1. Assimilation of crowdsourced data in hydrological modeling to improve flood prediction
Maurizio Mazzoleni, Leonardo Alfonso and Dimitri Solomatine
WMO MOXXI workshop – Innovation in Hydrometry, from ideas to operation
Geneva

2. Introduction Climate and floods are changing..
G. Blöschl el al., "Changing climate shifts timing of European floods," Science (2017).

3. Integrate real-time observations into water-system models
Introduction
 In order to reduce flood damage there is the need for
More observation systems
Better predictive modelling tools
Analytical methods to handle uncertainty
Changes in design and adaptive management practices …
Integrate real-time observations into water-system models

4. Static physical sensors
often data is transmitted automatically and stored in databases

5. Crowdsourced sensors

6. Sensors classification
Dynamic
Static
Space
Quantitative information
Qualitative information
Classic station

7. Physical sensors Vs Crowdsourced sensors
Traditional network of sensors:
Static
Low uncertainty
Expensive
High mainteinance
More consistent flow of information
Crowdsourcing sensors:
High and variable uncertainty
Dynamic in space
Cheap
Low mainteinance
Observations coming at random moment

8. Why using citizen observations for flood prediction?
Technical motivation:
Integrate numerical models and citizen observatories to fill the gap in hydrological applications regarding the optimal use of crowdsourced observations not only in post-event analyses but in also in real time by their optimal assimilation.
Social motivation:
Bring citizens closer to decision-making processes and to understand how their participation in the model development process could improve models.

9. Crowdsourced approachQ
Qobs(t)
Qmod(t)
QCS(t) t 1h 2h 3h 4h 5h

10. The WeSenseIt Project
The aim of the project it is to develop a citizen-based observatory of water, which will allow citizens and communities to become active stakeholders in information capturing, evaluation and communication. WeSenseIt will enable citizens to take a major role in prevention of flooding
Optimally design/upgrade the networks of physical and social sensors
Optimally incorporate Crowdsourced uncertain data into hydrological and hydraulic models

11. Case study – Bacchiglione catchment, Italy

12. Sensor characteristics

13. Hydrological and hydraulic models
Hydrological model
Hydraulic model

14. Model updating – Kalman filter
P-k = APk-1AT + Q
ŷk = ŷ-k + K(zk - H ŷ-k )
K = P-kHT(HP-kHT + R)-1
Pk = (I - KH)P-k
ŷ-k = Ayk-1 + Buk
Observational error

15. Observations characteristics

16. Assimilation approach for asynchronous observations
Updated
Model
Q (m3/s)
Q (t*)
Q+(t)
Qobs(t0* )
Observation
Q+(t0*)
Q (t)
Q (t0)
Q (t0*)
Update t0
t0* t t*
Prediction 16

17. Assimilation of asynchronous flow observations from StSc sensors in hydrological model
Experimental setup

18. Mazzoleni, M. , Verlaan, M. , Alfonso, L. , Monego, M. , Norbiato, D
Mazzoleni, M., Verlaan, M., Alfonso, L., Monego, M., Norbiato, D., Ferri, M., and Solomatine, D. P.: Can assimilation of crowdsourced data in hydrological modelling improve flood prediction?, HESS, 21, , 2017.
Results

19. Simplified method to represent citizen engagement
StSc and DySc sensors
Simplified method to represent citizen engagement
Mazzoleni et al.: Towards assimilation of crowdsourced observations for different levels of citizen engagement: the flood event of 2013 in the Bacchiglione catchment, HESS, accepted, 2017.

20. Results (I)

21. Results (II)
Effects of different values of maximum engagement (coefficient K)
Effects of different values of engagement over time (coefficient r)

22. Economic impact of data fusion of in-situ and crowdsourced data
Reliable information about the whole area
Robust (a failing sensor is not a problem)
Costly (operation and maintenance)
Citizens:
- could report on O&M issues
- could eventually replace some sensors
Non-reliable: areas without data
Non- Robust (a failing sensor is a problem)
Costly (consequences of bad decisions)
Citizens:
- could help augmenting network

23. The GroundTruth 2.0 Project
Citizen observatories:
More than just data collection
…from citizen-based data collection
to knowledge sharing
for joint decision-making, cooperative planning and environmental stewardship

24. Impact = Scosts current network - Scosts potential network
Approaches to evaluate economic impact
Compare the costs of the current monitoring network with a potential reduced / complemented future monitoring network
Impact = Scosts current network - Scosts potential network
Estimate the Value of Information of both current and potential networks in terms of the decision-making consequences
VOI = Utility decisions using potential– Utility decisions using current

25. Objective – Generic methodology
Cost (e.g. flood damage)
Economic impact
Network 1 (current in-situ)
Network 2 (current in-situ + CO)
Cost reduction
Collected data/Value of information
Benefit in data

26. Effect of different sensor network: An example
Assimilation of streamflow observations from an heterogeneous network of sensors in hydrological modelling
Observed
CS data
Optimal in situ
Non optimal in situ
CS data + Non optimal in situ

27. Conclusions
This research demonstrates that crowdsourced observations can improve flood prediction if integrated in hydrological and hydraulic models.
A given, limited, number of citizen observations are required in order to achieve stable model improvements.
Citizen engagement behaviour in providing water level observations plays an important role in flood predictions if assimilated in hydrological and hydraulic models.
The results demonstrated that a networks of low-cost social sensors can complement traditional networks of physical sensors and improve the accuracy of flood forecasting

28. Limitations Lack of real citizen observations
Difficult estimation of accuracy of citizen observations
Need to test the methodology in additional catchments and flood events
Involve and keep citizen engaged over time

29. THANK YOU