Using Citizen Science Data to support Flood Risk Modelling
November 2015
Dr Barry Evans and Dr Paul Brewer
Aberystwyth University
Email: [email protected]
Overview
• Background• Previous work• Application design• Research questions• Calculating Flood Extents - Theory• Maximise data usage• Ensure data quality• Initial trials• Next steps
Citizen Observatory Web (COBWEB)•European FP7 Funded Project•UNESCO World Network of Biosphere Reserves
– Welsh BR (Dyfi) – Pilot study area– Greek BRs (Gorge of Samaria and Mount Olympus)– German BR (Wadden Sea and Hallig Islands)
About COBWEB
1. Biological Monitoring2. Earth Observation3. Flooding4. In-Situ sensors
About COBWEB
The crowd
Validated Quality approved Compliant
Authoritative data
Data client
- Commercial- Government- Community
Sensors in the environment
Data
Fieldtrip Open COBWEB App
Key features•Capture information
– Images– Audio– Text– Location– Polygons
•High quality background maps•Saved maps for use “offline”•Custom data collection forms•Manual location correction•It’s free
COBWEB Flooding App
Key features•Capture information
– Images– Text– Location
•High quality background maps•Saved maps for use “offline”•Use of Phones internal sensors (Tilt, Pitch and Yaw)•Access to information from in-situ sensors•It’s free
The AppsThe Applications
COBWEB
AuthoritativeData
CitizenData
COBWEB
AuthoritativeData
CitizenData
Flooding (Native App) Biological monitoring and Earth observations
Web Server Visualisation
Other Applications QualityControl
Issues with managing flood risks
Local historical records can be poor•Ungauged or short duration gauging records
(typically within UK average record is ~50 years)
Current methods of data collection on real events for validation not always possible or complete•Orbit revisit time, vegetation, cloud cover effects ability to collect data satellite data
Limited budgets for local authorities•Cost implications in routine monitoring
How can we task citizens to collect data to help create a more comprehensive understanding of flood risks?
Designing a Citizen Science project
7 key principles to consider when designing a citizen science project (Environmental Observation Framework):
1.Consider data requirements2.Manage volunteers to get best data3.Ensure data quality4.Harness new technologies5.Manage data effectively6.Report and share data7.Evaluate and maximise data value
Data requirements
1. Flood risk mapping2. Strategic planning and development control3. Asset management and maintenance4. Flood forecasting and warning5. Flood incident management
Some of the ways in which flood risk is currently managed
North West Wales Catchment Flood Management Plan (2010)
Data requirements
Event Data Source Risk Management
Pre-Flood
Debris, blockages of channels and culverts
Citizen Science,Crowdsource,Smart citizen
2 and 3
Flood Location, inundation extent, flow velocities and water colouration
Citizen Science,Crowdsource,Smart citizen,Drone
1,4 and 5
Post-Flood
Trash lines, public response, Media
Citizen Science,Crowdsource,Drone
1,4 and 5
1. Flood risk mapping2. Strategic planning and development control3. Asset management and maintenance4. Flood forecasting and warning5. Flood incident management
Data requirements
From a simplified initial standpoint the following information would be useful
•GPS Location
•Textual information
•Photograph
•Uncertainty data
•User rating
Where data was captured
Description of observationImage for checking/analysis
Accuracy of data e.g. location uncertainty
Experience of user for confidence rating
Previous work
Creek watch Application for monitoring waterways (Kim et al. 2011)
Previous work
Capturing river level data via mobile phone (SMS) (Lowry and Fienen 2013)
Previous work
Fusing crowdsource data to improve flood hazard maps (Schneble and Cervone 2013).
Flood application design (Manage volunteers)
1 32
Flood application design (Manage volunteers)
Real-time feedback for data capture helps facilitate citizen to capture “good” data:
•Angle of tilt in limited range,
•GPS coverage,
•Zone information.
