Using rainfall thresholds and ensemble precipitation forecasts to issue and improve urban inundation alerts

Tsun-Hua Yang, Gong-Do Hwang, Chin-Cheng Tsai, Jui-Yi Ho
2016 Hydrology and Earth System Sciences  
<p><strong>Abstract.</strong> Urban inundation forecasting with extended lead times is useful in saving lives and property. This study proposes the integration of rainfall thresholds and ensemble precipitation forecasts to provide probabilistic urban inundation forecasts. Utilization of ensemble precipitation forecasts can extend forecast lead times to 72<span class="thinspace"></span>h, predicting peak flows and to allow response agencies to take necessary preparatory measures. However,
more » ... es. However, ensemble precipitation forecasting is time- and resource-intensive. Using rainfall thresholds to estimate urban areas' inundation risk can decrease this complexity and save computation time. This study evaluated the performance of this system using 352 townships in Taiwan and seven typhoons during the period 2013–2015. The levels of forecast probability needed to issue inundation alerts were addressed because ensemble forecasts are probability based. This study applied six levels of forecast probability and evaluated their performance using five measures. The results showed that this forecasting system performed better before a typhoon made landfall. Geography had a strong impact at the start of the numerical weather modeling, resulting in the underestimation of rainfall forecasts. Regardless of this finding, the inundation forecast performance was highly contingent on the rainfall forecast skill. This study then tested a hybrid approach of on-site observations and rainfall forecasts to decrease the influence of numerical weather predictions and improve the forecast performance. The results of this combined system showed that forecasts with a 24<span class="thinspace"></span>h lead time improved significantly. These findings and the hybrid approach can be applied to other hydrometeorological early warning systems to improve hazard-related forecasts.</p>
doi:10.5194/hess-20-4731-2016 fatcat:ugfy4m55o5etxljo2lhyzwr5e4