time equal to 24 h), i.e. random arrival frequency with fixed/controlled accuracy, the
average values of NSE, μ(NSE), are smaller but comparable with the ones obtained
in case of scenario 1 for all the considered flood events. In particular, scenario 3 has
lower μ(NSE) than scenario 2. This can be related to the fact that both scenarios have
random arrival frequency; however, in scenario 3 observations are not provided at
the model time step, as opposed to scenario 2. In scenario 4, represented using cold
blue colour, observations are considered coming at regular time steps but having
random accuracy. Figure 8 shows that μ(NSE) values are lower in case of scenario
4 rather than scenarios 2 and 3. This can be related to the higher influence of
Fig. 7 The experimental scenarios representing different configurations of arrival frequency,
number and accuracy of the streamflow observations [62]
Fig. 8 μ(NSE) values estimated for varying number of assimilated flow observations, for the
intermittency scenarios for the different flood events [63]
Exploring Assimilation of Crowdsourcing Observations into Flood Models
225