World Environmental and Water Resources Congress 2018 30

Defining Real-Time Flood Alerts with Multicriteria Analysis

A. S. K. B. Sosnoski1; M. T. L. Barros, M.ASCE2; F. Conde3; S. Pion4; and S. Uemura5
1
Engineer, Hydraulic Technology Centre Foundation (FCTH), São Paulo, Brazil. E-mail:
[email protected]
2
Professor, Dept. of Hydraulic and Environmental Engineering, Univ. of São Paulo (USP),
Brazil. E-mail: [email protected]
Downloaded from ascelibrary.org by University of Texas at Austin on 09/13/19. Copyright ASCE. For personal use only; all rights reserved.

3
Coordinator, Hydraulic Technology Centre Foundation (FCTH), São Paulo, Brazil. E-mail:
[email protected]
4
Engineer, Hydraulic Technology Centre Foundation (FCTH), São Paulo, Brazil. E-mail:
[email protected]
5
Engineer, Hydraulic Technology Centre Foundation (FCTH), São Paulo, Brazil. E-mail:
[email protected]

ABSTRACT
People confidence in flood forecasting systems is related to the quality and accuracy of the
forecasts. These two properties are basically function of the set of models that produces the
forecasts. If flood events are observed but not forecasted or forecasted but not observed, the alert
can be considered an error by citizens not well informed about how flood forecasting system
works. If a lot of forecasting errors occur, citizen confidence may be reduced for future alerts.
The main objective of this paper is to develop a methodology to produce flood forecasts risk
maps for different urban catchments. To implement the quality of the forecasts a multi criteria
analysis is proposed. This method can prevent the spread of misunderstanding information to
many citizens. The case study is a small catchment in São Paulo City named Jaguaré Basin. This
area was chosen due to its diverse floodplain occupation. The real-time flood forecast alerts can
be represented by thresholds that show a certain real time hydrological condition. This
information can be stated by flood maps. The methodology to define these thresholds is based in
a set of parameters: runoff depth, water velocity, flood probability, and construction
vulnerability. This set of parameters is used to define more precise alert boundaries. In this way
it is possible to have more reliable flood forecasts.
KEYWORDS: flood risk map, multi criteria analysis, flood forecasting systems.

INTRODUCTION
Flooding is one of the main environment problems for people living a in urban areas. In the
city of São Paulo floods occur mainly in the summer, from October to March. The basic reason
for this is the urbanization. São Paulo population increased a lot in the twenty Century, from
200.00 inhabitants to more than 20 million today. This geometric increase of population created
many social, economic and environmental problems, summer flood is one of them. In order to
minimize the impacts of the floods, the city has a flood forecasting system that started its
operation in 1976 with a very few telemetric gages. Today the flood forecast system has more
than 300 rain and stage stations and two weather radars. The real-time data are input for a set of
forecast math models which supplies a Decision Support System (DSS) accessed by many
private and public service companies (www.saisp.br). One of them is the São Paulo municipality.
Floods in the city of São Paulo have several causes, one of the most important being the
inadequate use of urban soil, mainly the irregular occupation of the flood zone of small rivers by

© ASCE

World Environmental and Water Resources Congress 2018

 World Environmental and Water Resources Congress 2018 31

the low-income population. In addition, the city's growth has greatly increased the index of
impervious areas, generating huge impacts on the hydrological cycle. Due to these diversity of
use and occupation of the soil the forecasting math models need to be adapted to different
conditions. The drainage hydraulic systems are basically composed of channels, polders and
detention ponds. In general, the drainage systems were designed many years ago for small return
periods (maximum 25 years). In some basins the present flow capacity is below 25 years.
Although Brazilian laws are advanced in water resource management, there is no explicit
Downloaded from ascelibrary.org by University of Texas at Austin on 09/13/19. Copyright ASCE. For personal use only; all rights reserved.

