The three calibrations of the LISFLOOD routine have mistakes because in some reservoirs the minimum volume ($V_{min}$) is larger than the normal volume ($V_n$).
The problem is that $V_{min}$ is fixed since the declaration of the reservoir, but $Vn$ varies with each iteration of the calibration through parameter $\alpha$. It was never checked that $V{min}$ must be at most equal to $V_n$.
I've found another possible improvement in the calibration of the LISFLOOD routine. At the moment, the flood outflow ($Q_f$) is a quantile of the observed inflow. This has the problem that it can be at most the highest recorded inflow, which could not be realistic if the time series is too short. Instead, it would be better to estimate $Q_f$ in the same way as it is done in the Hanazaki routine, i.e, as a factor of the 100-year return period:
It would be interesting to use similar model parameters in the LISFLOOD and Hanazaki routines.
The flood storage ($V_f$) is in the LISFLOOD routine a fraction of the total storage, whereas in the Hanazaki routine is a quantile of the observed storage. The second approach depends on the availability of records, whereas the first is independent.
The normal outflow ($Q_n$) is in the LISFLOOD routine a fraction of the flood outflow ($Q_f$), but in the Hanazaki routine is a fraction of the mean inflow. In this case, the inflow could be taken either from records or the GloFAS simulation. However, this second approach requires checking that $Q_n$ is smaller than $Q_f$, a check that is fulfilled by default in the LISFLOOD approach.
I declare $V_{min}$ as the minimum value between 0.1 times the total storage (LISFLOOD default value) and the minimum in the storage records. As the minimum value of $Vf$ in the calibration is 0.2 times the total storage, I skip the problem of $V{min}$ being smaller than $V_f$.
I've changed the estimation of $Q_f$, so it's identical to that in Hanazaki, i.e., a fraction of the 100-year return period.
I've changed the Hanazaki_calibrator in two ways to be similar to the Lisflood_calibrator:
$V_f$ is now estimated as a fraction of the total storage.
$Q_n$ is now estimated as a fraction of the flood outflow ($Q_f$).
The three calibrations of the LISFLOOD routine have mistakes because in some reservoirs the minimum volume ($V_{min}$) is larger than the normal volume ($V_n$).
The problem is that $V_{min}$ is fixed since the declaration of the reservoir, but $Vn$ varies with each iteration of the calibration through parameter $\alpha$. It was never checked that $V{min}$ must be at most equal to $V_n$.
I've found another possible improvement in the calibration of the LISFLOOD routine. At the moment, the flood outflow ($Q_f$) is a quantile of the observed inflow. This has the problem that it can be at most the highest recorded inflow, which could not be realistic if the time series is too short. Instead, it would be better to estimate $Q_f$ in the same way as it is done in the Hanazaki routine, i.e, as a factor of the 100-year return period:
$$Qf = \delta \cdot Q{100}$$