<p>A well-structured flood season staging system under the dynamic control of reservoir flood-limited water levels is essential. To achieve period division, the most widely applied Fisher optimal partitioning method measures the similarity between different periods by calculating the distance between indicators across various periods of the flood season. Currently, the Euclidean distance serves as the sole distance metric method in flood season staging, and existing research lacks systematic comparisons and discussions regarding distance metric methods. This study primarily analysed the impacts of the distance metric on flood season staging results under different temporal units, aiming to provide a scientific basis for selecting distance metrics in flood season staging. By employing daily precipitation records (1985–2022) from the Nandujiang River Basin, this study implemented the Fisher optimal segmentation method with four distance metrics (Euclidean, Manhattan, Chebyshev, and Langley) to divide flood seasons into two temporal units: 5 days (pentad) and 10 days (decade). Multiple division schemes were ultimately compared using the technique for order preference by similarity to the ideal solution (TOPSIS) method. The results indicate that in the 5-day (pentad) temporal unit, the staging scheme calculated with the Fisher method based on Euclidean distance and Manhattan distance performed the best, indicating that this combination has high applicability in tropical island regions. This study aimed to provide technical references for flood control and water resource management in tropical island basins.</p>