3.1. Probabilistic Flood Hazards
For each location along the Shanghai coastline, the return periods of flood levels (storm tides) under the NCAR/NCEP current climate are estimated through the flood probability analysis. The spatially continuous distribution of the estimated flood levels with various return periods (10-, 100- and 1000-year) along the coast is presented in Fig. 1. Results indicate that the flood level substantially increases with the increase of return period: it is between 3.77 and 6.93 m (with an average value of 4.37 m above Wusong Datum) for the 10-year scenario, 4.57 and 8.48 m (5.19 m) for the 100-year scenario, and 5.88 and 12.10m (7.03 m) for the 1000-year scenario. The highest flood return levels (η ) and the largest increases in η over the return periods can be observed along the northern coast of Chongming Island. This spatially continuous distribution of simulated flood levels along the coast provides greater detail and accuracy for flood hazard assessment compared to a previous study (Yin et al., 2020), in which flood return levels were interpolated from sparsely distributed tide gauges.
Fig. 2 displays the flood return level curves at selected gauge stations estimated by our analysis, compared to estimates by SWA using the officially recommended Pearson-III distribution based on annual maximum values of recorded historical water levels. Similar to that shown in Fig. 1, the predicted flood return levels vary greatly from place to place along the coast. For example, the 1000-year flood height is about 7.76 m at Jinshan Station but only about 6.10 m at Luchaogang Station. Only very low-probability events (over 3000 ~ 10000 year) will exceed the crest levels of seawall. Results also show a large uncertainty in flood levels with long return periods (> 100 years). SWA’s analysis predicts significantly higher flood levels, especially for small return periods (e.g., 5 ~ 6 m for the 10-year floods). Peak water levels recorded at the gauge stations during Typhoon Winnie indicate it to be close to a 100-year event by SWA’s analysis but an 800 ~ 900-year event by our analysis. This discrepancy can be primarily attributed to the different distribution models and sample data used in the flood probability analyses. The occurrence of Typhoon Winnie may have contributed significantly to lifting the return level curves estimated based on the limited (less than 60 year) historical data. Otherwise, using limited-length historical data often underestimates extremes (Lin et al., 2014; Xu et al., 2020); extrapolating SWA’s estimates for return periods greater than 1000 years would result in an underestimation of the extremes compared to our modeling-based estimates.