1 of 26

Flood forecasts tools for early warning systems

in the Milan urban area

A. Ceppi, E. Gambini, C. Corbari, G. Ravazzani, M. Mancini

Politecnico di Milano, Department of Civil and Environmental Engineering - Water Science and Engineering Division

Speaker: Alessandro Ceppi

Milano, 4 December 2023

Seminar CS4

2 of 26

The area of study

River

Basin Area

Seveso

277 km2

Olona

208 km2

Lambro

553 km2

Seventies

Today

Seminar CS4

3 of 26

The floods events in the last 15 years over Milan urban area

Major floodings and associated damages

Recent minor floodings (from 2018)

24/07/2020

15/05/2018

15/09/2023

  • Seveso flood (Milan)
  • Seveso flood (Milan)
  • Seveso flood (Milan)

21/09/2021

  • Olona flood (Canegrate)

15/07/2009

    • The Olona flood (Varese): 30 milion €

18/09/2010

    • The Seveso flood (Milan): 80 milion €

08/07/2014

    • The Seveso flood (Milan): 55 milion €

15/11/2014

    • The Lambro flood (Monza): 6 milion €

31/10/2023

  • Seveso flood (Milan)

Seminar CS4

4 of 26

The problem of frequent floods and damages in Milan urban Area

38%

51%

17%

2000

1980

1954

Ceppi et. al, 2022

The River Seveso in Milan city

On average, we notice a reduction of the return period of 6.9 years from 1954 to 1980, 7.2 years from 1980 to 2000 and about 1 year only in the last two decades (at Bovisio-Masciago city).

Seminar CS4

5 of 26

SOL

Flood forecasting system over Seveso, Olona,

Lambro

Counteraction measures to alert people and activities

Issue of warnings

by civil protection

Activation of local operative centre for natural disaster

time

Actions

MOCAP: an Italian acronym which stands for:

Sistema di MOnitoraggio Comunale Allerta di Piena (Bovisio-Masciago Municipality)

The flood warning system could be seen as a sort of “last mile” of the hydro-meteorological chain for civil protection purposes.

The SOL system works by coupling deterministic and probabilistic (ensemble) meteorological forecasts with the hydrological model FEST-WB

Real time hydro-meteorological tools for early warning

Ceppi et al., 2023 (under review)

Seminar CS4

6 of 26

  • 2 piezometric level measurement sensors by STS (a);
  • 2 data acquisition and transmission systems with Campbell Scientific datalogger, CR800 (b);
  • 1 Davis Vantage Pro 2 weather station (c);
  • 1 water level radar sensor VEGAPULS WL S 61 (d);
  • 1 infrared thermometer by Campbell Scientific IR100 (e) to monitor surface temperatures for road ice risk.

MOCAP: MOnitoraggio Comunale Allerta di Piena at Bovisio-Masciago

(a) (b) (c) (d) (e)

Hazard communication via sms/mail

Municipal monitoring for flood alerts

Seminar CS4

7 of 26

7

The river Seveso at Bovisio-Masciago

Discharge

H

Q

H

Q

Discharge values at the Bovisio-Masciago closure section along the river Seveso are calculated with two pressure transmitters installed to provide the water level measures and estimate the flow rate in real time using the "slope-area" method (Benson and Dalrymple, 1967), here, extended for continuous measurement (Continuous Slope Area, CSA) (Stewart et al., 2012).

MOCAP: MOnitoraggio Comunale Allerta di Piena at Bovisio-Masciago

Seminar CS4

8 of 26

  • SOL is an online tool which shows possible river floods over the Seveso, Olona and Lambro River basins (see acronym) with an advance of 24-36 hours;

  • Its technology is based on the sequential operation of meteorological, hydrological, calculation models of hydraulic engineering, and visualization on web platform;

  • SOL allows stakeholders (institutions, citizens and business companies operating in the flood risk areas) to early know how precipitation under different weather patterns can lead to dangerous flood alerts due to the disposal capacity of stream sections, soil moisture conditions and hydraulic structures present on each waterway.

