Model order selection based on the fitting error of the reconstructed point source model and the given visibility measurements (Dataset I). The model order is the minimum number of sources such that the fitting error reaches the noise level of the measurements (Background images of (b)–(d): CLEAN images; signal to noise ratio (S/N) in the visibilities: 0 dB; Actual model order: 3). Panel a: Evolution of the fitting errors against different point source model orders K in (8). Panel b: Reconstructed source models with the correct model order K = 3. Panel c: Reconstruction with under-estimated model order K = 1. Panel d: Reconstruction with over-estimated model order K = 5.