Continuous Super-Resolution of Copernicus Atmosphere Images Using a Graph-Based Diffusion Model

Image super-resolution aims to reconstruct high-resolution images from low-resolution inputs and plays a critical role in remote sensing. In environmental monitoring, particularly air quality assessment, super-resolution is essential for producing fine-scale pollutant concentration maps, since datasets, such as Copernicus Atmosphere Monitoring Service, generally provide coarse spatial resolution, limiting the capture of localized variations. In this context, this study proposes the Improved Graph-based Implicit Diffusion Model (IGIDM). It is a diffusion probabilistic framework, combining Implicit Diffusion Model with the novel Superpixel-based Graph Attention Network, to enable continuous-scale super-resolution prediction with an associated uncertainty estimation. Moreover, a novel Extended-Scale Optimization Strategy is incorporated to guide the super-resolution process, improving generalization and perceptual fidelity, particularly under extreme scaling factors. Experimental validations are conducted on publicdomain benchmarks properly dedicated to air quality super-resolution. The proposed model achieves competitive reconstruction accuracy compared with state state-of-the-art discrete super-resolution methods and substantially outperforms probabilistic baselines. Notable gains are observed when the Extended-Scale Optimization Strategy is employed to leverage ground-station data. Furthermore, IGIDM provides wellcalibrated uncertainty estimates across both standard and extreme scaling scenarios. The experimental results demonstrate that the proposed super-resolution model is a robust framework for enhancing environmental datasets, providing more reliable and fine-grained air quality measurements than its direct competitors.