We introduce a source localisation method dealing with an unknown reverberant environment. The measured acoustic field is decomposed as the sum of a diffuse field and a source field, using a priori knowledge on these field, i.e. structured sparsity deduced from the governing equations of acoustics. Simulations and experimental results in 2D domains, using greedy algorithms as well as optimisation-based methods, show that the method is robust with respect to noise and sensors localisation. Additionnaly, the density of sensors required can be lower than the Nyquist frequency. These facts make this method a interesting alternative to standard localisation methods particularly when a large number of narrowband sensors are deployed.