legal imposition that obligates municipalities to implement flood forecasting or flood mapping in
urban areas. This situation is different from other countries, where guidelines and regulations
require the use of hydraulic models to map flood areas. (FEMA 2002).
In the pursuit to obtain better flood protection São Paulo Flood Alert System (SAISP) is
using in real-time a mathematical simulation model to forecast runoff and floodplains for twenty-
five (25) of its most critical small catchments. The model uses observed weather radar data and
forecasted precipitation maps for one-hour ahead to predict the flood impacts.
SAISP DSS system models simulate the hydraulic of the drainage systems to produce flood
maps. The hydrodynamic simulation outputs are flow and water level along of the river channels.
Considering the hydraulic simulation of the drainage system the SAISP model produces flood
forecasts considering four levels of hydrological severity: “attention”, “alert”, “emergency”
and “flooded”. These levels are linked to the risk of flood in the next hour and they were
empirically estimated for all twenty-five catchments covered by SAISP. This methodology is
also used by the Advanced Hydrologic Prediction Service (AHPS) in the United States. AHPS
produces animated flood maps with high, medium, and low risks at different locations (NOAA
2011).
Flood forecast is a process that uses a set of parameters and data through rigorous
simulations and historic analysis to define and to inform citizens and companies about the
possibility or imminent risk of an area to be flooded. The forecasts have a series of
approximations many of them linked to the model accuracy and the uncertainty of the short-term
rain forecasts.
This paper describes the method to obtain the forecast flood maps. Basically, the forecasts
need to considerate: runoff depth, runoff velocity, soil use and losses scale. The risk of flood is
estimated by multicriteria analysis. This methodology can enhance the precision and the
confidence on flood alerts.

METHODOLOGY
Real-time flood forecasting
The real-time flood forecasting model used in this study is operated by the SAISP. It operates
a telemetry system with 300 rain and scale stations installed in the Upper Tiete River Basin
(UTRB) in the main river and its important tributaries. The SAISP also operates two weather
radars: a S-Band and X Band both installed in the UTRB.
The S-Band radar operates in a standard 2×2 km horizontal spatial resolution within a radius
of 240 km. When there is no precipitation detected, data is stored every 30 minutes, from the
moment it detects traces of precipitation every 5 minutes. The radar input data for the flood
model is the CAPPI (Constant Altitude Plan Position Indicator) observed at 3 km of altitude.
Spatial rainfall forecast is generated by the SHARO model (Short-Term Automatic Rainfall
Prediction) and is based on the special correlation between two CAPPI images at different

© ASCE

World Environmental and Water Resources Congress 2018

 World Environmental and Water Resources Congress 2018 32

moments and produces rainfall forecasts every 10 minutes for up to 1 hour ahead in 5 minutes
intervals. The viability of using radar data on modeling was described by Gonçalves et al.
(2004).
Downloaded from ascelibrary.org by University of Texas at Austin on 09/13/19. Copyright ASCE. For personal use only; all rights reserved.

Figure 1. Flood Severity Grid Classification (FEMA, 2014)

To reduce correlation errors from radar spatial data and local rainfall data observed in the
telemetric gages it is used the integration of rainfall measured by the rain gages with the data
from weather developed by Rocha Filho et al (2013). The method applied for the integration of
radar data and telemetry data is the Conditional Merging, which assumes that the weather radar
can accurately capture the spatial precipitation field, but not its intensity, while the rain gage is
able to measure accurately the rain in each position in the basin. These integrated data are
available as an input for the flood forecasting model.

Flood Mapping, Rainfall-Runoff model

The model applied for converting hydrological data into runoff and consequently flow, depth
and flooded areas is the Storm Water Management Model (SWMM), available by US
Environmental Protection Agency (EPA), with the PCSWMM interface, developed by
Computational Hydraulics International (CHI). The model runs every 5 minutes, independently
of rainfall detection by the radar. Due to this, it was necessary to implement in the model
groundwater flows representing the aquifer recharge.
Flood Forecasts
The usual process of developing a flood forecasts is to gather and interpret data regarding
actual and future basin parameters as rainfall, runoff, river levels, detention ponds levels and
vulnerable assets among others. SAISP produces two categories of flood forecasts:
 Water level (Hydrological Thresholds) – Due to the water levels in scale stations located
in some critical region of the basin;
 Dynamic flood maps – As a product of the Rainfall-Runoff model, flood extents are
calculated and presented in real-time in the SAISP DSS.

