Abstract
<jats:p>The article substantiates and develops a hybrid architecture for a system of integration and processing of heterogeneous data for environmental modeling and state forecasting tasks. The drawbacks of traditional monolithic and file-oriented approaches in working with mathematical models for predicting the dispersion of harmful emissions, which make flexible automation of environmental modeling impossible, are analyzed. To overcome the problem of format and semantic heterogeneity of input data, a three-tier hybrid data storage model is proposed. It effectively combines knowledge-oriented technology in the form of a domain ontology, a relational DBMS with a spatial extension for topological queries, and an object file storage for artifacts. A software system based on a service-oriented architecture using containerization in the Kubernetes environment is proposed. Specialized adapter services have been developed, ensuring the seamless integration of the forecasting module's computational preprocessors into a modern distributed web environment. The practical implementation of the solutions is carried out in a software system that automates the full calculation cycle and provides an interactive geospatial interface for visualizing pollutant concentration fields. The proposed approach creates a solid scientific and technical foundation for managing environmental risks, particularly in urbanized territories.</jats:p>