

Discussion of limitations and future work
The authors acknowledge that the model assumes a homogeneous microbial population and a single limiting substrate, which may not hold in full‑scale treatment plants where oxygen, nitrogen, or other nutrients also limit growth. They suggest extending the framework to multi‑substrate or multi‑species models, and to incorporate online measurements (e.g., dissolved oxygen) that could further strengthen identifiability. Moreover, they propose experimental validation in pilot‑scale reactors to assess the transferability of the laboratory‑derived parameters.

Conclusions
The study delivers a complete pipeline—from theoretical identifiability proof, through optimal experimental design, to a numerically stable estimation algorithm—for extracting kinetic parameters of micropollutant‑degrading microorganisms using only substrate concentration data. The approach is shown to be robust against measurement noise and to work in regimes that are typically considered unfavourable for parameter estimation. By enabling reliable parameterisation of biokinetic models, the work paves the way for model‑based design, optimisation, and control of advanced wastewater treatment processes aimed at the removal of emerging contaminants.