setting C. In case of high lead time value (12 h), results of setting C tend to be similar
to the ones obtained with setting B. As in case of scenario 10, also in case of scenario
11, the best results are achieved in case of setting E.
6 Conclusions
This chapter describes the novel methods mainly developed within the EU-FP7
WeSenseIt project, aimed to optimally assimilate heterogeneous intermittent observations, coming from static social sensors, to improve hydrological and hydrodynamic models for flood prediction. The proposed methods used to assimilate
crowdsourced observations are applied to the Brue and Bacchiglione catchments,
in which different hydrological and hydraulic models are implemented. A Kalman
filter and ensemble Kalman filter are used to assimilate flow observations in linear
and non-linear models, respectively. Observational error is assumed uniformly
distributed with multiplying factors of 0.1 and 0.3 as minimum and maximum values
for the static social sensors, respectively. It is worth noting that because real
crowdsourced observations from citizen were not available at the time of this
study, model-based synthetic realistic flow observations are used instead.
This study demonstrated that crowdsourced citizen-based observations can significantly improve flood prediction if integrated into hydrological and hydraulic
models. In addition, networks of low-cost static and dynamic social sensors can
actually complement traditional networks of static physical sensors, for the purpose
of improving flood forecasting accuracy. This can be one of the potential applications of increasing efforts to build citizen observatories of water. On the one hand,
citizens can play an active role in information capturing, evaluation and communication, and on the other hand, they can also help in improving models and increasing
flood resilience.
In particular, assimilation of streamflow observations from static social sensors
provides improvements in model performance which depends on the location of
such observations and the structure of the considered hydrological model. Flood
forecasts are influenced by the total number of social sensors and their locations in
the case of semi-distributed model with sub-catchments connected in parallel, while
results achieved with sub-catchment connected in series are more sensitive to the
locations of the static physical sensors but not to their number.
This research proved that assimilation of asynchronous observations results in a
significant improvement of NSE for different lead time values. Increasing the
number of assimilated crowdsourced asynchronous observations within two model
time steps induces an improvement in the NSE. However, after a threshold number
of crowdsourced observations, NSE asymptotically approaches a certain value
meaning that no improvement is achieved with additional observations.
Besides these important results, this work has still certain limitations which should
be mentioned. Additional analyses on different case studies and hydrological/hydraulic model have to be carried out to draw more general conclusions about assimilation
228
M. Mazzoleni et al.
to the ones obtained with setting B. As in case of scenario 10, also in case of scenario
11, the best results are achieved in case of setting E.
6 Conclusions
This chapter describes the novel methods mainly developed within the EU-FP7
WeSenseIt project, aimed to optimally assimilate heterogeneous intermittent observations, coming from static social sensors, to improve hydrological and hydrodynamic models for flood prediction. The proposed methods used to assimilate
crowdsourced observations are applied to the Brue and Bacchiglione catchments,
in which different hydrological and hydraulic models are implemented. A Kalman
filter and ensemble Kalman filter are used to assimilate flow observations in linear
and non-linear models, respectively. Observational error is assumed uniformly
distributed with multiplying factors of 0.1 and 0.3 as minimum and maximum values
for the static social sensors, respectively. It is worth noting that because real
crowdsourced observations from citizen were not available at the time of this
study, model-based synthetic realistic flow observations are used instead.
This study demonstrated that crowdsourced citizen-based observations can significantly improve flood prediction if integrated into hydrological and hydraulic
models. In addition, networks of low-cost static and dynamic social sensors can
actually complement traditional networks of static physical sensors, for the purpose
of improving flood forecasting accuracy. This can be one of the potential applications of increasing efforts to build citizen observatories of water. On the one hand,
citizens can play an active role in information capturing, evaluation and communication, and on the other hand, they can also help in improving models and increasing
flood resilience.
In particular, assimilation of streamflow observations from static social sensors
provides improvements in model performance which depends on the location of
such observations and the structure of the considered hydrological model. Flood
forecasts are influenced by the total number of social sensors and their locations in
the case of semi-distributed model with sub-catchments connected in parallel, while
results achieved with sub-catchment connected in series are more sensitive to the
locations of the static physical sensors but not to their number.
This research proved that assimilation of asynchronous observations results in a
significant improvement of NSE for different lead time values. Increasing the
number of assimilated crowdsourced asynchronous observations within two model
time steps induces an improvement in the NSE. However, after a threshold number
of crowdsourced observations, NSE asymptotically approaches a certain value
meaning that no improvement is achieved with additional observations.
Besides these important results, this work has still certain limitations which should
be mentioned. Additional analyses on different case studies and hydrological/hydraulic model have to be carried out to draw more general conclusions about assimilation
228
M. Mazzoleni et al.