Multicriteria Analysis
To implement a flood alert that considers all the parameters involved it is necessary to use a
tool that can translate the interaction and importance of each parameter. For this a multicriteria
model is applied following the method proposed by Zopounidis & Pardalos (2010).

© ASCE

World Environmental and Water Resources Congress 2018

 World Environmental and Water Resources Congress 2018 33

The parameters chosen for the multi criteria analysis were the probability of an event to
occur, velocity and water depth. The probability is estimated by the return period and in this
study the probabilities range from 0.5 to 0.01. The velocity from 0 to 3 m/s and water depth from
0 to 4.0 meters reference for these values is presented by FEMA (2014) Figure 1. Regarding
edification in riparian areas, the classification proposed by Edwards J. (2007) was used.
For the multicriteria analysis, it is important to develop a weighting method to describe
parameter variations and classify them in terms of flood risk. In this method flood losses have a
Downloaded from ascelibrary.org by University of Texas at Austin on 09/13/19. Copyright ASCE. For personal use only; all rights reserved.

scale from 1 to 5, from normal to extreme respectively. The weighs are presented in the Table 1.
Each set of the parameters defines the thresholds from low to extreme. Considering the real-time
operation every 10 minutes a new scenario is formulated and critical locations can be classified
by order of flood hazards. In this way flood alerts are highlighted only for the most critical areas.

Table 1. The losses scale adopted for the multicriteria analysis

Parameter grades 1 2 3 4 5
Probability 0.5 0.1 0.04 0.02 0.01
Velocity (m/s) 0 0.7 1.5 2.4 3
Depth (m) 0 1 2 3 4
Edification type Low (1) Medium (2) Hight (3) Very High (4) Extreme (5)

A flowchart of the methodology is shown in Figure 2. The steps of the methodology are
clearer in this Figure.

Figure 2. Flowchart of the methodology

© ASCE

World Environmental and Water Resources Congress 2018

 World Environmental and Water Resources Congress 2018 34

CASE STUDY
This method was applied in a typical small catchment in the city of São Paulo: the Jaguaré
River Basin. This basin has an area of 28.2 km² and represents 1.9% of São Paulo area. The main
channel is 11 km long and contributes to Pinheiros river, one of the main tributaries of Tiete
River. Figure 3 shows the map of Jaguaré River Basin.
Downloaded from ascelibrary.org by University of Texas at Austin on 09/13/19. Copyright ASCE. For personal use only; all rights reserved.

Figure 3. Jaguaré River Map (in detail and in the UTRB)

Figure 4. Edification and flood layer. (Jaguaré Basin)

A hydrodynamic model was applied to the Jaguaré river channel to estimate the floodplain.
Once the estimated floodplain is obtained the result is cross connected with the edification
(property) data base. This map helps the decision makers to identify the flood impacts in the
area, residences or public services, for example, and decide by the losses scale to be adopted.
The usual flood alerts are displayed when the predicted floodplain reaches specific levels and
when properties are inside the flooded area. This process is done with the intersection of water

© ASCE

World Environmental and Water Resources Congress 2018

 World Environmental and Water Resources Congress 2018 35

levels calculated in the river cross section with the lowest point in the property. Therefore, in the
traditional analysis there is no distinction between different types of property occupation and its
specific use. There is no reference to the risk involved with water velocity and human safety for
the alerts. Figure 4 shows an example of the flood area estimated by the model.
Downloaded from ascelibrary.org by University of Texas at Austin on 09/13/19. Copyright ASCE. For personal use only; all rights reserved.

Figure 5. (a) Water level, (b) velocity and (c) edification type and flood risk scales
In this way in the case of a flood event, properties with different water levels (WL) and water
velocity relationship will be included in the same alert category. As an example, in an event
where only water level is considered conjugated lots will be similarly included in the same flood

© ASCE

World Environmental and Water Resources Congress 2018

 World Environmental and Water Resources Congress 2018 36

alert.
In the case where water levels and velocity are considered, there is the possibility that one of
property is severe impacted with a combination of WL and Velocity and not for one with
moderate WL and low Velocity. The same idea can be applied to the situation where one of the
impacted properties is a vacant lot and the other is a nursery home. Both will receive the flood
alert, but it is more important to alert the critic nursery home and not the vacant lot.
To apply the multicriteria analysis to real-time flood alerts, the process runs every 10
Downloaded from ascelibrary.org by University of Texas at Austin on 09/13/19. Copyright ASCE. For personal use only; all rights reserved.