Flood application design (Manage volunteers)
• Immediate access to user submitted geo-tagged data
• Simple symbology
• Pan and Zoom (Pinch)
• Tap for more details
Image [1] 28/10/13 13:24
Research questions
Potential of smartphones to capture real-time, and high-resolution, geo-tagged and time stamped data relating to flood events:
•Flood limits to calibrate and validate existing flood hydraulic models.•Flood water colour to estimate river sediment loads, which will be used to quantify catchment erosion and sediment delivery patterns. ?
Calculating Flood Extent (Harness new technology)
• Maximise use of smartphone functionality
• Keep interaction quick and simple
Calculating Flood Extent (Harness new technology)
• User position• Estimated height• Orientation• Angle of device• Terrain
Draw edge of Water line
x
Mobile Geo-tagging from a distance using Line of Sight (Meek et al. 2013)
Calculating Flood Extent (Harness new technology)
• Original LoS approach yields location of central point in the image
• Mobile camera has pre-defined camera viewing angle attributes which relates to its field of view
• Potential to build up spatial grid relating pixels to their location
Calculating Inundation Extent (uncertainties)
x
One unique location for X becomes an area that depicts the likely locations for X based on uncertainties during capture.
Calculating Flood Extent (Harness new technology)
• Derive inundation extent from user values
• Multiple possibilities due to uncertainties
• Derived likelihood of inundation based on user data
Maximise data usage
Where does INSPIRE fit in?
• Building up large data repositories– Authoritative– Crowd-Sourced
• Maximise data usage– Education– Local Authorities– General Public
• Allow for expansion– Open Source approach– Cross platform communication– Bolt-On applications
Interoperability
About COBWEB
The crowd
Validated Quality approved Compliant
Authoritative data
Data client
- Commercial- Government- Community
Sensors in the environment
Data
Ensure data quality
Perceived Lack of confidence in Citizen Science data
• Compare data against existing flood maps
• Cross validate against other citizen captured data
• Check data against in-situ data (weather data, river gauges)
• Keep information (metadata) about quality on the data being captured with the data
Data
Meta-data
Data
Meta-data
Ensure data quality
COBWEB
Flood
Pre-Flood
Post-Flood
Other
AuthoritativeData
CitizenData
InitialValidation
Check
Cross-validation
Check Meta-data
Data
FurtherAnalysis
CS BasedFlood data
Meta-data
Devicechecks
Data
Meta-data
Ensure data quality
Initial Trials: Overview
• For each photograph we generate 100 different flood extents based on uncertainties present in the observation
• “Rolling Ball” technique used for surface flow routing and flooding
• Each flood model run produces a true or false flood map on a cell by cell basis
• Combination of the 100 outputs gives us a likelihood of a cell being flooded
Initial Trials: Model simulations
Initial Trials: Tal-y-bont Floods
Initial Trials: Test Case 1
Absence of flood events means initial tests carried out with historical photo records
Initial Trials: Test case 1
Initial Trials: Comparison
1
2 3
Initial Trials: Comparison
Photo 1
Photo 2
Initial Trials: Comparison
As we’re using Fuzzy Classification of flood likelihoods we can combine results from multiple observations to create a combine view of the flooding likelihood.
Initial Trials: Model simulations
Initial Trials: Simulation time
• Each simulation is independent of the other• Allows for parallel simulation• Simulation Run-time is scalable• 100 simulations on HPC ~ 30s
Next steps
• Initial deployment with Talybont Floodees (volunteers) to begin collecting data
References
• Kim, S., Robson, C., Zimmerman, T., Pierce, J., Haber, E. M., 2011. Creek watch: pairing usefulness and usability for successful citizen science.
• Lowry, C. S., Fienen, M. N., 2013. Crowdhydrology: Crowdsourcing hydrologic data and engaging citizen scientists. Ground Water 51 (1), 151156.
• Meek, S., Priestnall, G., Sharples, M., Goulding, J., 2013. Mobile capture of remote points of interest using line of sight modelling. Computers & Geosciences 52 (0), 334 344. URL http://www.sciencedirect.com/science/article/pii/S009830041200338X
• Schnebele, E., Cervone, G., 2013. Improving remote sensing ood assessment using volunteered geographical data. Nat. Hazards Earth Syst. Sci. 13 (3), 669677, nHESS.