There are at least two factors that make observations of tropical cyclones harder to effectively assimilate into numerical models than regular observations. First, the error covariance matrix describing the uncertainty of the first guess field used in data assimilation schemes is flow dependent and highly anisotropic in the vicinity of tropical cyclones. Current assimilation schemes make very poor estimates of these covariances, as a consequence, these schemes can produce analyses that can be worst than the first guess fields in some areas. Second, differences between the first guess field and observations tend to be very large near tropical cyclones; thus, four dimensional assimilation approaches are compromised by the fact that these large differences evolve nonlinearly. Here, we use ensembles of barotropic tropical cyclone forecasts to deduce ensemble based schemes for economically overcoming these difficulties. Schemes that are based on Kalman filter/smoother and 3D/4D Var schemes will be compared and contrasted