Abstract: Rapidly and objectively identifying flood-affected areas and quantifying damage severity are core needs of emergency response and post-disaster assessment. To overcome the limitations of pre-/post-event comparisons, including unavoidable seasonal fluctuations, subjective thresholding, and noise interference in single indices, this study proposed a graded flood disaster detection method based on harmonic baselines of optical–microwave remote sensing time series. Taking the August 7, 2025 extreme flood in Yuzhong County, Gansu Province as a case, 829 Sentinel-2 L2A and 649 Sentinel-1 GRD descending scenes from 2019 to July 2025 were retrieved via Google Earth Engine (GEE). Unified third-order harmonic regression baselines were built for seven flood-sensitive indices (NDVI, NDWI, NDMI, BSI, MNDWI, VV, SDWI) and packaged as a 10-m, 56-band GeoTIFF baseline product. The deviation of post-disaster measured values was standardized using Z-score, and a joint statistic χ² = ΣZᵢ² was constructed, which follows a chi-square distribution. The values were then combined with a directional consistency voting constraint based on physical flood priors, and classified into four levels according to critical values of the chi-square distribution with degrees of freedom (k=7). The results showed that the unified optical-microwave harmonic baselines fit all seven indices stably (median RMSE of NDVI, NDWI, VV: 0.075, 0.083, 0.039), thereby serving as reliable background baselines for flood anomaly detection. After phenological background removal, severe anomaly area accounted for 3.21% of the study area, distributed in strips along the mainstream and tributaries of the Yuanchuan River, which was highly consistent with the actual disaster conditions of five severely affected townships. The overall accuracy reached 86.8% with a Kappa coefficient of 0.78 over built-up, cropland, and vegetation areas, representing a 12%-14% improvement over single-index thresholding. This method provides a statistically rigorous and physically interpretable grading discrimination framework for near-real-time remote sensing monitoring of sudden environmental disasters such as floods, offering quantitative support for post-disaster environmental damage assessment and ecological restoration.