Детерміновано-Стохастичне Моделювання Забруднення Поверхневих Вод Пестицидами: Програмна Реалізація Та Обчислювальний Експеримент

Introduction. Agricultural activity is one of the principal sources of diffuse contamination of water bodies: less than 0.1 % of an applied pesticide dose reaches the target organism, while the remainder is redistributed within the environment. Existing models of pesticide transport are predominantly deterministic and operate with fixed parameter values, whereas every input parameter of the “field – water body” system carries substantial uncertainty. The aim of the work is to develop a mathematical model and software for modelling water resource contamination caused by agricultural activity, based on the integration of deterministic and stochastic approaches. Materials and Methods. Five physical processes are integrated into an end-to-end computational chain: pesticide degradation (first-order kinetics), sorption (a linear isotherm with a retardation factor), surface runoff generation (the SCS-CN method), retention in a vegetated buffer zone (an empirical exponential model) and transport within the water body (the one-dimensional convection–diffusion equation solved by an explicit finite-difference scheme under the Courant–Friedrichs–Lewy condition). The deterministic core is wrapped in a stochastic shell based on the Monte Carlo method (N = 2000 iterations, 7 uncertain parameters). The software is implemented in Python 3 (NumPy, SciPy, Matplotlib) as two modules and ten classes. Results. A computational experiment was carried out for an atrazine application scenario on a 50 ha maize field in Cherkasy Oblast (MAC-EQS = 2.0 µg/L). With a 20 m grass buffer zone the deterministic model yields a peak concentration of 0.396 µg/L (19.8 % of the MAC-EQS), whereas the stochastic simulation reveals a 1.7 % probability of exceeding the MAC-EQS, an upper bound of the 90 % confidence interval of 1.42 µg/L and a maximum simulated concentration of 4.41 µg/L (a 2.2-fold exceedance). Sensitivity analysis identifies the hydrological parameters as dominant: the Curve Number (ρ = +0.66) and rainfall intensity (ρ = +0.53). A 20 m grass buffer lowers the peak concentration by 77.8 %, and a 10 m forest strip is equivalent to 20 m of grass. Conclusions. Ignoring the variability of the input parameters leads to a substantial underestimation of environmental risk: a deterministic estimate that formally indicates an acceptable level conceals extreme scenarios with a multiple exceedance of the MAC-EQS. The transition from averaged to simulation-based models is therefore a necessary condition for an adequate assessment of the risk of water body contamination, while the proposed software ensures transparency of the computations and modular replaceability of the model components.