DSD-INT 2016 Real-time flood extent maps based on social media - Van Loenen

Real-time flood extent maps

based on social media

Arnejan van Loenen

Dirk Eilander

Patricia Trambauer

Jurjen Wagemaker (FloodTags.com)

Email: [email protected]

FloodControl 2015 – Dashboard Banjir Online

Pilot area: Jakarta

3

Frequent floods in

Jakarta

Feb 9th-11th 2015

• > 728 000 tweets

• peak 900 tweets/minute

• 2200 incl. water depth

• ~ 900 tweets (40%)

location

• For comparison: 10-20

water level gauges (not

all functioning)

“15.31 #Flood in Cipinang

Melayu where the water

level reaches up to +/ - 2m”

Flood observations in the “Twitter cloud”

Data integration platform: Delft-FEWS

Global Flood Monitor

5

Contents

• What social media can add to flood disaster management

• How to find the useful information

• Real-time flood mapping

• There is some uncertainty though

Information sources during floods

Telemetry CCTV

Rain radar Remote Sensing

Potential use of social media

8

-Finding flood prone areas

-validating hydrodynamic

models

- Currently flooded areas

- People in need

- Evacuation routes

Flood forecasting

Flood prevention

Disaster management

Disaster recovery - Flood Impact analysis

How to find the useful information

9

Is it a flood prone area?

Method 1: Height Above Nearest Drainage (HAND)

The HAND model normalizes topography according to the local

relative heights found along the drainage network

10

Is it a flood prone area?

Method 2: Slope

11

Is it a flood prone area?

12

• HAND (<3.5m) & Slope (<0.2 m-1) match well with flood

areas from MODISlance archive

• Uncertainty in threshold derived from bootstrapping HAND

and Slope flood areas

• Could be used to improve geo-location tweets

What is heavy rain?

13

90th percentile

99th percentile

Local hourly precipitation Accumulated daily precipitation

Contextual validation method

Static maps Dynamic maps

Weighted average taken as

Proxy for probability of flood

Real-time flood mapping

15

• Twitter Data

5 Min resolution

Water depth

Location

• February 5th 2014

Peak 250/5

minutes

Probabilistic approach

16

P = 0.3

P = 0.5

P = 0.2

x

x

x

Results 10 Feb 2015

17

York floods 2015: deterministic approach

29 m

0 m0 5 102.5 Km

Generate flood plane

19

• Group observations by flowpath

• Interpolate water levels on reach

! Observation0 3 61.5 Km

29 m

0 m0 5 102.5 Km

York floods, 2015

20

[1]

[3]

[5]

[2]

[6]

[9]

[11]

[10]

[8]

[7]

[4]

0 1 20.5 Km

Recorded Flood Extent

Mapped Water depth (m)

5 m 0 m

[4][1]

[3]

[5]

[2]

[6]

[9]

[11]

[10]

[8]

[7]

0 0.1 0.20.05 Km

Recorded Flood Extent

Mapped Water depth (m)

5 m 0 m

Uncertainty analysis

Uncertainty due to:

• Locational errors

• Elevation errors

• Water depth error

• Total uncertainty

21

[1] [2]

[3]

[4]

[5]

!

!

!!

!

!

!

!

!

!

!

!

!!

!

!

!

!

!

!

!

!!

!

!

!

!

!

!

!!

!

!

!

!!

!

!

!

!

!

!

!

!

!

!

!

!

!

!

!

!

!

!

!

!!

!

!

!

!

!

! !!

!!

!

!

!

!

!

!

!

!

! Validation Points

Mapped Water Depth

5 m 0 m

0 5 102.5 Km

Percentage Flooded

100% 50% 0%

0 5 102.5 Km

Percentage Flooded

100% 50% 0%

0 5 102.5 Km

Percentage Flooded

100% 50% 0%

0 5 102.5 Km

Percentage Flooded

100% 50% 0%

0 5 102.5 Km

Flood probability

Concluding

• There is a lot of useful information out there

• The information is real-time and (partly) publicly available

• The challenge is filtering out the useful data; real-time hydrological

data and tools can help

• Uncertainty in flood mapping due to location and observation error

• Quality of flood mapping increases by using hydrological

characteristics

• Useful flood maps can be generated using a low number of

observations

• Photos contain a lot of useful information

Questions?

Ideas for further development and possible

applications in projects are also welcome!

[email protected] +31(0)88335 8525

[email protected] +31(0)88335 7672