observation accuracy if compared to arrival frequency. The combined effects of
random arrival frequency and observation accuracy are represented in scenario
5 using a magenta colour (i.e. the combination of warm and cold colours) in
Fig. 8. As expected, this scenario is the one with the lower values of μ(NSE) if
compared to the previous ones. The remaining scenarios, from 6 to 9, are equivalent
to the ones from 2 to 5 with the only difference that they are non-periodic in time. For
this reason, in Fig. 8, scenarios from 6 to 9 have the same colour of scenarios 2–5 but
indicated with dashed line in order to underline their non-periodic behaviour.
Overall it can be observed that non-periodic scenarios have similar μ(NSE) values
to their corresponding periodic scenario. However, their smoother μ(NSE) trends are
due to lower variability of NSE values which means that model performances are
less dependent to the non-periodic nature of the crowdsourced observations than
their periodic behaviour. Overall, σ(NSE) tends to decrease for the high number of
observations.
5.2.2 Assimilation of Flow Observations from Both Physical and Social
Sensors
In the following, the contribution of assimilating synthetic flow data from a heterogeneous network of physical and social sensors on the semi-distributed model
implemented in the Bacchiglione catchment is analysed. Streamflow observations
from physical sensors are assumed to be synchronous with hourly frequency, while
social observations are considered asynchronous with higher and irregular frequency. Five different experimental settings are introduced and represented in
Fig. 9, corresponding to different types of sensors used.
The physical and social observations are assimilated in order to improve the poor
model prediction at the catchment outlet (city of Vicenza) affected by an underestimation of the 3-day rainfall forecast used as normal input in flood forecasting
practice in this area. Scenarios 10 and 11, described in Fig. 7, are used in this
experiment in order to represent an irregular and random behaviour of the social
observations.
Figure 10 shows the results obtained from the experiment settings represented in
case of observations from distributed physical and social sensors. One of the main
outcomes of these analyses is that the replacement of a physical sensor for a social
sensor at only one location (settings B) does not improve the model performance in
terms of NSE for different lead time values. Distributed locations of social sensors
(setting C) can provide higher value of NSE than a single physical sensor, even for
low number of observations in both regular and intermittent social observations. It is
interesting to note that in case of integration between physical and social sensors
(setting D), the NSE is higher than in case of setting C for low number of observations. However, with the higher number of observations, setting C is the one
providing the best model improvement for low lead time values. Best model
improvement is achieved in case of setting E. In case of intermittent observations
(d, e and f), it can be noticed that the setting D provides higher improvement than
226
M. Mazzoleni et al.
random arrival frequency and observation accuracy are represented in scenario
5 using a magenta colour (i.e. the combination of warm and cold colours) in
Fig. 8. As expected, this scenario is the one with the lower values of μ(NSE) if
compared to the previous ones. The remaining scenarios, from 6 to 9, are equivalent
to the ones from 2 to 5 with the only difference that they are non-periodic in time. For
this reason, in Fig. 8, scenarios from 6 to 9 have the same colour of scenarios 2–5 but
indicated with dashed line in order to underline their non-periodic behaviour.
Overall it can be observed that non-periodic scenarios have similar μ(NSE) values
to their corresponding periodic scenario. However, their smoother μ(NSE) trends are
due to lower variability of NSE values which means that model performances are
less dependent to the non-periodic nature of the crowdsourced observations than
their periodic behaviour. Overall, σ(NSE) tends to decrease for the high number of
observations.
5.2.2 Assimilation of Flow Observations from Both Physical and Social
Sensors
In the following, the contribution of assimilating synthetic flow data from a heterogeneous network of physical and social sensors on the semi-distributed model
implemented in the Bacchiglione catchment is analysed. Streamflow observations
from physical sensors are assumed to be synchronous with hourly frequency, while
social observations are considered asynchronous with higher and irregular frequency. Five different experimental settings are introduced and represented in
Fig. 9, corresponding to different types of sensors used.
The physical and social observations are assimilated in order to improve the poor
model prediction at the catchment outlet (city of Vicenza) affected by an underestimation of the 3-day rainfall forecast used as normal input in flood forecasting
practice in this area. Scenarios 10 and 11, described in Fig. 7, are used in this
experiment in order to represent an irregular and random behaviour of the social
observations.
Figure 10 shows the results obtained from the experiment settings represented in
case of observations from distributed physical and social sensors. One of the main
outcomes of these analyses is that the replacement of a physical sensor for a social
sensor at only one location (settings B) does not improve the model performance in
terms of NSE for different lead time values. Distributed locations of social sensors
(setting C) can provide higher value of NSE than a single physical sensor, even for
low number of observations in both regular and intermittent social observations. It is
interesting to note that in case of integration between physical and social sensors
(setting D), the NSE is higher than in case of setting C for low number of observations. However, with the higher number of observations, setting C is the one
providing the best model improvement for low lead time values. Best model
improvement is achieved in case of setting E. In case of intermittent observations
(d, e and f), it can be noticed that the setting D provides higher improvement than
226
M. Mazzoleni et al.