Quantum key distribution (QKD) enables two authenticated parties to share secret keys with the ability to detect any attempts to eavesdrop the keys theoretically. Currently, QKD systems can be categorized by its source of uncertainty in nature; single photon or coherent light. As coherent light is harnessed, the overall cost for composing systems drops in general. However, more elaborate post processing is required to extract and share errorless secret keys. Such process is called reconciliation, which is critical to performance. To extend the maximum distance to distill secret keys and prevent it from being bottleneck, we propose a reconciliation protocol that combines multidimensional reconciliation and polar codes. Especially, considering some practical issues, Gaussian approximation is chosen to construct polar codes. The simulation results show that the proposed protocol can support decent performance of coherent-light-based systems at long distance with reduced complexity.