Abstract We introduce a novel approach to modelling the nebular emission from star-forming galaxies by combining the contributions from many H ii regions incorporating loose trends in physical properties, random dust attenuation, a predefined Hα luminosity function and a diffuse ionized-gas component. Using a machine-learning-based regression artificial neural network trained on a grid of models generated by the photoionization code CLOUDY, we efficiently predict emission-line properties of individual H ii regions over a wide range of physical conditions. We generate 250,000 synthetic star-forming galaxies composed of up to 3000 H ii regions and explore how variations in parameters affect their integrated emission-line properties. Our results highlight systematic biases in oxygen-abundance estimates derived using traditional methods, emphasizing the importance of accounting for the composite nature of star-forming galaxies when interpreting integrated nebular emission. Future work will leverage this approach to explore in detail its impact on parameter estimates of star-forming galaxies.
Read full abstract