Vienna, 18-22 April 16 Session HS4.1/AS4.3/GM9.12/NH1.7 Flash floods and associated hydro-geomorphic processes: observation, modelling and warning Real time probabilistic precipitation forecasts in the Milano urban area: comparison between a physics and pragmatic approach A. Ceppi 1 G. Ravazzani 1, A. Amengual 2, G. Lombardi 1, V. Homar 2, R. Romero 2 and M. Mancini 1 1 Department of Civil and Environmental Engineering, Politecnico di Milano, Milano, Italy 2 Grup de Meteorologia, Departament de Física, Universitat de les Illes Balears, Palma de Mallorca, Spain Background and motivation of the study Background Precipitation forecasts from mesoscale numerical weather prediction (NWP) models often contain features that are not deterministically predictable. In particular, accurate forecasts of deep moist convection and extreme rainfall are arduous to be predicted in terms of amount, time and target over small hydrological basins due to uncertainties arising from the numerical weather prediction, physical parameterizations and high sensitivity to misrepresentation of the atmospheric state, therefore they require a probabilistic forecast approach. Objectives Explore different setups of ENSEMBLE simulations to detect what is the most reliable for real time flood forecasting of convective events in Milan Approach The first approach is based on a hydrological ensemble prediction system (HEPS) designed to explicitly cope with uncertainties in the initial and lateral boundary conditions (IC/LBCs) and physical parameterizations of the NWP model. The second involves a pragmatic postprocessing procedure by randomly shifting in space the precipitation field provided by the deterministic WRF control run in order to get a cluster of different simulations. 2
Area of study Ticino W Como 1 Olona N S 3 E Seveso 5 6 Pusiano Lambro River Tributary Waterways CSNO Lake Milano City River basin 2 Villoresi 4 7 Monza Area (km 2 ) Olona = 8 Seveso = 7 Lambro = 5 Total = 13 Naviglio Grande 1 3 Kilometers 8 Milano Naviglio Pavese Adda 3 Flood and urbanization problem Milan city is a flood prone area that has been frequently flooded in the past and recent years. 1954 198 17% 38% 51% List of main floods in the 7s, 8s and 9s 4
Recent floods: the seven analyzed events 5 Hydrological model: FEST-WB Snow Dynamics Meteorological forcings Spatial interpolation: Thiessen, IDW Soil Moisture update FEST-WB: Flash flood Event based Spatially distributed rainfall runoff Transformation including Water Balance Soil Parameters Vegetation Parameters Mancini, 199: PhD thesis Ceppi, A., et al. 13: NHESS, 13(4), 151-162. Ravazzani, G. et al., 14: Water Resources Management, 28(4), 133-144 Corbari, C., Mancini, M., 14: Hydrological science journal, 59 (1), 183-1843 DEM x = m t = 1 h Percolation Surface Runoff Definition of river network Subsurface routing groundw ater Surface flow routing Lakes and reservoires LEGEND Input Process Hydrograph Output Internal variable 6
Meteorological model: WRF Dynamical downscaling performed with WRF 3.4 with: 2.5 km grid spacing and 28 vertical levels Initial and Boundary Conditions coming from the global ECMWF HIGH RESOLUTION => CONVECTION EXPLICITLY SOLVED The WRF model domain Initial and Boundary condition: ECMWF.25 WRF grid: 2.5 km Forecast horizon: 48h Temporal output: 1 hour Starting run: : UTC Vertical levels: 28 Shortwave radiation scheme for the WRF: Dudhia; Dudhia (1989, JAS) Longwave radiation scheme for the WRF: RRTM; Mlawer et al. (1997, JGR) Microphysics scheme for the WRF: WSM6; Hong and Lim (6, JKMS) Land surface model: Noah Planetary boundary layer for the WRF: MYJ; Janjic (1994, MWR) 7 Different approaches: physics vs pragmatic physics pragmatic 8
