Striving Sufficient Lead Time of Flood Forecasts via Integrated Hydro-meteorological Intelligence Dong-Sin Shih Assistant Professor, National Chung Hsing University, Taiwan, Sep. 6, 2013
Outlines Introductions Integrated hydro meteorological simulations Ensemble Quantitative Precipitation Forecasts Experiment, Unified Regional Circulation Rainfall and Hydrological Watershed Model, Modeling Test-bed with Integrating Simulations/Observations/Assimilations in Demo Site, To improve our understandings from individual ensemble flood forecasts. Summary Page 2 / 32
Introductions Page 3 / 32
Introduction (Rainfalls) Taiwan, located in the subtropical zone of the west Pacific Ocean, often experiences typhoons accompanied by frequent flooding. During the summer season, an average of 3 to 4 typhoons make landfall, usually producing massive amounts of precipitation island-wide. The total amount of annual rainfall is more than 2,500 millimeters. The total amount of rainfall is almost 2.6 times the world s average. Page 4 / 32
Motive (Floods) During the rainy season, excessive torrential rain tends to cause flash floods, sometimes with devastating effects on people s lives and property. In recent decades, Taiwan has particularly suffered from severe attacks of torrential rain. Almost 73 percent of the population of the island are under threats from more than three of major natural disasters. Typhoon induced flood is the major natural threat to the island. Page 5 / 32
Introduction (Environment) The topography of Taiwan is steep and as a consequence its rivers are short and fastmoving, so are prone to frequent flooding following extreme precipitation. This is particularly the case in Taiwan which is covered by varied terrain with most watersheds being located in mountainous regions. Due to such steep landforms, we could only gain 3-5 hours of hydraulic response time when flash floods occur. It is necessary to establish a flood warning system to alert the population for possible evacuations. Page 6 / 32
TTFRI It is unquestionable that accurate rainfall prediction is the most crucial factor driving the hydrological response. It is also important to have very dense observations to provide the initial conditions via assimilation for hydrological modeling. How to integrate useful early information and condense the forecasting process is of great concern. This study aims at understanding the distinctive features of flood forecasts that are based on integrated rainfall-runoff simulations.. Page 7 / 32
Integrated Hydro meteorological Simulations Page 8 / 32
Field Observations Hydro meteorological System Ensemble Forecasts for Quantitatively Precipitation Experiments A quantitative precipitation experiment has been examined in Taiwan since 2010. Over 20 members of ensemble rainfall forecasts are generated by Weather Research Forecasting (WRF) Model. Hydro-Meteorological Simulation in Watershed Scale Hydrology A hydrology model, WASH123D, was sequentially employed to flood routing. Field observations are real-time received, and put into model simulations with feedback control. Warning Messages Display/Plot All procedures were scheduled and triggered in sequence with controlled by computer program. Simulation results are automatically extracted from modeling outputs and exhibited on the proposed platform. Page 9 / 32
track errors(km) The Experiment of Ensemble Quantitative Precipitation Forecasts 5 typhoons(36) 19/02LT 300 250 200 150 100 50 113 105 74 79 77 67 55 67 47 57 148 137 103 96 83 184 146 134 118 117 147 292 217 212 201 185 NOGAPS JMA UK NCEP CWB TTFRI MEAN 0 12h(36) 24h(35) 36h(31) 48h(28) 60h(25) 72h(23) Forecast time (cases) 10/18 18:48 10/19 20:50 10/20 18:36 10/21 18:07 10/18 0800 ~ 10/22 0800 10/19 0800 ~ 10/22 0800 10/20 0800 ~ 10/22 0800 10/21 0800 ~ 10/22 0800 Page 10 / 32
W a te r S ta g e (m ) W a te r S ta g e (m ) R a in fa ll in te n s ity (m m /h r) R a in fa ll in te n s ity (m m /h r) Hydrological Routing 16 0 Model configurations. 12 30 Combining instantaneous field observations to provide initial accurate modeling. 8 4 T y p h o o n A R E A ( A u g.2 3 rd ~ A u g.2 6 th, 2 0 0 4 ) O b s e rv e Sim ulate M A E : 0.2 4 m 60 90 120 R M S E : 0.3 0 m Estimating infiltration for determining effective rainfall. 0 8 / 2 3 / 1 4 :0 0 8 / 2 4 / 0 1 :0 0 8 / 2 4 / 1 2 :0 0 8 / 2 4 / 2 3 :0 0 8 / 2 5 / 1 0 :0 0 8 / 2 5 / 2 1 :0 0 T im e (m /d /h o u r) 150 16 0 Hydro-meteorological simulations. 12 30 Improving our understanding of Flood Forecasting. 8 T y p h o o n N A N M A D O L ( D e c.0 3 rd ~ D e c.0 4 th, 2 0 0 4 ) O b s e rv e Sim ulate 60 90 4 M A E : 0.2 2 m 120 R M S E : 0.2 7 m 0 150 1 2 / 3 / 0 1 :0 0 1 2 / 3 / 1 0 :0 0 1 2 / 3 / 1 9 :0 0 1 2 / 4 / 0 4 :0 0 1 2 / 4 / 1 3 :0 0 1 2 / 4 / 2 2 :0 0 T im e (m /d /h o u r) Page 11 / 32
Display http://140.110.147.168/wims/ Page 12 / 32
Operational Warning System 24 hr 48 hr 72 hr am08:00, hydro-meteorological simulation done am06:00, QPF-process done, and start hydrological routing am04:00, Get NCEP-gfs data, and start rainfalls generation am00:00, Start operational process Page 13 / 32
