Abstract: Community detection is an important topic in spatial network researches. It can discover interesting spatial patterns, which help understand the spatial structures hidden in the networks. However, community detection in spatial networks is more difficult than traditional networks. Since it has to consider some other spatial correlations, e.g., spatial contiguity, geographic distance. This paper proposes a new spatial community detection algorithm-geographial weighted central node distance based Louvain method (GND-Louvain). It uses a geographical and network dual-constrain to measure the distance and to calculate the distances decay effects between spatial nodes and meta-communities. It also extends the famous fast unfolding community detection algorithm-Louvain, by using the distance-modularity. In addition, a merge order is defined during the GND-Louvain's optimization process to get high quality and stable community results. Comparative experiments are designed on five different community detection methods based on five different spatial constraints, which are ① no distance constraint, ② spatial contiguity edge constraint, ③ geometry median center distance, ④ degree center distance, and ⑤ degree weighted geometry median center distance. The experimental data comes from the Chinese railway lines. And results prove that the newly defined GND-Louvain algorithm can produce more accurate spatial communities than others. And the merge orders also ensure the high quality and stable results.