The physics approach European Center for Medium range Weather Forecasts global Ensemble Predictions System (ECMWF-EPS) consists of 5 members, operating at T639 spectral resolution (~32 km), that are generated by perturbing an initial analysis. Perturbations are derived from flow-dependent singular vectors computed daily at ECMWF in order to span the synoptic-scale uncertainties of the day (Buizza and Palmer, 1995; Molteni et al., 1996). The ECMWF-EPS members exhibiting the largest diversity over our numerical domain are identified and used as initial and boundary conditions for the entire PILB ensemble. To this end, we applied to the 5 ECMWF-EPS members a k-means clustering algorithm using the Principal Components of the 5 hpa geopotential and 85 hpa temperature fields over the area spanned by the WRF domain. Microphysics scheme Planetary boundary scheme (PBL) Purdue Lin (Lin et al., 1983) Yonsei University (YSU; Hong et al, 9) Ferrier (1994) Mellor-Yamada-Janjic (MYJ; Janjic, 1994) WRF single-moment 6-class (WSM6; Hong and Lim, 6) Goddard scheme (Tao et al., 1989) New Thompson (Thomson et al., 8) Mellor-Yamada Nakanishi Niino level 2.5 (MYNN; Nakanishi and Niino, 6) Asymmetric convection model 2 scheme (ACM2; Pleim, 7) Ravazzani., G., Amengual, A., Ceppi, A., Lombardi, G., Romero, R., Homar, V., Mancini, M. A hydro-meteorological ensemble prediction system for real-time flood forecasting purposes in the Milano area. Submitted to Journal of hydrology. Special issue "Flash floods, hydro-geomorphic response and risk management" Reanalysis of the two major convective flood events 8 milion as total damages! 1
Results: Precipitation and Discharge PILB vs MPS Peak Box plot, Zappa et al, 13 (Hydrological processes) 14 1 event - Seveso at Cantù - Peak Flow Analysis Observed (18/9/1-14: - 18.25 m3/s) sim Fest-WB 35 Control Simulation PILB-HEPS Accumulated Precipitation [mm] 1 1 8 6 4 PILB MPS PILB MPS PILB MPS Seveso at Cantù Lambro at Peregallo Lambro at Milano WRF model at basin close section Deviation from observed peak discharge [m 3 /s] PILB-HEPS Median PILB-HEPS 25-75 3 PILB-HEPS Min-Max MPS-HEPS MPS-HEPS MPS-HEPS Median 25-75 25 MPS-HEPS Min-Max Warning Threshold 15 1 5-35 -25-15 -5-5 5 15-1 -15 - Delay from observed peak time [hour] Deviation from observed peak discharge [m 3 /s] 1 event - Lambro at Peregallo - Peak Flow Analysis Observed (18/9/1-17: - 91.56 mc/s) sim Fest-WB 1 Control Simulation PILB-HEPS PILB-HEPS Median PILB-HEPS 25-75 PILB-HEPS Min-Max MPS-HEPS 8 MPS-HEPS Median MPS-HEPS 25-75 MPS-HEPS Min-Max Warning Threshold 6 4-35 -3-25 - -15-1 -5 5 1 - -4-6 -8 Delay from observed peak time [hour] -1 Deviation from observed peak discharge [m 3 /s] 1 event - Lambro at Milano - Peak Flow Analysis Observed (18/9/1-: - 117.32 mc/s) sim Fest-WB Control Simulation PILB-HEPS PILB-HEPS Median PILB-HEPS 25-75 PILB-HEPS Min-Max MPS-HEPS MPS-HEPS Median MPS-HEPS 25-75 MPS-HEPS Min-Max Warning Threshold 6 4-35 -3-25 - -15-1 -5-5 -4-6 -8-1 Delay from observed peak time [hour] -1 11 Reanalysis of the two major convective flood events Return period of this meteorological event: 1- years 55 milion as total damages! 12
