From conventional monitoring to intelligent prediction: data-driven analysis of inorganic elements in atmospheric wet deposition at an urban site in Lanzhou.

Atmospheric wet deposition represents a key pathway linking atmospheric pollution to terrestrial ecosystems, with its chemical composition and deposition flux serving as important indicators of regional environmental quality. However, conventional monitoring approaches are labor-intensive and time-consuming, limiting their ability to support timely environmental assessment. In this study, systematic precipitation monitoring was conducted at an urban site in Lanzhou from 2018 to 2024. Based on analyses of concentration levels, deposition fluxes, reuse potential, and source contributions of 20 inorganic elements, machine learning techniques were further introduced to improve concentration prediction and monitoring efficiency. The results indicate that inorganic element deposition fluxes in Lanzhou generally fall within the typical range observed for Asian cities, while crustal-derived elements (e.g., Ca, K, and Mg) exhibit markedly elevated fluxes, reflecting a combined pattern of regional similarity and local characteristics. From a reuse perspective, precipitation shows potential for agricultural irrigation and emergency drinking water supply, although heavy metals such as Pb and Cd constitute the primary constraints on safe utilisation. Positive matrix factorization analysis revealed that the chemical composition of wet deposition is predominantly controlled by soil and construction dust, contributing approximately 68%, highlighting the combined influence of arid geographical conditions and urbanization. Furthermore, an XGBoost-based predictive model integrating conventional atmospheric pollutants and meteorological variables with newly introduced "source" and "transport" factors was developed. Incorporating information on key source regions and regional transport pathways substantially improved model performance, with the coefficient of determination (R2) increasing from 0.46 to 0.59. Overall, this modelling framework provides a promising approach for optimising wet deposition monitoring systems that rely primarily on manual sampling and laboratory analysis, and demonstrates the potential of machine learning for supporting more efficient and responsive environmental management.