This paper addresses the problem of decentralized estimation of node locations in wireless IoT networks. We propose utilizing an estimator with a distributed architecture based on a Bayes filter with Gaussian Sum parametrization of the posterior. The estimator's state is updated using a Covariance Intersection algorithm which handles missing cross-covariance information, inherent in a distributed architecture, in a consistent manner. Simulations, performed for the network of Contiki nodes using the Cooja environment and employing real-life ultra-wide band (UWB) ranging models, have demonstrated the feasibility of the proposed approaches.