Results: Precipitation and Discharge PILB, MPS, MPS1h Peak Box plot, Zappa et al, 13 (Hydrological processes) Accumulated Precipitation [mm] Deviation from observed peak discharge [m 3 /s] 1 1 8 6 4 PILB MPS1h MPS PILB MPS1h MPS PILB MPS1h MPS Seveso at Cantù Lambro at Peregallo Lambro at Milano WRF model at basin close section 14 event - Lambro at Peregallo - Peak Flow Analysis Observed (8/7/14-11: - 12.85 mc/s) sim Fest-WB Control Simulation PILB-HEPS PILB-HEPS Median PILB-HEPS 25-75 PILB-HEPS Min-Max MPS1h-HEPS MPS1h-HEPS Median MPS1h-HEPS 25-75 MPS1h-HEPS Min-Max MPS-HEPS MPS-HEPS Median MPS-HEPS 25-75 MPS-HEPS Min-Max 6 Warning Threshold 4-15 -1-5 5 1 15 - -4-6 -8-1 Delay from observed peak time [hour] Deviation from observed peak discharge [m 3 /s] Deviation from observed peak discharge [m 3 /s] 14 event - Seveso at Cantù - Peak Flow Analysis (8/7/14-2: - 48.9 mc/s) sim Fest-WB Control Observed Simulation PILB-HEPS PILB-HEPS Median PILB-HEPS 25-75 PILB-HEPS Min-Max MPS1h-HEPS MPS1h-HEPS Median MPS1h-HEPS 25-75 MPS1h-HEPS Min-Max MPS-HEPS MPS-HEPS Median MPS-HEPS 25-75 MPS-HEPS Min-Max Warning Threshold 1-1 -5 5 1 15-1 - -3-4 -5 Delay from observed peak time [hour] 14 event - Lambro at Milano - Peak Flow Analysis Observed (8/7/14-9: - 11.8 mc/s) sim Fest-WB Control Simulation PILB-HEPS PILB-HEPS Median PILB-HEPS 25-75 PILB-HEPS Min-Max MPS1h-HEPS MPS1h-HEPS MPS1h-HEPS Median 25-75 MPS1h-HEPS Min-Max MPS-HEPS MPS-HEPS MPS-HEPS Median 25-75 6 MPS-HEPS Min-Max Warning Threshold 4-1 -5 5 1 15 - -4-6 -8-1 Delay from observed peak time [hour] 13 Results Probabilistic PILB, MPS, MPS1h, Lagged and Lagged 1h This physics approach has a high computational cost! 14
The convective event of 8 July 14 on the Seveso basin Cantù gauging section Q observed: 48.9 m 3 /s The convective event of 8 July 14 on the Seveso basin Cantù gauging section Q simulated by the hydrological model FEST-EWB: 34.6 m 3 /s
The convective event of 8 July 14 on the Seveso basin Cantù gauging section Q forecasted by the meteorological WRF deterministic model: 8.2 m 3 /s Spatial comparison of the target rainfall: observed vs forecasted Observed precipitation Observed Cumulated rainfall: 4 mm 1 km Forecasted precipitation Forecasted Cumulated rainfall: 27 mm Distance: 28 km 1 km
Spatial comparison of the target rainfall: observed vs forecasted Observed precipitation Observed Cumulated rainfall: 43 mm 1 km Forecasted precipitation Forecasted Cumulated rainfall: 31 mm Distance: 28 km 1 km The pragmatic approach: Shift Target (ST) ensemble Conceptual scheme Theis et al., 5, Met. App. NW 5 km N 4 km 3 km NE 4 translations (N, NE, E, SE, S, SW, W, NW by 1,, 3, 4, 5 km) of the forecasted rainfall field of the WRF deterministic (control) run W 5 km 4 km 3 km km km 1 km E 1 km Real domain, no shift Shifted domain SW S SE
The pragmatic approach: Shift Target (ST) ensemble 4 translations of the forecasted rainfall field of the WRF deterministic run The convective event of 7-8 July 14 on the Seveso basin: shift-target simulations From rainfall Shift Target (ST) simulations to probabilistic hydrological ensembles: the Union Jack plot NO -5-4 -3 - -1 N 1 3 4 5 NE 5.93 2.87.54 5 4 1.25 1.64 1.5 4 3 5.67 1.41 2.65 3 2.93 2.9 5.12 1 8.12 3.92 7.5 1 O 3.3 1.42 3.22 18.71 16.7 8.18 25.6 11.16 14.28 4.73 5.77-1 12.33 6.39 24.63-1 - 6.58 8.28 9.53 - -3 7. 6.64 13.44-3 -4 7.65 8.26 1.21-4 E The values show each peak discharge for the 4 Shift-Target simulations Key: Not exceeding Exceeding threshold Exceeding rate: -5 3.24 4.35 3.9-5 SO -5-4 -3 - -1 S 1 3 4 5 SE 22
