Asp-CLEAN2026: CLEAN deconvolution with adaptive sky-model components for radio interferometric imaging

Complex radio sources often contain both compact and diffuse multiscale emission. Under incomplete uv sampling, convolution with the point spread function (PSF) couples these structures in the image plane, making deconvolution challenging for high-sensitivity observations. Scale-free CLEAN struggles to represent continuous extended emission, Multi-scale CLEAN (MS-CLEAN) depends on predefined scale sets, and Adaptive Scale Pixel CLEAN (Asp-CLEAN) relies on explicitly prescribed component shapes. These constraints reduce the flexibility of CLEAN-based methods in modeling complex source morphology. We aim to improve the flexibility of CLEAN-based deconvolution for complex multiscale radio sources by developing an imaging algorithm that constructs model components directly from residual information. We propose Asp-CLEAN2026, a CLEAN deconvolution algorithm that constructs adaptive sky-model components for radio interferometric imaging. The method forms each incremental model component from significant emission structures in the current residual image. The component support is determined by an adaptive intensity threshold, while its boundary and amplitude response are modulated by a smooth nonlinear weighting function. The threshold and transition-width parameters are jointly estimated in a residual-domain least-squares optimization, allowing each component update to adapt to the current residual structure. Comparative experiments with representative CLEAN-based deconvolution methods on simulated SKA datasets and real VLA observations indicate that Asp-CLEAN2026 improves the recovery of structural details and faint extended emission in controlled simulations and provides more coherent and structurally detailed model representations of the observed emission on real interferometric data. Asp-CLEAN2026 remains compatible with the CLEAN framework while improving modeling flexibility through adaptive sky-model components. Results on the tested simulated SKA and real VLA datasets indicate that the method can improve structural representation in radio interferometric imaging.