minutes. Therefore, the model produces simple and straightforward information that can be
easily interpreted. Flood maps is an easy way to present spatialized criteria, each parameter can
be plotted in a set of maps, where can be graduated by its parameter grades. In this study the
maps were done with GIS tools integrated in the interface of the Rainfall-Runoff model. Figure 5
shows a set of maps for different parameters.
Considering the intersection of this layers and using multicriteria analysis real-time risk maps
can be produce and used to identify specific areas and properties to be alerted. Parameters
obtained with the Rainfall-Runoff modelling are objectively graded. The model determines WL
and water velocity, consequently defines the risk scale. Grading parameters like property use is a
subjective process, they are entitled by decision and policy makers who define the criteria
adopted for the soil use classification. The analysis depends on how is accurate the classification
regarding property uses and vulnerability. So, it is important to determine grades for each
property according to a set of criteria defined the decision makers.
In the Jaguaré Basin, the round of property use and vulnerability evaluation have been made
by five engineers with knowledge of the basin flood condition and basin occupation.
Vulnerability evaluation was done by field data and thru Google Maps Street View. When
applying the multicriteria analysis the resulting risk map presents a pattern of alerts for the
region. The results are presented in Figure 6.

Figure 6. Flood Risk Map

To describe the status of alerts this method considers three categories:
 Alert : Inform property owner
 Attention : Hold the alert status until the next simulation
 Normal : No alert

© ASCE

World Environmental and Water Resources Congress 2018

 World Environmental and Water Resources Congress 2018 37

CONCLUSIONS
This study presented a methodology to map floods using multicriteria analysis. With this
methodology is possible to reduce the number of “fake alerts” and improve citizen confidence in
alert systems. Flood risk maps are generated every ten minutes and clearly indicate a shorter
number of affected properties. Although procedures are simple, there is the influence of multiple
decision makers which introduces subjective elements to the process. It is important to point out
that this procedure is only possible with the application of an intensive work on hydraulic
Downloaded from ascelibrary.org by University of Texas at Austin on 09/13/19. Copyright ASCE. For personal use only; all rights reserved.

modelling. Once the procedure is applicable, it is possible not only to improve the alert system
but also to point specific infrastructure problems. This can help to design new drainage works in
the basin.

ACKNOWLEDGMENTS
To SAISP people and to São Paulo city municipality who gave support for this study.

REFERENCES
Edwards J. (2007). Handbook for vulnerability mapping.
FEMA (2014). Flood depth and analysis grids, 2014.
FEMA. (2002). “Guidelines and specifications for flood hazard mapping partners appendix C:
Guidance for riverine flooding analyses and mapping. ”https://www.fema.gov/media-
library/assets/documents/13948”, (December 10, 2017).
Gonçalves, J. E. et al. Estudo da viabilidade da utilização de estimativas de chuva efetuadas por
radar na modelagem hidrológica. In: Congresso Brasileiro de Meteorologia, 13. 2004,
Fortaleza, CD-ROM
Handbook of Multicriteria Analysis, Zopounidis, C.; Pardalos, P. M. (Eds.) 2010, XXV, 455 p.
92 illus., Hardcover ISBN: 978-3-540-92827-0
NOAA (National Oceanic and Atmospheric Administration). (2011). “NOAA partnered
guidelines for the development of AHPS flood inundation mapping.
“http://water.weather.gov/ahps/NOAA_AHPS_Guidelines_Final_2011_v3.pdf “(December
10, 2017).
Rocha Filho, K. L;Conde, F.; Andrioli, C. P. Correção em tempo real da precipitação estimada
por um radar meteorológico com uma rede de superfície. In Anais do XX Simpósio
Brasileiro de Recursos Hídricos, Bento Gonçalves, nov. 2013.
United Nations Environment Programme (UNEP), Industry and Environment, 1998, Hazard
Identification and Evaluation in a Local Community, Technical Report nr. 12.

© ASCE

World Environmental and Water Resources Congress 2018