  • The SOL system is freely accessible online: https://www.sol.polimi.it/

The SOL warning system: a multi-model approach

DIFFERENT METEOROLOGICAL MODELS

MONITORED RIVER SECTION

FORECASTED DISCHARGE

FORECASTED RAINFALL

FORECASTED TEMPERATURE

Seminar CS4

9 of 26

GFS

Bolam

Moloch

50 km, Δt 3h, +144h

8.3 km, Δt 1h, +72h

1.25 km, Δt 1h, +45h

WRF

5.5 km, Δt 1h, +72h

8 ensembles

Cosmo-2I

Cosmo-5M

2.2 km, Δt 3h, +48h

5 km, Δt 3h, +72h

COSMO-LEPS

7 km, Δt 3h, +132h

20 ensembles

Meteorological observations and forecast models

Meteorological stations used in the current analysis. The map shows the location of rain gauges for precipitation measurements by the Meteonetwork Association in yellow (Giazzi et al., 2022), and ARPA Lombardy in green.

Seminar CS4

10 of 26

  • Mancini, 1990: Ph.D Thesis.
  • Ceppi, A., et al., 2013: NHESS
  • Ravazzani, G. et al., 2014: Water Resources Management
  • Corbari, C. & Mancini, M. 2014: Hydrological Science Journal

Input

Output

LEGEND

Process

Internal Variable

Spatial resolution of 200 m

FEST-WB: Flash – flood Event – based Spatially – distributed rainfall – runoff Transformation – including Water Balance

Spatial interpolation:

Thiessen/IDW

Snow Dynamics

Soil Type

Vegetation Parameters

Percolation

Runoff

Sub-surface propagation

Flow routing

Flow hydrograph for each River cross section

DEM

Groundwater

LST SM

Energy Fluxes

Meteorological data

The Hydrological model: FEST-WB

www.fest.polimi.it

Seminar CS4

11 of 26

The SOL performance [2019-2022] for events exceeding yellow discharge threshold

Multi-model scores for the day+1 forecast

(Precision-Recall plot or “performance plot”)

False Positives

(False Alarms)

Hits

False Negatives (Miss Alarms)

True Negatives

Exceeding of discharge treshold

Observed alarm

Yes

No

Predicted alarm

Yes

Hits

False Positives

No

False Negatives

True Negatives

 

 

Do we prefer a system which minimizes false alarms (precision) or missed alarms (TPR)?

1 - (Miss Alarm Ratio)

1- (False Alarm Ratio)

~ 45%

~ 35%

Seminar CS4

12 of 26

QPF (Quantitative Precipitation Forecast) vs QDF (Quantitative discharge forecast) relative error evaluation for all the observed events exceeding the first warning threshold at the day+1, separating between stratiform and convective events.

Assessing the overall reliability [2019-2022]: QPF vs QDF relative error plot

Seminar CS4

13 of 26

Conceptual scheme: from Theis et al., 2005

40 translations:

(N, NE, E, SE, S, SW, W, NW by 10, 20, 30, 40, 50 km) of the forecasted rainfall field of the deterministic run

Real domain, no shift

Shifted domain

E

Forty discharge forecasts obtained by forty precipitation domain shifts issued on 7 July 2014 over the Seveso basin closed at Cantù gauge section.

Union Jack plot for the Cantù section issued on 7 July 2014: the forty values are the maximum discharge forecasts over the entire simulation horizon according to latitude and longitude shift from 0.1 to 0.5 degree which is approximately 10 to 50 km.

Shift Target (ST) approach: Union Jack plots, are they useful for civil protection purposes?

Lombardi et al., 2018

Seminar CS4

14 of 26

Rainfall Threshold (RT) is defined as the minimum cumulated volume of a rainfall event which can generate a critical water level (or discharge) at a specific river section for a given duration and an initial soil moisture condition (Mancini et. al 2002, Golian et al., 2010, Priest et al., 2011, Jang, 2015, Dao et al., 2020).

Adapted from Martina et. al, 2005

Rainfall Thresholds – Offline tool for early warning

Seminar CS4

15 of 26

Rainfall threshold daily used by Lombardy Region for Hydraulic and hydrological risk

Hazard: thresholds for both hydraulical and hydrogeological risk made by ARPA-Lombardy.

They are divided into 3 durations (6, 12, 24 hours), but they do not consider the Antecedent Moisture Conditions (AMC).

