With the rapid development of sensing technologies, electronic noses have become an important tool for real-time environmental monitoring, but ensuring their applicability and accuracy across various scenarios remains a key challenge. In this study, an electronic nose system with multi-scenario applicability and enhanced accuracy was developed to measure four common key pollutant concentrations in three typical pollution scenarios: landfills, wastewater treatment plants and livestock farms. A scenario-adaptive strategy was proposed to minimize the impact of interferences on the measurement accuracy by constructing a hierarchically structured qualitative-scenario-specific qualitative sub-network to process the sensor response data. Random Forest and Support Vector Machine algorithms were used and evaluated in scenario classification, with the Random Forest model performing best, achieving 100 % classification accuracy for 176 samples across all scenarios. Subsequently, scenario-specific qualitative models and unified model were developed with Random Forest Regression (RFR) and Artificial Neuron Networks (ANNs) after eliminating sensor features affected highly by interferences with feature importance analysis. The scenario-adaptive strategy achieved R² values exceeding 0.88 in target pollutant concentration prediction across all scenarios, with a mean absolute percentage error (MAPE) reduction of at least 15 % compared with the unified model for the test set. Furthermore, by flexibly integrating the most applicable algorithms, the scenario-adaptive strategy allows the benefits of different algorithms to be fully utilized in various scenarios. This study highlights the effectiveness of the adaptive strategy in improving electronic nose performance across various scenarios, laying a foundation for robust, adaptive electronic nose systems.