Results Deterministic (Control Run) 23 Results Shift Target (ST) ensemble 24
Shift Target (ST) ensemble: Union Jack plot, are they useful for civil protection purposes? NO -5-4 -3 - -1 N 1 3 4 5 NE 5 33.87 24.81 12.33 5 4 49.3 26.96 15.65 4 3 47.37 29.19 18.51 3 45.71 3.15 21.72 1 4.58 27.4 27.5 1 O 87.62 62.34 74.69 76.53 33.16 41.76 59.26 25.47 24.4 4.69 53.7 E -1 66.1 5.52 42.74-1 - 79.59 37.58 33.88 - -3 59.58 52.72 33.93-3 -4 73.66 79. 69.79-4 -5 8.1 65.64 65.52-5 SO -5-4 -3 - -1 S 1 3 4 5 SE NO -5-4 -3 - -1 N 1 3 4 5 NE 5 1.81 6.93 43.88 5 4 2.58 9.99 64.45 4 3 4.43.14 18.87 3 19.32 21.59 13.41 1 24.63 33.22 74.37 1 O 2.61 5.53 28.55 17.32 22.64 31.64 93.96 11.95 57.94 96.98 1.87 E -1.92 35.78 87.8-1 - 13.84 26.4 59.5 - -3 37. 4.41 134.76-3 -4 39.38 34.3 185.24-4 -5 8.26 66.16 137.27-5 SO -5-4 -3 - -1 S 1 3 4 5 SE NO -5-4 -3 - -1 N 1 3 4 5 NE 5 33.87 1.81 1..81 24.81 6.93 1.88.82 43.88 12.33 4.87.9 5 4 49.3 2.58 1.93.81 26.96 9.99 3.99.81 64.45 15.65 1.67 2.17 4 3 47.37 4.43 3.86.81 29.19.14 2.96.81 18.87 18.51 2.76.91 3 19.32 45.71 12.64.81 3.15 21.59 51.45.82 13.41 21.72 28.19.89 1 24.63 4.58 9.97.83 27.4 89.7 33.22.82 95.83 74.37 27.5.84 1 O 168.49 87.62 2.61.81 12.29 62.34 5.53.81 146.39 74.69 28.55.81 139.84 76.53 17.32.95 33.16 76.78 22.64.9 41.76 83.38 31.64 1.37 59.26 83.85 93.96 2.28 11.95 25.47 98.36 1.36 57.94 75.82 24.4.88 96.98 21.4 4.69.82 1.87 53.7 27.61.81 E -1.92 66.1 54.79 2.22 5.52 4.26 35.78 2.65 42.74 58.93 87.8 3.61-1 - 13.84 28. 79.59 3.1 37.58 26.4 1.81 4.6 33.88 59.5 14.3.86 - -3 37. 11.82 59.58 1.24 52.72 4.41 5.32 2.84 134.76 33.93.81 4.56-3 -4 39.38 73.66.86 5.47 79. 34.3.94 3.85 185.24 69.79.93 4.64-4 -5 8.1 8.26.9 3.76 65.64 66.16.83 8.25 137.27 65.52 11.15.82-5 SO -5-4 -3 - -1 S 1 3 4 5 SE NO -5-4 -3 - -1 N 1 3 4 5 NE 5.81.82.9 5 4.81.81 2.17 4 3.81.81.91 3.81.82.89 1.83.82.84 1 O.81.81.81.95.9 1.37 2.28 1.36.88.82.81-1 2.22 2.65 3.61-1 - 3.1 4.6.86 - -3 1.24 5.32.81-3 -4.86.94.93-4 -5.9.83.82-5 SO -5-4 -3 - -1 S 1 3 4 5 SE Legend: Don t exceeding threshold Exceeding threshold Exceeding rate: NO -5-4 -3 - -1 N 1 3 4 5 NE 5 1. 1.88 4.87 5 4 1.93 3.99 1.67 4 3 3.86 2.96 2.76 3 12.64 51.45 28.19 1 9.97 89.7 95.83 1 O 168.49 12.29 146.39 139.84 76.78 83.38 83.85 98.36 75.82 21.4 27.61 E -1 54.79 4.26 58.93-1 - 28. 1.81 14.3 - -3 11.82 2.84 4.56-3 -4 5.47 3.85 4.64-4 -5 3.76 8.25 11.15-5 SO -5-4 -3 - -1 S 1 3 4 5 SE 25 E Conclusions and future developments 26
Conclusions and future developments 1) Despite structural measures, flood residual risk in Milan is still very high due to land use change in the past years that lead to an increase of flood frequency. Therefore, there is a need to test non-structural measures as real time flood forecasting systems. 2) The multiphysics forecast (MPS) gave better or equal performance to the PILB ensembles. 3) Although the physics-based approach needs a high computational cost, it outperforms the pragmatic set of configurations, which, however, turns out to be an acceptable low-budget alternative for real time flood forecasts over small urban basins when a single deterministic run is available. 4) Future developments involve the analysis of more events in order to detect if there are some physical schemes more capable than the others in simulating convective/stratiform events in Milan area. 27 Thank you for your attention alessandro.ceppi@polimi.it