Satellites constitute an essential source of observations in operational satellite data assimilation (DA). In this thesis, we investigate the impact of assimilating satellite observations at different spatial scales: is there a relative benefit in focussing on small rather than large scales (or vice versa)? In order to address this question without using complex and computationally expensive Numerical Weather Prediction (NWP) models, we conduct a series of idealised satellite DA experiments based on a modified shallow water model able to imitate convection and precipitation. The use of an isopycnal, single-layer version of the model (modRSW) is discussed first. A series of forecast-assimilation experiments are carried out using a Deterministic Ensemble Kalman filter (DEnKF). As a result, the filter performance and the relevance of the modRSW model for convective-scale DA in Numerical Weather Prediction systems are demonstrated and a protocol to extend a similar analysis to other idealised systems is presented. After establishing that the modRSW model is not suitable for satellite DA research, a new isentropic, 1.5-layer model (ismodRSW) is developed. The revised model is equipped with a fluid temperature definition and is therefore a better candidate for satellite DA experiments. The dynamics and the numerics of this model are discussed, and its numerical solver is verified against an analytical solution. In order to imitate closely an operational system, an idealised observing system comprising both ground and satellite observations is created, and pseudo observations mimicking microwave radiation measured by polar-orbiting satellites are generated, with clouds and precipitation implicitly taken into account within the new (and nonlinear) observation operator. Finally, a new series of forecast-assimilation simulations is run to obtain a well-tuned system which is used as a reference in a series of data denial experiments, where satellite observations at small and large scales are selectively excluded from the assimilation to evaluate their impact on the system. Preliminary results show a degradation of both the analysis and the forecasts when only large-scale satellite observations are utilised, although further work is needed to ascertain the robustness of these findings. All in all, this thesis shows that the idea of investigating satellite DA using a modified shallow water model is a viable strategy. By imitating closely several aspects of an operational system and by developing a more realistic model, we have demonstrated that large-scale satellite observations alone can have a negative impact on the quality of a DA system.