If your covariance matrices all have the same size and correspond to data which have the same statistics and the same physical sense (i.e. you cannot sum a covariance matrix of a population and one corresponding to a random signal for instance), then a weighted sum allows to merge the covariances:
C = w1C1+w2C2 + ... + wnCn,
where wi are the weights and Ci are the covariance matrices.