hydrograph (higher for the model structure 1 than for the structure 2). Intermittent
observations do not improve the model results in the same way that social observations coming continuously in time do.
Figure 6b confirms that similar improvements for the scenario 3 are achieved
assimilating observations coming from the optimally located static sensors running
continuously in time (scenario 5). In addition, a combined assimilation of intermittent observations (during daylight time) and static observations from optimal and
nonoptimal network of static sensors tends to slightly improve the model output.
Figure 6 demonstrates that considering this type of hydrological model in this
particular basin, in the case of an inappropriate distribution of static physical sensors
within the basin (scenario 10), the model performances can be improved. However,
there is an evident limitation of the model in providing biased hydrographs (especially for MS2), underestimated when compared to the observed one. Biased models
can affect the DA results [34].
5.2 Assimilation of Asynchronous Observations
In the previous analysis, social data are provided at the same time of the model time
step. However, in case of CS observations, the arrival moment might have lower
frequency than the model time step (asynchronous observations), as reported in
Mazzoleni et al. [62]. Various experimental scenarios representing different configurations of arrival frequency, number and accuracy of the flow observations are
reported in Fig. 7. In order to remove the random behaviour related to the irregular
arrival frequency and observation accuracy, different model runs (100 in this case)
are carried out, assuming different random values of arrival and accuracy (coefficient
α in Eq.10) during each model run, for a given number of observations and lead time.
The NSE value is estimated for each model run, so μ(NSE) represents the mean of
the different values of NSE.
5.2.1 Assimilation of Flow Observations Only from Social Sensors
A lumped hydrological model based on the KMN model is applied to the Brue
catchment in order to assimilate synthetic asynchronous observations using the
modified version of KF reported in Mazzoleni et al. [62]. Two flood events and
experimental scenarios from 1 to 9 (see Fig. 7) are considered in this section.
As it can be seen from Fig. 8, increasing the number of social observations within
the observation window results in the improvement of the NSE, but it becomes
negligible for more than ten observations. This means that the additional social
observations do not add information useful for improving the model performance.
From Fig. 8 it can be seen that, overall, assimilation of crowdsourced observations improves model performances in all the considered flood events. In the case of
scenarios 2 and 3 (represented using warm, red and orange, colours in Fig. 8, for lead
224
M. Mazzoleni et al.
observations do not improve the model results in the same way that social observations coming continuously in time do.
Figure 6b confirms that similar improvements for the scenario 3 are achieved
assimilating observations coming from the optimally located static sensors running
continuously in time (scenario 5). In addition, a combined assimilation of intermittent observations (during daylight time) and static observations from optimal and
nonoptimal network of static sensors tends to slightly improve the model output.
Figure 6 demonstrates that considering this type of hydrological model in this
particular basin, in the case of an inappropriate distribution of static physical sensors
within the basin (scenario 10), the model performances can be improved. However,
there is an evident limitation of the model in providing biased hydrographs (especially for MS2), underestimated when compared to the observed one. Biased models
can affect the DA results [34].
5.2 Assimilation of Asynchronous Observations
In the previous analysis, social data are provided at the same time of the model time
step. However, in case of CS observations, the arrival moment might have lower
frequency than the model time step (asynchronous observations), as reported in
Mazzoleni et al. [62]. Various experimental scenarios representing different configurations of arrival frequency, number and accuracy of the flow observations are
reported in Fig. 7. In order to remove the random behaviour related to the irregular
arrival frequency and observation accuracy, different model runs (100 in this case)
are carried out, assuming different random values of arrival and accuracy (coefficient
α in Eq.10) during each model run, for a given number of observations and lead time.
The NSE value is estimated for each model run, so μ(NSE) represents the mean of
the different values of NSE.
5.2.1 Assimilation of Flow Observations Only from Social Sensors
A lumped hydrological model based on the KMN model is applied to the Brue
catchment in order to assimilate synthetic asynchronous observations using the
modified version of KF reported in Mazzoleni et al. [62]. Two flood events and
experimental scenarios from 1 to 9 (see Fig. 7) are considered in this section.
As it can be seen from Fig. 8, increasing the number of social observations within
the observation window results in the improvement of the NSE, but it becomes
negligible for more than ten observations. This means that the additional social
observations do not add information useful for improving the model performance.
From Fig. 8 it can be seen that, overall, assimilation of crowdsourced observations improves model performances in all the considered flood events. In the case of
scenarios 2 and 3 (represented using warm, red and orange, colours in Fig. 8, for lead
224
M. Mazzoleni et al.