Hydraulic-Hydrogeological hazard codes

P Thresholds (mm) in 6h

P Thresholds (mm) in 12h

P Thresholds (mm) in 24h

Seminar CS4

16 of 26

“Antecedent Moisture Conditions”

Classes:

AMC I: Dry

AMC II - Average

AMC III: Wet

Luong, T.T., et. al. 2021

State of the art: the discrete «classical» approach

A small change in the value of the antecedent rainfall could cause a great change in rainfall threshold values to issue warnings.

For example: in case the observed antecedent rainfall is 13 mm, are you sure to use the AMC II threshold instead of AMC I? Are 0.3 mm (exceeding the 12.7 mm of the AMC I class limit) substantial to change threshold values? Probably not!

Seminar CS4

17 of 26

In the proposed methodology, the rainfall threshold is unique, and implictily considers both 5-days antecedent rainfalls (P5) as well as the CN. At the start of the event (i> 1mm/h) , the value of Peq0 is added to the forecasted rainfall. Peq0 is continuos with respect to P5 and CN.

«Base» equivalent rainfall 🡪 represents the worsen of the hydraulic potential risks due to past rains.

Forecasted rainfall over the basin

Gambini et al., 2023

Regional Directive n. XI/4114

 

Soil moisture accounting the «equivalent rainfall»: a new method

Seminar CS4

18 of 26

Cross section

N° of events

TPR

Prec.

F2*

TPR

(Eq. rainfall)

Prec.

(Eq. rainfall)

F2*

(Eq. rainfall)

Lainate

26

0.88

0.35

0.67

0.88

0.39

0.69

Milano Via-Feltre

16

0.88

0.38

0.71

0.88

0.55

0.74

Molteno

29

0.86

0.44

0.70

0.90

0.38

0.70

Palazzolo

24

0.86

0.41

0.74

0.92

0.44

0.76

Castellanza

12

0.92

0.59

0.82

0.92

0.63

0.83

Cantù

17

0.76

0.27

0.53

0.88

0.34

0.63

Peregallo

14

0.71

0.2

0.47

0.86

0.2

0.53

Summary of the results of empirical rainfall thresholds on SOL basins [1998-2022] for the orange threshold

 

TPR is weighted more than Precision

«I am looking to reduce the missed alarms»

Seminar CS4

19 of 26

Last flood event in Bovisio-Masciago municipality on 31 October 2023 at 5-6 am local time

MOCAP

SOL FORECASTS

UNION JACK

LIGHTNINGS

RADAR + OBS. RAIN GAUGE

Seminar CS4

20 of 26

Forecasting a hazardous event (H) with a sufficient time to save people (V) and materials (E)

RISK = H*E*V

…but reduce exposure (E) and vulnerability (V)

Do not reduce hazards, H…

Not to keep the water away from the people…

….but the people away from the water (Nemec, 1986)

Milano, July 2014

Milano, November 2014

Flood monitoring and forecasting in real time are fundamental tools to live in inclusive, safe, resilient and sustainable cities!

Conclusions

Seminar CS4

21 of 26

Thank you for your attention

The SOL System

The Real Time Hydrology group website, research team lead by Prof. Marco Mancini

Alessandro Ceppi: alessandro.ceppi@polimi.it

The MOCAP System

Seminar CS4

22 of 26

Seminar CS4

23 of 26

The slope-area method

Slope-area discharge calculation is based on the equations of energy conservation and Manning.

It requires to know:

  • Distance between two channel gauge sections
  • Measure of water levels in the two monitored sections, in order to calculate the friction slope
  • The flow area
  • The hydraulics radius
  • The Manning’s roughness coefficient

Seminar CS4

24 of 26

Seminar CS4

25 of 26

Il premio “Italiadecide”: SOL – Sistema di previsione di piena dei fiumi Seveso – Olona - Lambro

26 of 26

«Base» equivalent rainfall 🡪 represents the worsen of the hydraulical potential risks due to past rains.

Forecasted rainfall over the basin

From Gambini et. al 2023, under review

Regional directive n. XI/4114

Mishra & Singh, 2004

 

Soil moisture accounting – the «equivalent rainfall» method

 

Where M=f(CN,P5) is the “antecedent moisture” term (Mishra & Singh, 2004)

 

Seminar CS4