In this paper, a comparison of input-output pairing-based decomposition methods for distributed state estimation of large scale systems is presented. Three methods, namely Relative Gain Array, Niederlinski Index and Partial Relative Gain, are implemented as an initial step to decompose the system into subsystems. Subsequently the different subsystems configurations are compared by evaluating the centralized and local prediction error and the convergence of distributed Kalman-filter-based state estimators for each case. Simulation results are presented using a heat plate as test bed, spatially discretized, resulting in a large-scale linear system.