Study Site Study site The Langyang creek basin, situated in Northeastern Taiwan, covering an area of 978 km 2. Page 14 / 32
Model Configurations We discretized the computational domain with finite element meshes using line elements for rivers and triangular elements for land Table. Land use and Manning s roughness for the study site. Agricultural Forestry Traffic Water Buildings Others Land use 43.94% 19.08% 8.04% 4.65% 15.81% 8.48% Manning s N 0.200 0.350 0.100 0.050 0.150 0.250 surface. Four sorts of Manning s roughness are assigned in studied river area based on the official reports, varying from 0.030 to 0.038. Land-use is classified into six types, as shown above, where the sets of Manning s value are calibrated in coupled 1-D/2-D systems. Page 15 / 32
Model Calibrations Model calibrations using Typhoon Sinluku and Jangmi. Event RMSE CE VER EQp Typhoon Sinluku 0.27 0.89 0.01 0.08 Typhoon Jangmi 0.35 0.88 0.03-0.04 Page 16 / 32
These two cases both show very small errors in total volume during this simulation period and captured the peak time exactly. In general, both cases reveal very good results for the coefficient of efficiency. It is noted that there is great improvement in the rising period of simulated hydrograph when the mechanism of infiltration is included. The simulated hydrograph tuned by artificial adjustments could precisely capture measured flow patterns. Page 17 / 32
Typhoon Megi (2010) Page 18 / 32
Figure. Hydro-meteorological simulations from the run that began at 18:00, Oct. 19 th of 2010. Page 19 / 32
Figure. Hydro-meteorological simulations from the run that began at 00:00, Oct. 20 th of 2010. Page 20 / 32
Figure. Hydro-meteorological simulations from the run that began at 12:00, Oct. 20 th of 2010. They warned the approach of a considerable flood, but could not reach an agreement on the peak time and related quantity. Page 21 / 32
Simulation using rain-gauge Model calibrations for Typhoon Magi. It seems easy to obtain a reasonable response using the measured rainfall to drive the watershed simulation. A perfect guess in terms of the quantity as well as distribution of precipitation is not easy to obtain. Although precipitation estimates produced by the WRF can capture the total amount of rainfall, its related distributions are sometimes inadequate for numerical hydrological modeling. Page 22 / 32
We should shift our expectations to the outcome of ensemble quantitative rainfall predictions. It is impossible to obtain a perfect guess in advance due to limitations in our understanding of the aspects of atmosphere and hydrology. The key is how to understand these types of information, and how to integrate earlier useful intelligence into the watershed model. Page 23 / 32
Typhoon Nalgae (2011) Page 24 / 32
Figure. Hydro-meteorological simulations from the run that began at 00:00, Oct. 02 nd of 2011. Page 25 / 32
Each rainfall participants has been chosen because of their individual representatives. The key to a flood warning system is not to expect too much of highly accurate quantitative precipitation forecasts, but to improve our understandings from individual ensemble flood forecasts. Therefore all participants are necessary to this QPF project with their own specific representatives conducting individual analysis. Page 26 / 32
This member captures the shapes of the predicted hydrograph poorly in the first few runs. They implied that a considerable flow peak would occur, but missed the time, which was behind the actual occurrence. They did not provide sufficient lead time for flood forecasts here. This actually provided very slight help. Page 27 / 32
The basically reveal that a flood is going to pass, but the precise time and peak are constantly being adjusted. They indeed provided some useful information for flood mitigation, because of being able to send out an early warning of the event. We can obtain some warning from early operational hydro-meteorological runs. To Issue advance warnings of possible danger is better than doing nothing. Page 28 / 32
The water stage for this event was incredibly high from the first to fourth operational predictions. The advantage in this case was that they always issued warning messages. This is helpful for mitigation of flash flood danger. Page 29 / 32
Summary Page 30 / 32
Concluding Remarks A system integrating rainfall-runoff calculations, field observations and instant data assimilations, is presented. Simulation results indicate that well calibrated coefficients of the hydrological model were not satisfied for sequential flood predictions. This is because these two types of precipitation data are systematically different because of the aspect of their generation. Page 31 / 32
Concluding Remarks Each QPF participants has been chosen because of their individual representatives. We must further understand that the key for such a warning system is not to have greatly expectations of highly accurate rainfall predictions, but to improve our understanding from the point of view of individual ensemble flood forecasts. Page 32 / 32
Thank You.